UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Thermal remote sensing of urban surface temperatures Voogt, James Adrian 1995

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


831-ubc_1995-060780.pdf [ 8.09MB ]
JSON: 831-1.0088366.json
JSON-LD: 831-1.0088366-ld.json
RDF/XML (Pretty): 831-1.0088366-rdf.xml
RDF/JSON: 831-1.0088366-rdf.json
Turtle: 831-1.0088366-turtle.txt
N-Triples: 831-1.0088366-rdf-ntriples.txt
Original Record: 831-1.0088366-source.json
Full Text

Full Text

THERMAL REMOTE SENSING OF URBAN SURFACETEMPERATURESByJames Adrian VoogtB. Sc., Queen’s University, 1986M. Sc., University of British Columbia, 1989A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinTHE FACULTY OF GRADUATE STUDIESDEPARTMENT OF GEOGRAPHYWe accept this thesis as conformingto the required standardTHE UNIVERSITY OF BRITISH COLUMBIA1995© James Adrian Voogt, 1995In presenting this thesis in partial fulfilment of the requirements for an advanceddegree at the University of British Columbia, I agree that the Library shall make itfreely available for reference and study. I further agree that permission for extensivecopying of this thesis for scholarly purposes may be granted by the head of mydepartment or by his or her representatives. It is understood that copying orpublication of this thesis for financial gain shall not be allowed without my writtenpermission.Department of____________The University of British ColumbiaVancouver, CanadaDate £tç. i , iqqçDE-6 (2/88)AbstractThermal remote sensors have been used extensively to examine urban surface temperaturepatterns but without explicit consideration of the complex form of the urban surface. Thisresearch investigates the extent of possible biases which may be present due to the uniquestructure of the urban surface and is an attempt to address some of the issues raised byRoth et al. (1989).The research approach is primarily observational. An extensive measurement programme integrating ground and airborne remote sensor platforms as well as fixed monitoring sites was designed and implemented to obtain information from three of themost common urban surface types: downtown office/commercial, residential and lightindustrial. Both nadir and off-nadir high resolution thermal imagery was collected toobtain information on component surface temperatures, and to determine the extent ofanisotropy in urban surface emissions at the land use scale.Vertical facet temperatures are shown to exhibit large diurnal temperature rangeswhich strongly control the anisotropic longwave emission of the urban surface. Airborneobservations reveal strong anisotropy of surface longwave emissions over each of thesites at times of maximum temperature contrast between opposing canyon facets. Themagnitude of the anisotropy is sufficient to demand consideration whenever daytimethermal remote measurements are made over urban surfaces.Mixed distribution modelling has been explored as a technique to recover componentsurface temperatures from composite temperature frequency distributions. A low numberof component distributions can be used to successfully model the composite distribution,however, the successful extraction of individual component temperatures is more difficult11and depends upon the nature of the surface, and the temperature contrasts present.Estimates of the complete urban surface temperature have been made for the firsttime by combining remote and surface-based estimates of horizontal and vertical facettemperatures. The complete surface temperature is cooler than nadir or off-nadir remotemeasurements and is shown to be best approximated by off-nadir measurements in thedirection of the most shaded vertical facet.A simple two-dimensional geometric urban surface model (a slightly modified versionof that orignally presented by Sobrino and Caselles (1989) and extended by Caselles et al.(1992)) is used to investigate the nature of anisotropy over the study areas. Urban surfacetemperature components are specified from observed data but a modified version of theMyrup (1969) surface energy balance model is shown to be adequate for estimating surfacetemperatures in canyon environments for the extension of geometrical surface modellingto other times and locations. Results indicate that the two-dimensional representationof the urban surface is generally inadequate. The development of a three-dimensionalsurface geometrical model is recommended to better reperesent the structure of the urbansurface.111Table of ContentsAbstract 11rct&t of t€-sList of Tables xiList of Figures xivList of Symbols and Abbreviations xxvAcknowledgements xxxiii1 INTRODUCTION 11.1 Thermal Infra-red Remote Sensing of Urban Areas . 11.2 Limitations of TIR remote sensing. . 61.3 The Urban Surface 61.4 TIR Remote Sensing of the Urban Surface 91.5 TIR Remote Sensing of Agricultural and Forest Surfaces 101.6 Modelling of Thermal Emissions over Plant Canopies . 111.7 Objectives 141.8 Methodology 151.9 Study Sites 161.9.1 Industrial (False Creek South) 181.9.2 Downtown 181.9.3 Residential (Sunset) 201.10 Prevailing Climate 23iv2 AUTOMOBILE TEMPERATURE TRAVERSES2.1 Introduction2.2 Method2.2.1 Sensor Configurations2.2.2 Sampling Methodology2.2.3 Projected FOV2.2.4 Processing2.3 Surface Emissivity2.4 Everest - AGEMA Scanner Comparisons2.5 Temperature Distributions2.5.1 Traverse Configuration 12.5.2 Traverse Configuration 22.5.3 Modelled Temperature Distributions2.6 Summary28282929303234343941414646573 AVERAGE TEMPERATURES OF CANYON3.1 Distribution Truncation3.1.1 Surface/air Temperature Relations in the3.1.1.1 Industrial Area3.1.1.2 Downtown3.1.1.3 Residential3.1.2 Results3.1.2.1 Industrial Study Area3.1.2.2 Downtown3.1.2.3 Residential Area3.2 Mixed Distribution ModellingFACETSStudy Areas5959616165666868707575v4 AIRBORNE TIR OF SELECTED LAND-USE AREAS4.1 Introduction.4.2 Methods4.2.1 Thermal Imaging System4.2.2 Helicopter-AGEMA Installation4.2.3 Remote Sensing Flights4.3 Atmospheric Corrections.4.4 Surface Temperature Analysis.4.4.1 Industrial Area.4.4.2 Residential4.4.3 Downtown4.4.4 Summary4.5 Modelling the Distribution of Surface Temperature4.5.1 Introduction4.5.2 Modelling a Simple Case4.5.3 Modelling Road Surface Temperatures4.5.4 Modelling the Study Area Distributions4.5.4.1 Temperature Frequency Distributions4.5.4.2 Frequency Distribution Differences4.5.4.3 Summary4.6 Surface Temperature Anisotropy3. Surface TemperatureWall Surface TemperaturesA Single Model for Canyon Facets. .7985953.3 Summary 9598989999100101104104104116126129131131132132138138143149150viObservational ResultsComparison with Other SurfacesScale of Anisotropy.Anisotropy Relative to Other Influences Upon Remotely SensedSurface Temperature 1635 THE COMPLETE URBAN SURFACE TEMPERATURE 1671671681681691721731731731731751761841841851851875.1.1 Estimating the Complete Surface Area5.1.1.1 Buildings5.1.1.2 Trees5.1.1.3 Horizontal5.1.1.4 Top-of-Canopy Plan Area5.1.1.5 Complete Surface Area5.1.2 Complete Surface Areas of the Study Sites5.1.2.1 Industrial5.1.2.2 Downtown5.1.2.3 Residential5.2 A Complete Urban Surface Temperature5.2.1 Definitions5.2.2 Estimating the Complete Surface Temperature5.2.2.1 Nadir Airborne and Traverse Temperature Distributions5.2.2.2 Nadir and Off-nadir airborne5.3 Comparison of Complete and Incomplete Surface Temperatures5.3.1 Industrial Area5.3.2 Downtown Area4. Definition of the Complete Urban Surface189189193vii5.3.3 Residential Area5.4 Surface and Air Temperature Relations5.5 Summary6 MODELLING URBAN SURFACE THERMAL EMISSIONS 2096.1 Introduction 2096.2 Geometric Models to Estimate Longwave Surface Anisotropy6.2.1 Extension to Urban Surfaces1992032082092106.3 Model Description 2106.4 An Energy Balance Model to Estimate Surface Temperatures 2166.4.1 Urban Surface Temperature Models 2166.4.2 The Myrup 1-D Surface Energy Balance Model 2176.4.3 Modifications to the Myrup Model 2206.5 Model Tests Against Observations 2236.5.1 Road surface temperatures 2246.5.1.1 Sensitivity Analysis 2246.5.1.2 Industrial 2286.5.1.3 Residential 2306.5.1.4 Downtown 2306.5.2 Roof Temperatures 2326.5.3 Grass Surfaces 2346.5.4 Facet Temperatures 2366.5.4.1 An Individual Building 2366.5.4.2 Canyon Facets in the Study Areas 2386.6 Summary 2456.7 Application of the SC Model 246viii6.7.1 Application to the Study Areas: Overflight Days6.7.1.1 Downtown6.7.1.2 Industrial6.7.1.3 Residential6.7.2 Summary7 CONCLUSIONS7.1 Summary of Conclusions7.2 Suggestions for Future WorkReferences 267A AGEMA 880 LWB INFRARED SCANNERA.1 Instrument DescriptionA.2 Instrument TheoryA.3 Instrument CalibrationA.4 Radiance-Photon-Temperature RelationsA.5 Spectral Response ConsiderationsA.5.1 IntroductionA.5.2 LOWTRAN 7 ParametersA.5.3 ResultsA.6 ConclusionsB ATMOSPHERIC CORRECTIONS OF THERMALB.1 IntroductionB.2 MethodologyB.3 LOWTRAN 7 Atmospheric Radiation ModelB.4 Composite Atmospheric Profilesix246247250255259260260264280280282285288290290290293294IMAGERY 295295296297298B.4.1 Local Radiosonde Profiles.298B.4.2 Climatological Radiosonde Profiles 300B.4.3 Climatological Data 305B.5 Airborne Imagery vs. Surface-Based Measurements 306B.5.1 Discussion 313B.5.1.1 Sea Surface Temperatures 320B.5.1.2 Grass Temperatures 323C EVEREST INTERSCIENCE MODEL 4000A IRT 325C.1 Instrument Description 325C.2 Instrument Operation 325C.3 Instrument Calibration 328D TIR REMOTE SENSING COVERAGE OF THE STUDY AREAS 331xList of Tables1.1 Summary of past urban TIR remote sensing studies 21.2 Examples of observed longwave anisotropy over agricultural and forestsurfaces 122.1 Sampling intervals per urban surface structural unit 322.2 Surface emissivities for surface materials in urban areas 352.3 View factors for the mid-point on the floor of a canyon and a point low onthe canyon wall for canyon H/W representative of the three study areas 362.4 Facet temperatures (estimated from traverse data for the Industrial studyarea) with calculated apparent temperature for wall 2 using canyons withH:W = 1:1 and 1:2 372.5 Difference between true and apparent wall temperatures at select timesfor a H:W = 1:4 canyon 382.6 Model parameters used in EIRT distribution modelling 532.7 Model parameters for EIRT distribution modelling 553.1 Description of symbols used in Figures 3.2, 3.3 and 3.4 623.2 Temperature distributions assigned to a simple model of EIRT response,and results from mixed distribution modelling on the EIRT model output. 944.1 Flights conducted to acquire airborne thermal imagery 1034.2 Comparison of component temperature distributions for an image of asimple scale model 134xi4.3 Percentage of the FOV occupied by component surfaces of a N-S transectin the Industrial area 1444.4 Temperature differences between means for pairs of view directions. . 1564.5 Magnitude of temperature differences produced by processes affecting remote thermal measurement of an urban surface 1644.6 Satellite parameters for satellites commonly used in thermal analysis ofurban areas 1665.1 Definition of area component symbols 1685.2 Major surface component areas and percentages of the complete surfacearea for the Industrial area 1745.3 Major surface component areas and percentages of the complete surfacearea for the Downtown area 1765.4 Building types in the Residential study area 1795.5 Statistical summary of estimated tree structural parameters 1835.6 Major surface component areas and percentages of the complete surfacearea for the Residential area 1836.1 User input required for the SC nEiodel 2146.2 Myrup model input requirements 2186.3 Myrup model output 2196.4 Base input parameters: asphalt sensitivity tests 2266.5 Reference values of asphalt thermal properties used in the Myrup model 2286.6 Model input: Summer Equipment building facets 2366.7 Model input: Industrial, Downtown and Residential Sites 2386.8 SC model surface parameters for the Study areas 246xiiA.1 Technical specifications: AGEMA 880 LWB TIR system 282A.2 Scanner constants: AGEMA THV88O LWB (AGEMA, 1991) 284A.3 LOWTRAN 7 input parameter summary 292A.4 Sensor normalized photon emissions calculated from LOWTRAN 7 radianceestimates 293B.1 Local radiosonde (Airsonde) launch times and maximum altitude. est.indicates maximum altitude was estimated based upon assumed ascentrate and comparison of temperature and humidity profiles with other flights 299B.2 Summary of radiosonde data collected 306B.3 Summary of reporting differences for radiosonde measurements 308B.4 Ground calibration sites and surface types sampled 309B.5 Surface emissivities for ground calibration surfaces 318B.6 Reported values of e for surface types tested 319B.7 Solar position and AGEMA scanner and Everest azimuth angle direction:Flights 1—5 322C.1 Everest Interscience Model 4000A Infrared Transducer Specifications. . . 325C.2 EIRT calibration results 329xli].List of Figures1.1 Various definitions of the urban surface 81.2 Length scales used to simplify and generalize the three-dimensional urbansurface 91.3 Location of the three study areas within the city of Vancouver, B.C. . 171.4 The Industrial area 191.5 The Downtown study area 211.6 The Residential study area 221.7 Pacific Weather Centre surface analysis for August 15, 1992 241.8 Pacific Weather Centre surface analysis for August 16, 1992 251.9 Pacific Weather Centre surface analysis for August 17, 1992 261.10 General weather conditions on the three study days: August 15 - 17, 1992. 272.1 Traverse configurations used to sample surface temperatures 292.2 The traverse vehicle outfitted in traverse configuration 2 312.3 Projected EIRT FOV dimensions and overlaps 332.4 Vertical transect of Tr up the face of a north-facing wall of a glass-cladoffice building 392.5 Surface temperature time series using EIRT and AGEMA scanner 402.6 Apparent surface temperature distributions for the AGEMA scanner andEIRT obtained from the traverse tests 422.7 Frequency distributions of Tr obtained during a mid-morning traverse ofthe Industrial study area 44xiv2.8 North and south facet temperature distributions in alleys and streets inthe Industrial area 452.9 Facet temperature distributions for streets and alleys in the Residentialarea. Note variation in frequency scales 472.10 Frequency distributions of Tr from the right-facing EIRT obtained duringa mid-morning traverse of the Downtown study area 482.11 Frequency distributions of Tr from the left-facing EIRT obtained during amid-morning traverse of the Downtown study area 492.12 Cumulative frequency distribution of building heights for Downtown (traversed northeast facets only) and Industrial (north street and alley facets) 502.13 Modelled and measured EIRT temperature distributions for north facetsduring a mid-morning traverse in the Industrial study area 512.14 Modelled and observed EIRT temperature distributions for a mid-morningtraverse in the Downtown study area (Southwest facets only) 522.15 Modelled and observed 100 left-facing EIRT temperature distributions fora mid-morning traverse in the Industrial study area 563.1 Temporal variation of canyon floor surface temperatures in an E/W streetcanyon compared with 0.5 m air temperature 603.2 North-facing alley facet temperatures and air temperatures in the Industrial study area 633.3 Subset of northeast-facing facet and air temperatures in the Downtownstudy area 653.4 Subset of north-facing facet and air temperatures in the Residential studyarea 67xv3.5 Mean facet temperatures and standard deviations following distributiontruncation in the Industrial study area 693.6 Mean facet temperatures of each street in the Industrial study area. . 713.7 Mean facet temperatures and standard deviations following distributiontruncation in the Downtown study area 723.8 Mean facet temperatures of blocks or block sequences in the Downtownstudy area 733.9 Mean facet temperatures of blocks or block sequences in the Downtownstudy area for each EIRT 743.10 Mean facet temperatures and standard deviations following distributiontruncation in the Residential study area 763.11 Mean facet temperatures of each street in the Residential study area. 773.12 Temperature distribution from a satellite image 783.13 Distribution of Troad in the Industrial study area 813.14 Troad statistics for the component populations derived from mixed distribution modelling for the Industrial area 823.15 Troad statistics derived from mixed distribution modelling for the Downtown study area 833.16 Troad statistics derived from mixed distribution modelling for the Residential study area 843.17 Morning east-facing facet temperature distribution with component populations from the Industrial study area 863.18 Late-afternoon, west-facing facet temperature distribution and componentpopulations from the Industrial study area 873.19 Morning southeast-facing facet temperature distribution from the Downtown study area 88xvi3.20 Facet surface temperature statistics derived from mixed distribution modelling for the Industrial study area 893.21 Facet surface temperature statistics derived from mixed distribution modelling for the Downtown study area 903.22 Facet surface temperature statistics derived from mixed distribution modelling for the Residential study area 924.1 AGEMA 880 system components 994.2 Thermal imaging system installed in the helicopter 1004.3 Scanner mounting arrangement 1024.4 Thermal image of the Industrial study area 1054.5 Apparent surface temperature distributions for each view direction overthe Industrial area during Flight 1 1074.6 Modelled composite temperature distribution 1094.7 Broadening of the apparent surface temperature distribution 1104.8 Comparison of apparent surface temperature distributions before and afterimages with a non-standard block orientation or surface cover were removed.1114.9 Same as Figure 1.5 except for Flight 2 1134.10 Same as Figure 1.5 except for Flight 3 1144.11 Temporal development of the distribution of Tr as indicated by the difference between surface and canyon level air temperature. Industrial area 1154.12 Thermal image of the Residential study area, Flight 7 1174.13 Surface temperature frequency distributions for the three flights over theResidential study area 1184.14 Same as Figure 1.5 except for Flight 7 over the Residential study area. . 120xvii4.15 Comparison of Tr distributions for morning flights over the Industrial andResidential study areas 1214.16 Same as Figure 1.15 except for early afternoon flights 1224.17 Same as Figure 1.15 except for late afternoon flights 1234.18 Comparison of Tr distributions for two primary block orientations in theResidential study area 1254.19 Thermal image of the Downtown study area 1264.20 Same as Figure 1.5 except for Flight 4 over the Downtown study area. . 1284.21 Comparison of Tr distributions for the Downtown, Industrial, and Residential areas 1294.22 Same as Figure 1.20 except for early (Downtown) and late afternoon flights(Industrial and Residential areas) 1304.23 Temperature frequency distribution for a concrete scale model representingthe Residential area 1334.24 Transects of apparent surface temperature across shadow boundaries. . 1354.25 Fitted distributions to road surface temperatures 1374.26 Fitted component distributions to Flight 2 over the Industrial area. . . 1394.27 Summary of fitted and observed distributions applied to the morning flightover the Industrial area 1404.28 Same as 1.27 except for Flights 2, 4, and 7 1414.29 Summary of fitted distributions applied to temperature frequency differences for the morning flight over the Industrial area 1454.30 Same as Figure 1.29 except for: Flights 2, 4, and 7 1474.31 Distribution of image mean apparent temperature for nadir and opposingview directions over the Industrial study area 1514.32 Same as Figure 1.31 except for the Downtown study area 152xviii4.33 Same as Figure 1.31 except over the Residential study area 1534.34 Mean apparent surface temperature and standard error of each view direction for each flight 1554.35 Calculated mean apparent temperatures and standard errors of each viewdirection at the TOA for NOAA-11 AVHRR Channel 4 1595.1 Illustration of component areas 1675.2 Definition of tree structural parameters 1705.3 Modelled shapes for the tree types in the study area 1715.4 Building heights (stories); Industrial study area 1745.5 Frequency distribution of building heights in the Industrial study area 1755.6 Building heights (stories); Downtown study area 1775.7 Frequency distribution of building heights in the Downtown study area 1785.8 Building heights (stories); Residential study area 1805.9 Frequency distributions of estimated building structural parameters (Residential area) 1815.10 Frequency distributions for the estimated tree structural parameters (Residential area) 1825.11 Component and composite surface temperature frequency distributions.. 1865.12 Traverse and image-extracted vertical facet temperature distributions: Flight1, Industrial area 1885.13 Estimated T distributions; Industrial area 1905.14 Traverse and image-extracted vertical facet temperature distributions; Flight2, Industrial area 1915.15 Traverse and image-extracted vertical facet temperature distributions; Flight3, Industrial area 192xix5.16 Component temperature distributions (from vehicle traverses); Industrialarea 1945.17 Complete and off-nadir apparent surface temperature distributions; Industrial area 1955.18 T distributions; Downtown study area 1965.19 Traverse and image-extracted vertical facet temperature distributions; Downtown study area 1975.20 Traverse and image-extracted vertical facet temperature distributions; Downtown study area 1985.21 Apparent surface temperature distributions from traverse and image sourcesfor building walls and trees; Flight 6, Residential area 1995.22 Estimated T distributions; Residential area 2015.23 Apparent tree canopy temperatures extracted from off-nadir imagery; Residential area 2025.24 Comparison of mean image apparent surface temperatures by view direction, traverse air temperatures (Tat, denoted by solid diamonds), andcomplete apparent surface temperatures (Ta,, solid triangles labelled C)for all study areas 2045.25 Tat compared with the lowest modelled temperature distribution fitted tothe composite frequency distribution 2065.26 Same as Figure 5.25 except showing means only, and with data subdividedinto shaded and sunlit facets 2076.1 Two-dimensional model representation of an urban surface showing inputdimensions 2116.2 Detail of the component angles calculated in the SC model 212xx6.3 Example of SC model output 2156.4 Sensitivity of K. to D and w 2226.5 K4. differences and w 2236.6 Observed and modelled incoming global solar irradiance for August 15, 1992.2246.7 Comparison of mid-facet estimates of net shortwave radiation (K*) generated using the mid-point model and the Arnfield (1982) canyon radiationroutines 2256.8 Comparison of mid-facet and facet average estimates of K* generated usingthe mid-point model and the Arnfield (1982) canyon radiation routines. 2266.9 Sensitivity analysis of asphalt road surface temperature 2276.10 Modelled and observed Troad in the Industrial study area 2296.11 Modelled and observed Troad in the Residential study area 2306.12 Modelled and observed Troad in the Downtown study area 2326.13 Roof surface temperatures 2336.14 Modelled and measured surface temperatures and net radiation for a grasssurface in Trafalgar Park 2356.15 Modelled and observed surface temperatures for the Summer Equipmentsite (Industrial area) 2376.16 Comparison of modelled and traverse facet temperature for the Industrialstudy area 2396.17 Comparison of modelled canyon facet temperatures obtained using hourlyand daily input data (Industrial area) 2416.18 Comparison of modelled and traverse facet temperature for the Downtownstudy area 2426.19 Estimated temporal variations of shadow ratio for vertical facets in theResidential area 243xxi6.20 Comparison of modelled and traverse facet temperature for the Residentialstudy area 2446.21 Surface profiles for the NE portion of the Downtown study area 2486.22 Surface proportions viewed: Downtown, Flight 4 2496.23 Modelled AGEMA scanner temperature for the Downtown study area. . . 2516.24 Surface profiles for the Industrial study area 2526.25 Modelled temperature seen by the AGEMA scanner for the Industrialstudy area including along-canyon building spacing 2536.26 Modelled temperature seen by the AGEMA scanner temperature for theIndustrial study area, Flight 2 2556.27 Surface profiles for the Residential study area 2566.28 Modelled temperature seen by the AGEMA scanner for the Industrialstudy area, Flight 6 2576.29 Modelled AGEMA scanner temperature for the Industrial study area,Flight 7 258A.1 AGEMA 880 LWB system response. Data supplied by Linnander (pers.comm.) 281A.2 UBC Soil Science blackbody calibration facility 285A.3 Relative position of the image mode for complete and subset areas of thecalibration images 287A.4 Comparison of AGEMA scanner temperature with average blackbody cavitytemperature. (a) Calibration plot, (b) Temperature differences 288A.5 Relation between AGEMA Thermal Value and photon emission over thecalibration range 289xxiiA.6 Ratio of reflected solar radiance to Earth-emitted radiance for three surfaceemissivities 291B.1 Atmospheric profiles of temperature and humidity for each flight 301B.2 Location of the Vancouver study area and upper air reporting stationsPort Hardy, B.C. (YZT) and Quillayute, WA (UIL) 305B.3 Comparison of the LOWTRAN 7 Midlatitude summer atmosphere modelwith measured profiles from station YZT 307B.4 Look-up table plots of atmospheric corrections 311B .5 Comparison of ground-sampled Tr with corrected AGEMA scanner Tr; Flights1—12 314B.6 Modelled variations in apparent radiative surface temperature for a 60°FOV Everest IRT viewing a water surface 321C.1 Radiant power distribution received by EIRT from a surface 326C.2 Block diagram of EIRT . 327C.3 Temperature differences due to conversion of Tev from e = 0.98 to 1.00and differences due to using Tdet +1,2,5°C in place of Tdet 330D.1 Study areas in Vancouver, B.C 332D.2 Scanned areas (by direction) and composites for Flight 1 333D.3 Scanned areas (by direction) and composites for Flight 2 334D.4 Scanned areas (by direction) and composites for Flight 3 335D.5 Scanned areas (by direction) and composites for Flight 4 336D.6 Scanned areas (by direction) and composites for Flight 5 337D.7 Scanned areas (by direction) and composites for Flight 6 338D.8 Scanned areas (by direction) and composites for Flight 7 339xxiiiD.9 Scanned areas (by direction) and composites for Flight 8 340nivList of Symbols and AbbreviationsABBREVIATIONSA/D Analog to digitalAVHRR Advanced Very High Resolution RadiometerEIRT Everest Infra-red TransducerFOV Field of ViewGRE Ground Resolution ElementH:W building height to street width ratioHCMM Heat Capacity Mapping MissionHCMR Heat Capacity Mapping RadiometerIFQV Instantaneous field of viewITOS Improved TIROS Operational SatelliteLAT Local Apparent (Solar) TimeLDT Local Daylight TimeLST Local Standard TimeLN Liquid nitrogenLUT Look-up tableLWB Longwave Band (Scanner type used with AGEMA system)LAI Leaf area indexxxvMSS Multispectral scannerNOAA National Oceanic and Atmospheric AdministrationPDT Pacific Daylight TimeRMS root mean squareSC sensor-surface model (modified from Sobrino and Caselles (1990))SR Scanning radiometerSST Sea surface temperatureSUHI Surface urban heat islandTIR Thermal Infra-redTIROS Thermal Infrared Operational SatelliteTM Thermal Mapper (of the Landsat satellite)UBL Urban Boundary LayerUCL Urban canopy layerUHI Urban heat islandYD Year daySYMBOLSAreas (m2)A,, building area (roof + wall)A complete surface areaAL Lot areaA0 horizontal ground-level areaxxviA plan areaApr plan or apparent roof areaplan (apparent) vegetation areaAr roof areavegetation area (3-D)wall areaBL building length (m)C subsurface volumetric heat capacity (J m3 K1)c ellipse semi-axis (Chapter 5)c speed of light (2.998 x 108 m s’)D dust particles (cm3)DPD Dewpoint depression (°C)E adjusted voltage for EIRT calculations (mV)Fhf fraction of total foilage height (m)Fgap ratio between 3-D vegetation area and geometric canopy bounding areaf fractional componentH building height (m)h Planck’s constant (6.6262 x iO J s)hf total foilage height (m)hr height of maximum crown radius (m)tree height (m)xxviihtk vegetation trunk height (m)sensor height (m)IU isotherm units (AGEMA system)iups AGEMA internal constantK . incoming global shortwave radiation (W m2)K t reflected global shortwave radiation (W m2)net shortwave radiation (W m2)k Boltzmann’s constant (1.281 x 1O_23 J K—’)L radiance (W m2 sr’)L radiance of the complete surfaceL3 sensor detected radianceL0 surface blackbody radiancehemispheric incoming sky radiance at the groundLa atmospheric path emission (radiance)L sensor-detected radiance -Lay EIRT-detected radianceLag AGEMA-detected radiancen counterAlley counter (SC model)street counter (SC model)P photon emission (Photons s2 cm2 sr’)xxviiiT) spectral photon emission for specified temperaturePscan AGEMA scanner normalized photon emission (Photons _2 cm2 sr’)Fsolar AGEMA scanner-detected photon emission due to reflected shortwave radiation (Photons _2 cm2 sr’)Q energy of a photon (J)net radiation (W m2)R actual emitted radiation (W m2)EIRT measured apparent radiationRH Relative Humidity (%)maximum crown radius (m)SV sample valuesT temperature (°C or K)TV thermal values (IU)Ta air temperatureT complete apparent surface temperatureT1 complete surface temperature estimated using traverse facet temperaturedataT2 complete surface temperature estimated using image-extracted facet temperature datablackbody calibration cavity temperatureTd dewpoint temperatureTdet detector temperaturexxixTev EIRT apparent temperatureTmod modelled temperature distributionT0 thermodynamic (kinetic) surface temperatureTr apparent surface temperatureT3k apparent (radiative) sky temperature/Tu_r urban rural temperature differenceAGEMA scanner voltage (V)W street width (m)W street width (SC Model) (m)WAp alley width (SC Model) (m)w precipitable water (mm)X distance from sensor sub-point to beginning of IFOV (m)z length of (truncated) lower vegetation ellipse semi-axes (m)Z Solar zenith anglesurface roughness (m)Zmax maximum radiosonde reporting height (m)Greekalbedoangular portion of IFOV occupied by ground‘Yr angular portion of IFOV occupied by roofs7w angular portion of IFOV occupied by wallsxxxend angle for x component angle calculationstart angle for x component angle calculationdifferenceemissivity0 sensor view angle from nadirO angle between normal to the surface (of arbitrary slope) and the solarbeamsubsurface thermal diffusivity (m2 s’)wavelength (,um)p reflection coefficientstandard deviationT atmospheric-path transmissionrelative instrument spectral response with wavelengthSolar azimuth anglecanyon azimuth anglesensor azimuth angleview factorsky view factorMiscellaneousdegreespatial averagemimeanSubscriptsa airc complete (height of FOV centre; Chapter 2)f canyon floor (fixed site - second subscript; Chapter 2)g ground teamh FOV height (Chapter 2)t traversew canyon wall (FOV width; Chapter 2)muAcknowledgementsWithout the support and assistance of many persons, this thesis could not have beenundertaken.My research supervisor, Dr. T.R. Oke, has shown unwavering support of the projectfrom the beginning, and his encouragement has allowed me to continue to pursue thisresearch through many difficult times. His guidance and teaching have extended beyondthe academic realm and have been instrumental in my development as teacher, researcher,and person.Thanks are also due to the roles played by the other members of my supervisorycommittee: Dr. P. Austin provided guidance on the atmospheric corrections and theimplementation of the LOWTRAN model, as well as enduring all my questions on variousaspects of workstation computing. Dr. M. Church provided a thorough review of thethesis draft and gave useful and realistic guidance in the practical matters of completinga thesis when it was most needed. Dr. R. Cole was a continual source of inspiration onthe role of building-climate interactions, and our conversations always served to -motivatemy interest in the field.A great many other people have helped with various stages of this thesis: Specialthanks are due to Dr. S. Grimmond of Indiana University for carrying out the localradiosonde launches, and operating a second energy balance site during the observationprogramme. Dr. R.Spronken-Smith endured me as a lab-mate for many years and helpedout with many of the observational details of the ground-truth programme. Thanks alsoto Drs. H.P. Schmid and M. Roth for their support and encouragement of this research.T. Cheong, A. Mclean, T. Newton, B. Rehwald, K. Richards, and S. Smith of UBCxxxiiiGeography, and M. Demanes of Indiana University provided assistance with the field observations. Pilots Graham, Steve, and Barry of Vancouver Helicopters provided smoothflights and Chris Baille patiently endured the many noise complaints during the remotesensing flights. The Ontario Laser and Lightwave Research Centre made available theAGEMA scanner and Victor Isbrucker provided extensive assistance with the operationof the instrument. Dr. A. Black of the Soil Science Department made available the temperature calibration facility and thanks are due to R. Adams for numerous consultationson the use of the Everest IRT. Jan Skapski applied his technical expertise to the installation of the AGEMA scanner in the helicopter and to the EIRT setup on the traversevehicle. Vincent Kuj ala provided significant computer consultations and constructed theprimary AGEMA data conversion program.Funding for this research has been provided to Dr. T.R. Oke by the Natural Sciencesand Engineering Research Council of Canada and the Atmospheric Environment Serviceof Environment Canada. Personal funding for this project was provided by the Universityof British Columbia Graduate Fellowships, Veterans Affairs Canada, and Teaching andResearch Assistantships in the Department of Geography.Some special friends have helped me to make it through some very difficult periodsduring the thesis. Thanks are especially due in this regard to T. Cheong, S. Rice and M.Brown. To all the others in the Department with whom I have shared lunch, icetime, orbeer, many thanks. My family has provided support throughout this project. To Sandra,whose support and encouragement was immeasurable, and who sacrificed too much tohelp me, thank you for what you were able to provide. And finally, thanks to S. Fairburnfor her encouragement during the final stages of this research.xxxivChapter 1INTRODUCTION1.1 Thermal Infra-red Remote Sensing of Urban AreasRemote sensors operating in the thermal infra-red (TIR) portion of the electromagneticspectrum provide detailed and spatially-continuous records of the radiative temperatureof urban surfaces, information that would otherwise be difficult to acquire. Surfacetemperature strongly affects the lowest air layers and thus constitutes an importantboundary condition for studies of the behaviour of the urban atmosphere.TIR remote sensing has been applied to the study of urban climate almost since theadvent of the first sensors. In particular, the urban heat island (UHI) phenomenon, longknown to be a characteristic of the urban atmosphere, was shown to exist in patterns ofsurface temperature even by early, coarse resolution satellite imagery (Rao, 1972). Laterseries of satellite-based sensors including the (A)VHRR (Advanced Very High ResolutionRadiometer) on board the NOAA series of satellites, the limited duration HCMR (HeatCapacity Mapping radiometer) carried by the HCMM (Heat Capacity Mapping Mission)satellite, and the Landsat TM (Thermal Mapper) have all been utilized in studies of urbanthermal patterns as have numerous aircraft-scanner combinations. A brief summary ofsome of the published studies is presented in Table 1.1.These studies focus upon three primary areas. Firstly, the relation between landsurface cover and surface brightness temperature as it relates to the surface urban heatisland (SUHI); secondly, the relation between remotely-measured surface temperature1Chapter 1. INTRODUCTION 2Table 1.1: Summary of past urban TIR remote sensing studies. ITOS = ImprovedTIROS Operational Satellite, NOAA = National Oceanic and Atmospheric Administration, HCMM = Heat Capacity Mapping Mission, SR = Scanning Radiometer, (A)VHRR= (Advanced) Very High Resolution Radiometer, HCMR = Heat Capacity Mapping Radiometer, MSS = Multispectral Scanner, AGEMA = thermal scanner (trade name), TM= Thermal Mapper (on board Landsat), EIRT = Everest Infrared Radiation Transducer.Tr is the apparent surface temperature, Ta air temperature.Study Platform Sensor ObjectiveRao (1972) ITOS-1 SR Identification of SUHI (Surface Urban Heat Island) with coarse resolution thermal imagery.Pease et al. (1976) Aircraft M7 MSS Surface energy balance calculation and modellingusing thermal imagery.Carlson et al. (1977) NOAA-3 VHRR Analysis of urban heating patterns. Surface energy balance calculations from day/night satellitepairs.Matson et al. (1978) NOAA-5 VHRR Identification of nocturnal SUHI in 50 U.S. cities.Price (1979) HCMM HCMR Extent and intensity of daytime SUHI in NewEngland.Matson & Legeckis NOAA-5 VHRR Daytime SUHI analysis of New England cities.(1980)Carlson et al. (1981) HCMM HCMR Estimation of surface energy fluxes, moisture- availability and thermal inertia using thermal imagery and 1-D boundary layer model.Foster et al. (1981) HCMM HCMR Strength of day/night SUHI in English cities.Goward (1981) Aircraft MSS Investigates thermal inertia as a cause for UHI.Artis & Carnahan Aircraft M7 MSS Variation of roof emissivity in urban areas.(1982)Vukovich (1983) HCMM HCMR Day/night, summer/winter SUHI analyses for St.Louis.Birnie et al. (1984) Aircraft MSS TIR survey.Hoyano (1984) Aircraft MSS Summer/winter Tr relations with type of residential area, seasonal land coverage changes.Chapter 1. INTRODUCTION 3Table 1.1 continuedStudy Platform Sensor ObjectiveBarring et al. (1985) Aircraft Scanner Relation between canyon geometry, Tr, and Ta.Goldreich (1985) Aircraft RS-25 Spatial structure of pre-dawn SUHI.Kidder & Wu (1987) NOAA-7 AVHRR SUiT under snow-covered conditions.Balling & Brazel NOAA-9 AVHRR Relation between land cover and surface thermal(1988) patterns.Schmid (1988) Aircraft MSS Structural analysis of Tr and surface cover.Lewis & Carison HCMM HCMR Analyze spatial variations of energy fluxes, sur(1989) face moisture availability and thermal inertia;comparison with land surface cover.Henry et al. (1989) HCMM HCMR Compare and model ground-level and traverse Ta,and remote Tr.Roth et al. (1989) NOAA- AVHRR Day/night, summer/winter SUHI in three coastal7- 9 cities.Dousset (1989) NOAA- AVHRR Difference between Tr and Ta; study of microcli6,7,9,10 mates.Carnahan & Larson Landsat-5 TM Analysis of morning urban cool island.(1990)Dousset (1991) NOAA- AVHRR Examine relation of remotely determined land6,7,9,10 cover with T.Eliasson (1991) Helicopter AGEMA Relation between Tr, Ta and surface geometry.Eliasson (1992) Helicopter AGEMA Effect of building geometry upon nighttime Tr.Kim (1992) Landsat TM Examine causes of urban heating. Energy balanceformulation using remotely sensed products.Stoll & Brazel (1992) Helicopter EIRT T,. and Ta relations; representativeness of groundvs. remote Tr.Chapter 1. INTRODUCTION 4Table 1.1 continuedStudy Platform Sensor ObjectiveGab et al. (1993) NOAA-11 AVHRR Assessment of Normalized Difference VegetationIndex (NDVI) as an UHI indicator.Lee (1993) NOAA- AVHRR Comparison of satellite T,. with Ta, Tr in South9, 10 Korea.Nichol (1994) Landsat TM Use of thermal imagery in tropical city planning.Quattrochi & Ridd Aircraft TIMS Variation of thermal response of different urban(1994) surface types.Shoshany et al. (1994) Aircraft RS-25 Extraction of roof-top areas for UHI analysis.and screen-level air temperature; and thirdly, the use of remotely-measured surface temperature to estimate surface energy fluxes in conjunction with numerical models. Basedupon these studies a number of consistent observations can be made regarding the natureof the SUHI:• Warmest daytime surface temperatures are foundin industrial-commercial land usezones.• Daytime SUHI is much stronger than the air temperature UHI.• Daytime SUHI and intra-urban variability of Tr are greatest in the warm season.• Nocturnal SUHI is weaker than the air temperature UHI.• Nocturnal SUHI and intra-urban variability of Tr are much smaller than in daytime.• Strong relation between vegetation, surface moisture and surface temperature patterns.The results from the numerous studies attempting to relate surface and air temperatures are more difficult to interpret. Lack of correlation between these measurementsChapter 1. INTRODUCTION 5has been attributed by Roth et al. (1989) to three main factors: lack of simple couplingbetween the surface and air due to advection in the urban canopy layer (UCL); biasedspatial sampling of surface temperature by remote sensors which preferentially view horizontal, unobstructed surfaces; and mismatches in the scales of measurement used. Ingeneral, aircraft-based, high-resolution thermal imagery shows better agreement withnear surface air temperatures, perhaps because the scales of observation are more closelymatched, than when lower resolution satellite imagery is used. Birnie et al. (1984),and Balick and Wilson (1980) note that trees display surface temperatures related tothe profile of atmospheric temperature, thereby implicitly recognizing that the surfacesobscured may have different temperatures.Modelling of surface energy balances using remotely-measured radiative surface temperature (Pease et al., 1976; Carlson et al., 1981; Lewis and Carison, 1989; Kim, 1992)is an example of the inverse problem (Price, 1982, 1989; Becker and Raffy, 1987) inwhich surface parameters are inferred from the observed temperature, which is itself afunction of the radiance measured by the sensor. This technique has been used to derivevalues of surface thermal inertia for use in geological applications (Kahle et al., 1976;Kahie 1977) and to estimate evaporation from surfaces (see, for example, Huband andMonteith, 1986b; and Taconet et al., 1986). Reviews of the application of remote sensorsto determine surface energy fluxes and evaporation are given by Price (1982), Jackson(1988), Choudhury (1989) and Kustas et al. (1989).Studies over urban areas generally find higher sensible heat flux in more built-upareas, and that surface moisture availability exerts strong control on the partitioningof energy fluxes. Spatial patterns of thermal inertia are found to be weak. There hasbeen no validation of such modelling over urban areas, although Pease et al. (1976)found maps of modelled and measured surface temperature to be in general agreement.Difficulties in matching the surface represented in the model with that viewed by theChapter 1. INTRODUCTION 6satellite have been advanced as restrictions to the testing of such models (Roth et aL,1989).1.2 Limitations of TIR remote sensingThe use of remote TIR sensors over any surface presents a unique set of problems whichrestrict the interpretation and application of the measurements obtained. Among theseare:• atmospheric attenuation of the transmitted radiance and re-emission;• non-blackbody surface emissivity;• spatial averaging of the signal over heterogenous surface areas;• directional variations of the measured radiance (anisotropic radiance distribution;also referred to as anisotropy) which arise due to the form of the surface and thedistribution of fluxes emitted (i.e., patterns of surface temperature) and/or reflectedby the surface.Over urban areas, these problems are compounded by the complex nature of the urbansurface-atmosphere system: the surface is heterogenous down to relatively small scales,it is composed of a wide variety of materials with a range of radiative properties, it isessentially three-dimensional, and spatial variations in the urban atmosphere complicateatmospheric transmission.1.3 The Urban SurfaceFrom a climatological perspective, the surface is an important concept. The surface isthe site of radiant energy receipt. Properties of the surface control the partitioning andChapter 1. INTRODUCTION 7conversion of the energy received, so the nature of the surface strongly conditions thebehaviour of the lowest layers of the urban atmosphere. Specification of surface conditionsand properties is thus an important objective for study and to gain understanding of theurban atmosphere.The complete (or true) urban surface is composed of a myriad of individual surfaceelements, ranging from a blade of grass or tree leaf to a sidewalk or road. These elementsmay be combined to form larger component surfaces. The description of the completeurban surface is hindered because of its complexity (Oke, 1984) but this can usually besimplified by considering only some of the elements, combining elements, or generalizingto consider only their macroscopic surface structure. The definition and area! limitationof the urban surface may be dependent upon the process(es) under study and the scaleof the investigation, both temporal and spatial (Grimmond, 1988).Several possible urban surface definitions are presented in Figure 1.1 (Oke, pers.comm.). The position of observation may affect the parameterization of the urban surface.For example, a representation of the urban surface from a ground-based observation pointmay not include roof tops (Figure 1.lb), and that taken from an overhead aerial positionmay omit vertical surfaces (Figure 1. ld). Very simplified conceptual definitions of theurban surface have been made in order to simplify the urban surface in numerical models.These conceptions differ considerably from the complete surface, and in some cases, theymay not actually be true atmosphere/solid interfaces at all (e.g., Figure 1.le). Onedimensional UBL models often represent the urban surface in this way (e.g., Carlson etal., 1981).The regular geometric patterns of common elements which characterize urban areashave been used to simplify and generalize the three-dimensional urban surface throughthe use of various length scales which represent the height, spacing and density of those elements. These are defined for an idealized three-dimensional urban surface in Figure 1.2.Chapter 1. INTRODUCTION 8(a)(b)(c)(d)(e)Ideal / completeground-levelroof-topbird’s eye / planzero-planedisplacementFigure 1.1: Various definitions of the urban surface. (a) ideal/complete, (b) ground-level,(c) roof-top, (d) bird’s eye/plan view, (e) zero-plane displacement.Chapter 1. INTRODUCTION 9Figure 1.2: Length scales used to simplify and generalize the three-dimensional urbansurface: H - building height, W - street width, BL - building length, Ar - roof (plan)area, A1 - Lot area, A - wall area.1.4 TIR Remote Sensing of the Urban SurfaceIn order to fully understand remotely-sensed data, an understanding of the surface(s)viewed by the sensor is needed. Over urban areas, the complex three-dimensional natureof the surface poses a difficulty in determining what the sensor views. The qualitativeappraisal of Roth et al. (1989) suggests there may be potential for significant biasesbetween remotely-observed surface temperature and the complete or “true” surface temperature over urban areas, which actually conditions the urban atmosphere. These biasesare likely to arise because remote sensors, when viewing straight downwards (at nadir),preferentially view horizontal surfaces and undersample vertical surfaces and obstructedhorizontal surfaces (e.g., beneath tree canopies). Alternatively, for off-nadir viewing angles, a biased subset of vertical surfaces may be included in addition to the plan view.B[Chapter 1. INTRODUCTION 10Resultant biases will also depend upon the Sun-sensor-surface geometry, as this controls the temperature of the various component surfaces. Preferential viewing of surfaceswith particular thermal properties (e.g., roofs and tree tops which are characterized bylow thermal inertia) can also contribute to potential biases; in the case of low thermaladmittance surfaces, the diurnal range of surface temperature will be unduly large.Understanding directional variations in space and time is therefore important for theinterpretation of remotely-sensed measurements and identification of potential biases. Italso allows the expansion of spatial and temporal coverage via off-nadir sensing, and mayyield improved or additional information about the surfaces viewed (Kimes et at., 1984).While direct investigation of the existence and magnitude of the suspected biases hasnot been carried out over urban areas, there have been a number of studies over agricultural and forest canopies to address similar questions. The following section reviews theresults of these studies.1.5 TIR Remote Sensing of Agricultural and Forest SurfacesAgricultural crops generate composite scenes in remotely-sensed imagery because thesensor’s field of view is a mixture of vegetation and substrate components and the regular spacing of row crops lends a three-dimensional structure to the crop surface, notunlike that of a street canyon. In order to infer the properties of a crop, the effects of thedifferent components must be separated. Efforts have been directed to resolving component temperatures of the different surfaces. When near-nadir measurements are taken,the resultant temperature is a composite of the vegetation and soil or substrate materialbetween the rows. Thus the objective, which is to extract vegetation temperature fromthe composite response, cannot be achieved under most conditions (Kimes and Kirchner 1983). One approach to this problem is to use multiple measurements at differentChapter 1. INTRODUCTION 11viewing angles in order to infer attributes of the scene components, and to calculate thedirectional response of sensors to crop types (Kimes and Kirchner, 1983; Kustas et al.,1990). The multiple measurements are then combined in a modelling framework in whichphysically-based models of sensor response are devised (Kimes, 1981, 1983; Otterman etaL, 1992.)Sensor-surface relations have been investigated over vegetated surfaces, especiallycrops and, more recently, forest canopies (Sader, 1986; McGuire et al., 1989). Studies ofthe response of sensor radiance to view angle, azimuth and crop geometry for differentvegetation covers have been conducted by a number of research groups. Reviews of manyof the experiments conducted are given by Nielsen et al. (1984), Boissard et al. (1990)and Paw U (1992). Table 1.2 summarizes the results of a number of studies investigatinglongwave anisotropy over a variety of agricultural and forest surfaces.In general, effects of longwave anisotropy giving directional and/or angular variationsin apparent radiative temperature are minimized over full-cover, homogenous crop orforest canopies, and are maximized when a row structure dominates. This conclusionsupports the suggestion by Roth et al. (1989) that such effects may be present overurban areas.1.6 Modelling of Thermal Emissions over Plant CanopiesThe geometrical description of vegetative surface elements used in agricultural studiesis more advanced than that of urban surfaces due to the many modelling studies of radiative transfer in vegetation canopies. In detailed vegetation surface parameterizations,canopy leaves may be modelled as randomly distributed elements of a given shape, withprescribed distributions of inclination and orientation angle over a canopy volume, sothat probability theory may be used to describe the radiation field statistically (Norman,Chapter 1. INTRODUCTION 12Table 1.2: Examples of observed longwave anisotropy over agricultural and forest surfaces. Max. - Maximum observed temperature difference with view direction/angle,RMS - root mean square difference of temperatures, Azim. Max. - maximum differencedifference during azimuthal rotation.Authors Surface type Anisotropy ViewMax. RMS Azim angleaMaxMonteith and Szeicz (1962) Grass 3 1 1 0-80Fuchs et al. (1967) Sudan grass 0.6 0.3 15-75Alfalfa/bromegrass 1.8 0.7 15-75Soybeans 1.7 0.7 15-60Kimes et al. (1980,1983) Cotton rows 16.2 0-80Wheat rows 13.0Lodgepole pine 2.2 0.7 5-85Nielsen et al. (1984) Soybeans 1.5 0-75Balick and Hutchison (1986) Oak-hickory forest 7 5-10 0-70Paw U et al. (1989) Sunflower 0.4 9.3 0-90McGuire et al. (1989) Oak-hickory forest 3.2 6.8 10-85Lipton (1992) Mountainous 3b 46-65terrain 6cameured from nadirbrealistic viewing conditionsCrelative bias; 2 satellitesChapter 1. INTRODUCTION 131975). This model type is most useful for homogeneous crop covers (Boissard et al.,1990). The primary structural information needed is the total leaf area per unit soil areain the different canopy layers. This, together with some additional information on standparameters such as plant distribution and density, can be used to derive the probability ofinterception or non-interception of radiation by an element in a given layer of vegetation.These probabilities have been used to formulate a thermal infra-red exitance model of aplant canopy (Kimes et al., 1981). The Kimes et al. (1981) model has been extended andrefined by Smith et al. (1981) and McGuire et al. (1989) to incorporate the variation ofthe probability of interception due to changes in view angle and azimuth (although thethree-dimensional surface function was not actually incorporated in the model).Still more complex models of radiative transfer in plant canopies, using a weighted-random approach, are available (e.g., Norman and Welles, 1983) which combine probability theory with a simplified geometric representation of individual vegetation elements(such as spheres, ellipsoids etc.) in order to model discontinuous vegetation canopies.From an energy-balance approach, a number of models are available which can estimate longwave emissions. The simplest of these are single or multiple-layer “big-leaf”models which can be used to describe anisotropic emissions. Complex models couplingparameterized plant canopy turbulence and energy budgets with view angle geometryhave also been developed (Paw U et al., 1985).In contrast, very simplified abstractions of vegetative surfaces have been made bySutherland and Bartholic (1977), Kimes and Kirchner (1983), and Caselles and Sobrino(1989) in considering thermal radiance models of orchards and row crops. These studiesuse extended rectangular surfaces to represent the overall vegetation structure. Theseso called “geometric projection models” combine canopy structure information and component temperatures with a given remote sensor IFOV (Instantaneous Field of View) tomodel apparent sensor temperature.Chapter 1. INTRODUCTION 14Geometric projection models have been devised by Jackson et al. (1979), Kimes etal. (1981) and Caselles and Sobrino (1989). Kimes and Kirchner (1983) have validateda simplified version of the Jackson et al. (1979) model. Agreement between measuredand modelled sensor response was within 1°C RMS deviation. The Caselles and Sobrino(1989) model is used to derive the temperature of oranges in an orchard using thermalinformation from the different components of the orange grove, and is used for frostprediction purposes.Geometric projection models can be used to optimize the directional view of sensorsfor specific canopy structures and to investigate optimal inversion strategies for inferringcomponent temperatures (Kimes, 1983; Kimes and Kirchner, 1983). Geometric projection models form an appropriate basis for developing a model for urban surfaces, due tothe similarities exhibited between crop row structures and urban canyons. The development of such a model is detailed in Chapter 6 of this thesis.1.7 ObjectivesThe overall objective of this thesis is to investigate how the three-dimensional formof the urban surface temperature distribution affects observations of radiative surfacetemperature by thermal infra-red remote sensors. Specific objectives of this research areto:1. collect direct observations of the temperatures of vertical surfaces (typically undersampled in remotely sensed data) in selected urban areas, and characterize theirspatial and temporal variations;2. compile complete (or “true”) three-dimensional surface temperature informationfor selected urban land-use areas;Chapter 1. INTRODUCTION 153. use thermal remote sensing observations over selected urban land-use areas to determine the extent of directional variations (anisotropy) of urban surface thermalemissions;4. investigate the suitability of available two-dimensional geometric projection modelsto the prediction of apparent sensor temperature for remote sensors viewing typicalurban surfaces.1.8 MethodologyThe objectives set out in Section 1.7 are implemented using an intensive observationalprogramme of selected sites in Vancouver, B.C. The observational programme consistsof three main components.1. Remote observation of urban surface temperature, at different viewing angles andazimuths, using a portable thermal scanner carried on board a helicopter. Theseobservations were made at times when preliminary ground-based data suggestedthe potential for maximum variation.2. Mobile sampling of the surface temperatures of vertical building facets using vehicle-mounted infra-red transducers. Traverses were made on a regular basis throughoutthe daytime period to determine the heating and cooling patterns on the different facet orientations. Traverses also monitored road surface temperature andin-canyon air temperature (at a height of approximately 1.5 m).3. Sampling of component surface temperatures with hand-held infra-red thermometers by ground teams, to determine the variation of temperature at the microscaleat the time of remote sensing data acquisition.In addition, a number of supporting observations were made or obtained including:Chapter 1. INTRODUCTION 16• the vertical profile of atmospheric pressure, temperature and humidity in order toallow correction of the remote thermal imagery for atmospheric absorption andre-emission;• surface temperature measurements of homogeneous calibration surfaces for comparison with remote observations;• emissivity measurements of the calibration surfaces;• concurrent surface energy balance measurements (at two sites);• thermal satellite imagery (NOAA-11) as available.To support model development, and for comparison with observations over the studysites, a brief scale modelling exercise using concrete blocks to represent buildings wasperformed. In this component of the study, the thermal scanner was mounted on a mobileplatform and viewed simple surface geometries constructed in an open, fiat parking lot.1.9 Study SitesThe observational programme was carried out using three primary study areas withinthe city of Vancouver, B.C. (Figure 1.3).The sites were chosen based upon a number of considerations including:• the importance of the surface type as a fraction of the total urban area, (and implicitly, its associated weight with respect to its influence upon the urban atmosphere);• variations in the magnitude of surface structure as expressed by measures such asthe plan to active surface area, and street height to width (H:W) ratios;• areas where topographic surface variations are small with respect to the urbansurface structure;00z0 0C 00.CDci) CD CDCI)CD01E1[WWWW[‘EWUIi[“E1fl100wwW00I.Chapter 1. INTRODUCTION 18• the ability to define a representative sample of the surface for detailed study. Forexample, the representative area concept introduced by Schmid (1988) for residential land use types in which the contribution to the total variance attributed tospecific surface structures is analyzed;• structural simplicity of the surface structure; a real surface is selected to closelyresemble model surface representations in order to provide a link between the modelling and observational components. Such a surface is characterized by box-likebuildings with fairly homogeneous facet material distributions, a lack of vegetationand a regular surface geometry;• previous remote sensing coverage, database development, or coincident parallelresearch.1.9.1 Industrial (False Creek South)This area is an inner city light industrial district of fairly limited extent (Figure 1.4).The area is typified by very rectangular, fiat topped buildings of between 1 and 3stories. Commonly, buildings adjoin one another and are arranged on blocks with theirlong axis oriented E-W. Blocksare separated by relatively wide streets. Alleyways parallelthe long axis of the block. There is very little vegetation, particularly trees, which leadsto relatively simple geometric street canyon configurations. The topography shows agentle slope trending upwards to the SE.1.9.2 DowntownThe downtown study area fulfills the requirement for an area with very large verticaldevelopment where vertical walls become a large fraction of the total surface area (Figure1.5).Chapter 1. Introduction 19___L’.: *14--k: Y’/‘L - c‘ - im.i4_ _______—___-V.— F-Zi - —__ _ _-.- Y—-kct!2’__.Figure 1.4: The Industrial study area (False Creek South). View is to the west.Chapter 1. INTRODUCTION 20Building heights approach 150 m for some office towers. Building facades are complex,and contain large amounts of glass. Large office towers dominate the southern part ofthe study area while blocks to the north are characterized by older, adjoining 5 - 7 storybuildings. Blocks are oriented NE-SW with some square blocks and some with theirlong axis in the NE-SW direction. Some blocks are characterized by NE-SW alleyways.Vegetation in the form of trees is fairly prevalent along the streets, although for the mostpart it is immature. The topography slopes downwards to the water in a NE direction.1.9.3 Residential (Sunset)The Sunset area is an urban residential neighbourhood located in South Vancouver withsome commercial and institutional developments interspersed, particularly along majorstreets (Figure 1.6).The housing is primarily 1 or 2 story detached dwellings, relatively closely spacedalong the street. In the area most extensively studied for this project, two block orientations are present; an E-W long axis alignment south of 49th Aye, and a N-S longaxis alignment north of 49th. Buildings are set back significantly from the street andgarages border alleyways which divide the blocks along their long axis. Roofs generallyhave a shallow pitch; many have a simple peak roof with the long axis perpendicular tothe block axis. Vegetation of all forms is extensive.The Sunset study area has a long history as the base for many local atmosphericstudies (Steyn, 1980; Schmid, 1988; Grimmond, 1988; Roth, 1989) and is the location ofa free-standing tower used as an instrument platform for those studies.Chapter 1. Introduction 21I4c,.7;,Figure 1.5: The Downtown study area. View is to the west.Chapter 1. Introduction 22_____________-..w -t ——_____ __-I : “I __.i ‘tL¼us- ‘--.- SSSI— i .Figure 1.6: The Residential study area (Sunset). View is to the northwest.Chapter 1. INTRODUCTION 231.10 Prevailing ClimateThe Vancouver area is characterized in summer by persistent anticyclonic systems whichyield extended periods of fine weather suitable for remote sensing work. Surface analysesfor the primary study period of August 15-17 1992 are presented in Figures 1.7, 1.8, 1.9.During this period an offshore ridge of high pressure gave rise to clear skies and a synoptically induced northwesterly flow at the surface.Local observations of radiation, temperature and wind (Figure 1.10) show practicallyidentical radiation regimes for the three study days.Incident shortwave radiation at the surface approaches 850 W m2 and daytime netradiation (taken from a 22 m tower in the Residential area) is close to 500 W m2. Airtemperatures from two of the study sites (measured at 9 m and 10 m at the Industrial,and Residential sites respectively) and the Vancouver airport (screen-level observations)show the third day to be somewhat warmer than the first two, especially in the Residentialarea and the second day to be slightly cooler than the first. August 15 shows evidenceof a local seabreeze development with evening and morning easterly winds shifting towesterly or southwesterly in the afternoon and remaining steady until evening. August16 is characterized by a stronger northwesterly flow influenced by a steepening of thecoastal pressure gradient associated with the offshore ridge and inland low pressure areas(Figure 1.8). August 17 shows daytime winds from a more westerly or southerwesterlydirection (especially at the residential site) and with lower velocities than for August 16.Chapter 1. INTRODUCTION 24ENVrn090EMENT CANADA- V ——‘ ENVIRONM6NTCANAQA• j• 005 v AUGUST 00002• 7 ISSUED AT 0320Z QV2 3 122 1) 2_SSA\• 8 )0A 134 24 060 017 1 108•S 8 8 • Ii / 4 o ..08 05 64 5.7 10710 0020 (1O517 \10••072I ?..• V• S..7/• 81 0 I1TM0012328 62o\ :: ;—>08046 56 05\. 1°SV•j0• . 12 128120 0.1•136 3CYX 2 174 (11 80 1 •••S•••...._f460330801 05 ! : 1 12 411 :13 02 •_SVVi •••5 N.FL5 I •.. . 0•° . 003 8•••\%%/;;12:1 V(13)J0CF VV07578D9\5 13P((5) —. . — 2 V V V V f?4 . A 80VV 11VVO.MGj’V ___V•VV(12) /6 V V VV(15).62 10’ V 20V ( V -k’j VS—cV OUV V VVV00 i\ 17 t1\ .os- \ICCC V • V 1 215 86.1 V 1427 0 22A 1/:415/:o, 2I 745,10/ 8 915Figure 1.7: Pacific Weather Centre surface analysis for August 15, 1992.Chapter 1. INTRODUCTION 2500- ENVGNMENTCAN0D0ENVj00900MENT CANADAPACIFICWE*T99001NTRE7 999 j joe SURFACE ANALYSIS V 32 ‘071 AUGUST 16 9992 0000ZISSUED AT Q32OZ- 204054...‘—‘ I89 94\ . -..;‘ \;---7 7 29 1116 029 o—.- 10\10 43 g’12222 - 8 189 \. 32 8 148v3 \ a 240 \ o ‘- , ‘.ze_ 06.:‘-•.. 80•• Q3( • I 2 0 23 -.2i l22 1663 4243‘.•BA<,-40+ \.•3 1 05\ 9 •••• 1 32 -. 136.: :0104°o8 20 138 0Ii05122 4 13‘28-iS9 6Figure 1.8: Pacific Weather Centre surface analysis for August 16, 1992.Chapter 1. INTRODUCTION 26• 00 O0 o ONVIWONMENT CANADAINVIRONNEMENT CANADA.SPACIFIC WEATHER CENTRe8 050 21 038 SURFACE ANALYSIS V 3/ 63 • 15 A 136J 7 AUGUST 17 /992 0000Z1 008VISSUED AT O32OZIO\ a:.,23 0 4&6 5. .5 19I740 3 0 .Ii \TL.-a03\62 / .- ...5 . 22 .. 160 21 1 N16 A OAR \24.—o 03 1. I.Z4_13 D• . .(121 ID322 •S 1 co9004 25—‘6 6 28 201 20061. 173• . .. 14 25 CDI 1, 24 -/205 3’)505s4 2I..,—T 074 0 17 \ \ 9 74 0 02 20 01 2‘22 276 0 c155— o_ f • io4 : 8 80012 74\07 4 440 24 03 26 14 12o AR’• 050 •uo 8 13./6o1 O 06)3 140 02912)2;‘:: :‘042 1 0 13 /480332902 . 20 14• 4O3. \. f ç14/ 80 I2 1 125....(121 • • . S S •• • 10 6/4o18 0S .• / • • • 20 76 •.Z 01 143S •. .15- 4 _.J 4ç-u lAS- S • ( (1 •. 441 . • •\ ‘61• • S • •5• . / . • l7T • . 142 \ . \(• S . •- I? 26 2O\ —5 .5 5 5 . . . /. • • 1671 d I22P9EC_ S ‘°4’..-r3(1__IFigure 1.9: Pacific Weather Centre surface analysis for August 17, 1992.Figure 1.10: General weather conditions on the three study days: August 15 - 17, 1992.(a) incident shortwave radiation and net radiation from the Residential site, (b) airtemperatures (Residential - 10 m, Industrial - 9 m, airport - 1.5 m), (c) windspeed anddirection (10 m).Chapter 1. INTRODUCTION 279008007006005004003002001 000—100(G)E>‘C,,CC,)xD>‘C,)Cw(b)0 12 24 12 24 12 24Time (PDT)(C)a-o0 12 24 12 24 12 24Time (PDT)7807a)04aU,— Resideotial— —— Airport ,/0 12 24 12Time (POT)24 12 24Chapter 2AUTOMOBILE TEMPERATURE TRAVERSES2.1 IntroductionThis chapter presents the aims, methodology, and results of vehicle surface temperaturetraverses. The primary aim of the traverses is to specify, for a particular land use area,typical surface temperatures of the vertical building facets (i.e., walls), surfaces whichare often undersampled in remotely-sensed imagery. In addition, the roadway surfacetemperature and UCL air temperature (at approximately 2 m) were also monitored duringthe traverses.Traverses were conducted along major streets and alleyways in order to obtain representative samples from a range of buildings. Traverses were designed so that all majorfacet orientations (4 are assumed) were sampled. Stratification of the data by facetorientation allows identification of times when strong azimuthal variations in surfacetemperature between different facets occur. This information is used to schedule remotesensing missions at times when differences of observed temperature with view directionare maximized.Facet orientations are defined by their relative position to the centre of a building(e.g., the east wall faces east), which appears to be the most conventional definition. Notehowever, that if the point of observation is within a street canyon, it is not unreasonableto define facet orientations with respect to the canyon centre; (e.g., the north canyonfacet is actually a south-facing building wall). For consistency and to reduce confusion28Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 29the former definition is used throughout this thesis.2.2 Method2.2.1 Sensor ConfigurationsSurface temperature measurements were made using an array of 15° FOV infrared transducers (Everest Interscience Model 4000A, hereafter referred to as EIRT) mounted on apickup truck. Technical details of the EIRT are provided in Appendix C.Two primary configurations were employed (Figures 2.1 and 2.2): in the first, 4 EIRTswere mounted facing to one side of the vehicle at angles of 0, 15, 30, and 45° above thehorizontal. The second had two pairs of EIRTs at angles of 0 and 10-15° mounted facingoutwards to the right and left. Both configurations also have a single EIRT facing to therear and downward (at approximately 45° below the horizontal) to measure road surfacetemperature, and a shielded and aspirated thermocouple (26 - 30 awg) to monitor airtemperature.(a) (b)Figure 2.1: Traverse configurations used to sample surface temperatures. (a) configuration 1, (b) configuration 2.Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 30The first configuration is optimized for use in urban areas characterized by large H:Wstreet canyons. This configuration obtains temperatures from only one canyon facet andtherefore requires two traverses to obtain complete canyon facet temperature information.The second configuration allows expanded coverage, or more frequent traverse intervalswithin areas characterized by low H:W street canyons.2.2.2 Sampling MethodologyA wheel-mounted magnet-sensor array interfaced with a Campbell Scientific (CS) 21Xdatalogger was used to generate digital pulses at a rate of one per full revolution ofthe truck wheels. Surface temperatures were sampled when the accumulated pulse countreached a pre-set number. The minimum spatial sampling interval is set by the circumference of the wheel, nominally 2.38 m. The magnet-sensor array and datalogger samplinginterval limit the maximum useful traverse speed to approximately 60 km hr’. However,the EIRTs have an internal calibration procedure which obscures the field of view each 0.5s for 0.25 s during which the temperature value is not updated. This limits the maximumtraverse speed to 17 km hr’ when one sample per wheel rotation is desired. Generally,traverses were conducted with sampling conducted after every second wheel rotation,allowing traverse speeds of approximately 34 km hr’, i.e., samples spaced at 4.76 m.The number of samples obtained for different structural units of the urban surface in thethree study areas is outlined in Table 2.1. A written record of the truck position andtime during traverse was kept, with entries made at the start and end of each street, oralleyway. To more closely establish position along the street a digital record was kept bytoggling a switch on the datalogger when the truck was in an intersection between blocks.This procedure also allows those observations to be removed from the record if desired,although, as will be demonstrated later, lag in the response of the instrument results inthe displacement of within-intersection temperatures to outside the flagged boundaries.Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 31Figure 2.2: The traverse vehicle outfitted in traverse configuration 2.Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 32Table 2.1: Sampling intervals per urban surface structural unit.Study Area Structural Sampling IntervalsUnit per structural unitResidential building 2block 40 (N/S), 30-40 (E/W)Downtown building 2 - 4block 17, 32Industrial building 2 - 4block 17 (N/S), 37 (E/W)2.2.3 Projected FOVWhen the EIRT FOV is projected onto street canyon walls, the area viewed is that of anellipse. In the limiting case of the horizontal EIRT the projected FOV is a circle (if theground surface is not “seen”). The size and elongation of the ellipses increase with sensorangle above the horizontal (Figure 2.3a). Ellipse dimensions for the sensor to surfacedistances along the traverse routes are presented in Figure 2.3b. As the building setback distance increases, overlaps between adjacent samples can occur, otherwise adjacentsamples are separated by gaps (Figure 2.3c). Because of the manner in which an EIRToperates, the portrayal of the FOV as an ellipse is true only for a stationary sensor. Withmovement, the signal is obtained as the FOV moves across the building facets during thetime the EIRT optical aperature is open for viewing (0.250 s each 0.5 s, see AppendixC).Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 33( ‘1 3.76m\Q) <E6If)163 mDistance from facet; 10 m( 1.) 14612-o• 108C5)60U--D 45)C-)a)0C0 01 3 5 791113151719212325Distance from Facet (m)(c)........ 0.58 Sensor Angle /7—e---o° / 0.4.--s-- 15° ,0 5-C 6--G—30° / / -o° 45° e 0.35 5 / >> / oo 4 \ / / /p3 // ;;: 0.2 o.I.1 0 / ‘‘ 0.1 °0___________________________________________0.01 3 5 7 91113151719212325Distance from Facet (m)Figure 2.3: (a) Projected EIRT FOV for traverse configuration 1 and 10 m from wall.(b) Projected FOV dimensions with distance from facet. (c) Projected FOV gaps (left)and overlaps (right) for given sensor-facet distances.width: 0°width:15°- —— width: 30° 0—— width; 45° 0S length: 1 5°C length: 30° 0o length: 45° 0—— —Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 342.2.4 ProcessingThe data were corrected using calibration information for each EIRT. The record wasalso checked to ensure samples were not taken more frequently than that limited bythe chopper frequency. Samples taken at more frequent intervals were considered as“missing” values and assigned a code. The written traverse record was used to search forthe nearest actual sampled value and a file (navigation file) was created of those times,street positions, and facet orientations viewed. Using the navigation file and the cleaneddata file, the data were separated according to facet orientation.2.3 Surface EmissivityNaturally occuring surfaces have emissivities less than unity. Therefore, measurementsmade using thermal infrared radiometers incorporate both thermal emissions from thesurface and reflected radiance. The range of emissivities in urban areas is relatively largedue to the variety of surface materials used in building contruction. In general we mayexpect to view surfaces with emissivities ranging from 0.7 or less, for window materialswith special coatings, to 0.98 or higher for some painted surfaces and vegetation. Table2.2 presents a summary of the emissivities present in urban areas according to valuesin the literature. When infrared surface temperatures of a horizontal surface are viewedfrom above, surface emissivities less than unity lower the apparent surface temperature(Tr). This arises because the source of reflected radiance is primarily the sky (assumingthe surface is relatively unobstructed and behaves as a diffuse reflector). This is typical ofairborne or spaceborne observations of surface temperature over relatively flat areas. Inurban areas, horizontal within-canyon surfaces have an reduced sky view factor (b8), sothat radiance emitted from buildings replaces that from the sky. The reduction in surfacebrightness temperature is therefore less, because some sources of reflected radiance haveChapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 35Table 2.2: Surface emissivities for surface materials in urban areas. After Oke (1987);additions from Arnfield (1982).Surface Material emissivity,concrete 0.71 - 0.9asphalt 0.95brick 0.9 - 0.92stone 0.85 - 0.95wood 0.9glass 0.87-0.94 zenith angles<40°0.87-0.92 40° <Z <80°paint: black 0.9 - 0.98paint: white, red, 0.85 - 0.95brown, greenroofing shingles 0.9 - 0.92tar-gravel roof 0.92tile roof 0.90slate roof 0.90yard (90% lawn, 10% soil) 0.968short grass 0.97 - 0.99urban areas 0.85 - 0.96; average 0.95Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 36a radiative temperature close to that of the surface viewed. Another way to interpret thiseffect is to consider the canyon to increase the effective of the system due to internalmultiple reflections (e.g., Sutherland and Bartholic, 1977; Arnfield, 1982; Caselles etal. 1992). For the case of vertical canyon facets viewed from a position low within thecanyon, the source of incident radiance upon the facet includes significant portions fromthe ground (canyon floor), the opposite canyon wall, and relatively less from the sky(Table 2.3). The canyon H:W data of Table 2.3 are representative of the three studyareas: the Residential values are H:W 1:4—1:2 and the Industrial area values are H:W1:2—1:1 for streets and alleyways respectively. The Downtown area has H:W ratios whichvary significantly; values for several of the canyons traversed ranged from 1:1-2:1.Table 2.3: View factors for the mid-point on the floor of a canyon and a point low (3 mabove floor) on the canyon wall for canyon H/W representative of the three study areas(for a discussion see text). Canyon widths are 25 m (H:W 1:1, 2:1), 35 m, (H:W 1:4), 20m (H/W 1:2).View factor Canyon H:W1:4 1:2 1:1 2:1of walls for floor 0.10 0.30 0.55 0.75of sky for floor 0.90 0.70 0.45 0.25of floor for wall 0.45 0.43 0.44 0.44of opposite wall for wall 0.12 0.24 0.39 0.50of sky for wall 0.43 0.33 0.17 0.06To determine the effect of < 1 on the traverse measurements, view factors typicalof the canyon geometries in the study areas are calculated and surface temperaturesassigned based upon an examination of the initial temperature histograms. Using thesetemperatures as “true” values and applying different €, the wall Tr is calculated includingChapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 37a single reflection event from the other canyon facets.Results for canyons typical of the Industrial study area (Table 2.4) mostly show Trdecreases, but some increases occur when the EIRT views a cool wall and the opposingwall is hot. Differences vary with time and facet orientation. As canyon H:W decreases,Table 2.4: Facet temperatures (estimated from traverse data for the Industrial study area)with calculated apparent temperature for wall 2 using canyons with H:W = 1:1 and 1:2.Apparent sky temperature, (T3k) is set at —20°C, = 0.95 all facets. T01,T02,T0f arethe true temperatures of wall 1,2 and the canyon floor respectively. Trw2 is the calculatedapparent temperature for wall 2.Time Wall T01 T02 T0f Trw2 öTrw2 Trw2Pair H:W = 1:1 H:W = 1:20745 S-N 17 22 17 21.5 +0.5 21.2N-S 22 17 17 16.8 +0.2 16.5E-W 19 19 18 18.7 +0.3 18.41035 N-S 26 21 32 21.0 0.0 20.7S-N 21 26 32 25.7 +0.3 25.4E-W 21 32 32 31.4 +0.6 31.2W-E 32 21 32 21.2 -0.2 20.81405 N-S 35 23 44 23.4 -0.4 23.0S-N 23 35 44 34.6 +0.4 34.3E-W 28 26 42 26.1. -0.1 25.7W-E 26 28 42 27.9 +0.1 27.61740 N-S 25 24 39 24.0 0.0 23.7S-N 24 25 39 25.0 0.0 24.7E-W 26 22 36 22.1 -0.1 21.8W-E 22 26 36 25.8 +0.2 25.5Tr values are generally less than true surface temperatures (T0), due to the greaterThe decrease in Tr with lower H:W may be somewhat offset in heavily vegetated areasChapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 38where the overall canyon c may be higher (Table 2.5).Table 2.5: Difference between true and apparent wall temperatures at select times for aH:W 1:4 canyon (typical of the Residential area) for = 0.95 and 2 = 0.97. Inputtemperatures obtained from traverse data.Time zTEl /XT2 Time zT1 ZTE2 Time zT1 tT62 Facet Pair0830 0.7 0.4 1030 0.4-0.7 0.3-0.4 1330 0.6-0.7 0.3-0.4 N-S0.8 0.5 0.7-0.9 0.4-0.5 0.9-1.0 0.5-0.6 S-N1.0 0.6 1.0-1.1 0.6-0.7 0.8-0.9 0.4-0.5 E-W0.7 0.4 0.5-0.7 0.3-0.4 0.7-0.8 0.4-0.5 W-EA case where effects could lead to substantially lower Tr arises with specular reflection of sky radiance from the facets of buildings (especially the Downtown site) for theEIRT mounted at 30° and 45°. This effect has also been observed in other field studies(Verseghy and Munro, 1989). Figure 2.4 presents a transect of temperatures at increasingelevation angles, using a hand-held IRT with a narrow FOV, for an office tower clad incopper-coloured reflective glass.Correction of surface temperatures for on a point by point basis is impossible becausethe e of the viewed surface is unknown, the temperature of the surroundings must beestimated, and the exact canyon geometry is usually unknown. Application of emissivitycorrections is deferred until summary statistics of apparent temperature for each facetare available. Then, using the mean apparent temperatures for opposing facets, anestimated sky radiant temperature, and approximate values for canyon floor temperatureand canyon geometry, the actual emitted radiation of a facet is estimated usingR2 = — (1 — +8kSky,W2 + ff,W2)] (2.1)w2where R is the apparent radiation determined from the apparent temperature measurements made by the EIRT, R is the actual emitted radiation, and are view factorsChapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 3922- A20-A18ci) A0A. 141 2reflected sky AAci 10o A— 8Ao 62 ALt 42 reflected sunlit buildingAA0I I • I I I I9 10 11 12 13 14 15 16 17Apparent Radiative Temperature (° C)Figure 2.4: Vertical transect of 7’,. up the face of a north-facing wall of a glass-clad officebuilding.of the first subscripted facet for the second subscripted facet. Facet designators arew1 and w2 for wall 1 and 2 respectively, f for the canyon floor, and sky for the sky.Radiation values are calculated using the Laguerre-Gauss quadrature method (Johnsonand Branstetter, 1974) and T0 determined from R are obtained by interpolation in apre-calculated look-up table.2.4 Everest - AGEMA Scanner ComparisonsA limited number of short traverses were conducted in which the AGEMA THV 880 LWBscanner used for acquiring airborne measurements was placed alongside an EIRT mountedon the traverse vehicle. Various tests were conducted; those described here utilize asampling frequency of 12.5 Hz for the AGEMA scanner and a single pulse interval forChapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 40the EIRT sampling. EIRT observations more frequent than the 0.5 s limit are discardedfrom the analysis. The AGEMA scanner output (140 x 140 pixels) is averaged over theimage with the EIRT FOV weighting applied to the image. Comparison of the timeseries from the two instruments (Figure 2.5 shows a short section of one such test)demonstrates generally good agreement, although the EIRT response is slightly lagged.The presence of a lag is important only if the temperature value measured is to be relatedto structure on the ground. For example, the elimination of samples taken during passagethrough intersections may result in the removal of some observations near the start ofthe intersection which actually apply to buildings sensed just prior to the intersection.The EIRT tends to underestimate (overestimate) the highest (lowest) scanner values.40101 3.695 1 3.697 1 3.699 1 3.701 13.703 13.705 1 3.707 1 3.709Time (Hours; PDT)Figure 2.5: Surface temperature time series using EIRT (dashed line) and AGEMA scanner(solid line).Where these are related to small surface features, they are not picked up in the EIRTtrace. However the response to features of the scale of typical buildings is, in most casesreplicated by the EIRT. Distributions of sampled temperatures for the AGEMA scannerand EIRT for two traverses (Figure 2.6) show generally good agreement. It was expectedAGEMA ScannerEIRTChapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 41that the AGEMA, with its slightly smaller FOV and fast sampling period, would showgreater frequencies at high and low temperature classes relative to the EIRT. There issome evidence of this, but it is not strong, and based upon the tests conducted, theEIRT is accepted as providing an adequate representation of the vertical surface facettemperature.2.5 Temperature Distributions2.5.1 Traverse Configuration 1The traverse system samples temperatures, irrespective of the position of the projectedFOV at the sampling time. Thus, samples often include combinations of surfaces, including the sky. Plotting traverse results in the form of distributions illustrates this effect.In Figure 2.7 results are separated by facet orientation (panels a - d), and further brokendown into the angle of the EIRT, and the side to which the EIRT was facing with respectto the direction of the traverse (left or right; recall that two EIRT are oriented in eachdirection in traverse configuration 1; Figure 2.1). Also plotted is the distribution of airtemperatures (broken down into classes of 0.25°).A common feature of the four plots is the presence of a bimodal distribution with asmall peak at temperatures between 0—12°C and a larger peak at higher temperatures.The cooler set is interpreted as originating from samples taken when a large portion ofthe projected FOV is occupied by sky. The warmer set is comprised of samples whichview non-sky surfaces; here assumed to be primarily vertical surfaces. The shape of thewarmer distribution varies with facet orientation. Shaded facets (north facets (a)) arecharacterized by narrow, sharply peaked Tr distributions. Facets exposed to direct beamsolar radiation show a dramatic increase in the range of surface temperatures. In thiscase, sunlit or shaded surface possibilities exist resulting in a broad peak of temperatures>‘C-)CDci-ci)U-Test 3—2—a— AGEMA ScannerEIRTi410 20 30 40Apparent Surface Temperature (0 C)Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 42(a) II I30201000(b)181614Figure 2.6: Apparent surface temperature distributions for the AGEMA scanner and EIRTobtained from the traverse tests.50Test 3—7—a--— AGEMA ScannerEIRT0 10 20 30 40Apparent Surface Temperature (0 C)50Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 43with a tail towards higher values (east facets (c)).As the angle of the EIRT above horizontal increases, there is a shift towards a greaterproportion of observations in the lower portion of the distribution, because more samplesinclude sky in the FOV. Separation by direction of the EIRT accounts for the differencesin the distance to the facet between the left- and right-facing EIRT; the left-facing EIRTis further from the facet and therefore has a greater chance of having some portion ofthe FOV occupied by sky (since the projected FOV is larger). This effect is only weaklyexpressed in the results; it is best viewed as an increase in the frequencies of occurencesof left-facing samples in the lower distribution.In the Industrial and Residential study areas, traverses were conducted on both streetsand alleyways. In both areas, the geometry of the alleyways differs from that of the street(generally lower H:W for the street canyon compared to the alley), because alleyways arenarrower and generally have a smaller building set-back distance (although this distancetends to be much more variable than for the street canyon in these areas). Figure 2.8compares alley and street temperature distributions in the Industrial study area for a midmorning traverse. Results from left- and right-facing instruments are combined because,due to the manner in which the traverse was conducted, relatively few observations wereobtained for some facet-instrument pairings.South facets show slightly greater proportions of higher temperatures in the streetcanyons compared to the alleys, especially for the 100 EIRT. This may be due in part to(early morning) differences in solar access because of the different canyon H:W and/or asmaller and lower projected FOV within the alley canyon which may be more likely toview the shaded portions of walls. This may also account for the enhanced frequenciesof sky or mixed sky and building observations in the alleys which occur counter to thatexpected due to canyon H:W differences. The overall shapes of the distributions aresimilar and the apparent position of the peaks are not too dissimilar.Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 44Figure 2.7: Frequency distributions of Tr obtained during a mid-morning (10254048PDT) traverse of the Industrial study area for: (a) north, (b) south, (c) east, (d) westNorth Facets-—.— LeftlO°•—-— Left 0°LeftT1—0—— Right 10°—e--— Right 0°Right T0(0)0 10 20 30 40 50Apparent Surface Temperature (° C)>‘0ca)0a)U>‘0ca):30a)U.. Facets (c)—e— Left 10°—-— Left 0°——— LeftT01—0---— Right 10°—e--- Right 0°RightO5ooooao r0 10 20 30 40 50Apparent Surface Temperature (° C)West Facets (d)—.-— Left 10°—a-— Left 0°——— LeftT01—0--— Rightl0°—&--— Right 0°— RightT0South Facets (b)—•— LeftlO°-—0-— Left 0°LeftT—0—-- Rightlo°—0---— Right 0°Right0 10 20 30 40 50Apparent Surface Temperature (° C)0 10 20 30 40Apparent Surface Temperature (° C)50facets.>‘0Cci)ccci)Figure 2.8: North and south facet temperature distributions for 0 and 100 EIRT in alleysand streets in the Industrial area. Note the variation in ordinate scale.North facets have similar distributions for the 00 EIRT. The 100 EIRT shows a tendency for streets to have more observations which view sky or mixed sky and building asmay be expected from the H:W differences, but the distribution shapes are similar, especially in the range of likely surface temperatures. Early and late evening traverses werechecked to determine if any enhancement in distribution differences occurred becausehigher local zenith angles at these times could potentially create greater solar access differences. No significant differences in the shapes of the distributions were noted. BasedChapter 2. AUTOMOBILE TEMPERATURE TRAVERSES(a) (c)452 6 10 14 18 22 26 30 34 38Apporent Surface Temperature (° C)0.5Streets North Facets, GIRT 0°Alleys0.4>0.30D00.20.10.02 6 10 14 18 22 26 30 34 38Apparent Surface Temperature (° C)(d)0.5Streets North Facets, EIRT 10°Alleys0.40.30Dcc0. 6 10 141822 26303438Apparent Surface Temperature (° C)(b)‘ Streets South Facets, EIRT 1 mAlleysnon2 6 10 14 18 22 26 30 34Apparent Surface Temperature (0 C)38Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 46on these results it is concluded that alley facet temperatures need not be separated fromstreet facets in the Industrial study area.In the Residential area, results from an early morning traverse (Figure 2.9), showmore significant differences, especially for the directly irradiated east facets which showa strong tendency for alleys to be significantly warmer than the streets for both the 0and 100 EIRT. A similar difference exists for north and south facets as viewed by the 00EIRT. The shaded (west and north) facets as viewed by the 100 EIRT at this time showstreets to be slightly warmer, the difference between streets and alleys for west facets areespecially striking. This effect may be due to the presence of many large trees within thenorth-south street canyons which are viewed by the 100 EIRT.2.5.2 Traverse Configuration 2.Figures 2.10 and 2.11 illustrate results obtained using traverse configuration 2 in theDowntown study area. The general trend of the results is similar to that describedearlier; the addition of the EIRTs at larger elevation angles yields a greater range of lowtemperature values. Separate peaks relative to the sensor angle are sometimes apparent(Figure 2.11 (b) northeast facets). These are related to the decrease of apparent skyradiative temperature with increase in elevation angle.2.5.3 Modelled Temperature DistributionsA simple model of Tr as obtained from the traverse sampling methodology was formulated to create hypothetical temperature distributions for comparison with the observeddistributions. The model requires surface structural information on the typical height,length, and spacing of buildings along a city block. Each block contains a number ofbuildings equally spaced along the length of the block plus an “intersection” with nobuildings. Each building is assigned a height drawn from a distribution with a mean andChapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 47NorthFot.0lRT0Alleys Streets rightJL0.014 15 16 17 10 19 20 21 22 23 24 25 26Apparent Surfoce Temperature (°C)14 15 16 17 18 19 20 21 22 23 24 25 26Apparent Surface Temperature (0 C)4 16 16 20 22 24 26 28 30 32 34 36 38 40 42 44Apparent Surface Temperature (°C)0.6 Streets Westrosets. 0610°Alleysa. 1516 17 18 1920 21 22Apparent Surfoce Temperature (°C)>. 0.1400.12O 0.104.. FOOots. EIRT1 0°Streelo rightAlleysllelt2 6 10 14 18 22 26 30 34 30 42Apparent Surface Temperature (°C)(h) 0.4__Streets WestForets.EIRTIS°Alleys0.3100iI20Apparent Surtoce Temperature (°C)Figure 2.9: Facet temperature distributions for streets and alleys in the Residential area.Note variation in frequency scales.(a)>‘0.34.-(c) 0.4>‘0.2(b) 0.30Streets0 25Alleys0.200.150) SouthFuoe0).EIRTS°AlleysilL..0.1(d)e(f)0.002 4 6 8 10 12 14 16 18 20 22 24Apparent Surface Temperature (°C)0.25I Streets Sooth Fosete. 610110°0lresls elIAlleys Alleys1 right0.200.150.10 L0.052 4 6 8 10 12 14 16 18 20 22 24Apparent Surface Temperature (°C)0.21EAst Forets 610110°StreetsSlreeto rigileys lettk ysAll0. ts EAst Fouls. 01010°All Streets. rrghtOr0C>‘L.Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 48(a)>‘UCa,Sa,U-“\ N(c)Northwest Facets (Right)Ta,/ \10 12 14 16 18 20 22 24 26 28 30Apparent SurfaceTemperature (° C)(ci)10 12 14 16 18 20 22 24 26 28 30Apparent Surface Temperature (°C)(b)00 /\ Southwest Facets (Right)0.3i 0.2: /7J10 12 14 16 18 20 22 24 26 28 30Apparent Surface Temperature (0 C)(e)0.100.09 Northeast Facets (Right)0.08 — . —: 3000.0745e‘ 0.06c‘ 1.0.05g 0.04‘ IL 0.03‘,0.02 ‘ \., /\/ !Apparent Surface Temperature (°C)Figure 2.10: Frequency distributions of Tr obtained during a mid-morning (1025-1048PDT) traverse of the Downtown study area. Right-facing EIRT for: (a) northeast, (b)southwest, (c) northwest, (d) southeast facets, (e) northeast facets (low temperaturerange).Apparent Surface Temperature (° C)Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES>C)CC)Sa>C)Ca)S)1)49Figure 2.11: Frequency distributions of Tr obtained during a mid-morning (1025-1048PDT) traverse of the Downtown study area. Left-facing EIRT for: (a) northeast, (b)southwest, (c) northwest, (d) southeast facets, (e) northeast facets (low temperature(a) (c)>‘UCC)Sa)U-10 12 14 16 18 20 22 24 26 28 30Apparent Surface Temperature (°C)(b)10 12 14 16 18 20 22 24 26 28 30Apparent Surface Temperature (0 C)(d)——150—— 30°450• TSoutheast Facets (Left)Apparent Surface Temperature (0 C)Apparent Surface Temperature (° C)(e)0.500 Northeast Facets (Left)—— 15°0.4 ——30°4500.3C)0.2ItIt 1’0.1 .‘E4\/j—20 —10 0 10Apparent Surface Temperature (° C)range)Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 501.00.9 V0.8> 0.7C-)c 0.6ci)Dcrci)LL0.3V0.2 V0.10.0100 2 3456 101 2 3456 102Building Height Class (m)Figure 2.12: Cumulative frequency distribution of building heights for Downtown (traversed northeast facets only) and Industrial (north street and alley facets).standard deviation based upon measured height classes (Figure 2.12). Temperatures ofbuilding facets (Tm) and sky (Tk) are generated from a normal distribution with a specified mean and standard deviation. Modelled Tr is obtained by weighting the equivalentbuilding radiation and sky radiation by the the proportion of the projected FOV thatis occupied by building and sky components to produce an equivalent temperature. Forsimplicity, the FOV is assumed to be rectangular, rather than elliptical and the size andprojected height of the FOV is set by the distance between the EIRT and the buildingfacet, and the EIRT angle.Modelled temperatures are calculated along the block using a given traverse samplinginterval with a random starting position (between 0 and one sampling interval). Theblock-intersection sampling sequence is repeated for a given number of iterations. Two1’IIIIIII I I I Ii I I• DowntownIndustrial2 3Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 510.40.3>(-)CC)D 0.2crC)0.10.0Figure 2.13: Modelled and measured EIRT temperature distributions for north facetsduring a mid-morning traverse in the Industrial study area. Model 1 sets all buildingheights to a minimum of 3.66 m. Model 2 uses height classes as measured.sequences are used: one to represent sampling conditions on main streets, the other foralleyways (e.g., for east-west traverses in the Industrial study area). Model parametersused in the case studies are presented in Table 2.6 and the measured -and modelledtemperature distributions for shaded facets in the Industrial and Downtown study areasare shown in Figures 2.13 and 2.14 respectively. The “building temperature” is thedistribution which would be measured if each sample contained a FOV consisting of onlya building.Building height classes in the Industrial area were categorized into the number ofstories (0 - 4, in 0.5 increments) and converted into heights (assuming 3.66 m (12’) perstory). Assignment to the 0 and 0.5 classes may be deceptive, because, despite theabsence of a building in the foreground, background buildings may fill the projected2 4 6 8 10 12 14 16 18 20 22 24Apparent Surface Temperatre (° C)Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 52(a) 0.6__NEFoaets(Left0e)Modelled0.5 — —— BuiIdig Ternperoture0C24 26Apparent Surface Temperature (° C)(b) 0.6 NEFoets(Left-15°)Modelled0.5 — — — Building Tenperoture0.40.0______________________Apparent Surface Temperature (° C)(c) 0.6NE Foe do (Left — 30)— Modelled0.5— — — Building Ternperoture00110—4 2814 20 26Apparent Surface Temperature (° C)(d) 0.6 NE Fonda (Left — 45)Modelled0.5 — — — Building lereperoture0.0LkAppareOt Surface Temperature (° C)Figure 2.14: Modelled and observed EIRT temperature distributions for a mid-morningtraverse in the Downtown study area (Southwest facets only). (a) 00 EIRT, (b) 15° EIRT,(c) 30° EIRT, (d) 450 EIRT.Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 53Table 2.6: Model parameters used in EIRT distribution modelling. A - Industrial: EIRTangle 100; dif (distance from facet) 12. 5 m (streets) and 5 m (alleys). B-E Downtown:EIRT angles 0,15,30,45° dif 15 m.Parameter Description (Units) A B C D EFOV1 Field of View width (m) 3.34 3.95 4.09 4.57 5.63FOVh1 Field of View height (m) 3.40 3.95 4.24 5.30 8.04FOV1 Height of FOV centre (m) 4.20 2 6.02 10.6 17FOV2 1.35FOVh2 1.35FOV2 2.88BLKL Block Length (m) 125 83 83 83 83INTL Intersection Length (m) 20 20 20 20 20BLDGL Building Length (m) 15 21 21 20 20NBLDG Number of Buildings 8 4 4 4 4BLDGSP Building Spacing (m) 0.6 0.8 0.8 0.8 0.8NIT1 Number of Iterations 60 100 100 100 100NIT2 50T (°C) 20.5 20 20 20 20(°C) 1.0 1.5 1.0 1.0 1.0T3k (°C) 4.0 16 -10 -18 -23sky (°C) 0.5 0.5 0.5 0.5 0.5FOV. Similarly, intersections, assumed to have 0 building height, may include horizontalor vertical surface elements within the FOV such as when distant buildings are viewed orif the topography slopes upwards. In the Downtown area, building heights were obtainedfrom inclinometer measurements and put into a number of classes.North facets have been chosen for the initial modelling efforts as they exhibit a narrow unimodal distribution of building temperatures, due to the absence of direct solarradiation (Figure 2.7). Results from the Industrial area show a strong sensitivity to thespecification of the building height classes. Using the classes as measured, the modelgreatly overestimates the proportion of observations with temperatures near that ofT3k,Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 54underestimates those near T, and slightly underestimates intermediate temperatures.Assuming a minimum building height of 3.66 m results in a much better agreement between the observed and modelled distributions with the exception that there is a localpeak near 10°C. This peak may be a result of excess observations in the 3.66 m class, incombination with the surface structure.Results from the Downtown study area show a similar overestimation of the skytemperature class and underestimation of the building temperature. Again, this mayreflect some bias in the low building height classes: for DIRT angles of 30 and 45° weexpect negligible viewing of any surfaces in the intersections, yet, the overestimationin this class remains high or even increases relative to the observed distribution. Theoffset between the modelled and observed building temperatures for the 45° EIRT mayrepresent an actual temperature decrease on the upper portion of the buildings (the 45°right-facing DIRT shows good agreement with the building temperature). Note that thenumber of observations for the left-facing EIRT in this portion of the traverse is not large(54). Extending the modelling analysis to other facets, Figure 2.15 presents results forthe south, east and west facets in the Industrial area. The surface structure differs forthe east and west facets because the block length is shorter, and each block contains analleyway as well as an intersection. Up to three temperature distributions are used inthe model. Proportions for each temperature population are listed in Table 2.7. Meantemperatures, standard deviations and proportions are estimated in order to achieve a“best fit” with the observed data. The model generally performs well for the east andwest facets, despite the fact that the building height distributions were measured only onthe north and south sides of the street, therefore assuming the distribution of buildingheights by street direction is invariant. T8k is under-represented in the observed datafor the south facets. Block length, building length, and the number of buildings wereadjusted to represent the differing character of the north-south streets.Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 55Table 2.7: Model parameters for EIRT distribution modelling. Industrial area,mid-morning traverse, south, west, east facets.Parameter (Units) South West EastFOV1 (m) 3.34 3.34 3.34FOVh1 (m) 3.40 3.40 3.40FOV1 (m) 4.20 4.20 4.20FOV2 (m) L35FOVh2 (m) 1.35FOV2 (m) 2.88BLKL (m) 125 83 83INTL (m) 20 20 20BLDGL (m) 15 18 18NBLDG 8 4 4BLDGSP (m) 0.6 2.8 2.8NIT1 60 110 110NIT2 50T1 (°C) 23.5 20.3 23.2(°C) 1.25 0.55 0.25T2 (°C) 29.0 37.5w2 (°C) 1.75 2.0T3 (°C) 46.0(°C) 2.5% Bldgl 45 100 5% Bldg2 55 75% Bldg3 20(°C) 5.0 5.0 5.0sky (°C) 1.5 1.5 1.5(b) 0.60.5> 0.4C)C5)D05)U- 4 8 1216202428323640444852Apparent Surface Temperature (° C)Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 56(a)>05)D00)U-1 5 13 17 21 25 29 33 372 West Facets (Left 100)Modelled- — —— Building TemperatureL,..0 2 4 6 81012141618202224Apparent Surface Temperature (° C)East Facets (Left 100)Modelled- —— Building Temperature(c) 0.3> 0.2C)C5)Dcr0)- 0.10.0Figure 2.15: Modelled and observed 100 left-facing EIRT temperature distributions for amid-morning traverse in the Industrial study area. (a) north, (b) east, (c) west facets.Chapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 572.6 SummarySpatial sampling of the surface temperatures of vertical facets, the road surface temperature and air temperature was carried out using automobile traverses in three land-useareas of Vancouver. This is the first such study to characterize street canyon temperatureson this scale. Previous studies generally were confined to single street canyons. Traversesincluded both streets and alleyways within the study area. Results are summarized inthe following list.• Distributions of apparent surface temperature show changes with time and facetorientation; facets under direct solar irradiation show a much broader range oftemperatures than do shaded facets which are characterized by very narrow distributions. All distributions are complicated by the presence of observations fromIFOV which view the sky or mixed building and sky scenes.• There is a lag in the response of the EIRT as shown by tests against the AGEMA fastresponse thermal scanner, so that the resultant temperature is not a direct functionof the projected IFOV at that instant. This has implications in the removal ofobservations using the navigation record.• Surface emissivities are not considered on a sample basis but rather are applied tothe spatial averages using a simple model of canyon radiative exchange.• Spectral reflection by low emissivity surfaces may pose a problem in the Downtownarea as evidenced by vertical transects of apparent surface temperature made byground teams using hand-held EIRT.• A simple model of the apparent surface temperature distributions was shown to givequalitative agreement with observed distributions. More precise agreement wouldChapter 2. AUTOMOBILE TEMPERATURE TRAVERSES 58be expected with improvements in the representation of the surface structure.Chapter 3AVERAGE TEMPERATURES OF CANYON FACETSIn order to determine a representative temperature for the vertical facets of a givenorientation, it is necessary to remove from the sampled data those points for which theprojected FOV contains sky radiation. Two approaches were attempted: truncation ofthe surface temperature distribution followed by the compilation of summary statisticson the remainder of the distribution, and mixed distribution modelling with or withoutpre-truncation of the temperature distribution (statistical separation of the compositedistribution into a number of component populations).3.1 Distribution TruncationA method to remove surface temperature observations which originate from sky, or mixedsky and building, scenes is to truncate the distribution and discard all observations below the threshold. This approach assumes the combination of sky and building surfaceradiance within the instrument FOV reduces Tr below that which might be reasonablyexpected, whereupon it is discarded. A difficulty with this method is that a combinationof a small fraction of sky radiance, combined with large fraction of high surface temperature, cannot be distinguished from a surface having a slightly lower temperature. Thismeans that the resulting truncated distribution may be biased towards slightly lowertemperatures than are actually present (although model results tend to show this is nota large problem). The use of air temperature (Ta) data as the truncation point was investigated. This is based on the assumption that surfaces cannot be far below the ambient59Figure 3.1: Temporal variation of canyon floor surface temperatures in an east-westoriented street canyon compared with 0.5 m air temperature. (After Nakamura and Oke,1981).Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 60air temperature. Therefore, relationships between Tr obtained from automobile traverses,Ta and remotely-sensed surface temperatures are investigated for selected shaded facetsto determine typical minimum surface temperatures.The surface temperature is controlled by the surface energy balance as influenced bythe properties of the surface material. Directly irradiated facets are expected to showsubstantial positive differences between surface and air temperatures. For shaded facetsthe differences are likely to be much smaller, or even negative (Nakamura and Oke, 1988)due to the lag in the thermal response of building materials with high thermal inertia(Figure 3.1). Similar results were found in the scale model canyon of Voogt (1989) whereT0- Ta was —0.8 to —1.3°C soon after sunrise.I I IC)0UiHUiLiH-5040302006 12 18 24August 29 TIML (h)06 12 18 24August 30Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 613.1.1 Surface/air Temperature Relations in the Study Areas3.1.1.1 Industrial AreaAs part of a parallel study on the storage heat flux in the Industrial area, the surfacetemperatures of selected building facets were monitored for several days. These providean independent set of facet measurements which may be compared to the traverse measurements. Other independent sources of data include: apparent surface temperaturemeasurements made by ground team members equipped with hand-held IRTs, and facettemperatures from the airborne scanner. Each set is compared to the select subset oftraverse temperatures. The traverse temperatures are screened by hand with “obvious”mixed-pixel values removed, and statistics calculated on the remaining points.Figure 3.2 presents the results for an alleyway within the Industrial study area; symbols are defined in Table 3.1. Error bars are used to plot +lu.The temporal trend of Tf shows a smooth curve excepting a period following sunrisewhen some direct solar radiation is received by the wall resulting in a sharp rise in thesurface temperature followed by a leveling off for about two hours. for the same blockis warmer (by 1-2 °C) for most of the day, but the standard deviation (plotted as errorbars) indicates that the traverse sample contains a reasonable number of observationswith temperatures similar to that of the building wall. Twa is slightly lower thanand has a larger standard deviation. Images show coolest temperatures near the base ofthe shaded walls which extends onto the ground surface so that including these in thesample may lower the mean. The greater variation may be due to the inclusion of a fewpixels from warm roofs, or smearing of roof pixels into upper wall areas.Minimum surface temperatures extracted from the airborne imagery are substantiallybelow both air temperatures and Tf and except for the first flight, are lower than Twgmjn(although the latter is a very small sample). Twgmin correlate reasonably well with thoseChapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 62Table 3.1: Description of symbols used in Figures 3.2, 3.3 and 3.4.Symbol Descriptionw facet (wall)f fixed observing sitea air (first subscript), airborne platform (second subscript)t traverseg ground teamdenotes a spatial averageTf apparent building facet surface temperature obtained from the fixed monitoring sitemean apparent building facet surface temperature from the traverse vehiclefor the same block (screened as described above);Tat the modal air temperature class (0.5 °C width) obtained from the vehicletraverses done in the study area (approx. 2 m level)Taf air temperature at 9 m above the groundTwgmim minimum recorded apparent surface temperature from north-facing wallsobtained from samples taken by ground teamsT9 mean apparent surface temperature from north-facing walls taken by groundteams1’wa mean apparent facet temperature of selected buildings taken from off-nadirairborne imageryTwamin minimum recorded apparent facet temperature of selected buildings obtained from the off-nadir airborne imageryChapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 632624C)0—‘ 22Figure 3.2: North-facing alley facet temperatures and air temperatures in the Industrialstudy area.of the fixed monitoring site in the morning, but show more scatter later in the day. Themeans of these points generally fall within +1 of i’wt. Tat is slightly less than foralmost all traverses but is higher (by 1-2°C) than 1’wt at all times, except for the firstand last traverses when the two values are in close agreement. The 9 m air temperature(Taf) agrees well with Tat up until approximately 1030 PDT, when Taf shows a sharpdrop, possibly associated with the onset of a local sea breeze. Following this drop, Tafremains close to the north-facing building surface temperature until approximately 1600when it becomes warmer for approximately 2 hours. The difference between the Taf andTat points to a possible decoupling of the canyon level (microscale) air volume from theUBL (mesoscale) air.By itself, Tf provides an independent measure to truncate the distribution, under6 8 10 12 14 16 18 20Time (Hours, PDT)Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 64the assumption that it is representative of the coolest surfaces available. For any otherfacet orientation, this limit also holds (i.e., for shaded portions of directly irradiatedfacets). However, the airborne and hand-sampled data appear to indicate significantlycooler surfaces do exist in the study area. The relation between Tat and Tf indicatesthat Tat is generally warmer requiring the specification of an offset of 1-2°C for mosttimes in order to utilize Tat as a truncation measure for this facet. For directly irradiatedfacets, the specification of an offset is not as critical because the bulk of the surfacetemperature distribution lies well above air temperature.The use of minimum temperature as a truncation point may be suitable for the 0°EIRT as they rarely view the sky (except at intersections), so the statistics are relativelystable. However, for the 10° EIRT, this cutoff may include significant numbers of mixedsky-building pixels. Using Tat with an additional (negative) offset (denoted by DTA (°C))the effect on the statistics of the traverse record was investigated as DTA is changedfrom 0 to 5. The results indicate that for the 0° EIRT, DTA greater than 2 incorporatemost of the data set and statistics tend to stabilize. For DTA less than 2, increasesexponentially; differences in between DTA = 0 and DTA = 2 are not large in absoluteterms (generally less than 1°C). The 100 EIRT tend to show a more linear decrease of‘tat as DTA increases, absolute differences are again less than 1°C. There are differencesbetween the EIRT with angle; for directly irradiated walls, the 0° EIRT tend to be coolerthan the 10° EIRT, possibly due to increased shading by awnings, vegetation lower onwalls, and decreased solar access nearer the ground. For north-facing facets, traversesduring the day show good correspondence between 0 and 100 EIRT for DTA 2. At thistime of day little or no temperature variation with height on this wall is anticipated sothis may be a satisfactory truncation point. Earlier and later in the day, when this wallmay be exposed to short periods of direct solar irradiance, the upper EIRT may showpreferentially warmer values. Nevertheless, based upon the above considerations, Tat — 2Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 65(-)0L)0ci)cii6ci)F—is used as a truncation limit for all facets, at all times, for this traverse. DowntownNo fixed monitoring of building surfaces was conducted for the other study areas socomparisons are between traverse, hand-sampled, and airborne data. The Downtowntraverse data is for two blocks of northeast-facing facets (Figure 3.3). The plot shows 1’atto be slightly below the average (except in the late afternoon, early evening period).But the ±lo- of surface temperature often falls below the ±la (Tat) again indicating thatsome surfaces fall below air temperature.lwtlotTwy mmIwgIomr (YVR)TwoT0 mm I262524232221 —e--f201918171615I I I I I I5 7 9 11 13 15 17 19 21Time (hours, PDT)Figure 3.3: Subset of northeast-facing facet and air temperatures in the Downtown studyarea.Airport air temperature (YVR) compares well with that measured in the study areaexcept for the morning and late evening traverses. These may be associated with thex+-I-Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 66nocturnal UHI. Hand-sampled values for this area show minima which tend to be closeto T lu and means generally warmer than + lu. The number of points involvedis small, but the relatively warm average value may point to the presence of surfaceswith low f (such as glass) affecting Facet temperatures derived from the airborneimagery tend to be substantially higher than those obtained from the traverse subset.This may be due to a greater proportion of pixels originating from the upper part of wallsin the imagery, compared to the traverse subset which uses temperatures from near thebase of walls. Specular reflection from low emissivity surfaces may also contribute to ahigher mean for the airborne imagery. The source of the reflected radiance is the surfaceor opposite canyon wall, which in this case is quite warm. For the Downtown area, thetruncation measure used is the minimum of either 1wt — Li or Tat — lu for each traverse. ResidentialThe results from the Residential area (north-facing side of 50th Aye) (Figure 3.4) consistently show Tat warmer than at all times, with differences of up to 3°C in thelate afternoon. Air temperature measured at the 10 m level of the Sunset Tower (îwa) ismuch lower than Tat and is lower than the until late afternoon, when it approachesand then exceedsMean airborne facet temperatures agree well with the traverse values for the morningand afternoon flights; the early afternoon flight shows means slightly lower thanStandard deviations are much larger, but are probably related to difficulties associatedwith extracting wall temperatures from the image data. Hand-sampled minimum temperatures show significant scatter and are, in several cases, substantially below the traversesurface temperatures, but are in approximate agreement with Twamin. This finding isnot surprising when the FOV of the traverse EIRT is considered. The sensor to facetdistance is large in the Residential area because of the small canyon H/W, which meansChapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 67312927(-)02321EcDH- 191715Figure 3.4: Subset of north-facing facet and air temperatures in the Residential studyarea.the projected FOV is large relative to the size of the buildings. This is likely to give abetter spatial average. The hand-sampled observations involve a narrow FOV radiometerand are taken mUch closer to the target facet. They are therefore more likely to registeranomalously warm or cool spots. The truncation point for this site is specified as-2a. Because of the small H/W and building set-back from the road, 100 EIRT rarelyview walls alone. The distributions contain most of the observations that are well belowair temperature, probably as a result of viewing mixed vegetation and sky. They are notconsidered in the determination of averages for this area.6 8 10 12 14 16Time (hours, PDT)20Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 683.1.2 ResultsUsing the distribution truncation limits defined previously, the traverse data are averagedby facet orientation for each portion of the traverse route as defined by the navigation file(a block or sequence of blocks along a street) and also for the entire traverse. Observationsmade within an intersection, and the first observation following an intersection, werediscarded, regardless of whether they met the truncation criteria. The overall facet meanfor this area is obtained by combining the 00 and 100 EIRT. In the figures which followthe solid symbols are mean temperatures with the emissivity correction applied. Industrial Study AreaFigure 3.5 presents the mean facet temperatures for east-west and north-south facet pairs,and the difference between the corrected mean temperatures. The east-west facet pairsshow two periods with large temperature differences between the means. The afternoondifferences are slightly smaller, due to the overall warming of the non-directly irradiatedfacets. The north-south facet pair show a single peak with differences maximized near1400 PDT, which lags solar noon (1317 PDT) by approximately 45 minutes. Standarddeviations of the shaded facets are much smaller than those from the irradiated facets.The latter typically show standard deviations of 4-5° C. Shaded facets tend to closelyfollow the trend of air temperature, except for the west facets in the afternoon, whichare slightly warmer, probably due to the release of heat stored when they were sunlit inthe morning.Averages for each street in the traverse (3-4 blocks for north-south facets, 5 blocks foreast-west pairs) are shown in Figure 3.6. Again variation between blocks is highest whendirectly irradiated. Three north facets are cooler than the remainder; two of these arealleys and the other is a street (8th Aye). It is suspected that the cooler temperaturesC)0a)Da,a)0Ea)(c)C)0a)a,C5)5)05)0‘Ua,E5)I-Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 69(a)I E I I I I20—e—— West Facets (open — T,, solid — T0)1 5 —s-—— East Focets (open — T,, solid — T)6 8 10 12 14 16 18 20Time (PDT)(b)c South Facets (open— Tr, solid — T0)1 5 —&--— North Facets (open — T, solid — T)—— T6 8 10 12 14 16 18 20Time (PDT)1050—5—.—— West— East—10—-a-- South — North6 8 10 12 14 16 18 20Time (PDT)Figure 3.5: Mean facet temperatures (apparent and corrected) and standard deviationsfollowing distribution truncation in the Industrial study area for (a) east and west facets,(b) north and south facets and (c) mean temperature differences.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 70are due to these alleys being narrower than the others, and the street may be somewhatcooler due to the presence of a large park for one block so that the recorded temperaturesmay be that of distant buildings and some ground surface. DowntownResults for the Downtown site differ from the Industrial area because of the differentstreet orientation (Figures 3.7 and 3.8).The northeast-southwest facet pairing (the 15° EIRT data are plotted) shows onlysmall differences in the early morning followed by a reversal and the creation of largedifferences in the mid- to late afternoon. Northwest-southeast facet pairs show only alarge late-morning to noon peak. Shaded facets are slightly warmer than the air temperature for the most part, except late in the day. Standard deviations are large fordirectly irradiated facets and overlap those of the shaded facets, probably due to mixedshade-sunlit facets in the deep canyons.There is a large variation in the mean block to block sequence temperatures especiallyfor the irradiated facets (Figure 3.8). This is attributed to the relatively large range oflocal canyon H/W ratios and possibly to € variations. Figure 3.9 embellishes Figure 3.8by including each of the EIRT angles used. In general, the warmest facet temperaturesare obtained with the 30 or 45° EIRT when viewing an irradiated facet. Again, thisis probably due to shading in the lower parts of the street canyons because of canyongeometry and the presence of awnings and building overhangs at lower levels. Whenfacets are shaded, there is some tendency for the larger angle EIRT to view a slightlylower temperature; this may be a function of specular reflections from the sky or a truedecrease in temperature with height.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 71(c)liii0335 000 Oo§ 0U)- 31 0 8:36’ 029 027 02523 021oci 0 West Facets17 EastFocets15________________________________I I I I I6 8 10 12 14 16 18 20Time (PDT)(b) 37 I I I’III•o 35 000— 33q)31°8 029 () 000 027E 250ciU)23CU) 210 19ci10 South FacetsNorth Facets15I I I I I6 8 10 12 14 16 18 20Time (PDT)Figure 3.6: Mean facet temperatures of each street in the Industrial study area; (a) eastand west facets, (b) north and south facets.U0a)D0a,0E0)(C)U0a)0Ca)a)a)aa)Ea)I-Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS(a)72—a)— SW Facets (open — T,, solid — T,—&-— NE Facets (open — T,, solid — TU0a)0a)0Ea)(b)353331292725232119171535333129272523211917157654320—1—2—36 8 10 12 14 16 18 20Time (PDT)i :i— ——a)-— SE Focets (open— Tr, solid — T0)—o—— NW Facets (open — T, solid — T,)— Ta,3 8 10 12 14 16 18 20Time (PDT),—a--’I //A / -I - -/Nj/—.— SW-NE--A-- SE-NW6 8 10 12 14 16 18 20Time (PDT)Figure 3.7: Mean facet temperatures (apparent and corrected) and standard deviationsfollowing distribution truncation in the Downtown study area for (a) northeast and southwest facets, (b) northwest and southeast facets and (c) mean temperature differences.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 73(a) 33310029 0 0ci) 00 0D-- 27 0 0o 0 0c) 25 08o0 08,23 00F—° c_21,1900 Southwest Facets1 7 Northeast Facets15_______________________________6 8 10 12 14 16 18 20 22Time (PDT)(b) 33o 31 ° °0 029ci)27 0 0025 000 008 Oo cP1— 08 23 021 0191 70 Southeast Facetst Northwest Facets15 I I I I6 8 10 12 14 16 18 20 22Time (PDT)Figure 3.8: Mean facet temperatures of blocks or block sequences in the Downtown studyarea; (a) northeast and southwest facets, (b) northwest and southeast facets.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 74Figure 3.9: Mean facet temperatures of blocks or block sequences in the Downtown studyarea for each EIRT. (a) northeast, (b) northwest, (c) southeast, (d) southwest facets.(a)Southwest Facets oo 00 EIRTo 15°EIRTO 300 EIRT* 45° EIRT*0000)aa)aE0)I—Ca)aaab0 Southeast Facets+ 0o + 0 0°EIRT150 EIRTO g a o 30°EIRTO + 45°EIRTOQf* 00i 0 648 10 12 14 16Time (PDT)(b)35 —333129272523211917156(d)18 20 223533s— 31a)D 2927a)aEF! 232119a< 17152524a, 23- 222120- 196 8 10 12 14 16 H8 20Time (PDT)(c)8 10 12 14 16 18 20Time (PDT)22*6Northwest Facets: 0E1ETo 30° EIRT+ 450 EIRT0 °Northeast Facets252423- 222120‘ :*+0 0° EIRTo 15° EIRTO 30° EIRT* 45° EIRT22 6 8 10 12 14 16 18Time (PDT)20 22Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 753.1.2.3 Residential AreaResults from the Residential area follow the general pattern described for the Industrialarea, but show somewhat smaller differences, and greater variation. Overall mean differences (Figure 3.lOc) are smaller, especially for the north-south facet pairing, and for theeast-west pairing in the afternoon. Mean facet temperatures (Figure 3.11) show somedependency upon the orientation of the EIRT relative to the facet: south facets viewedby a left-facing EIRT and north facets viewed by the right-facing EIRT, each have awarm bias. For the former case, the EIRT may be viewing some low-sloping roofs, andin the latter, the shorter facet to sensor distance increases the likelihood of a warmertemperature.3.2 Mixed Distribution ModellingMixed distribution modelling allows component populations to be extracted from a composite dis&ibution. It has a close analog in the separation of cloudy, or partly cloudy,pixels from clear pixels in remote observations of sea surface temperature (SST). Figure 3.12 illustrates typical temperature histograms from a clear, and a partially cloudcontaminated, thermal image.Several methods are available to estimate the temperature of the sea surface fromthe mixed distribution (Crosby and Glasser, 1978; Smith, 1985). All methods generally assume that the sea surface temperature is only slowly varying in space, that itcan be represented by a Normal probability density function, and that the presence ofclouds within the sensor FOV will lower the measured radiance. The techniques attemptto determine the statistics of the SST using either the maximum slope of the Normalprobability density function (using a system of equations to solve for the statistical parameters), or by fitting (by least-squares) a Normal probability function to points at theC-)05):30ci)0E5)F—(c)C)0a)C)Ca)a)ccci)0ci)0Eci)I-Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 76() 403530252015——e--— West Facets (open — T,, solid — T0)—s—— East Focets (open — T,, solid — T)——. Tt8 91011121314151617181920Time (PDT)(b) 4035C-)0a) 30:3025Eci)‘ 2015—9--— South Facets (open — T, solid — T)—&--— North Facets (open — T, solid — T)—— T8 9 1011121314151617181920Time (PDT)8 91011121314151617181920Time (PDT)Figure 3.10: Mean facet temperatures (apparent and corrected) and standard deviationsfollowing distribution truncation in the Residential study area for (a) east and west facets,(b) north and south facets and (c) mean temperature differences.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 77(ci) 40 I I I00000— 35 084—4-, A2300225-4-,C4--200 West FacetsEast Facets15 I • I I8 10 12 14 16 18 20Time (PDT)(b) 40C)0‘— 35ci):30230 000 00ci)ci o225-4-,CU)4--0 South Facets20North Facetsci15 I I I I I8 10 12 14 16 18 20Time (PDT)Figure 3.11: Mean facet temperatures of each street in the Residential study area; (a)east and west facets, (b) north and south facets.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS>-C-)zwCuJU-78Figure 3.12: Temperature distribution from a satellite image characterized by: (a) clearsky, (b) cloud contaminated. From (Smith, 1985).end of the observed frequency distribution.The case of the vehicle traverse observations differs from the cloud removal case inthat the underlying surface temperature distribution shows much greater spatial variation than does the SST. For shaded facets, the range of surface temperatures is quitenarrow, but the surface temperature of directly irradiated facets is strongly conditionedby the thermal and radiative properties of the materials and any local shading. Becausecanyon facets are generally distinctly inhomogeneous the resultant temperature distribution becomes very broad. In the limit, where different surface types are separated,and repeated temperature measurements made of each type, it is anticipated that eachsurface type would be characterized by a Normal distribution of temperatures. If, in thecanyons traversed, the facets are composed of only a few major classes of surface types,it may be possible to represent the observed temperature frequency distributions (or atleast the upper portion away from the mixed sky-building part of the distribution) with(a) CLEAR ATMOSPHEREAND NO SEA—SURFACETEMPERATURE GRADIENT(b) CLOUD CONTAMINATED SAMPLE302520151050285 290 295 300 305 260 265 270 275 280 285 290 295 300 305TEMPERATURE (K)Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 79a combination of normal probability curves. This method is referred to here as mixeddistribution modelling. The technique is not new. Harding (1948) describes a graphicalmethod for separating a number of mixed populations with examples drawn from animal science and Sinclair (1976) describes geochemical applications. Presently, computerapplications of the technique are available which allow selected probability distributionsto be fit to observed data in order to represent the overall distribution. The successfulapplication of mixed distribution modelling allows decomposition of a multi-modal datadistribution into its component populations, and to define thresholds which separate thedata into groups corresponding to the component populations. It assumes there is anunderlying physical model for the process generating the data. The goal is to identifythe values of the parameters in the model equation because they have physical meaning.The prime requirement is that the model equation be correct, otherwise the values of theparameters will not be physically meaningful.Here, mixed distribution modelling is applied to the temperature frequency distributions obtained from the vehicle traverses to recover the component populations represented in the temperature distributions, and in some cases to remove mixed sky/buildingpixels from the analysis. For the Downtown study area, in which traverse configuration1 was used, the distributions of the 15, 30 and 45° EIRT were pre-truncated to removethe obvious sky and mixed sky pixels prior to implementing the analysis. The mixeddistribution modelling procedure used was the computer program PROBPLOT (Stanley,1987). The number of classes was selected in order to achieve a good visual fit. The fitwas then optimized using a maximum likelihood procedure.3.2.1 Road Surface TemperatureRoad surface temperatures (Troad) offer the simplest application of mixed distributionmodelling to the observed temperature frequency distributions. Assuming that the roadChapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 80surface material is reasonably constant over the traverse route, the main control uponthe measured surface temperature is the presence or absence of shading. However, thereare probably some variations in road surface properties due to asphalt of differing ageswithin the study area, and variations in the local slope of the road surface. In addition,the length of time the road surface has been shaded results in further variations in surfacetemperature. Figure 3.13 presents a road surface temperature distribution from a morning traverse in the Industrial area. The distribution is well represented by three Normalcurves. The lowest temperature distribution has a fairly narrow standard deviation, andis thought to represent those sections of the road surface not yet heated by direct solarirradiance. The uppermost distribution comes from the road surface which has undergone direct heating, and the middle distribution may represent surfaces in transition. Inthis example, the curves were fit using the PEAKFITTM non-linear curve fitting programwhich implements the Marquardt-Levenberg algorithm for minimizing the sum of thesquared deviations. This program provides significant advantages over the PROBPLOTsoftware in terms of the types of distributions which may be used and the control overthe fitting procedure.Results from the mixed distribution modelling of the road surface in the three studyareas are presented in Figures 3.14, 3.15, and 3.16. Results are separated by streetorientation. The figures give the number of components, and the mean and standarddeviation of each of the component populations for each traverse. The position of themean is indicated by the plotted number which represents the percentage of the totalpopulation made up by that component.In general, the distributions show fewer component populations early in the day with astronger distinction between the means. This is probably due to the strong temperaturecontrast between shaded and irradiated portions of the road surface in the morning.Later in the day, when previously shaded portions have been warmed, and previously10 15 20Appa rent RoadChapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 81N/s Streets 0950 PDTElGaussian Curves(a)(b)(c)SumEl Observed25 30 35Surface Temperature (° C)405040100Figure 3.13: Distribution of road surface temperatures in the Industrial study area froma morning traverse showing bimodal distribution characterized by shaded and sunlit roadportions.irradiated portions are shaded (especially evident for N/S street orientations and theDowntown street orientations), the distinction between the means is less. Often there arethree or even four component populations. These generally constitute a relatively smallpercentage of the total data and are thought to be due to areas undergoing transitionfrom shaded to irradiated, or in some cases small areas of extreme temperature, perhapsdue to variations in road surface material (i.e., new “blacker” asphalt).The Industrial area shows a high proportion of the E/W streets in the uppermosttemperature class. This is because the small canyon H/W means few shadows are castonto the road, except early and late in the day. The N/S streets show a higher percentageof shaded temperatures early in the day, but shift to almost entirely high temperaturesChapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 82(a) 60 II nd ustra IO N/S Streets50 110215- 40 17 ii9 718 11 8j0 12E 1 T2.2911 1 1I’T618 iIohI3 110 919121:20 ‘17 3125191 0191178310 i I I I I8 10 12 14 16 18 20Time (PDT)(b) 60Industriolo E/W Streets317 45.11 1- 110397j TE 40 718 I ±8108j3 1112E 812 215 L613Q 315 1 8122092 115 115 112 1o 11918 112 110 I’ 191116 O1I41 2 14 1 6 1 8 20Time (PDT)Figure 3.14: Road surface temperature statistics for the component populations derivedfrom mixed distribution modelling for the Industrial area: (a) N/S streets, (b) E/Wstreets. The number of bars at each time indicates the number of component populations,the plotted numbers are percentages of the total population and are located at the meanof each component. Error bars are ±lu.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 83(a) 50’ I I I I IDowntown0 Nw/SE Streets 11040D410310 3TI 1356Io(1) 9 2T4i814110916 7T4ci)213613 T -‘- 512916 1- 6. .0 -‘ 217810 611i20 115 1160915 910 016010: I I I8 10 12 14 16 18 20Time (PDT)(b) 50 I IDowntown0 NE/Sw Streets0216 130I.ci340) 1615E 214 1 17 113ci0) 116_816 I 817 212 41881714 87C20 414 72 26 1±70o 916 816 819ci 9ci110110 01610 I I I I8 10 12 14 16 18 20Time (PDT)Figure 3.15: Road surface temperature statistics derived from mixed distribution modelling for the Downtown study area: (a) NE/SW streets, (b) NW/SE streets.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 84(a) 5o ‘ I I I IResidentaI 75o N/S Streets 8166o611110 710 116540- 1 2.0 T 2d&0610118 2Ii.5j143oE 610) 17103561 91259319 210112 IU) 20 13 Lo 76cL21710__________________________________8 10 12 14 16 18 20Time (PDT)(b) 50 I I I IResidential 1 656j00 E/WStreets65 Io) 4Q 15 1 10813±T T 418o6 3120) 816ilo 1i71Ii 47° 30 711 T-10 216411E 1818 112114U)C20 81 -o iI810 I I -8 10 12 14 16 18 20Time (PDT)Figure 3.16: Road surface temperature statistics derived from mixed distribution modelling for the Residential study area: (a) N/S streets, (b) E/W streets.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 85toward mid-morning in accord with increased solar access to the canyon floor. TheDowntown study area, with its deep street canyons shows a generally higher percentageof the low temperature distribution except for a short period when the solar azimuth isaligned with the canyon axis and permits direct solar access to the canyon floor. In theResidential area, the N/S streets show a decrease of the percentage contained in the lowesttemperature class through the morning and an increase in its standard deviation. At thesame time there is an increase in the percentage contained in the highest temperatureclass. From early afternoon, the warmest temperature class dominates but falls off later.The means of the lowest temperature class remain approximately constant but theirpercentage increases, and the distinction between the class means decreases with time.The E/W steets show a relatively small shaded fraction in the morning, with someincrease in the afternoon. A survey of the study area indicates much greater shading ofthe road surfaces by mature trees on the N/S streets compared with the E/W streetsso direct comparison of these street orientations is limited by the fact that the shadingregimes differ.3.2.2 Wall Surface TemperaturesWall (canyon facet) surface temperatures present a more complicated application forthe mixed distribution modelling technique. Here it is assumed that there exists atleast one, and in some cases multiple populations of wall surface temperature, togetherwith sky, and mixed sky and surface, temperature observations. The case of directlyirradiated facets poses a particular challenge. The long tail extending towards highersurface temperatures probably represents multiple surface types with high temperatures,and is also probably a function of the FOV of the instrument since that incorporates someaveraging of the surface temperatures. In order to better separate these populations,truncation of the distribution at the lower end can be utilized in cases where sky andChapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 860.200.1 80.1 60.1 4>0 0.12000.08U- 3.17: Morning east-facing facet temperature distribution with component populations from the Industrial study area.mixed FOV temperatures are obvious.Figures 3.17, 3.18 and 3.19 present results for three traverses which view facets experiencing direct solar heating. Figure 3.17 is for a west (east-facing) facet ii the Industrialarea taken in the early morning (0845 PDT). Three main components represent the majority of the distribution. A fourth, small, high temperature distribution is also added,but it is not well represented in the measured distribution. Figure 3.18 is a west-facingwall taken during the late afternoon (1.730 PDT) and again is well represented by threecomponent distributions.Figure 3.19 illustrates a particular case in which the number of individual distributionsbecomes large, and/or there is significant overlap between distributions, possibly createddue to averaging effects within the sensor FOV. The distribution may be fit by a finite20 30 40 50Apparent Facet Surface Temperature (° C)Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 870.200.1 80.1 60.1 4>C-) 0.12CcJ-) 3.18: Late-afternoon, west-facing facet temperature distribution and componentpopulations from the Industrial study area.number of Normal distributions, in this case 4, or, alternatively the overall distributioncan be represented fairly well by a single exponentially-modified Gaussian. The use ofexponentially-modified Gaussian may thus be more appropriate when the sensor IFOVis large with respect to the areas of component temperatures.The simple model for estimating EIRT temperatures (Section 2.5.3) was used todetermine if the mixed distribution modelling could accurately recover the model temperatures. Table 3.2 compares the model input and mixed distribution results for a 100EIRT viewing each facet in the Industrial area during a mid-morning traverse. As a testof the procedure, the simple EIRT response model was used to generate some model temperature distributions. Input was provided by several normal distributions of selectedmean and standard deviation to represent different surface temperature classes. TheIndustrial AreaWest Facets (00 EIRT)745PDT I-//I :.\Gaussian Curves(a)(b)(c)SumObserveda20 30 40 50 60Apparent Facet Surface Temperature (° C)Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 880.40.3>C-)C(1) 0.20Q)L0.10.0Figure 3.19: Morning southeast-facing facet temperature distribution from the Downtown study area showing representation by multiple normal distributions and a singleexponential Gaussian distribution.output was submitted to the PROBPLOT program to see if the original input statisticscould be recovered in the presence of the mixed sky and building pixels. The results(Table 3.2) show that the procedure can recover fairly closely the original temperaturedistribution statistics, but that the number of classes needed to achieve a good visualfit is often higher than the number of classes specified. This is probably due to themixed FOV temperatures. Standard deviations are generally larger than those specified,although truncating the distribution results in a better agreement between the recoveredand specified values.Results from the three study areas are presented in Figures 3.20, 3.21 and 3.22. Pretruncation was performed for the Downtown study area only. In the Industrial area, north15 20 25 30 35Apparent Facet Surface Temperature (° C)Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 890E200.1o29272505 23211917155CC-),o 403020IC50C,40302040309 11 13 15 17 19Time (POT)7 9 11 13 15Time (POT)9 11 13 15Time (POT)Ii 13 15 17Time (POT)17 1917 19 219 2140C)30E 201003002520E 15C 100.)C0E00.409 11 13 15 17 19Time (POT)7 9 11 13 15 17 19Time (POT)6CEast 1 0° EIRT50403020 °i’ .{s0010 0 + J. + +7 9 11 13 15 17 19 2Time (POT)500030E— 20109 Il 13 15 17Time (POT)19 21Figure 3.20: Facet surface temperature statistics derived from mixed distribution modelling for the Industrial study area separated by facet and EIRT angle. No pre-truncationof the distribution was carried out.North 0° EIRT2 +‘+ 1: {+ ::North 1 0° EIRT: } ‘1J’ ‘{• + + +:: + +012 H + + + + i.South 0° EIRT0}2{5 :.:South 1 o° EIRTi 51. ‘1° 1° ‘i °i’51 51 5]0 °io‘i 14 110 151 I° 010 ‘ 11’ ‘i oj +51’East 0° EIRT1 i’+410 +West 0° EIRT°+++f’rWest 1 0° EIRT‘°+1 51° s{s ++ 514 + °1’ ‘1° +Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 90(3S36 SEO EIRT24+6.1611 101616 L18 154 11616148 10 12 14 16 18 20Time (POT)3D330315 29272523t 211917150SF0a0aS06 8 10 12 14 16 18 20Time (POT)45SWO° 51ST635 I -30 10 1121161+ ;25 ‘I .1s2 1620 r4o1 210+11+ 611122Ps 2fOoejs2106 8 10 12 4 16 18 20Time (POT)282624222019161412109 19 12 14Time (POT)16 18 2054(101 44S 30241433Z? 312927O 252321— 195 170a is< 130S0SE 15° 51ST116111Iil}lt+++H12}4o6l1 o{z°i’{‘ II,436 9 19 13 14 16 18 20Time (POT)NW15° 51ST JIWIll ii ‘110’F49p216214o5]:6 9 19 12 14 16 18 20Time (POT)seSW 1 50 51ST2}II1,+4114+2. ?Time (POT)403429143432305 293 26242261 24185 16a 141210NEI5°EIRT1’11 I1 16 1°II’4o12 ‘ii 210 t ‘F ++‘ 14 4 a I’ ‘1-’9 10 12 14Time (POT)16 18 20Figure 3.21: Facet surface temperature statistics derived from mixed distribution modelling for the Downtown study area separated by facet and EIRT angle. DistributionsNWO° 51STJ2110 +110010 1:F :}0 :1: ‘1o1j2h10NE 0° EIRT}l}illlo 4+ ip i11o+t1+ 211010were pre-truncated.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 9140302010SE 30° EIRT5]4lb J. 010 °1’ oFoleofo +°I 11° 2i ?t:0 oto0)40a 30E.0 200a10SE 45° EIRT°t :r++ j4hilTs +1°j 010++ “° ?8 10 12 14 16Time (POT)18 200)aE00aEC—0a0SC—0.00SI0.0NW 450 EIRTTTT of19T0T0ioJo ;0i412 ol E8 10 12 14 16 18 20Time (POT)NW 30° EIRT32 0302826 o[o24 F: i:’i° oioioj.o 41o of.% oI:t:14 0161 oF12 fe108 10 12 14 18 18 20Time (POT)41.SW 30° EIRT+f 1ofoI 513j6:: +i11004o o35° 55‘18 10 12 14 16 18 20Time (POT)NE 30° EIRT32 0N toit910+ °b4°++°1 64353025201510310° 285 2725a 2321185 17a 15< 13II500o 40aS5 200a.010501)a 30SI—5 200.0108 10 12 14 16 18 20Time (POT)SW 450 [IRTon°‘:‘p,,°ii 010!° ‘b fsill i5o5 8 10 12 14 16 18 20Time (POT)NE 45° SIRT510ofo1’+715 n .7)4: °“°!of; ijo CD 013 ‘6 8 10 12 14 16Time (POT)18 20Figure 3.21 (Continued). Mixed distribution statistics8 8 10 12 14 16Time (POT)18 20for the Downtown study area: 30 and 450 EIRT.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 9250136a 20•040(3352 306•2 252015(-36a 20I-)2 402230a 20in40308 9 10 Ii 12 13 14 15 16 17 18 19 20Time (POT)North 0° LINTIi :: sij:1: ;It8 9 1011 121314151617181920Time (POT)5040308 9 1011 121314151617181920Time (PDT)605021J4qs8 9 10 11 12 13 14 15 16 17 18 19Time (POT)000402 3062 20100300e 252022 15104003201—2 10050030a.6202 10•0030408 9 1011121314151617181920Time (POT)8 9 1011 121314151617181920Time (POT)8 9 1011121314151617181920Time (POT)8 9 1011 121314151617181920Time (POT)Figure 3.22:modelling forFacet surface temperaturethe Residential study areastatistics derived from mixed distributionseparated by facet and EIRT angle. NoSouth 0° LiNTuiT TTlii 55 317 212201 ±3.* 1° :1:i4 343 4u6& +4 3FhI2 fsSouth 100 LIRTJt613i 8jii4&1f3 } ‘1 { :1: :3{848 2i6t6 1112149‘47141pf3214fNorth ISO LiNT 115217240o4oIf4IN14024534140 1 340 3182471 241 3J &JT6 I:6 I24i2I 1 0151 N3f3ijs 3112 ifs 217 tju if 113110 1 ifu fuWest 0° LiNT TTilt u:J;u ‘2?tb 41iij45J sfu110 tls sb 27ofu [944West 10° LINT of.fuI6313 of217 7 e1+ 415tu. tt1°° }ff ::2fLust 0° LiNT20Lust 10° LINTofJOfifiIs‘Is4o A14. 3164225+ sft je f’ ;3fjtf62f5Jof isis ifspre-truncation of the distribution was carried out.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 93and south facet temperature distributions are generally characterized by two populations,while the east and west facets show one or two narrow distributions during the time theyare shaded and an increase in the number and standard deviation of the components whenirradiated. Interestingly, the north facets (south-facing), which are directly irradiated formuch of the day, do not show as great a variation or number of component populationsas do the east and west facets when they are directly irradiated. This may be due to thereduced influence of direct-beam radiation due to the small local zenith angle at thesetimes, or perhaps to greater use of shade devices which reduce the area directly irradiated.The 100 EIRT differs from the 00 in that there is an obvious separation of the distributioninto facet and non-facet temperatures; with the lower temperature distribution assumedto be representative of sky or mixed sky/building temperatures. The percentage of thispopulation varies with facet orientation; it is approximately 20-30% for north and southfacets and 35-50% for east and west facets. This difference is ascribed to a greaterbuilding set-back and the absence of alleyways in the traverses observing east and westfacets. There also appears to be a tendency for the facet temperatures to be representedby a single component with a fairly large standard deviation; this may be a function ofthe resolution of the mixed distribution program during the initial fitting of componentpopulations. Pre-truncation of the 100 EIRT may allow further separation of the facettemperatures into further components.In the Downtown traverse (Figure 3.21) facets under direct irradiance show three orfour well separated component populations. Shaded facets sometimes have a small warmcomponent the origin of which is uncertain. In some cases it may be due to reflectedradiation or in others to temporary obstruction of the instrument FOV by a passingtall vehicle, which was known to occur. Small, low temperature components may be anartifact of the truncation procedure, which used a relatively conservative truncation pointto minimize the loss of any real facet temperature data. Results from the Residential areaChapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 94Table 3.2: Temperature distributions assigned to a simple model of EIRT response (tm,am), and results from mixed distribution modelling on the EIRT model output using theentire temperature distribution (7, o, ) and a truncated distributionFacet Assigned Mixed Distribution Modelling- Entire Set Truncated SetT a T TWEST 5.0 1.5 5.4 1.7 28.9 1.318.3 6.5 36.4 2.637.5 2.0 36.4 3.2 45.8 2.246.0 2.5 46.2 1.6EAST 5.0 1.5 6.5 2.2 16.6 0.313.2 2.7 17.8 0.320.2 0.7 19.9 0.9 19.9 0.8NORTH 5.0 1.5 5.7 2.0 18.5 1.213.5 2.8 23.3 1.523.5 1.25 23.1 2.0 29.2 1.729.0 1.75 29.4 1.6SOUTH 4.0 0.5 4.1 0.6 10.8 0.711.2 2.9 14.3 1.520.5 1.0 20.1 1.4 20.2 1.2(Figure 3.22) show a tendency for a greater number of components required to representthe total distribution. The 100 EIRT shows a very small percentage of the observationsin component populations whose temperature is above air temperature, indicating thisinstrument provides little information about canyon facets. The behaviour of the eastand west facets, as they change from shaded to sunlit, is similar to that described for theIndustrial area.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 953.2.3 A Single Model for Canyon FacetsAn alternative to extracting model components is to model the complete distributionunder the assumption that the small scale heterogeneity is too great to allow adequateseparation of individual components. Such an approach has been adopted by Holbo andLuvall (1988) in which Beta probability distributions were used to model surface temperature distributions obtained from an airborne thermal scanner over a forested region.Their observed distributions often showed non-Normal shapes which were attributed tosmall scale variations in the surface type and orientation; e.g., openings in the forestcanopy which reveal isolated very warm soil or rock surfaces of differing orientations.Surfaces, if sufficiently categorized, would be represented by Normal distributions.However the number of components can become very large so that overlap betweendistributions creates an essentially smooth continuum of temperatures. The Beta distribution is used to represent these distributions because it allows a wide variety of shapesto be represented. This approach makes no assumptions about the number of underlyingcomponent distributions, instead it considers the spread of temperature to be related tosmall scale variations in wall properties which are too numerous to subdivide. Distributions at any time or for any facet orientation may be modelled using a single model,whereas the number of component populations required is higher for strongly irradiatedsurfaces. The difficulty in using this approach is relating the parameters of the Betadistribution to the more commonly used statistical moments of the mean and standarddeviation of each facet.3.3 SummaryTwo methods were examined in order to remove those surface temperature observationsfrom the mobile traverses which have mixed building and sky or sky in their FOV:Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 96truncation of the distribution, or mixed distribution modelling. Following the applicationof these methods, the spatial and temporal variation of the mobile traverse temperaturedata were examined for each of the study areas in order to determine the magnitude andtiming of maximum temperature differences between vertical facet pairs. The resultsindicate the following.• Distribution truncation investigated a subset of surface temperatures from a northor northeast facet which was shaded for most of the day and compared these withvarious parameters including: air temperature from the traverse vehicle, fixed airtemperature sensors, hand-sampled temperatures from ground teams, and temperatures from the airborne thermal imagery in order to determine a reasonable lowerlimit for the apparent surface temperature.• Truncation measures differed for each study area: in the Industrial area Tat — 2°Cis used, in the Downtown study area the minimum of — u, Tat — a, and in theResidential area — 2g.• Temperatures of the facets show large increases in variance when directly irradiatedand means for each block show much greater variation under these conditions.• Temperature differences between facet pairs are greater than 10°C in the Industrial area and slightly less in the Residential, with the largest difference occurringmid-morning between east and west facets. Other times of maximal temperaturedifferences between facet pairs occur within approximately 1 hour after solar noonbetween north and south facets and in the late afternoon between east and westfacets. In the Downtown area, the different street orientation yields one time periodwith a large facet temperature difference and a second where the differences areminor, with results showing a fairly high degree of symmetry.Chapter 3. AVERAGE TEMPERATURES OF CANYON FACETS 97• Mixed distribution modelling is demonstrated for road surface temperatures wherethe number of component populations is related to areas of the road which havebeen shaded or irradiated for some time, and areas which have recently changedfrom one radiation regime to the other. Morning distributions in particular arewell represented by two component populations, later in the day generally threepopulations are required.• Modelled surface temperature distributions were used to determine how the technique would behave in the presence of mixed sky and building temperatures. Anybias towards low temperatures caused by mixed pixels not discarded appears to beminor.• Application of the mixed distribution modelling to vertical facet temperatures results in most distributions requiring between two and four component populations.The technique is less successful when facets are directly irradiated because thenumber of underlying distributions of temperature is potentially very large due todifferences in surface material combined with varying radiation regimes.• Results from the Industrial area show, without pre-truncation of the distribution,there is less sensitivity in the method for the 100 EIRT, but there is a clear separation of a low temperature population.• The Downtown area is characterized by a greater separation of component populations than the other two study areas; the Residential area requires more components and results show the 100 EIRT to be of limited use for representing facettemperature.Chapter 4AIRBORNE TIR OF SELECTED LAND-USE AREAS4.1 IntroductionAn airborne thermal scanner was used to provide detailed surface temperature information in each of the study areas to complement that obtained by ground-based sampling.The airborne scanner allows direct observation of the extent of anisotropy in surface long-wave emission over selected urban land-use areas and also allows the compilation of acomplete urban surface temperature distribution, through the acquisition of images fromdifferent view directions. The approach adopted provided benefits in terms of flexibility,cost and detailed coverage of small areas.This chapter details the methods and equipment used to obtain the imagery, post-acquisition corrections to the imagery, and the results of analyses performed. Detailsof the instrumentation and correction techniques are provided in appendices and arereferred to where appropriate. It is not the aim to provide an exhaustive accountingof the thermal response of different surface types (see e.g., Quattrochi and Ridd, 1994).Instead, the emphasis is upon a description of how the distributions of temperature for theland-use areas vary with time and view direction and how the temperature distributionsof component surfaces combine to form the overall distributions.98Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 994.2 Methods4.2.1 Thermal Imaging SystemThe thermal imaging system used was the AGEMA Thermovision 800 BRUT Systemequipped with a model 880 LWB scanner. System components are illustrated in Figure 1.1; full description of the specifications and operating details of the scanner areprovided in Appendix A. A brief summary of the main features is provided here.VGA Monitor12° Lens LW 880 Scanner\________I_____ fl ii ii_H II I U L1__System Controller Keyboard with trackballFigure 4.1: AGEMA 880 system components.The scanner uses a cryogenically-cooled detector which is sensitive primarily in the8—14 um waveband, and is capable of measuring surface temperatures between —30°Cand 1300°C with a sensitivity of 0.05°C at 30°C. The scanner was fitted with a 12° FOVlens with a geometric resolution of 1.2 mrad. Images were obtained by scanning a setnumber of lines (140 or 280) across the FOV. Each line consists of 140 pixels. The imageswere stored digitally on an internal hard disk and later transferred to other computersystems for processing and analysis.Chapter 4. AIRBORNE TIR OF SELECTED LAND- USE AREAS 1004.2.2 Helicopter-AGEMA InstallationThe AGEMA system was installed in a Bell 206B JetRanger helicopter. The systemcontroller and monitor were placed in the front passenger seat together with a portablepower supply consisting of two deep-cycle 12 V batteries and a DC-AC power inverter.The system operator occupied the left back seat and directed operations via the systemkeyboard and the internal helicopter communication system (Figure 1.2).1.28m11.28 mlFigure 4.2: Thermal imaging system installed in the helicopter. Top: top view, Bottom:side view.The scanner was mounted on a custom-built harness worn by a second operator.The harness was strapped over the operator’s shoulders and the scanner was mountedSystemSystem Operator Station3.2mChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 101on an adjustable camera mount fixed to a plate projecting forward from the harness(Figure 1.3). Using body position and an attached level, the scanner operator directedthe scanner towards the ground at right angles to the flight path at the required viewingangle. The scanner operator was in communication with the system operator and pilot,so that by adjustment of the flight path and scanner direction, the required ground areaswere imaged. To avoid viewing the helicopter skid during nadir flight lines, the scanneroperator stood on the skid and leaned outwards to ensure an unobstructed view of theground.4.2.3 Remote Sensing FlightsEight primary flights were made in support of the objectives outlined in the introduction. A series of four additional flights was made in support of other research objectives.Details of each flight are listed iii Table 1.1. Flights originated and terminated at Vancouver International Airport. Times listed for Flights 1—8 are for the period of imageacquisition, including imaging of the ground calibration sites carried out prior to, andfollowing coverage of the selected land-use area. For Flights 9—12 the times refer to the sequence of continuous scanning between the rural area (Delta) and downtown Vancouver.Overflights of several urban parks were carried out following this sequence.The time of the flights was determined from an analysis of preliminary auto-traversedata which identified the times at which vertical facet temperature differences were maximized for the canyon orientations in each of the study areas.On each flight, the nadir view flightlines were flown first, followed by the off-nadirlines. Because the scanner was fixed on one side of the aircraft, the four off-nadir viewdirections were scanned in a rotational sequence to conserve flight time, (e.g., east, south,west and north). Two to four rotations were necessary to achieve the required coverage.The ground speed of the helicopter varied depending upon wind direction and speedChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 102Table 4.1: Flights conducted to acquire airborne thermal imagery.Fit Date Time (PDT) Location(s) Alt (m) SA (0)1 08/15/92 1000—1100 False Creek S. 647, 457 0, 452 08/15/92 1345—1430 “ 647, 457 0, 453 08/15/92 1705—1745 “ 647, 457 0, 454 08/16/92 1115—1210 Downtown 689, 488 0, 455 08/16/92 1605—1645 “ 689, 488 0, 456 08/17/92 0940—1005 Sunset 975, 548 0, 457 08/17/92 1345—1415 “ 975, 548 0, 458 08/17/92 1705—1730 “ 975, 548 0, 459 08/24/92 1520—1550 Urban/rural 2134 010 08/24/92 2015—2045 “ 1524 011 08/24/92 2345—0010 “ 1524 012 08/25/92 0450—0520 “ 1524 00SAscan anglebut was generally between about 60 and 80 km hr’. Images were- sampled so that aminimum 75% overlap was maintained. Maps of the actual area covered by the imagesobtained are presented in Appendix D.Each flight included coverage of one or more ground calibration sites, general1,r bothprior to and following the coverage of the selected land-use area. These sites, locatedat urban parks, were imaged and compared with ground-based observations of surfacetemperature over several surface types in an effort to confirm the appropriateness of thecalculated atmospheric corrections. Surfaces scanned included water, grass, concrete andasphalt in order to obtain a range of surface temperatures. Details of the methods andresults are presented in Appendix B.Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 103Figure 4.3: Scanner mounting arrangement.Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 1044.3 Atmospheric CorrectionsThe airborne thermal imagery is influenced by the processes of atmospheric absorptionand re-emission by atmospheric gases, primarily water vapour, but also ozone, carbondioxide, nitric oxide, carbon monoxide and methane, which serve to alter the apparenttemperature measured by the scanner compared to that observed at ground level. In general, this leads to an underestimate of the surface temperature, for most surfaces undertypical daytime conditions (high radiant input, lapse profile). Correction for these processes used the single infrared channel method (Becker and Li, 1990) in which a verticaldescription of the atmosphere (in terms of pressure, temperature and humidity) is usedin conjunction with an atmospheric radiation transfer model to estimate atmosphericabsorption and re-emission in the spectral range of the instrument. The model usedwas the LOWTRAN 7 Atmospheric Transmittance/Radiance model (Kneizys et al., 1988).Measured vertical profiles of pressure, temperature and humidity were made by radiosondes. The lowest layers of the atmosphere were measured by kcally launched radiosondes,while upper atmosphere information was obtained from one of the two nearest upper airreporting stations, Port Hardy B.C., or Quillayute WA. Details of the measured profiles,methods and results are presented in Appendix B. Final corrections, in the form of apolynomial approximation, were determined for each altitude/scanner angle combinationand applied to images prior to further processing.4.4 Surface Temperature Analysis4.4.1 Industrial AreaA thermal image taken at a 45° off-nadir angle towards the south during the morningoverflight of the Industrial area is shown in Figure 1.4. The original 280x 140 image hasChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 105been re-configured to a square 280x280 image for display purposes.Figure 4.4: Thermal image of the Industrial study area, Flight 1 (1030 PDT). View issouthwards at a 45° off-nadir angle.The flat-topped, rectangular buildings clearly show very warm roof areas; roof temperatures are generally homogeneous for individual buildings but differ between buildings,probably due to material and construction differences. Isolated darker features on roofsare generally related to low emissivity (aluminum) heating/cooling equipment or ducts.Where adjacent buildings are of different heights, some areas of the roof are also cast inshadow.Road surfaces tend to show a fairly narrow temperature range, except on the N/Sstreet where cooler areas near shadows exist. These are thought to be related to areasformerly in shadow which have recently become directly irradiated. Some vehicles may289A Apparent Temperature (K) 311.2Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 106be identified as very dark (cool) areas (one is located on the E/W street near the middleof the image); these are trucks with polished aluminum cargo boxes. Painted vehiclesare not so easily identifiable; trucks parked at the loading dock of the building in thelower right-hand corner of the image display temperatures as warm as, or warmer thanthe road surface.The associated temperature distributions, separated by view direction, are presentedin Figure 1.5. The composite image temperature distributions (solid line) are overlaidwith temperature distributions from the major surface types present in the study area(flat roofs, roadways, walls and shaded ground). These were extracted from select imagesby digitizing the component areas and aggregating the temperature values. Areas weredefined solely on the spatial pattern of temperature. This may lead to some bias, basedupon expected thermal patterns relative to the surface structure, e.g., defining the verticalextent of the walls can be difficult when both they and the ground are shaded. Eachfrequency distribution is normalized by its maximum class frequency. The position of thecomponent distributions may then be used to determine which surfaces are responsiblefor the shape of the composite distribution.The plots show a strongly demarcated low temperature peak for the south, east, andvertical view directions, which is related to shaded wall and ground surfaces. In contrast,the north and westward views show only a minor peak or shoulder at these temperatures.All view directions exhibit a second, clearly defined mode as well as a shoulder of highertemperatures. The distribution of road surface temperatures approximately matches themain peak of the composite distribution but is shifted towards warmer temperatures,especially for the south and east view directions. The upper shoulder is due to roofsurfaces, and in the westward viewing case, west (east-facing) walls. The position ofthe wall temperature distribution varies with view direction and it therefore alters thesubsequent shape of the composite distribution. In the east and south view directions wallFigure 4.5: Apparent surface temperature distributions for each view direction over theIndustrial area during Flight 1 (morning).107Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREASApparent Surface Temperature (0 C)UcQ):3-o0)aEaz1.> 0.90.80)070.6o>‘C-)cq):300).0: Surface Temperature (° C) Apparent Surface Temperoture (0 C)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 108temperatures lie very close to that of the shaded ground resulting in a sharply definedpeak in the composite distribution, (more so for the east than the south direction).The warm west facets contribute to the upper temperature shoulder in the compositedistribution, enhancing the effect of the roof temperatures. North facet temperatures liemidway between the shaded ground and sunlit streets, resulting in a broadening of themain peak of the composite distribution.There exists a gap in the component distributions as they relate to the composite fortemperatures less than the sunlit road surface temperature, but warmer than the shadedground. For some view directions (north, south), wall temperatures occupy part of thisintermediate position, but there is clearly a large fraction of surfaces in a temperaturerange not incorporated by any of the component surfaces extracted from the images. Thisis further confirmed in Figure 1.6 which uses approximate weightings derived from imagesfor each of the extracted components and sums the weighted component distributions incomparison with the observed South 45° composite distribution.Examination of the range of “missing” temperatures on select images reveals that it isassociated with what may be characterized as partly shaded surfaces. A large fraction ofthese are road surfaces which have recently changed from shaded to sunlit, while othersare seen near boundaries between walls and roofs.Differences between the frequency distributions for opposing view directions and nadirversus off-nadir fiightlines are summarized in the final plots of Figure 1.5. These showthe temperature ranges in which the frequencies differ the most. Differences are stronglyrelated to the presence or absence of shaded surfaces, as evidenced by the strong peak at18°C (i.e., close to air temperature). Differences between the off-nadir and nadir viewsshow large frequency differences for temperatures in the range 27—29° C.The highest frequency peak shows some offset between the four off-nadir view directions and the vertical view; this may be a result of warming during the time needed toChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 1090.1 00.090.08>N0.070.060.0500. 4.6: Modelled composite temperature distribution using component temperaturedistributions weighted by their occurrence within the sensor FOV.cover each of the nadir flightlines. (The vertically scanned flight lines were flown beforethe off-nadir). There is little or no offset among the off-nadir view directions becauseof the rotational sequence in which they were sampled. Warming or cooling during thetime required to complete the -flightlines may act to broaden the distributions somewhat;warming/cooling rates calculated from ground and remote observations can be large forroad and roof surfaces, especially in the morning and afternoon.The degree of broadening in the composite distribution due to warming in the Industrial area during the morning flight (Flight 1) is investigated in Figure 1.7.Assuming similar surface structure and materials among the flightlines, differencesin the frequency distribution are attributed to warming between runs. The main andupper temperature peaks associated with road and roof temperatures respectively showa substantial offset (taken to be associated with a warming of these surfaces) between the20 30 40 50Apparent Surface Temperature (° C)Chapter 4. AIRBORNE TIR OF SELECTED LAND- USE AREAS0.110.1 0.05U->N0.05cJ) 30 40 50Apparent Surface Temperature (° C)110Figure 4.7: Broadening of the apparent surface temperature distribution attributed dueto surface warming during data acquisition of the morning flight over the Industrial area.20 30 40 50Apparent Surface Temperature (° C)Line 1 4 (1 037)Line 1 8 (1 044)Line 1 0 (11 00)All South(b)4’’:qi:J! •\\ - -!/ I •,\•.1/ I ‘r’-/.,, I ‘/- ,-, •1 --“‘-—//Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 111first and last nadir flightlines (Figure 1.7a), while frequencies of the lowest temperaturesdecrease, in proportion to the reduction of shaded areas. Distributions for the northview direction (Figure 1.7b) show a shift in the main peak (road temperatures) towardswarmer temperatures with time, a reduction (increase) in the frequencies of the coolest(warmest) temperatures associated with shaded (roof top) areas.No attempt was made to correct the imagery for this warming or cooling during thetime of the overflight because different surfaces warm and cool at different rates dependingupon their orientation and thermal properties. However, because some surfaces undergosubstantial warming with time, comparisons between ground and remote observationsshould take the time periods over which they have been made into consideration. Asecond factor to be considered is the inclusion of areas of “non-standard” surface cover(here defined as block orientations different from the rest of the study area, or areassuch as parks or large open areas not included in the categorization of the area). Thiseffect is investigated in Figure 1.8 which compares the complete composite distributionwith a distribution obtained when all images containing non-standard block orientations(present in the NW portion of the study area), Jonathan Rogers Park, and areas to thesouth of the study area (many large trees) are omitted from the analysis.Figure 4.8: Comparison of apparent surface temperature distributions before and afterimages with a non-standard block orientation or surface cover were removed.0.07> 0.06U0.050.04‘-i- 30 40 50 20 30 40 50Apparent Surface Temperature (° C) Apparent Surface Temperature (00)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 112The distributions show a reduction in frequencies at low temperatures (associatedwith the removal of the vegetated areas), a slight narrowing and increase in frequenciesof the mid-range peak, and enhancement of the upper portion of the distribution (for alloff-nadir view directions). The effects are best seen in the east and west view directionsas these flightlines had proportionally a greater number of non-standard images.The early afternoon flight (Figure 1.9) shows distributions with features similar tothose described for the morning flight but with somewhat less well defined peaks. Thesmearing of the peaks is attributed to the overall warming of all surfaces and the transitionof many surfaces from shaded to sunlit status which gives more surfaces with intermediatesurface temperatures.The “shoulder” of warmer temperatures above the main peak is still present butless well defined than in the morning flight. This appears related to a greater overlapin the temperature distributions of these components; similar results were obtained byQuattrochi and Ridd (1994).The east and west view directions present almost identical distributions, and differfrom the vertical only by slightly higher frequencies of low temperatures and lower frequencies of warm temperatures (Figure 1.9). Shaded areas are minimized at this timegiving a smaller low temperature peak for the southward view. Images obtained witha northward view show an almost complete absence of temperatures in the range forshaded ground, and the composite distribution is a unimodal, almost symmetric shape.The late afternoon overflight (Figure 1.10) shows much less distinction between roofand road surface temperatures and between roof and sunlit wall surface temperaturescompared with previous flights.This is evident by the absence of the warm temperature “shoulder” of the morningand early afternoon flights. The low temperature peak associated with shaded surfacesis less well defined at this time; it is to be anticipated that the range of temperatures in>‘0 0.01Ca. 0.00a)‘-i-—0.01—0.02Figure 4.9: Same as Figure 1.5 except for Flight 2 (1345-1430 PDT).Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 113>‘0Ca,Da,-oa,aE0z1. 30 40 50 60Apparent Surfoce Temperature (° C)>‘0Ca)D0a,-oa,0E0z>‘0Ca,a0a)a,aE0z1.0 1.0j100.03 — N—S ( d) 0.03-0.03—0.03-w (g)V—E-—./-20 30 40 50 60Apparent Surface Temperoture (° C)20 30 40 50 60Apparent Surface Temperature (° C)>‘‘U3aa)0a)0E0z0.—0.01—0.02—0.03Figure 4.10: Same as Figure 1.5 except for Flight 3 (1705—1745 PDT).Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 114a)’UCa)0a)Va)0E0z1 .’(b)Apparent Surface Temperoture (0 C)1 . .0>. 0.90.8a)’a)3a-a,U0.030.02—1N—S3V—N j.! ‘0.010.00—0.01—0.02—0.03E=W20 30 40 50Apporent Surface Temperature (° C)20 30 40 50Apparent Surfoce Temperoture (° C)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 115areas classified as shaded will be larger in the afternoon because many of these surfaceswere previously sunlit and are in the process of cooling; the range of temperatures willbecome narrower with time as cooling progresses. This effect extends to the shadedground on the south sides of E-W canyons which become exposed to direct radiationin the late afternoon and undergo some warming, further smearing the low temperaturepeak.Temporal development of the temperature distributions is summarized in Figure 1.11.which shows the difference between apparent surface temperature, and the canyon-levelair temperature from the traverse vehicle, for each view direction.—10 0 10Figure 4.11: Temporal development of the distribution of Tr as indicated by the differencebetween surface and canyon level air temperature. Industrial area.North20 30 40—10 0 0 20 30T, — T. (° C),—T.(°C)—10 0 10 20 30 401,— T0 (° C)The shift and change in the features of the distributions is easily traced. At all times,Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 116the majority of the surface temperatures greatly exceed the canyon-level air temperature.Those features of the distributions attributed to shaded surfaces tend to have temperatures slightly less than that of the air (2.5 degrees). The lower bound of the distributionsis fairly consistent with time; the cutoff is approximately 6 degrees below air temperature. This may represent the temperature of those surfaces shaded throughout the day.The upper bound varies strongly with time of day.4.4.2 ResidentialIn the Residential study area, thermal images clearly show the cool vegetated surfaces,including tree crowns and grass, and a greater variety of roof temperatures, due in partto variations in the pitch and orientation of the mostly non-flat roofs (Figure 1.12).The ability to view building walls depends upon the building and block orientation.North-facing walls are clearly evident on buildings aligned N/S along E/W streets, butthe small inter- building spacing hides most of the N-facing wall from view for buildingsaligned E/W along N/S streets. The temporal development and shape of the temperature distributions (Figure 1.13) is similar to that described for the Industrial area: Themorning flight shows a bimodal distribution of temperatures with a sharply defined lowercutoff and a more gradually decreasing tail of warmer temperatures. The peaks are lesswell separated than for the Industrial area and there is no distinct plateau or shoulderof warm temperatures as seen in the Industrial area. This appears to be related to agreater range of temperatures for the building roofs, which occurs because many roofsare pitched rather than flat. This serves to increase the range of temperatures above thatdue solely to material or construction differences.Component temperature distributions observed during Flight 7 (1345-1415 PDT)(Figure 1.14) are presented for 8 major component surfaces in the N-S block orientationChapter 4. AIRBORNE TB? OF SELECTED LAND- USE AREAS 117294A Apparent Temperature (K) 318.3Figure 4.12: Thermal image of the Residential study area, Flight 7 (1400 PDT). View issouthwards at a 45° off-nadir angle.Chapter 4. AIRBORNE TIR OF SELECTED LAND- USE AREAS0.110.1 00.090.08‘—10 0 10 20 30 40T,—To (°C)0.110.1 4.13: Surface temperature frequency distributions for the three flights over theResidential study area: solid line (0945 PDT), dash-dot (1400 PDT), dash (1730 PDT).North— 1, — To (0 C)‘ 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 119area: shaded ground, tree crowns, grass, streets (asphalt), walls (separated by orientation), fiat roofs and north and south pitched roofs. To clarify presentation, thosecomponents common to all view directions are shown together with the nadir composite. Each off-nadir view direction adds only those components unique to that direction;primarily the walls, but also the north and south-pitched roofs.The range of temperatures for some components is large, especially roof areas. Aclear separation is evident between roofs classified as north- and south-pitched, but eachcontains significant overlap with the fiat roof category. North-pitched roofs, which arecooler than flat or south-pitched roofs have temperatures very similar to those obtainedfor road surfaces and south-facing walls. The low temperature peak of the bimodal distribution appears to be a combination of three surfaces: tree crowns, shaded groundareas, and, where visible, north-facing walls. Grass surfaces show similar frequenciesover a fairly wide range of temperatures (7°C) which overlaps the range of north, eastand west walls, and of the north-pitched roofs, and leads to the relatively high frequencies in the composite distribution between the two peaks. East and west walls havevery similar distributions. West walls show a slightly narrower distribution and a tailof warm temperatures as these facets cool, and east walls warm, with time in the afternoon. The similarity in their composite distributions is confirmed by the difference plot(Figure 1.14(g)) which shows essentially random variations around 0.Differences between nadir and east and west distributions are also very similar. Therelative reduction in low temperature frequencies appears to be correlated with the eastand west wall temperature distribution; the source of enhancement for temperatures at40°C is less clear, presumably it is related to a more direct view of the pitched roofsurfaces. North-south differences are stronger and show a narrow peak at approximately25°C related to the presence or absence of north-facing walls and shaded ground betweenthe two views, and also to temperature differences across the tree crowns. The moreChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS>‘0Ca)00a)U00)aEaz>‘0Ca)3a-a)Ua)CE0z0.040.030.02>0.010). 0.00U- —0.01—0.02—0.03—0.04>0Ca)0a-a)U--o0)aE0z120Figure 4.14: Same as Figure 1.5 except for Flight 7 (1345—1415 PDT) over the Residentialstudy area.Apparent Surface Temperature (° C)1 .—0.01—0.02—0.03—0.04SE-W (9)• taV—E-—. taV—W1’’ ‘20 30 40 50 60Apparent Surface Temperature (° C)20 30 40 50 60Apparent Surface Temperature (° C)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 121exposed view of the south-pitched roofs accounts for the plateau of positive differencesat high temperatures. The enhancement of temperatures between 40 and 50°C appearsto be linked to the south-facing walls.’ 30 40 5020 30 40 50Apparent SurfaceTeroperature (0 C)Figure 4.15: Comparison of Tr distributions for morning flights (Flights 1 and 6) overthe Industrial and Residential study areas.Comparison with the Industrial area (Figures 1.15, 1.16, 1.17) shows that the Residential area exhibits greater frequencies for the low temperature peak, presumably dueto greater amounts of vegetation and/or shading. The relative magnitude of this peakis almost comparable to, or in the case of the eastward-viewing direction, exceeds thefrequency of the warm temperature peak.North0.090.084’ Surface Temperature (0 C) Apparent Surface Temperoture (0 C)South0. 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS0. 4.16: Same as Figure 1.15 except for early afternoon flights (Flights 2 and 7).20 30 40 50 60,West Early Aftersoos Flight NadirlI_20 30 4 50 60Apparent Surface Temperature ( C)EastApparent Surface Temperature ( C)Apparent Surface Temperature (° C)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 123The early afternoon flight shows evidence of a shoulder of warm temperatures developing. Vertical, north and southward distributions are otherwise generally similar exceptfor a greater frequency of low temperature classes. Peaks in the distribution are offsetdue to the warmer temperatures on this day. East and westward views are again similarat this time, and the relative magnitude of the peaks is almost comparable, in contrastto that of the Industrial area.Figure 4.17: Same as Figure 1.15 except for late afternoon flights (Flights 3 and 8).In the late afternoon, the north-, south- and eastward distributions are characterizedby bimodal distributions with the classes between the peaks exhibiting relatively highfrequencies (Figure 1.17). Compared to the Industrial area, the warm temperature peakis reduced in magnitude and the cool temperature peak enhanced. Sharp drop-offs in the0100.090080.070.060.055’l500.055 30 40Apparent Surface Temperature (° C) Apparent Surface Ten-iperature (DC),”,”0.010.0020 30 40 50Apporent Surface Temperature (DC)60Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 124distribution shape are evident at both warm and low temperature ends of the distribution.A complicating feature of the Residential area is the presence of two distinct areas ofdiffering block orientations. North of 49th Ave. blocks have their long axis orientednorth-south. Buildings are oriented E/W with a small inter-building spacing in the N/Sdirection. Two-sided pitched roofs are preferentially oriented N/S because the long axisof most buildings is east/west. Streets are lined with large mature trees. South of 49thAve. most blocks are oriented east-west and there is less mature vegetation.Images for Flights 6—8 were categorized according to the primary block orientationand the frequency distributions were recalculated (Figure 1.18). Images with mixedorientations were removed from the analysis. Nadir images consistently show higherfrequencies at low temperatures for the N/S oriented blocks.This is taken as evidence of larger amounts of vegetation in this portion of the studyarea. Morning eastward and afternoon westward views in the N/S block orientationsexhibit higher frequencies of low temperatures which are attributed to the greater apparent wall area exposed along the streets compared to that in the E/W oriented blockswhere east and west walls are less easily viewed due to the small inter-building spacing.Northward-looking views from the morning flight over E/W blocks were characterizedby higher frequencies of warm temperatures, possibly due to the more open south-facingfacets in this area. The late afternoon flight shows a similar effect but for southwardviews. East and westward views during the early afternoon flight show higher frequenciesof warm temperatures in the E/W blocks which are attributed to fairly equal temperatures on the pitched roofs in this block orientation compared to those in the N/S blocks,where north-pitched roofs are cooler than the south-pitched ones.Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 1250. 30 40 50 60 7020 30 40 50 60 70Apporeet Surface Temperoture (°C)20 30 40 50 60 70Apparent Surface Temperature (° C) 4.18: Comparison of Tr distributions for two primary block orientations in theResidential study area. Data for the early afternoon flight (Flight 7).NorthApparent Surface Temperature (° C)East0.,,ISouthChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 1264.4.3 DowntownThe large building structure in the Downtown area is well illustrated by Figure 1.19. Thisimage was taken looking towards the southwest, as evidenced by the shadow patternsvisible on the street on the left-hand side of the image. Note that the tall buildings for themost part obscure the streets which parallel the flight direction, and the roofs of shorterbuildings are obscured by tall buildings. This leads to a reduction in the frequencies ofwarmer temperatures in these view directions. Building roofs tend to be more structurallycomplicated than the other two land-use areas, so that the roofs often contain areas ofshadow or apparent temperature differences due to low emissivity materials. The relativeproportion of roof area to the entire image is smaller than in the other two areas.303.4Figure 4.19: Thermal image of the Downtown study area, Flight 4 (1130 PDT). View isto the southwest at a 45° off-nadir angle.Apparent Temperature (K)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 127The image was collected using scanner mode 3, in which only 140 x 140 pixels weresampled, thus no interpolation has been required but the image is somewhat coarser thanthat of corresponding Figures 1.4, and 1.12. Temperature distributions from the morningDowntown overflight (Figure 1.20) show distributions dominated by low temperatures,except in the direction of the most directly sunlit facet (NW at this time). The nadirview shows a bimodal distribution with the lower temperature peak slightly larger. Thelow temperature peak is related to shaded areas on the ground, and the warm temperature peak is derived from the warm roof and street areas. In this figure, temperaturedistributions for roadways are not limited to sunlit areas, samples were taken along thelength of the visible streets (in the direction of the sensor view) and thus include bothshaded and sunlit road surfaces.There is strong differentiation between the NW and SE view directions. The NW distribution is unimodal; the frequency of shaded surfaces viewed by the sensor is very small.There is little difference between the NE and SW distributions at this time, each having alarge low temperature peak and a weakly expressed warmer mode. The dominance of thelow temperature classes occurs because the walls and streets include a high proportion ofshaded areas. The view angle in combination with the tall buildings effectively restrictsviewing NW/SE streets which are characterized by relatively little shading at this time.The NE/SW oriented streets are primarily shaded except at intersections.Temperature distributions from the afternoon flight (not shown) are reversed fromthose of the morning. The largest difference is the shape of the nadir temperaturedistribution which is no longer bimodal; the upper peak is replaced by a gradual tail-offof frequency.Comparison with the other sites is difficult because the different street orientationsalter the time of the maximum vertical facet temperature differences. An alternativeis to use times at each site when the relative solar azimuth with respect to the street0.06 0.060.04 0.040.02 0.02‘ 0.00 0.00—0.02—0.02D—0.04—0.04—0.06—0.06—0.08—0.08—0.10—0,10—0.1 2—0.12—0.1 4_______________________________________________—0.1 4Figure 4.20: Same as Figure 1.5 except for Flight 4 (1130 PDT) over the Downtownstudy area.128Chapter 4. AIRBORNE TIR OF SELECTED LAND- USE AREASApparent Surface Temperature (° C)>‘0CeaeUVaE0z1 .>‘0.8VD 0.70’0.6U.0.5V.‘‘,8NL—SWAV—NE- .—. ev—sw10 20 30 40 50 10 20 30 40 50 60Apparent Surface Temperature ( C) Apparent Surface Temperature (° C)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 129orientation is similar for the Downtown and another site. However, it should be notedthat since solar zenith angles are different there are differences in the angle of incidenceof the solar beam with the building facet. The closest comparisons possible, based uponrelative solar azimuth to the canyon facets, are shown in Figures 1.21 and 0.060.07 0.12 0.070.06 0.10 0.060.05 0.08 0.050.04 0.06 0.040.03 0.020.01 0.02 0.010.00 0.00 0.0010 20 30 40 50 60 10 20 30 40 50 60 10 20 30 40 50 60Apparent Surface Temperature (° C) Apparent Surface Temperature ( C) Apparent Surface Temperature ( C)Figure 4.21: Comparison of Tr distributions for the Downtown (Flight 4), Industrial(Flight 1), and Residential (Flight 6) areas. Comparison based upon relative solar azimuth to canyon orientation.The dominance of shaded surfaces in view directions SE -(Flight 4) and SW (Flight5) is very apparent. A much smaller fraction of the FOV intersects the ground or roofsurfaces compared with the other study sites, and, even when horizontal surfaces areviewed they are in many cases shaded by tall buildings. The increase in shaded groundarea is also apparent in the enhanced frequencies of low temperatures in the nadir views(Figure 1.21(a), 1.22(a)).When viewing the directly sunlit facet, the distributions appear more similar to thosefrom the other study areas; and become more unimodal.4.4.4 SummaryHigh resolution thermal images of the three study areas were used to show spatial temperature patterns. Road surface temperatures tend to be relatively homogeneous except forNar j\ Morr4eg Flights/ — Downtown• lndutriul— .—• ReulduntiulEast. SE FacetsChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 130Downtown (o=43c1=-23)Industrial (e=57,o=—21)Residential (o=56,cl=-18)0.06 .....,. 0.1 0.100.05 0.05 0080.04 0.020.01 0.01 0.020.00 0.00 0.0010 20 30 40 50 60 10 20 30 40 50 60 10 20 30 40 50 60Apparent SurlaceTemperature (° C) Apparent Surloce Ternperoture (° C) Apparent Surface Temperature (0 C)Figure 4.22: Same as Figure 1.20 except for early (Downtown) and late afternoon flights(Industrial and Residential areas).shaded areas. Roof temperatures show variations related to materials and constructionand especially to roof pitch. Distributions of surface temperature tend to be bimodal;the distributions are most sharply defined during the morning when differential heatingcreates the largest temperature differences between surfaces.The lower peak of the distribution is associated with shaded surfaces and varies instrength depending upon the view direction. Shaded surfaces exhibit their narrowesttemperature range in the morning.The composite temperature distribution may be used to discriminate between land-use types: in Residential areas characterized by a wide variety of surface covers, thetemperature distribution is less well defined and shows more constant frequencies acrossa wide range of temperatures. In commercial/Industrial land-use areas, the larger roofand paved areas enhance narrow ranges of temperature.— Downtown Nadir—— IndustrialResdwntioISouth, SW Facets— Downtown-—— Industniwl-—. RnsidnntiulChapter 4. AIRBORNE TIR OF SELECTED LAND- USE AREAS 1314.5 Modelling the Distribution of Surface Temperature4.5.1 IntroductionThe composite image surface temperature frequency distributions show features relatedto the surface structure and materials. Where the surface structure is relatively simpie (e.g., the Industrial area), the distributions are shown to be composed of relativelyfew individual distributions, each arising from a component surface (e.g., wall, roof,road, shaded ground). In more structurally complex areas (e.g., the Residential areas),the overall distributions show fewer distinct features, because the number of componentdistributions increases and there is greater overlap between distributions. This sectioninvestigates the possibility of modelling the composite distributions as a mixture of distributions using the techniques described in Chapter 3. Of particular interest is whether thecomposite distribution can be modelled so that the component surface temperature distributions can be recovered without having to resort to intensive extraction of small scalefeatures within an image (e.g., see Quattrochi and Ridd, 1994), or the use of coincidentmultispectral imagery.In some respects, the question is: How successfully may surface type be classifiedupon the basis of temperature ? Where different surface components exist with similartemperatures, discrimination will not be possible. Where a surface type is representedby a wide temperature distribution due to variations in surface properties at small scales(e.g., different roof types), problems will also arise due to substantial overlap with othersurface categories (e.g., walls or road surfaces). Discrimination between components isalso expected to depend on the time of day; relative differences in thermal and radiativeproperties produce the most sharply defined composite distributions during the morningheating period.Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 1324.5.2 Modelling a Simple CaseA simple composite temperature distribution, obtained from a thermal image of a set ofconcrete blocks on an asphalt surface, is shown in Figure 1.23a.This model simulates the general surface structure of a Residential area in terms ofthe dimensions and spacing of the blocks and buildings. It obviously greatly simplifiesthe surface in terms of material type; only asphalt and concrete are present. Temperaturevariations are therefore largely constrained by the surface orientation and shading. Over-lain on the composite distribution are the component distributions obtained by extractingselected surfaces from the image; in this case the image is subdivided into roofs (top ofconcrete blocks), walls (only the west wall is viewed), and the asphalt surface (shadedand unshaded components). The form of the composite distribution is most differentfrom that obtained over the Residential area in that the roof temperatures are muchlower than either the walls or unlit asphalt surface. The composite distribution was thenfitted with four component curves using the PEAKFITTM program (Figure 1.23b,c). Inthe first example, all component curves were Gaussian. In the second, Beta distributionswere used for the wall and shaded surfaces, to better represent the asymmetry displayedby those distributions as extracted from the image. Comparison between the statisticsof the extracted components and modelled components is given in Table 1.2. The agreement is very good; (R2 values are 0.993 and 0.963) however, the most accurate fittingrequires tuning the position and shape of the individual components prior to starting thefitting iterations. This is especially true with the Beta distributions.4.5.3 Modelling Road Surface TemperaturesA second example examines (as in Chapter 3) mixed distribution modelling of the roadsurface temperatures. The airborne thermal imagery is used this time instead of theChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS(a)>.C,Ca,D0a)(b) 2>‘o 0.10C0. 60.140.12>. 0.10C,C0.0800. 28 30 32 34 36 38Apparent Surface Temperature (° C)28 30 32 34 36 38Apparent Surface Temperature (° C)133Figure 4.23: Temperature frequency distribution for a concrete scale model representingthe Residential area showing the complete image distribution (solid line) and selectedcomponent surfaces. View is to the west.Residential Model— compositeView West: 11 30 PDT!!I ‘ I,hode Roofs 1Rooc Eo$t w91126Apparent Surface Temperature (0 C)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 134Table 4.2: Comparison of component temperature distributions for an image of a simplescale model. (All values °C).Extracted Modelled a Modelled bComponent Mean SD Mean SD Cntr Widl Wid2 Wid3cWall 35.82 0.75 35.49 0.76 35.70 13.01 18.01 3.51Roof 30.19 0.53 30.10 1.18 30.05 1.08Asphalt (sunlit) 32.89 0.55 32.98 0.68 32.95 0.73Asphalt (shaded) 27.75 0.67 27.71 0.53 27.76 13.01 2.51 14.51aji surfaces Gaussianbwall and shaded surfaces BetacWidl, Wid2, Wid3 are adjustable paramters for the Beta Distributionvehicle-mounted EIRT. Road areas were selected from an image, (in this case a N/S roadfrom the morning flight over the Industrial area which shows both shaded and non-shadedareas). As described in Chapter 3, road surfaces are relatively homogeneous in terms oftheir thermal properties and orientation (this can be important at small scales due tothe camber of the road, and at larger scales due to topography), so that temperaturedifferences are most strongly conditioned by shading. The surface temperature of bothshaded and sunlit streets may be assumed to be represented by a Normal distributionif they have not recently undergone a change from one irradiance regime to the other(analogous to the approach taken in Chapter 3). However, inspection of the imagesshowed significant areas with temperatures intermediate between that of the shaded andsunlit portions. These sections appear to have recently become sunlit (or shaded). Theyexhibit a temperature distribution where the frequency of temperatures is related to thetime since the change in irradiance. The temporal response is expressed spatially as anon-linear change in temperature across a shaded/non-shaded boundary. The rise is mostrapid near the boundary and tails off with distance (Figure 1.24(b)); more of the area ischaracterized by temperatures close to those of unshaded than shaded areas. PhysicallyChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 135this may be related to the large temperature gradient imposed near the surface whenthe surface becomes sunlit. The temperature rise will be initially be rapid as the largetemperature gradient is confined to a shallow layer near the surface. However, as theasphalt layer warms, the temperature gradient will be reduced and the temperature willmore gradually approach the new value. Note that this effect appears to be limited tomorning hours on the north-south streets where the area of shadow changes most rapidly;on east-west streets, the area undergoing a shading change is smaller and appears to haveless bias towards either shaded or unshaded temperatures.32 35(a) 3331282926 27C) C)24 2522 23212019171 62 4 6 8 1 0 1 2 1 4 1 6 1 8 2010 5 1 0 1 5 20 25 30Distance (pixels) Distance (pixels)40 3838 363634 3432.— 32° 3028I—26 282422 2620 2418___________________________________222 4 6 8 10 12 14 16 18 0 5 10 15 20 25 30Distance (pixels) Distance (pixels)Figure 4.24: Transects of apparent surface temperature across shadow boundaries.Industrial Area: 1 030 POTN—S shadow transects(b)Industrial Area: 030 POTE—W shadow transect(c).7///7,/ // ..-,‘ • inctustrioi Area: 1 730 POT— ,-‘ N—S shadow tronsects(d)Industrial Area: 1 730 POTE—W shadow transectsThe afternoon extension of shadows onto the N-S streets yields a much more linearChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 136transect; several profiles show a more rapid drop off from the unshaded temperaturewhich is the inverse of the results from the morning, but the effect is subtle. To representthe observed morning bias in the modelled distributions two approaches may be adopted:First, the composite distribution may be represented by two distributions in accordancewith the number of radiation classes (i.e., shaded or sunlit). Assuming the pixel size issufficiently small to make the number of mixed shaded/sunlit pixels insignificant, eachclass can be represented by a Beta distribution. This accounts for the asymmetric shapeincurred by those areas which have recently undergone a change in radiation regime; i.e.,shaded areas have a warmer tail, and sunlit areas have a cool tail. Results (Figure 1.25a)indicate this approach has some trouble achieving a fit to temperatures intermediatebetween the two modes.Second, shaded and non-shaded surfaces which are not undergoing rapid temperaturechanges with time due to a recent change in the radiation regime can each be representedby Gaussian distributions. A third distribution, representing those areas recently shadedor non-shaded (the latter dominates at this time but is not the sole class because theshadow position rotates and shortens as the solar zenith angle decreases), is added. Thisthird distribution may be represented by a Beta distribution. The combined curves(Figure 1.25b) yield a good fit across most of the observed temperature range, exceptfor temperatures from 21—25°C where observed frequencies are under-represented. Theobserved frequencies in this and other examples show a range of surface temperaturesbetween the two main peaks, for which almost equal frequencies were observed, andincreasing frequencies towards the main, high temperature peak. This suggests a generalshadow pattern which consists of a linear portion beginning at the shaded/sunlit border,and a decrease in the slope as the directly-irradiated surface is approached. In this casethe lower temperatures might be better represented by a rectangular distribution, andthe upper portion by a Beta distribution, while retaining Normal distributions for theChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 137(a)_____________500 Industrial Area 1030 POTGauss 1400—— Gauss 2Beta14 16 18 20 22 24 26 28 30 32 34Apparent Road Surface Temperature (°c)(b)500 Industrial Area 1030 POTBetol400—— Beta2I14 16 18 20 22 24 26 28 30 32 34Apparent Road Surface Temperature ( C)(c)___500 Industrial Area 1030 PDTGauss 1400 Gauas2BetaiI100 A0______ _14 16 18 20 22 24 26 28 30 32 34Apparent Road Surface Temperature (° C)Figure 4.25: Fitted distributions to road surface temperatures (a) using Gaussian distributions for shaded, and sunlit surfaces and a Beta distribution for areas recently unshaded, (b) using Beta distributions for shaded and unshaded surface classes, (c) sameas (a) except that a rectangular distribution has been subtracted from the observeddistribution between 18.5 and 25.5°C.Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 138shaded and sunlit surface temperature classes. To implement this, a constant countwas subtracted from the temperature classes between 18.5 — 25.5°C to represent therectangular distribution, and two Gaussian and a single Beta distribution were fitted tothe adjusted data (Figure 1.25c). The agreement between the combined curves and theobserved results is excellent.4.5.4 Modelling the Study Area Distributions4.5.4.1 Temperature Frequency DistributionsSelected composite distributions were analyzed to determine the success of modellingwith mixed distributions and the relation between these distributions and the statisticsof selected component surfaces obtained by extraction from images or from the traversedata.Two examples of fitted component distributions are shown in Figure 1.26. In thefirst, component distributions were selected as Gaussian or exponential Gaussian (extreme value) which has a negative skew to the distribution (a tail of higher frequenciestowards higher temperatures). The use of the exponential Gaussian provides a good fitto the upper portion of the composite distribution. Both distributions are fitted withthree adjustable parameters: the amplitude, centre and width of the distribution. Theadvantage of using an exponential Gaussian distribution over a Beta distribution is thatit is fully described by a mean and standard deviation, while the Beta distribution requires four parameters. The second example retains a Beta distribution and includes anextra Gaussian distribution to better represent the shape suggested by the main peak ofthe observed data.Comparison with component surface temperatures extracted from the image suggestthat the uppermost distribution is related to roof surface temperatures, the main peak>()C:5U)U->NUCU):5crU)U-10 20 30 40 50 60Apparent Surface Temperature (° C)(a)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 139(b) 4.26: Fitted component distributions to Flight 2 (early afternoon flight over theIndustrial area). (a) using only Gaussian or exponential (extreme value) Gaussian distributions, (b) including Beta distributions.10 20 30 40 50 60Apparent Surface Temperature (° C)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 140(upper peak of the two in the second example) relates to sunlit road and the low temperature peak is a composite of the shaded horizontal and south wall surfaces. Theintermediate peaks are less easily correlated with any one component surface.Results from the application of the technique to composite distributions from Flights1 and 2 (Industrial study area) Flight 4 (Downtown) and Flight 7 (Residential) aresummarized in Figures 1.27 and 1.28.Traverse DataW S EFacet H--llI I I MeasuredRoad N < I x I ModelledImage DataNadir I I I I INorth I I • I• ISouth I I I I ModelledEast l—+-1I • ii.iiWest F--1I • ii. II •iExtrac ted Image AreasNt MeasuredW acet S facet E facetI I IsFe E/St fIt roof1I I I10 20 30 40 50Apparent Surface Temperature (° C)Figure 4.27: Summary of fitted and observed distributions applied to the morning flightover the Industrial area. Facet traverse data are means and standard deviations of thetruncated data set for each facet, road data are the means and standard deviations ofthe fitted distributions. Image data are fitted distributions to the composite surfacetemperature frequency distribution for each view direction. Extracted Image areas areobserved frequency distributions for select pixel categories.The fitted distributions (described by a mean and standard deviation) for each viewdirection are plotted horizontally in the middle of the figure. Generally, between threeChapter 4. AIRBORNE hR OF SELECTED LAND-USE AREAS 141(a) Traverse DataEFacet I )I )( II It I MeasuredRood N W f—Il—-I 41-I ModelledImoge DataNadir F—---I I •f I Ill—4---INorth f—.-—II •i e if—---tSouth f—I—-I I I I—IH ModelledEast l44—•-—f I • If—I—-4West f—I—I I II •i •1 IF—I-HExtrac ted Image Areasse MeosuredWet E/StNet S facet10 20 30 40 50 60Apparent Surface Temperature (°C)( b) TroverseData SEFacet sPJ MeasuredRood i—o—l I—IE----I ModelledImage DataNadir 14-f—•---f 14 • INW 11,11 ISE ill i. I I I I I ModelledNE K-I-4-—-f F I I f—I-—ISW I-I-fl—f I I f f-—I-—HExtrac fed Image AreasNW I eel MeasuredStriS facet150Apparent Surface Temperature (° C)(c ) TraverseDotoFacet II It I MeasuredRood F-Il—fl-Il-f- ModelledImage DataNadir I-I-I I I —I--fill F---.—!North--I F—-—ii I ISouth I--I I —I—--! I—+--f ModelledEast F4-f---—II • III IWest iii. II • 1.111 I IExtrac ted Image AreasIe N S t Measuredslocet IIItOOrlOfSI20 30 40 50 60 70Apparent Surface Temperature (° C)Figure 4.28: Same as 1.27 except for (a) the early afternoon flight (Industrial), (b) noonflight (Downtown), (c) early afternoon flight (Residential).Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 142and five distributions are required to achieve a close fit. Some “tuning” of the procedure(initial adjustment of individual distribution position and parameters) was carried outprior to allowing the fitting procedure to iterate. R2 values for the modelled curvewere always greater than 0.95 and generally above 0.99. Only Gaussian and exponentialGaussian curves were used; the use of Beta distributions requires substantial tuning of theshape to fit the curve and was generally not well related to any one component surface.Summary statistics for component surfaces extracted from select images are plotted atthe base of the figure and the means and standard deviations for facet temperaturesobtained from the traverse vehicle and road surface temperature components (estimatedvia mixed distribution modelling of the road surface temperature) are provided at thetop of the figure for comparison. These plots allow inspection to assess the agreementbetween positions of the modelled, component, and traverse distributions.The plot for the morning flight over the Industrial area (Figure 1.27) shows good correspondence between the means of the flat roof temperature and the uppermost modelleddistribution, the sunlit street temperatures and the second highest modelled distribution,and the shaded areas with the lowest modelled distribution. There is a tendency for themodelled distributions corresponding to the street and shaded temperatures to have agreater standard deviation than that of the extracted components. There is no clearcorrespondence between the facet temperature components and any of the modelled distributions; in part this is because several of the facet temperatures closely overlap othercomponent surface temperature distributions. For example, the south and east canyonfacets are very close to the shaded ground temperature, and the west facet overlaps theroof temperatures.Results from the early afternoon flight also show fairly good correspondence betweenthe roof and sunlit street components with the two warmest modelled distributions,although the modelled distribution most closely matched with the street is represented byChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 143a lower mean and larger standard deviation than indicated by the extracted component.Shaded surfaces match well for all view directions which have a significant shaded surfacearea (i.e., not including the north view direction). At this time there is a better matchbetween modelled components and facet temperature distributions for the east, west andnorth view directions. For several view directions, the number of modelled componentsexceeds that of the assumed component surface types. The additional distribution hasa wide standard deviation and has a mean value several degrees below the distributionmost closely associated with the street temperatures. Frequency Distribution DifferencesMixed distribution modelling was also applied to the frequency distribution differences.It is hypothesized that facet temperature frequency distributions should be more apparent when differences between view direction are considered. A simple two-dimensionalapproximation to the average street canyon sequence (N-S transect across a building-canyon-building sequence) in the Industrial area was used to determine the changes inthe fraction of the FOV subtended by each component surface (Table 1.3). Shadingeffects were included by using a solar zenith angle representative of the time of Flight 2.Between opposing off-nadir (north or south) view directions, there are significant changesin the total angle subtended by the sunlit and shaded components of the canyon floor, aswell as the change in the canyon facet viewed. The angle subtended by roof areas remainsconstant; however, if the assumption of flat roofs is not met, (e.g., there are structureson the roof which cast shadows), then the roof contains shaded and sunlit areas whichvary with view direction.Differences in the component angles for nadir versus off-nadir views show large changesin the fraction of wall viewed as well as changes in the area of sunlit floor (especially forthe south view direction) and the shaded canyon floor (north view direction). ChangesChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 144Table 4.3: Percentage of the FOV occupied by component surfaces of a N-S transect inthe Industrial area. Z = 36°, FOV = 12°.View DirectionSurface Nadir North SouthRoof 65.2 67.9 67.9Wall 1.1 14.9 14.9Floor (shaded) 22.6 17.2 6.6Floor (sunlit) 11.1 0.0 10.6in apparent roof area are rather small for the conditions tested. These results suggestthat at least three and perhaps four or five component distributions should be apparentin the difference distributions due to the change in apparent area of major componentsurfaces.Mean and standard deviations of the fitted distributions are presented in Figures 1.29and 1.30, along with those for the extracted component surfaces and the overall meanand standard deviation of apparent facet temperatures obtained from the vehicle traverse.The sign of the difference for the fitted distributions is represented by the symbol: anopen symbol indicates a negative difference and a solid symbol a positive difference.The number of distributions required for Flights 1 and 2 is generally 3 or 4, in agreement with that expected due to differences in the apparent component surfaces. Theidentification of facet surface temperature distributions is inferred by comparison withthe component distributions and the traverse statistics.For Flight 1, the south, west, and east facet temperatures appear to be represented.The west facets appear warmer when viewed by the airborne scanner rather than theEIRT; this is probably due to the ability of the airborne scanner to view the differentialwall areas which occur along the blocks where buildings of different heights are adjacent,Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 14510 20 30 40 50Apparent Surface Temperature (° C)Figure 4.29: Summary of fitted distributions applied to temperature frequency differencesbetween pairs of view directions for the morning flight over the Industrial area. Opensymbols for the fitted distributions indicate a negative value for the difference betweenthe view directions, solid symbols represent a positive difference.but which are not seen from the traverse route, which is limited to the exposed facetsof buildings at the ends of the blocks. The east and south facet components viewed bythe airborne scanner tend to be cooler. This may be due to inclusion of some surfacetemperatures at the base of the walls which are cooler and/or the limitation of theEIRT to view only a portion of the wall. The N/S and nadir/S differences show goodagreement between the lowest modelled difference distribution and the extracted southfacet component. Similarly, the E/W and nadir/E differences are characterized by a lowtemperature distribution which correlates well with that of the east facet component.Vehicle traverses obtain mean values for the east and south facets which are slightlyTraverse DataW S EI+——I1I Ix IxII )xIIMeasuredModelledFacetRoadImage Data— SV— NAV— SAE-WV— EAV- WExtrac[—RH I II—RH I I ModelledHRH H-RHH.H I • IHK—1H-RH F.HI—.H IF.HH-Hted Image AreasNF?t MeasuredWfacet S facet EfacetI Ise E/St fIt r’oof1HI 0 IChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 146warmer than either the modelled or extracted distributions. The uppermost distributionof the E/W and nadir/W differences is assumed to be related to west facet temperatures,but the mean is substantially lower than that of the extracted west facet temperatures,and warmer than that obtained by the ground traverse. North facet temperatures arenot as obvious in the N/S and nadir/N differences; the best match is with the secondlowest distribution of each, but these are somewhat cooler than suggested by the traverseand component surface temperatures.All the view direction difference pairs have a peak associated with the sunlit roadsurface temperature; this occurs due to the reduction or increase in apparent area of thissurface with view direction. Somewhat surprisingly, the differences obtained with theN,S and nadir directions also contain a distribution most closely associated with the roofcomponent. This may be due to different viewed proportions of sunlit and shaded roofareas, caused by roof structures or buildings of different heights, or perhaps due to thetemperature distributions on pitched roofs.The results for Flight 2, (Figure 1.30a) are in general agreement with those suggestedby the 2-D analysis. All facet temperatures except for north facets appear to have a component distribution well correlated with both the traverse and extracted temperatures.North facet temperatures obtained from the vehicle traverse and extracted from selectimages show good agreement but are not well correlated with the component distributions used to model the differences between north/south and nadir/north view directions.The reasons for this are not clear.Road surface temperatures correlate reasonably well with one of the components ineach of the view direction pairs; generally this is the second highest temperature distribution. However, traverse results suggest a component of road surface temperaturesexceeding the temperature of the flat roof component so the uppermost modelled distribution may also be related to changes in the apparent area of this component.Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 147(a) Traverse DataFacet N E S MeosuredRoad I I I ModelledW b—of--I FaImoge DataAN—S e—t F—e—---I i—.—l l—.--—ILW—N Ii • I i p—a——IILV—S i—a--i I I l——--l I—.----1ltE—W--i f—---I I—•-1 ModelledI i---i i-.--itV—W I—a- 1—a----l I-.--1 i—1Extracted Image Areasshe Efocet MeosuredVt fot E/StNE1 S facet10 20 30 40 50 60Apparent Surface Temperature (00)(b) :rse DataMeosuredRood l—e-—l ModelledImage DataNW—SE-e-- I II1W—NW F-.--l 1010111tV—SE i-a-i 1•1 S IANE—SW-a-1 U [*1—S--I I S Modelled1W-NE i-el [.I—*H F-S-H1W—SW WI-a--I [*-S--lI--S-HExtracted Image AreasNWj_cet MeosoredN , etS a et fliroofStr) Stree focet110 20 30 40 50 60Apparent Surface Temperature (° C)(c ) Traverse DataFacet N E s MeoouredRood ii II ModelledF-a-H FImage Data1N—S 1-a-I F-a-H 1 [—S--I -S I1W—N I—i [—S—I [91 Ut-a-If1V—S 101 0 I [*1 —S—-HModelled1W—E I-el-i-a—I -*1 I-S-ILW—W 101 0 I ++ i- 4Extracted ImogeAreasIçe N S st MeosoredEfat Sectfitrjof s20 30 40 50 60 70Apparent Surface Temperature (0 C)Figure 4.30: Same as Figure 1.29 except for: (a) the early afternoon flight (Industrial),(b) noon flight (Downtown), (c) early afternoon flight (Residential).Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 148More modelled distributions are required to fit the frequency distribution differencecurves from the Downtown and Residential areas than were required for the Industrialarea. Assignment of modelled components to surface types is therefore more difficult,especially where several modelled distributions lie close together and where some component surfaces (e.g., sunlit canyon facets, building roofs) may have a wide range oftemperatures which could be modelled by multiple distributions. Facet temperaturesobtained from the vehicle traverse, and by extraction from individual images, show goodagreement for the three facets not directly irradiated. These can also be relatively easily identified in the modelled distributions. Overlap between the temperature distributions of shaded road surfaces and shaded facets (refer to the extracted image componenttemperatures and traverse temperature distributions) makes distinction between thesecomponents difficult, although several pairs distinguish two or three low temperaturecomponents. Of these, the nadir-NE difference most closely matches the extracted component temperatures and has the expected sign of the change correct (i.e., the shadedroad surface percentage is higher for the nadir view and the shaded facet componentpercentage is greater in the off-nadir view).The temperature of the most directly irradiated facet (NW) shows a significant difference in the mean temperature as described by the traverse data compared with thatextracted from select images (the traverse data are much cooler). This may be due inpart to spectral emissivity differences and the fact that the traverse instrument pointsupwards (the 300 EIRT results are shown), while the airborne sensor is pointing downwards. This means that the reflected component of radiance from low emissivity surfacesoriginates from the sky for traverse data but from within the canyon (probably the road)for the airborne data. Fitted distributions for the NW-SE and nadir-NW differencesshow fairly good agreement between the uppermost modelled distribution and the NWfacet image-extracted temperature.Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 149Sunlit road surface temperatures obtained by extraction and from the traverse datacorrelate well with modelled distributions. The roof surface temperatures show a highvariability in the study area due to the complex structure of many of the building roofsand the wide variety of roofing materials used. The best match between modelled andextracted roof temperatures is with the nadir-NE and nadir-SW differences.In the Residential study area, the large number of component surfaces and their overlapping temperature distributions complicates the task of assigning modelled distributions to a surface component. The results shown in Figure 1.30(c) are from the northernpart of the study area which is characterized by a north/south block orientation.East, west and south facets show generally good correspondence between modelledand observed distributions. Road surface temperatures overlap with several other components and are somewhat more difficult to distinguish but appear to be linked withmodelled components in several of the view direction pairs. The most directly irradiatedfacet (N) again shows a large difference in the mean temperature obtained from the traverse results compared with those extracted from images. This may be due to a bias inthe selection of facets when analyzing the images and/or the inclusion of surfaces otherthan the north facets by the traverse vehicle. Where the interbuilding spacing is small,the apparent viewed area of north walls may be higher for nadir views than for off-nadir,and the sign of the modelled distribution difference which matches the traverse resultsmost closely supports this. SummaryThe use of mixed distribution modelling on image frequency distributions and distribution differences suggests some component surface temperatures can be identified. Whilea statistically good fit is obtained with a relatively small number of distributions tothe temperature frequency distributions, the apparent link between these distributionsChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 150and component surface temperatures desired is weak. Comparison of Figures 1.28 and1.30 suggest that the use of the temperature frequency differences allows better correlation between the modelled distributions and the component and traverse estimates offacet temperatures, i.e., the hypothesis holds. The results improve as the simplicity andregularity of the surface structure increases.The ability to confirm the results is limited by the differences in the view between thetraverse vehicle instruments and that of the airborne scanner, and with difficulties in obtaining a representative sample from the images for some component surfaces, especiallycanyon walls in the Residential area, and roof tops in the Downtown area. Particular difficulties are noted for several of the flights in the relation between airborne andground-based estimates of the most directly irradiated wall surface temperature.4.6 Surface Temperature Anisotropy4.6.1 Observational ResultsFor each of the three study areas, distributions of image means, taken from a set ofimages covering the study area, were constructed (Figure 1.31, 1.32, 1.33) to illustratethe anisotropy of surface longwave emissions.These results show the degree of anisotropy which may be experienced by sensorswith IFOV covering a block or more (i.e., those that sample an area which contains mostof the variance over that land-use area (Schmid, 1988)). It is assumed that each imagefulfills this requirement, so that the variation between image means is weak. As theIFOV decreases (averaging the image over a reduced subset of lines and columns), theoverall mean should become more dependent upon the position of the projected FOV onthe ground and the distributions of the means show a much wider range. Performingthis operation for several images showed differences between means of images remained>‘C)Ca)D0a)U>‘(CC0)Zrora)Ua)’C-)C0)Zrora)U-Figure 4.31: Distribution of image mean apparent temperature for nadir and opposingview directions over the Industrial study area. (a) Flight 1, (b) Flight 2, (c) Flight 3.Bracketed numbers in this and subsequent figures are the number of images used to createthe frequency distribution.151Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS(a) 28 30 32 34 36 38 40 42 44Image Mean Apparent Surface Temperature (° C)26 28 30 32 34 36 38 40 42 44Image Mean Apparent Surface Temperature (° C)0.5____Cost (38)West (35)0.4 — Nadir (88)0.20.3 A.0.10c . flu26283032343638404244Image Mean Apparent Surface Temperature (°C)26 28 30 32 34 36 38 40 42 44Image Mean Apparent Surface Temperature (ac)(c) 730 PDTNorth (39)South(41)a Nadir (80)-fiji_0. (36)West (30)fi Nadir (80)26 28 30 32 34 36 38 40 42 44Image Mean Apparent Surface Temperature (° C)26 28 30 32 34 36 38 40 42 44Image Mean Apparent Surface Temperature (0 C)Chapter 4. AIRBORNE TIR OF SELECTED LAND- USE AREAS(a)(b)>‘0CVDa152Figure 4.32: Same as Figure 1.31 except forFlight 5.the Downtown study area. (a) Flight 4, (b)>‘0Ca)D‘a)1) 0.120 22 24 26 28 30 32 34 36 38 40Image Mean Apparent Surface Temperature (0 C)20 22 24 26 28 30 32 34 36 38 40Image Mean Apparent Surface Temperature (‘a C)Image Mean Apparent Surface Temperature (0 C) Image Mean Apparent Surface Temperature (° C)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS>‘0C0)00) 71 5 PDTNorth (39)South (33)Nadir(46)24262830 3234 36384042 44Image Meon Apparent Surface Temperature (° C)Figure 4.33: Same as Figure 1.31 except over(b) Flight 7, (c) Flight (31)West (28)• Nadir (35)LA__2426283032343638404244Image Mean Apparent Surface Temperature (° C)the Residential study area. (a) Flight 6,0.40.324 26 28 30 32 34 36 38 40 42 44Image Mean Apparent Surface Temperature (° C)24 26 28 30 32 34 36 38 40 42 44Image Mean Apparent Surface Temperature (0 C)1 400 P01North (32)South (29)• Nadir (35)0.50.4>‘0C0)aa)U0-.C)Ca)Da.a)U-24 26 28 30 32 34 36 38 40 42 44Image Mean Apparent Surface Temperature (° C)24 26 28 30 32 34 36 38 40 42 44Image Mean Apparent Surface Temperature (0 C) 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 154small until the image FOV covered approximately 80 m2.Results are separated into opposing view angle directions. In each figure, the distribution of image means for the nadir view angle is also presented for comparison. Theplots have been constructed by averaging the image photon emittance (corrected for atmospheric influences) and converting to an equivalent radiant (apparent) temperature.This may be considered to be a measure of the “true” anisotropy of the surface; i.e.,that which would be observed in the absence of an atmosphere. Actual observations bysensors with a FOV equivalent to, or larger than the image size may differ because thesensor averages the radiance over the IFOV and atmospheric corrections would be applied to the apparent temperature derived from the total radiance received. Correctionsvary depending upon the spectral response of the instrument.Images containing non-standard surface cover (parks, irregular block orientations)have been removed from the analysis. Clear separation of the two distributions and theirmodes is taken as evidence of anisotropy; (i.e., the distributions are characterized bydifferent mean apparent temperatures).Differences between the means of the distributions are summarized in Figure 1.34which gives the mean temperature and standard error for each viewing direction of eachflight, and Table 1.4 which summarizes all the differences between pairs of viewing directions.The majority of differences between pairs of viewing directions were found to besignificant to a high level of confidence as described by the T-test for two samples withunequal variance (Table 1.4). Of the 80 pairs tested, 6 fail the significance test at the0.1 level, 7 at the .05 level and 13 fail the significance test at the 0.01 level.Results between opposing off-nadir viewing directions show the anticipated differences. Strong east/west differences occur in the morning and late afternoon flights andthere is little difference for the early afternoon flight. North-south differences follow anChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREASMean Image Apparent Temperature (° C)155Figure 4.34: Mean apparent surface temperature and standard error of each view direction (V = nadir, C = complete temperature) for each flight.SE NW+HøCVI I I I I I I II!II!II.II,I!I! IS E V N+Io IC WI I I I I I—ISW NEø- aC Vp i I I I I I24 26 28 30 32 34 36 38 40 42IllIIIIII ISWNE V NW-{f-H-e-I I-elEI I I I I I Iwmz0-JU-2345678[ I!I!1II111!1 ISW NWV NEIH -C SETime(P DI)1 0301400 Industrial1 7301145Downtown1 63009501400 Residential1 71 524 26 28 30 32 34 36 38 40 42IESV WM- NC NSWE V N+ NNN N NCWS NVENNCI I I I24 26 28 30 32 34 36 38 40 42Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 156Table 4.4: Temperature differences between means for pairs of view directions; differencesNOT significant (0.01 level) are italicized.FLT N-S E-W V-N V-S V-E V-W N-E N-W S-E S-W1 3.71 -3.46 -3.06 0.65 -0.06 -3.52 3.00 -0.46 -0.70 -4.172 6.68 0.98 -2.09 4.59 1.51 2.49 3.60 4.59 -3.08 -2.093 2.33 2.31 0.29 2.62 0.41 1.90 0.70 -1.61 3.03 -0.726 1.34 -5.53 -0.51 0.83 1.77 -3.76 2.28 -3.25 0.94 -4.597 5.21 0.55 2.03 3.18 1.75 2.29 3.77 4.32 -1.44 -0.898 1.43 3.74 0.56 1.99 -0.81 2.93 -1.37 2.37 -2.80 0.94FLT NW-SE NE-SW V-NW V-SE V-NE V-SW NW-NE NW-SW NE-SE SW-SE4 9.59 1.04 -6.98 2.61 1.29 2.33 8.27 9.31 1.32 0.285 2.67 7.53 1.36 4.03 -3.87 3.66 -5.23 2.30 7.90 0.37inverse pattern, (i.e., strongest differences at midday, less in the morning and later) butshow clear differences between distributions at all times, with the exception of the lateafternoon flight over the Residential area. The east-west differences are larger in theResidential area and north-south differences are greater in the Industrial area during themorning and late afternoon flights. These differences may be a result of the relativelylow east/west wall areas in the Industrial area where most of the buildings directly joinwith their neighbours. East and west walls are therefore only visible for buildings at theend of blocks and where adjoining buildings have a height differential. The preferentialE-W block orientation with alleys also increases the visible north and south wall areasrelative to the east and west walls.The images from the Residential area have not been stratified according to the blockorientation, and the strength of the observed differences with direction was somewhatChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 157surprising, given the low building height, small inter-building spacing and greater fraction of vegetated surfaces. The strong east-west differences are maintained through acombination of large temperature differences on east-west oriented peaked roofs in thesouth portion of the study area, and the differences in vertical facet temperatures whichare seen in the north part of the study area.Distributions of nadir means are generally normal, although Flight 1 exhibits somepositive skewness. This was found to be a result of images from the last flightline takenover the northern boundary of the study area which was characterized by a few large, hotbuildings in the sample images. It is also the flightline which has had the most time towarm relative to the other nadir flightlines. The position of the nadir distributions relativeto the off-nadir varies with time, due to the large temperature changes on the verticalfacets. Morning flights generally show a closer correspondence with the shaded facet, andthe late afternoon flights show better agreement with the heated facet. Midday flightsshow the nadir mean to be substantially cooler than the north view direction, but warmerthan all other view directions. Because of the dynamic nature of the temperatures of thedifferent facet orientations (e.g., see facet temperatures changes from truck traverses),these results likely differ substantially with time.In the Downtown area, the morning flight shows the most directly irradiated facet tobe most different in temperature from the facets seen in the other three view directions,the most shaded of which are very close in temperature. The afternoon flight shows agreater separation between the temperatures of each view direction but a reduction inthe difference between the warmest view direction and nadir.As an example of observed anisotropy by a satellite sensor, the analysis for Flights 1—3were repeated using channel 4 of the NOAA-11 AVHRR. Image apparent temperatureswere corrected to true apparent temperatures using the AGEMA LUT relations derivedin Appendix B and subsequently converted to apparent surface temperatures using aChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 158calculated LUT for the satellite sensor. Results are shown in Figure 1.35.The effect of the atmosphere is to reduce the apparent temperature differences. However, significance as calculated by T-tests remains identical because the same transformation is applied to each view direction and the ratio of variances remains constant beforeand after the transformation.4.6.2 Comparison with Other SurfacesThese observations constitute the first attempt to directly measure the anisotropy ofthermal radiation over an urban surface. Prior investigations have studied the anisotropyat local scales over agricultural (row crops) or natural surfaces (forest canopies) and atlarger scales in mountainous terrain (c.f. Chapter 1). Comparison with Table 1.2 showsthe observed variation with view direction over cities is large relative to the natural andagricultural surfaces listed, being similar in magnitude to those reported for an oak-hickory forest and a sunflower crop. Two major factors are responsible for the degreeof observed anisotropy: the regularity of the surface structure, and the creation of largetemperature differences on the different surface components viewed by the sensor.Anisotropy for vegetated surfaces without a regular geometric surface structure alsocan be large, when canopy cover is incomplete and the sensor IFOV includes portions ofvegetation and soil which change with sensor viewing angle (more vegetation is seen atlarger zenith angles, i.e., longer paths through the canopy).In urban areas, anisotropy is dominated by the regular geometric structure of the surface. The anisotropy arises due to temperature differences on the vertical facets generatedby differences in absorbed solar radiation and thermal properties and due to variationsin the amount of shaded and sunlit horizontal surfaces viewed. Based upon these considerations, anisotropy should be maximized in areas of the city where the block andbuilding orientation is very regular, the area of vertical facets is large, and considerableChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREASLImzH0-JU-Air from Nadir (° C)1400CD-alAflfl159Figure 4.35: Calculated mean apparent temperatures and standard errors of each viewdirection at the TOA for NOAA-11 AVHRR Channel 4. Flights 1—3 only.1 0301 0302233SVE NWFe-H-edI I I • I II•I•I•I’I’III• ISE NWVI I I I I I I I—4 —3 —2 —1 0 1 2 3 4rIS W V NF-eH HH FeI I I • I I II•III•I•I•II• IS WV NI-8H feEI I I I I • I I—4 —3 —2 —1 0 1 2 3 4I•I•I•II•II•I IS W NVEtel FeiFteteI I I I I I •I•III’I•I•I•I ISW NEVI I I I1 7301 730—4 —3 —2 —1 0 1 2 3 4Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 160differences in the amount of shaded and sunlit horizontal surfaces with view direction canbe observed. The results from the three study areas confirm the Downtown area has thelargest apparent temperature differences with direction, although because of the differentstreet orientation, direct comparison with the other study areas is not straightforward. Amitigating factor in both the Downtown and Industrial areas is the close spacing of buildings which limits the view of wall areas and/or cast shadows onto facets which wouldotherwise be directly irradiated. These effects reduce anisotropy. The spacing of thebuildings is therefore an important factor in the creation of large apparent temperaturedifferences.In the Residential area, small scale structural features (peaked roofs) appear to enhance anisotropy which otherwise was expected to be relatively small because of smallerwall areas relative to horizontal areas, and reduced regularity in the block and buildingstructure. The occurrence of large vegetative elements in the Residential area enhanceanisotropy because they cast shadows and because the tree crowns have spatial temperature variations associated with the receipt of solar radiation. Image analysis showstree crowns are warmer by approximately 2°C when viewed on the side receiving directsolar radiation compared with the shaded side. Similar results were obtained by Balicket al. (1987) over a mixed deciduous forest canopy. Like buildings, the spacing of thetrees is probably important if the tree crowns are to develop temperature differences andground-shaded areas remain distinct. As the canopy closes a more uniform top surface ispresented, the temperature differences are reduced, and shaded areas beneath the canopyare obscured.The presentation of the frequency distribution of image means (Figs. 1.31— 1.33) andthe tabulation of the difference between the overall means by direction (Table 1.4) providea more conservative statistical measure than the use of simple maximum differences withdirections often reported for other studies (Paw U, 1992). An estimate of the maximumChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 161difference to be expected can be obtained from the frequency distributions. As observedin previous studies, differences vary with time and sensor orientation. In contrast torow crops, which show a single temporal maximum, urban surfaces with street canyonsoriented in orthogonal directions show large differences at two or more times during theday, depending upon the street orientation and sensor view direction.Differences due to the variation in viewing angle are restricted to nadir and 45° off-nadir comparisons, therefore no comment can be made regarding the angle at whichdifferences are maximized from the nadir. This angle depends on the local surface geometry and the solar position as it affects the pattern of surface heating. Qualitatively, it isanticipated that with increasing angle from the nadir when viewing a shaded facet, themean temperature will drop due to an increase in the view of the shaded wall and loss ofview of a portion of the canyon floor. The image mean temperature will likely decreaseuntil the view of the shaded wall/floor area becomes restricted, at which time the meantemperature should rise again. For a view direction towards a sunlit facet the trend willbe reversed.4.6.3 Scale of AnisotropyThe observed biases are a function of the resolution of the sensor. As described by Lipton(1992) for a satellite sensor viewing mountainous terrain, the observed bias will increaseas sensor resolution decreases and the proportion of steep slopes viewed increases. Thelimiting case suggested for a sensor where the IFOV consists entirely of terrain made up ofparallel ridges aligned NE/SW was estimated to be 19°C during the morning. Increasingproportions of terrain with different slope orientations decreases the observed anisotropy.The local control on the anisotropy is the temperature difference between the differentslope orientations of the terrain as determined by the energy balance of those surfaces(higher for less vegetated, drier surfaces).Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 162In urban areas, anisotropy is also conditioned by surface structure. In this casethe regularity and relative extent of vertical (or sloped) surfaces to horizontal surfacesis important (e.g., proportion of wall or sloped roof surfaces to horizontal surfaces).Anisotropy is reduced as irregularity in the surface structure (e.g., street pattern) increases, so proportions of cities characterized by twisting streets are expected to showless anisotropy than those with a more regular street pattern. Increase in the verticalextent of buildings increases anisotropy because a greater proportion of surfaces havelarger temperature differences. It is assumed the remote sensor has an IFOV sufficientto view an area large enough to contribute most of the temperature (and by associationsurface structural) variance. It does not make sense to define the anisotropy at verysmall scales: high resolution instruments, will, in the limit, resolve individual surfaceelements and the variations in surface temperature become visually apparent (as in thethermal images presented here). Successive degradation of images showed evidence ofthe warm roofs and cool shaded areas at pixel sizes of up to approximately 6 m2 (a peakin the local variance occurs; see Woodcock and Strahler (1987)) which is indicative ofmaximum variation in the temperatures of adjacent pixels. As pixels become larger thanthis size, they begin to average across the scale of maximum temperature differences andthe variance between adjacent pixels decreases. Of importance is the anisotropy at scaleswhere the surface structure is unresolved by an individual IFOV.It seems likely that there will exist a range of scales (surface area) for which theanisotropy remains constant or very nearly constant; these correspond to areas of similarsurface structure which generate a particular frequency distribution of temperatures; i.e.,homogeneous areas. The lower bound of this range of scales is probably no larger thanthe scale suggested by Schmid (1988) to be representative of the surface temperature andstructural variance (200 m diameter in a Residential area similar to that studied here). Itmay be substantially smaller. Schmid (1988) found the main contribution to the spatialChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 163variance of surface temperature occurred at 25 m (street/alley to house row spacing) and50 m (street to alley spacing).Anisotropy is expected to remain relatively constant as scales increase up to the pointwhere the IFOV begins to cover different land-use areas (and/or different surface structurepatterns) at which point it is expected that compensation of directional variations shouldbegin to take place. The similarity of single image temperature frequency distributionswith that for all images within a study area lends some support to the hypothesis thatanisotropy is constant over the range of scales encompassed by a single image (150 m2)to that of the entire study area (1 km2).4.6.4 Anisotropy Relative to Other Influences Upon Remotely Sensed Surface TemperatureThe directional temperature variations observed in this study are compared with themagnitude of other factors which influence a remote measurement of surface radiativetemperature; atmospheric absorption, and surface emissivity (Table 1.5). Both factorsmay show spatial variability in cities.In their consideration of errors, Roth et al. (1989) suggest spatial temperature errorsof up to 1.5 K are possible due to variations in urban-rural surface emissivity. Horizontalvariability in atmospheric water vapour and pollutant concentrations is suggested to yielderrors of 1 K (Carison, 1986).It is worth noting that the assumption that atmospheric corrections need not beapplied to determine urban/rural temperature differences is not true; calculated LUTcorrections show strong variations with temperature. Appendix B indicates correctionsof —1—+8°C are required over the range of observed surface temperatures for the AGEMAscanner (8—12pm). Applying atmospheric corrections to Figure 1 of Roth et aL(1989)Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 164Table 4.5: Magnitude of temperature differences produced by processes affecting remotethermal measurement of an urban surface.Process Magnitude (°C) CommentsSurface 2—2.5 average urban emissivity of 0.95emissivity (Arnfield, 1982)Atmospheric 4 — 7 mid-latitude summer atmosphereabsorption/emissionAnisotropy up to 10 45° off-nadir, coinciding withup- and down-Sun directions(Vancouver daytime heat island; NOAA- 9 AVHRR Channel 4) for the standard mid-latitude summer atmosphere (contained within the LOWTRAN program) increases theapparent surface temperature by 4 K in rural areas and up to 7.5 K in the heat islandcore. Urban - rural surface temperature differences are therefore enhanced due to thesecorrections; in this example an increase from 11 to 15.5°C is calculated. Errors dueto anisotropic emissions will likely vary on a pixel by pixel basis depending upon thesurface structure and relative sensor position, and can be large within individual land-use areas. In industrial/commercial areas where daytime surface heat island values arelarge, the observations here suggest directional variations of up to 6 K are possible (Table1.4). The effect of anisotropic emissions on the determination of heat island magnitudes(assuming a consistent, regular surface structure throughout the urban area, and a fiatisotropically emitting rural surface) can lead to either a reduction or enhancement ofthe heat island signal, depending upon the time of day and direction of view. Thisis potentially important for studies which use the maximum Tu_r value in furtheranalyses.Chapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 165The relative magnitude of errors will be scale dependent. The scale dependence ofanisotropy has already been noted. Emissivity differences are maximized at scales on theorder of metres, and decreases with averaging across a ground resolution element (GRE)of the remote sensor. As GRE decreases the relative error in emissivity will increase assmall, low emissivity areas become more important. Atmospheric variability is small andprobably operates at scales on the order of hundreds of metres to a few kilometres.Consideration of these effects from an operational perspective indicates that the effectof anisotropy is likely most important for aircraft-based sensors which combine a smallIFOV and large range of off-nadir scanning angles to yield pixels for which anisotropy islikely to be high.For the commonly used satellite-based thermal scanners employed in studies of theurban heat island (Table 1.6) only the NOAA series of satellites incorporates a largeoff-nadir scanning capability which can be affected by anisotropy. These polar-orbitingsatellites have ground tracks of roughly NNE-SSW (descending node) or SSE-NNW (ascending node), and scan at right angles to the sub-orbital track. Under these conditionsanisotropy for daytime, early afternoon passes is likely to be relatively minor in areaswith a preferred N/S-E/W block structure, but may be large for areas where the blockstructure is rotated 45° (e.g., Flights 4—5 over Downtown Vancouver), although the effectof anisotropy is probably reduced due to the large GRE of these sensors at off-nadir angles. The morning passes made by NOAA-6,8,1O and 12 may be subject to the influenceof anisotropy during the summer at higher latitudes, where east-facing facets are exposedto direct irradiance early in the morning. Results from the vehicle traverses indicate thedevelopment of significant temperature differences by 0730 LST between east and westfacets during the study period which was almost 2 months from the summer solstice.For satellite-based sensors which have restricted off-nadir capabilities, anisotropy willnot be significant. However, the apparent surface temperature must still be reconciledChapter 4. AIRBORNE TIR OF SELECTED LAND-USE AREAS 166Table 4.6: Satellite parameters for satellites commonly used in thermal analysis of urbanareas. Ascending and descending node times in LAT (Local Apparent (Solar) Time).Satellite Ascending Descending Scan Angle IFOVNode Node (nadir)TIROS-N 1500 0300 ±57. 1.1 kmNOAA-6,8,10,12 1930 0730NOAA-7 1430 0230NOAA-9 1420 0220NOAA-11 1340 0140Landsat (TM) 0945 ±7.7 120 mHCMM 1330 0200 ± 600 mwith the fact that only a limited subset of the complete urban surface is viewed sothat the observed temperature is not necessarily representative of the complete surfacetemperature. This effect is considered in Chapter 5.Chapter 5THE COMPLETE URBAN SURFACE TEMPERATURE5.1 Definition of the Complete Urban SurfaceSurfaces may be represented in meteorological applications by a variety of forms as discussed in Section 1.3. The actual surface which forms the boundary between the atmosphere and the substrate is often very complex. Surface representations simplify andapproximate the actual surface, often using surfaces “seen” by a sensor, or planes ofobservation which coincide with the measurement level of a sensor (Figure 1.1). Thesurface representation adopted generally is scale dependent; details of the surface structure are increasingly simplified as the total area increases. The surface representationadopted in this thesis includes major surface structural features such as buildings andtrees (including sizable shrubs), and horizontal surface categories such as roadways andgrass. Component areas are defined in Figure 5.1 and Table 5.1.Figure 5.1: Illustration of component areas.A0167Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 168Table 5.1: Definition of area component symbols.Symbol Description4, Plan (horizontal) area (obtained from map or air photo);also the area as seen by a nadir-pointing remote sensorA0 Horizontal ground-level area (grass, roads, gardens)Ar Roof area (actual)Apr Plan or apparent roof area: for flat roofs Ar = Apr;equivalent to building plan areaA Wall area (total or with an additional subscript denoting facet direction)Ab Building area (Sum of roof and wall areas)A1, Vegetation area (three-dimensional tree representation)Plan or apparent vegetation area (horizontal projection)(Subdivided into trees with a trunk heights (htk) = 0, > 0A Complete surface area (Sum of A0, A,,, A,,)In general the surface representation does not include details at length scales less thanthat of a building or tree. Landform relief is ignored.5.1.1 Estimating the Complete Surface Area5.1.1.1 BuildingsThe three-dimensional area of buildings (A,,), made up of roof area (Ar) and wall area(A,,,) was calculated using building outlines digitized from high resolution (1:2500) aerialphotography. Large roof-top structural elements such as elevator shaft housings wereincluded but structural details with length dimensions less than one half the shortestfacet were omitted. Roofs may be planar or consist of 2 or 4 angled surfaces.The area A, of a polygon represented by a series of x, j points (i = 1, N) can beChapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 169determined from Stoke’s theorem as1N-1A=Xjj1 — yjXji) + XNY1 — X1YN (5.1)i=1where the zy values are the digitized points for each building or surface object. Verticalfacet areas were obtained by calculating the length of a vector (side of the polygon) andmultiplying this by the assigned height of the building. Where adjacent buildings sharea common wall, common vectors were identified. If the heights of the buildings differ,the area of the exposed wall was calculated. Wall orientation was derived from the signof eq. 5.1 in combination with the slope (determined from the coordinates of the pointsrelative to a reference point) of the line representing the wall segment (a slope near 0indicates a north or south wall, a very large slope indicates an east or west wall). TreesTrees were digitized as points using air photo analysis and ground surveys. They wereincluded only in the Residential study area; the Downtown and Industrial sites are largelydevoid of trees. Trees were categorized into 5 types, based upon geometric form andrelative abundance: B - bushes, C - evergreen (coniferous), D - broad-leaved (deciduous),E - evergreen (non-coniferous), F - flowering deciduous.For surface area calculations, trees have been represented by cones (coniferous) (e.g.,Li and Strahler, 1985), spheres or cylinders (deciduous), (Jupp, Walker and Penridge,1986; Goel, 1988) or, more generally by ellipsoids (Campbell and Norman, 1989) for surface area calculations. A wide variety of tree forms may be represented by ellipse parameters, if the possibility of truncated ellipses is included (Charles-Edwards and Thornley,1973; Goel, 1988). Comparison of surface areas calculated using ellipsoidal representations versus those based on cones, cylinders and spheres (assigned to representativetree types), yield total tree surface area estimates within approximately 4%. EllipsoidalChapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 170hhrFigure 5.2: Definition of tree structural parameters.representations are used for the estimates in this chapter.Tree height (hi), maximum crown radius (re) and height to the base of the foliage(equivalent to trunk height htk) were estimated from ground surveys for each of the treesin the study sub-area. The position of the ellipse centroid (height of maximum crownradius hr) was estimated as a fraction (Fhf) of the total foliage height (hf) wherehf = — htk. (5.2)Fhf was estimated to be 0.25, 0.1, 0.5, 0.25, 0.1 for types B,C,D,E, and F respectively.Tree structural parameters are graphically portrayed in Figure 5.2. The ellipse semi-axes are represented by c and r (equivalent to a). When z = c the tree canopy isrepresented by a complete ellipse; when z < c the ellipse is truncated.Calculated shapes for selected examples of each tree type in the Residential studysub-area are presented in Figure 5.3. Crown radius is assumed to be symmetrical (i.e.,the crown circular) in the zy planes; the tree shapes are elliptical in the xz plane only,and may therefore be more precisely described as spheroids.hfh tkChapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 17115mB CDEj2Figure 5.3: Modelled shapes for the tree types in the study area. B- Shrubs/Bushes, C- Coniferous, D - Deciduous, E - Evergreen (non-coniferous), F - Flowering (Cherry).The representation of trees as simple geometric objects fails to account for gaps inthe foliage which reduce the projected surface area. The actual “viewed” or apparentsurface area directly emitting in a particular direction is theoretically defined as theprojection of the total canopy foliage onto a plane orthogonal to the direction of view.Lang and McMurtrie (1992) describe the theoretical basis for the commonly requiredcase of foliage projected onto a horizontal plane below the canopy. Chen et al. (1993)present projections based upon computer simulations.The complete surface area of a tree canopy (An) may be defined as the area of foilageprojected onto the bounding surface of the geometric shape representing the tree. Thisis the area of foilage which emits directly to the surroundings. The ratio (Fgap) betweenA,, and the area of the bounding geometric shape defines the reduction factor required toaccount for the gaps in the canopy foliage. Where available, Fgap allows the calculation ofthe complete canopy area from the simple geometric area, which can be estimated frombasic structural parameters. Calculation of Fgap is theoretically difficult and requiresdetails of the canopy foliage density, orientation and clumping beyond the scope of thisstudy.Observational alternatives exist (see e.g., Lang (1991) for a discussion), and thoseChapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 172employing photographic techniques appear to be most appropriate for this application.To obtain the projected area of the entire tree they need to be extended to side-views; planand side views of some trees are hypothesized to show significant differences in Fgap. Inthis work, a very crude approximation is made for Fgap: based upon field observations,estimates of 0.15, 0.2, 0.3, 0.2, and 0.45 were made for tree types B—F, respectively.Where individual trees differed significantly from the average type value, an estimatedfield value replaced the default value.Values of Fgap obtained from the literature often refer to canopies of tree types ratherthan single trees, and are generally based upon a cumulative projection of LAI ontothe horizontal plane below the canopy. These values therefore do not account for theanticipated variations in Fgap for projections in the vertical plane. Fgap estimates formodel poplar stands based upon downward cumulative leaf area index (Chen et aL, 1993)were in the range of 0.3 - 0.2 for a deciduous LAI of 4, the estimated maximum in thestudy area (Grimmond, 1988; Kramer and Kozlowksi, 1979). Gap estimates for lodgepolepine are also reported as 0.2-0.3, (Sampson and Smith, 1993), but these estimates are ofgaps between individual tree canopies rather than gaps within any particular tree canopy. HorizontalThe area of horizontal surfaces (A0; grass, roads etc.) is determined as a residual from:A0 = A— (Apr + Apv(htko)) (5.3)where A is the total plan area (calculated from a topographic map), Apr is the plan areaof roofs (equivalent to buildings), and Apv(hko) is the plan area of vegetation for whichthe canopy interesects the ground (most common for bushes but also occurs for someconiferous trees). A0 includes horizontal surfaces below tree canopies which have htk > 0(Figure 5.1).Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 1735.1.1.4 Top-of-Canopy Plan AreaThe apparent plan area which is visible to a nadir-pointing remote sensor is the sameas the map plan area, i.e., it is the horizontal projection of surface components onto ahorizontal plane above the tallest structure. It is denoted by A,, and is the equivalent ofthe bird’s eye view of the system (Figure 1.ld). In this projection, lower surfaces areobscured by parallel upper surfaces. The plan area generally includes the building planarea, the vegetation plan area, and that portion of the horizontal area not obscured bytree canopies (i.e., A0— Apv(hk>o)). Complete Surface AreaThe complete surface area A is estimated by adding the three-dimensional areas ofvegetation (Au) and buildings (A,,) to that of the horizontal area (A0):A=A+Ab+A0. (5.4)5.1.2 Complete Surface Areas of the Study Sites5.1.2.1 IndustrialComponent surface areas calculated from the surface structural database (Figure 5.4) arepresented in Table 5.2. Building heights were estimated in stories (to the nearest 0.25)with a story assigned as 3.66 m (12’). The building height frequency distribution is shownin Figure 5.5. A total of 733 buildings is included. However, many of these are adjoiningbuildings with common walls. Heights are distributed approximately normally, with onlya few instances of tall (>4 story) buildings. The mean building height is estimated to be7.3 m with a standard deviation of 3.2 m.Combined wall areas constitute 29.4% of the complete surface area and horizontalsurfaces (including roof tops) make up the remaining 70.6%. Greater areas of north- andChapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 174Table 5.2: Major surface component areas and percentages of the complete surface areafor the Industrial area.A A0 Ar A(N) Am(s) A(E) A(w) A(M) A A/AArea 617.5 380.2 237.3 68.76 69.77 55.38 54.46 8.68 874.5 1.42(m2 x 10)% of A 70.6 43.5 27.1 7.9 8.0 6.3 6.2 1.0INDUSTRIAL BUILDING HEIGHTS (STORIES)- —— 1_______200m—— S!05 -1 49___r I I: i i— 2.5 _ e 4I 1w_s I—3 a—-—— 3.5bI Si’: I1I’P—5- U • 14:: --— 8 —______Pa,*_____ _____N__W9brIIIsIII- —— ‘_ ___.aFigure 5.4: Building heights (stories); Industrial study area.Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 1750.4Industrial Areaz 0.20/yU-0.1 ///0.0------- 4 4 , .0 4 8 12 16 20 24 28 32Building Height (m)Figure 5.5: Frequency distribution of building heights in the Industrial study area.south-facing facets are exposed compared to east- and west-facing ones because many ofthe buildings along the blocks share common east and west walls and therefore have noeast or west exposure (Figure 5.4). There are relatively few trees in the study area; theseare not included in the surface area calculations. DowntownThe Downtown area represents the maximal extent of surface vertical structure in anurban area. Individual buildings may exceed 100 m in height (Figure 5.6). Heightsfor the 274 digitized buildings were obtained from City of Vancouver (1984) planningmaps (in stories) updated by field observations of height estimated using Abney levelsfor buildings along the traverse route. Where heights in stories were used, conversion tometres assumed 3.66 m story’.Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 176Table 5.3: Major surface component areas and percentages of the complete surface areafor the Downtown area.A A0 Ar A(NE) A(sw) A(Nw) A(sE) A A/AArea 453.3 286.8 166.5 138.1 136.9 133.5 133.3 995.1 2.20(m2 x iOu)% of A 45.6 28.8 16.7 13.9 13.8 13.4 13.4The distribution of building heights (Figure 5.7) shows a strongly asymmetric distribution with greatest frequencies in classes centred between 5 and 15 m, and a long tailof frequencies extending towards greater building heights. The mean building height is26.6 m, whilst the median height is 14.6 m.Complete surface calculations have been restricted to areas of NW/SE and NE/SWstreet orientation (Figure 5.6) to conform to the traverse route and reduce the numberof extracted facet temperature distributions required. Here, vertical facets combine toform 58.4% of the complete area, a value greater than the fraction of horizontal surfaces(41.6%) and much greater than the horizontal roof area (16.7%, Table 5.3). The variationof area with facet orientation is much less than for the Industrial area, but shows slightlygreater values for the NE/SW facets. Trees are omitted from surface area calculations. ResidentialThe Residential study area is characterized by one and two story single family housesgenerally arranged with the long axis of the building at right angles to the street (Figure 5.8). Garages and ancillary buildings are located at the back of the lot facing thealleyway. In the 8 block sub-area chosen for detailed analysis, 271 houses (an average of34 block—’), and 139 garages (approximately 17 block’) were digitized. The height (inChapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 1770S.•..I III•LIIstories, to the nearest 0.25, using 3.05 m story’), roof type, and pitch were estimated foreach building. Frequency distributions for the estimated building structural parametersare shown in Figure 5.9. Almost all garages are included in the 3 m height class (1 story)as might be expected. House wall heights are concentrated in the 5 and 6 m categories.Roof pitches for both building types are most frequently estimated to be between 10 and20 degrees. The distribution is skewed, with substantial numbers of both houses andgarages observed to have higher roof pitch angles. Roof types are fairly equally splitamong gabled and 4-sided types with only a minor proportion having fiat roofs.The mean building (house) height was 5.5 m, and the mean roof pitch was approximately 20°. This estimate is low in comparison to a previous estimate for the area (8.5m, Steyn 1980), but note that the current estimate is actually the height of the verticalU 10011? *1‘IIi’øiirFigure 5.6: Building heights (stori.es); Downtown study area.Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE0.>C-’ 0.12C0.1000.08U- 5.7: Frequency distribution of building heights in the Downtown study area.walls, rather than the complete building height. For non-flat roofs the building heightincreases in relation to the tangent of the roof pitch. Using the average pitch angle and anestimated building width of 10 m, an additional 1.8 m may be added to obtain a buildingheight estimate. Vertical facet surface area calculations for buildings with gabled roofsinclude the gables (triangular wall area above the level of the eaves on the end walls).Houses may be grouped into five major categories, (Table 5.4) according to their height,roof pitch and general layout. The most frequently occurring house type (37%) is the“Vancouver Special” a rectangular, 2 story building with a shallow (< 25°) roof pitch ona gabled (two-sided) roof (Type II).Garages have a mean wall height of 3.1 m and a mean pitch angle of 20°; adding in atypical roof height yields a total height estimate of 4.0 m, which compares favorably tothe 3.5 m estimate of Steyn (1980).In contrast to the Industrial and Downtown sites, large numbers of trees are present.0 15 30 45 60 75 90 105 120 135 150Building Height (m)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 179Table 5.4: Building types in the Residential study area.House TypeParameter I II III IV VHeight (stories) 1.5—2.5 1.5—2.5 1.5 2—3 misc.Bldg Layout rectangular rectangular square rectangular miscRoof pitch (°) 15—25 15—25 15—25 > 25 misc.Roof sides 4 2 4 2 misc.Number 62 101 39 53 16Frequency (%) 22.9 37.3 14.4 19.6 5.9- Garage TypeParameter I II III IVHeight (stories) 1 1 1 1Bldg Layout rectangular rectangular rectangular rectangularRoof pitch (°) 0 0—25 > 25 > 0Roof sides 0 2 2 4Number 12 33 33 60Frequency (%) 8.7 23.9 23.9 43.5Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 180Figure 5.8: Building heights (stories); Residential study area.Most streets have regularly spaced trees (street trees) along the roadway. Trees are alsopresent in backyards and along alleyways, although less frequent and more randomly located. A count of the trees and large shrubs/bushes in the 8 block sub-study area yielded385 trees and 41 shrubs. These were subdivided into shrubs/bushes (B), coniferous (C),deciduous (D), evergreen (non-coniferous) (E) or flowering deciduous (F) categories. Frequency distributions for the four most frequently occurring types (very few examples oftype E are present) are presented in Figure 5.10. Descriptive statistics of the estimatedtree dimensions are presented in Table 5.5Deciduous trees are the most frequently occurring type, accounting for 59% of theRESIDENTIAL BUILDING HEIGHTS (STORIES)1.25— 1.5— 1,752.252.5• 2.7550mChapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE>‘C)Ca):30G)C)>‘C-)Cci)30a)C)-181ResidentialHousesGoroqes(a)0.60.5a.0.4C)- 0.60 1 23456Wall height (m)7 8 9 15 25 35 45 55 65 75Roof Pitch (Degrees)(c) 2 4Number of Roof FacetsFigure 5.9: Frequency distributions of estimated building structural parameters (Residential area). (a) height, (b) roof pitch, (c) number of roof pitch azimuths; (0 = flat).t—93 4 5 6 7 8 9 10Max. Crown Radius (m)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 1820.(a) (m)>‘UC0U-0.01 .0(b)0.80.7>‘0C0.5a0) Area—e-— Bushes--& Coniferous—c-- Decidiuous0- — Flowering01 21 .>‘0C33):3a31)U-—e-— Bushes•• Coniferous—LI— Decidiuous- Ø- — Flowering0.10.00 1 2 3 4 5 6 7 8 9 10Trunk Height (m)Figure 5.10: Frequency distributions for the estimated tree structural parameters (Residential area). (a) height, ht; (b) maximum canopy radius, r0; (c) trunk height, htk (heightto base of canopy foliage).Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 183Table 5.5: Statistical summary of estimated tree structural parameters.Tree h (m) r (m) htk (m)Type n mean a med. mean o- med. mean a med.B 81 2.47 1.20 2.44 1.07 0.43 0.91 0.09 0.28 0.00C 78 8.55 4.13 9.14 3.09 2.18 2.44 1.77 1.73 1.37D 289 7.73 3.18 7.62 3.41 1.61 3.05 3.16 1.81 3.05F 43 6.24 1.81 6.10 3.00 1.31 3.04 2.74 1.51 2.13Table 5.6: Major surface component areas and percentages of the complete surface areafor the Residential area.A A0 Ar Apr A(htk = 0)Area 170.0 121.3 52.73 48.23 51.75 13.02 22.90(m2 x i0)% of A 55.5 39.6 17.2 15.8 16.9 4.3 0.2AW(N) Am(s) A(E) A(w) A A/A22.90 22.94 17.20 17.23 306.1 1.807.5 7.5 5.6 5.6total. The overall mean tree height (not including type B) is 7.73 m, which is approximately 50% of the height adopted by Schmid (1988). This difference is attributed tothe small spatial domain and abundance of relatively small street trees which make up alarge fraction of the total number of trees.The component surface percentages are presented in Table 5.6. A two-dimensionalprojection of the canopy area, less the 2-D area of trees where foliage intersects the groundyields the obscured horizontal surface area (12547 m2), 4.1% of the complete area).Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 1845.2 A Complete Urban Surface Temperature5.2.1 DefinitionsThe complete urban surface temperature (7) is an area-weighted temperature. Thecomponent surface temperatures are combined in proportion to their areal fraction ofthe complete surface. In the recently proposed terminology of Norman et al. (1995),the complete surface temperature is similar to the mean kinetic temperature, with theexception that the complete surface temperature used here is a brightness temperatureuncorrected for surface emissivity, rather than a thermodynamic temperature.In radiance form, the complete surface temperature, as defined above, is expressed as(5.5)for n component surfaces, where f are the fractional areas with emission L, and Lis converted to an equivalent temperature (Ta), i.e., assuming blackbody emission. Lrequires the specification of surface classes, their representative temperatures and thefractional area of each surface class. T is not directly observable, although it may bepossible to approximate it using hemispherical (or, preferrably, wide FOV) estimates ofupwelling longwave radiation. These may provide a useful approximation because theyreduce the directionality of the measurement and integrate both horizontal and verticalsurfaces of all orientations. However, view factors for individual surfaces will be biasedtowards horizontal surfaces and those surfaces most directly beneath the sensor.Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 1855.2.2 Estimating the Complete Surface Temperature5.2.2.1 Nadir Airborne and Traverse Temperature DistributionsOne method of estimating T is to combine the apparent surface temperature distributions obtained at nadir with the airborne scanner with the vertical facet distributionsobtained from the vehicle traverse. This horizontal/vertical combination circumventsthe need to subdivide the horizontal surface into component fractional areas and estimate mean temperatures or temperature distributions for each. The observed nadirtemperature distribution is assumed to represent the various components in their correct proportions. The component frequency distributions (nadir and four vertical) arecombined with weights according to their fraction of the complete surface area.Figure 5.lla illustrates the component distributions for the morning flight over theIndustrial area, where vertical facet distributions are from the vehicle traverse. Theresultant composite distribution is presented in Figure 5.llb.The mean temperature of the complete distribution is estimated using eq. 5.5, exceptf, are the frequencies for each temperature class (rather than fractional areas of emission)and L is the radiance for that temperature class. A disadvantage of this approach is theneed to truncate or otherwise modify the distribution of surface temperatures obtainedfrom the vehicle traverse to remove mixed building and sky values and the likelihood thatthe resultant distribution slightly underestimates areas of high surface temperature. Themethod also assumes that the distribution of vertical facet temperatures is representativeof all the vertical surfaces. The traverse methodology restricted observation to facetswhich may be viewed from positions along streets or alleyways and within the range ofelevation angles of the EIRT. Facets orthogonal to the street are not sampled excepton the ends of the block; the close inter-building spacing in the study areas means thatthese facets have a greater direct solar exposure on the ends of the blocks (where theyChapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE(a)1 030 PDTNadir (Airborne)North FacetsSouth FacetsEast FacetsWest Facets186Figure 5.11: (a) Component surface temperature frequency distributions for the Industrial area, Flight 1 (1030 PDT). Vertical facet distributions are from the vehicle traverse.(b) Composite temperature distribution.0.50.4>‘U 0.3 HCci):3L0.20.1(b)> 0.06C-)C fl5ci):3a- 0.04ci)L 30 40 50Apparent Surface Temperature (0 C)6020 30 40 50 60Apparent Surface Temperature (° C)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 187are sampled) than for similar facets within the block. Calculations for the Residentialarea indicate 57—80% of the area of inter-building walls are shaded during the morningand late afternoon flight, depending upon the building height and spacing. During theearly afternoon flight, 35—55% are shaded. Nadir and Off-nadir airborneAn alternative to the use of the vehicle traverse data is to use vertical facet surfacetemperature distributions extracted from the off-nadir airborne scanner imagery. Theseovercome some of the difficulties with the traverse vehicle data, but have their ownlimitations. When using the extracted data, results are dependent upon the samplingof the areas; recall that temperature patterns are the sole means of defining the spatialdimensions of the facets (c.f. Chapter 1). In the Residential and Downtown areas,temperatures of facets where inter-building spacing is small may be difficult to obtaindue to the off-nadir angle used and the small area of these surfaces in the direction ofview.Figure 5.12 compares the vertical facet temperature distributions obtained from thetwo sources for the morning flight over the Industrial area. Note the underlying facetorientation convention: east facets face east. East facets are, however, viewed by a remote sensor with a west view azimuth (the convention used in the presentation of theremotely-sensed temperature frequency distributions). The vehicle traverse data combines EIRT observations at both 0 and 100 elevation angles, and has not been truncated.The agreement between the distributions is generally good except for the east facet (thedirectly irradiated facet) where the image-extracted temperatures are much warmer. Thisis probably due to sampling biases relative to the building geometry: east facets at midblock have greater solar access earlier in the morning than do the end-of-canyon walls,which are more subject to shading by buildings on the opposite side of the street, andChapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE(a)>‘U0a,:3aa,(c)C,C1)Saa,188(b) 0.220.20 South0.18 Vehicle Traverse0.1 6 :: 1 - —— — Imoge xtrocted0.140.12002Apparent Surface Temperoture (°C)(ci) 0.30West000Figure 5.12: Traverse and image extracted vertical facet temperature distributions; Industrial area, Flight 1 (1030 PDT). Facet orientations are: (a) N, (b) S, (c) E, (d) W.Apparent Surfoce Temperature (0 C)Apparent Surface Temperature(0C)20 30 40 50Apparent Surface Temperature (00)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 189thus mid-block facets are warmer.5.3 Comparison of Complete and Incomplete Surface Temperatures5.3.1 Industrial AreaComplete surface temperature frequency distributions for the three flights made over thestudy area are presented in Figure 5.13. Two estimates of the complete surface temperature (Ta) distribution are made. In the first (T1), the vertical facet temperatures areobtained from the traverse vehicle, and the second, (T2), from off-nadir airborne thermalimagery. The T1 distributions exhibit a sharp break at 20°C due to truncation of the lowtemperature tail of the traverse temperature distributions. The two distributions differprimarily in the relative frequency of low temperatures; those obtained using the traversevehicle enhance low temperature frequencies. Differences between the distribution meansdecrease from 0.6°C for the morning flight to 0.1° for the late afternoon flight. The decrease is due in part to better agreement between vehicle and image-extracted verticalfacet temperature distributions of the most directly irradiated facets for the early andlate afternoon flights (mostly westerly) (Figures 5.12, 5.14, 5.15).North facets are equally well sampled by the two techniques, (except possibly in narrower alleyways where traverse data may be cooler) and so better agreement is expected.The anticipated bias in sampling east facets (traverse likely undersamples) yields smallerdifferences between the techniques than for the morning case with west facets, perhapsbecause relative solar access differences have not yet fully developed.The most apparent difference between the complete and nadir temperature distributions is the enhancement of low temperature frequencies. This occurs because, except forthe most directly irradiated facet, all vertical facets have temperature distributions whichhave greater frequencies of low temperatures compared to the horizontal (Figure 5.16).(a)>C.)Ccrci,U-(c)>C)Cci):3crci,U-Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 1900.1 30 40 50 60(b)> 1400PDT- - -- Complete (Traverse data) Tnociir = 39.2Complete (Image extracted) T = 35.720 30 40 50 600.1 Surface Temperature (° C)Figure 5.13: Estimated T distributions; Industrial area. (a) Flight 1 (1030 PDT), (b)Flight 2 (1400), (c) Flight 3 (1715).20 30 40 50 60>‘0Ca,0a,L>‘0Ca,cva)>‘0Ca)D0a,1910.300.250.200.1 50.100.05Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE(a) (b) 0.12 South0.110.10 a, VehicleTraverse0.09 1 - — - — irvoge Extractedo.o8Apparent Surface Temperature ( C)(d) 0.30NorthVehicle Traverse— — —— Image ExtractedApparent Surfoce Temperature ( C)(c) WestVeRic le TraverseIrraage Extracted10 20 30 40 50 60Apporent Surface Temperature (° C)10 20 30 40 50 60Apparent Surface Temperature (0 C)Figure 5.14: Traverse and image extracted vertical facet temperature distributions; Industrial area, Flight 2 (1400 PDT). Facet orientations are: (a) N, (b) S, (c) E, (d) W.Figure 5.15: Traverse and image extracted vertical facet temperature distributions; Industrial area, Flight 3 (1730 PDT). Facet orientations are: (a) N, (b) S, (c) E, (d) W.Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 192(a) (b)>UC5,0U->C,C5,a0)Li(c)5’C)C0)0Li-Apparent Surface Temperature (°C) Apporent Surface Temperature (° C)’C0C,1i0. 60.140120.,,verseIroogeEotrccted102030405060Apparent Surface Temperature (0 C)10 20 30 40 50 60Apparent Surface Temperature (° C)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 193Temporally, the difference between the nadir and vertical distributions is strongest nearsolar noon. Then, because of the relatively small zenith angle, even the most directlyirradiated facet is cooler than the horizontal (Figure 5.16b). At times earlier and laterin the day, the most directly irradiated wall has a frequency distribution only slightlycooler than that of the horizontal, so T estimates are closer to Tr for nadir observations.Comparison of the complete and off-nadir temperature distributions is presented inFigure 5.17. In each case, the complete distribution enhances the low temperature frequencies, resulting in a lower mean temperature. Agreement is best for off-nadir viewangles in the direction of a shaded facet.5.3.2 Downtown AreaComplete surface temperature distributions for the Downtown study area (Figure 5.18)show temperature decreases of 1-3°C compared with that of the nadir, depending uponthe source of the vertical facet temperature distribution.Large discrepancies are observed between extracted and traverse estimates of thetemperature distribution for the most directly irradiated facet for both Flights 4 and 5(Figures 5.19, 5.20) resulting in T differences of approximately 2°C for each flight.Each platform suffers a bias due to viewing position: traverse temperature estimatesinclude all viewing angles weighted in proportion to the number of “valid” (i.e., above thetruncation level) observations for each EIRT angle, but are biased towards low temperatures for the most directly irradiated facet for the reasons discussed previously. Airborneobservations preferentially view the top portion of walls, cannot “see” below the level ofany awnings present, may have some of the lower wall obscured when H/W geometry islarge, and the source of emission for specular reflections is the warm street.>0C5)Dcr5)L>0Ca):3a)U-0.4(c)0.3>0C0.2ora)LL0.10.0Apparent Surface Temperature (° C)Figure 5.16: Component temperature distributions (from vehicle traverses); Industrialarea. (a) Flight 1 (1030 PDT), (b) Flight 2 (1400 PDT), (c) Flight 3 (1715 PDT).Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 1940.5tC) 1O3OPDT0 4 Nodir (Airborne)- - -- North Focets—South Facets0.3 ——— EostFocetsI — West Facets0.2 K0.10.0____________________________(b) 0.420 30 40 50 601400PDT— Nadir (Airborne)North FacetsSouth Focets— East Facets— West FacetsI - Ii0. 30 40 50 6020 30 40 50 600.1 5. THE COMPLETE URBAN SURFACE TEMPERATURE 195(a)>.C-)Ccr0U(b)>C)C‘I):3C0U(c)>-.C-)C0.)DC)1)U-Figure 5.17: Complete and off-nadir apparent surface temperature distributions; Industrial area. (a) Flight 1 (1030 PDT), (b) Flight 2 (1400 PDT), (c) Flight 3 (1715 PDT).20 30 40 50 6020 30 40 50 60Apparent Surface Temperature (° C)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE(a) 0.200.1 80.1 60.1 4>UC0.100.08Lj>(-)CciCTci-)U-0.1 60.1 40.1 20.1 Surface Temperature (° C)Apparent Surface Temperature (° C)196Figure 5.18: T distributions; Downtown study area. (a) Flight 4 (1130 PDT), (b) Flight5 (1630 PDT).20 30 40 5020 30 40 50197Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE(a) (b) Traverse— — —— Image xtroc tedJ)Apparent Surface Temperature (° C)>‘0Ca):30a)U->0CeD0a)(c)>‘0Ca)Da)U-Apparent Surface Temperature (0 C)(d)uCa)0aU-20 30 40 50Apparent Surface Temperature (0 C)20 30 40 50Apparent Surface Temperature (°C)Figure 5.19: Traverse and image-extracted vertical facet temperature distributions;Downtown study area. Flight 4 (1130 PDT). Facet orientations are: (a) NW, (b) SE, (c)NE, (d) SW.Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 1980. 30 40 50Apparent Surface Temperature (° C)>‘UO 0.15a1. 30 40 50Apparent Surface Temperature (0 C)20 30 40 50Apparent Surface Temperoture (ac)Figure 5.20: Traverse and image-extracted vertical facet temperature distributions;Downtown study area. Flight 5 (1630 PDT). Facet orientations are: (a) NW, (b) SE, (c)NE, (d) SW.(b)0.250.20(a)C,a.U(c)— VehcleTt dNE— Vehicle Traverse— — —— Image Exlrac led(d)>‘0CaaU- Surtace Temperature (° C)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 199Figure 5.21: Apparent surface temperature distributions from traverse and image sourcesfor building walls and trees; Residential area; Flight 6 (0945 PDT). Facet orientationsare: (a) N, (b) S, (c) E, (d) W.5.3.3 Residential AreaThe first complete temperature estimate (T1) uses, as previously, the temperature distributions obtained from the traverse vehicle. In this instance, the 0 and 100 EIRT arecombined, and it is assumed that an adequate sampling of both the house and tree temperatures is obtained. Truncation of the traverse distribution is specified using a graphicanalysis of the traverse and extracted temperature distributions (Figure 5.21) and lookingfor evidence of the tree canopy signal in the traverse (especially 100 EIRT) distribution.This was related to a local minimum in the frequency distribution (Figure 5.21, east(a)C0U-(b)00.4 North South———— Traverse (0° EIRT) 04 Traverse (0° EIRT)—— Traverse (1 0° EIRT) Traverse (10° EIRT)• Wall (Eatrocted image) Wall (Extracted image)Tree (Extracted image)‘ 0.3 Tree (Extras ted image)02:10.2::10 20 30U- :i030 40 50Apporent Surface Temperature (° C) Apparent Surfoce Temperature ( C)(d): East West: - — —— Traverse (0° EIRT) 0.5 - — — — Traverse (0° TIRT)• — .—. Traverse (10° EIRT) — . —. Traverse (10° EIRT)— Wall (Extracted image) o 4 — Wall (Extracted image)Tree (Extras ted image) Tree (Extras ted image)1: II:10 20:50Apporent Surtoce Temperature (° C) Apporent Surface Temperature (°C)(C) 0.40.3>‘° 0.20U-0.10.0Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 200and south facets).The traverse data undersample the north and south facets and this may be importantfor the north facet case, as these are much more likely to be shaded at mid-block than atthe ends of the block. Sampling by image extraction yields little improvement becauseof the very narrow inter-building spacing and pixel smearing. Comparison of imageextracted and traverse distributions (Figure 5.21) show warmer temperatures for theextracted pixels, the opposite of that expected considering the inter-building spacing.As in the case of the Industrial site, the complete surface temperature of the Residential area (Figure 5.22) is cooler than that obtained by nadir observations. Differences aregreatest for the early afternoon flight when T is 4 - 5°C cooler than nadir Tr. The twoestimates of T are similar, with T1 showing a slight enhancement at higher temperaturefrequencies. Sensitivity tests of the effect of Fgap upon T indicated increases in T of0.5 - 0.7°C when Fgap was reduced from 1.0 for all tree types to the estimated valuesgiven in Section 5.1. Compared with the Industrial area, the effect of trees increases thedifference between nadir and complete apparent surface temperatures.Complete surface temperature estimates (T2) combine extracted temperature distributions for vertical facets and tree canopies. Off-nadir tree canopy temperature distributions show variations in the mean of approximately 3°C between the most directlyirradiated direction and the most shaded (Figure 5.23) for each flight over the Residentialarea.Weighting the tree canopy temperatures by tree canopy area is accomplished by calculating the projected area in each of the north, south, east, west, and nadir-viewingdirections and subdividing the complete tree canopy area by the relative projected area.Because trees are elliptical in the xz and yz plane, projected areas are greater for thevertical planes than the horizontal. A weighting is also calculated for the underside ofthe tree, for trees with htk > 0.Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE(a)>‘()1=a):3a)U(b)>‘0a):5cra)U(C)>‘3-)a)a)U-0.1 20.1 20.1 30 40 50 60Apparent Surface Temperature (° C)201Figure 5.22: Estimated T distributions; Residential area. (a)Flight 7 (1400 PDT), (c) Flight 8 (1715 PDT).Flight 6 (0945 PDT), (b)20 30 40 50 601715PD1 T —nadir —= 33.3= 32.4Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE>U0ci,202Figure 5.23: Apparent tree canopy temperatures extracted from off-nadir imagery; Residential area. (a) Flight 6 (0945 PDT), (b) Flight 7 (1400 PDT), (c) Flight 8 (1715PDT).,0945 PDT TreesI33North(a)>‘(-ICci):3a)(b)>‘C-)Ca):30a)(c) 20 22 24 26 28 30 32Trees 1 400 POTI\ IC182023032Trees 1 71 5 PDTNorth /— — —— South V,:18 20 22 24 26 28 30 32Apparent Surface Temperature (° C)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 2035.4 Surface and Air Temperature RelationsRemote measurements of apparent surface temperature are often compared with surface-layer air temperature measurements (Lee, 1991; Stoll and Brazel, 1992; Dousset, 1989;Henry et al. 1989) with the goal of generating estimates of air temperature from thermalimagery. Results vary for the reasons discussed by Roth et al. (1989): remote sensorsbias the surface observed; air and surface temperatures have a complex coupling throughflux divergence of the lowest layers of the atmosphere; and there are mis-matches inthe scales of observation used for remote and in-situ measurements. The estimation ofcomplete surface temperatures addresses the problem of observational bias in the remotethermal measurements.Complete mean apparent surface temperatures (T1), and traverse air temperatures(Tat) have been added to Figure 4.32 to produce Figure 5.24. This figure summarizesthe mean apparent temperatures of nadir and off-nadir viewed surfaces, the estimatedcomplete surface temperature and canyon-level air temperature for each of the studyareas.Figure 5.24 demonstrates that the use of complete surface temperatures yields only amarginal improvement in the comparison of air and apparent surface temperatures. Atthe scale of the images used in this analysis, daytime mean apparent surface temperaturesare substantially warmer than canyon-level air temperatures at all times, for all sites.The closest match occurs for Flight 6 (0945 PDT) in the Residential area where both offnadir and complete surface temperatures are within 2°C degrees of the canyon-level airtemperature. Much better agreement between air and surface temperatures is achievedby comparing the coolest modelled distribution obtained from the mixed distributionanalysis of the surface temperature data (i.e., the results in Figure 1.27 derived from thedistributions in Figure 1.5). Results are displayed by view direction (Figure 5.25) and asChapter 5. THE COMPLETE URBAN S URFACE TEMPERATURETemperature (° C)204Figure 5.24: Comparison of mean image apparent surface temperatures by view direction, traverse air temperatures (Tat, denoted by solid diamonds), and complete apparentsurface temperatures (T1, solid triangles labelled C) for all study areas.I I]VII•I•I5• I I’I•I’I•I I. 4 ACVI I I I I I I I I I I IIV s E1,NV 4 AIeC WI I I I I I I I I I I I IT0 SW NEI•II!I! IC VI I I I I I I I I I I18 20 22 24 26 28 30 32 34 36 38 40 42 44I!Ir•II!I!I•I!II!I•r•r•T0 SWNE V NWV.VEI I I I I I I I I I Iwz0-JU-2345678.CSETime(P DT)1 0301 400 Industrial1 7301145Downtown1 63009451 400 Residential1 71 5ES N. ie 0CV-1ot SWE, A 000 0 0C Vt WS N E. 40 000C VI I I I I I I I18 20 22 24 26 28 30 32 34 36 38 40 42 44Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 205a composite for shaded and sunlit facets (Figure 5.26).Very good agreement between Tmod and Tat is observed between the nadir and viewdirections towards shaded facets in the Downtown study area; the most directly irradiatedfacets (NW Flight 4 and NE Flight 5) substantially overestimate Tat. Winds were strongeron this day and the NW direction of the flow may have helped to equalize building and airtemperatures on this day. The other Flights display a tendency for Tmod to underestimateTat. The underestimate may be due in part to low emissivity surfaces (recall that in thedowntown area, airborne observations of vertical facet surface temperatures influencedby spectral reflectivity lead to warmer apparent temperatures) while in the other areas,low emissivity surfaces are primarily confined to rooftops and would therefore lead tolower apparent temperatures. The Industrial area is characterized by greater numbersof low emissivity surfaces and also shows the largest difference between Tmod and Tat.Alternatively, the results may be a true indication of the temperature difference betweenthe most shaded facets and air temperature.The poorest observed agreement for shaded facets occurs in the Industrial area; especially during the morning and early afternoon flights (Flights 1, 2). Tat is substantiallywarmer than the coolest extracted distribution (Figures 5.25 and 5.26). Reasons for thisare not obvious but it may be due in part to the ability for the remote imagery to viewthe bottoms of the street and alley-canyons along with lighter winds leading to greatertemperature differentiation than for the Downtown study area. Comparison of theseresults with Figure 3.5 shows that Tat and (traverse air and building temperaturesfor North-facing walls) and the mean airborne apparent surface temperatures are closelyrelated but Tramin and are substantially cooler than Tat, confirming the presence ofsurfaces much cooler than air temperature.Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE302826242220181616 18 20 22 24T,(eC)26 28 30.30282624.-.221201816206Figure 5.25: Traverse air temperature compared with the lowest modelled temperaturedistribution (Tmod) fitted to the composite frequency distribution. Plots separated byview direction for all flights. Error bars are ±lu. Bracketed directions refer to Flights 4and 5 (Downtown study area).03028262422201816West (OW)INsdir HF616 18 20 22 24 26 28 30T1(°C)[set (Ito) -16 16 20 22 24 26 26 30T ( C)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATUREC)0SI—C)0I—207Figure 5.26: Same as Figure 5.25 except showing means only, and with data subdividedinto shaded and sunlit facets. Bracketed directions refer to Flights 4 and 5 (Downtownstudy area).24l• 1111111Shaded Facets Sunlit Facets28 ONodir ..• 28 VLJNorth(NW) . 0 F4 F26 South(SE) . V V 26 V0 East (NE) . F8 V .24V West (SW)F7. F5 F7VVVV22 F5 . F3 22 F6.F6 8 F3 0 Nadir20 F4 20 . D North (NW). .F2 South (SE)v . 0 0 Lost (NE)1 8 •. 0 Fl 1 8 .. Fl V West (SW)16.16 .16 18 20 22 24 26 28 26 28 30T5 (0 C)30 16 18 20 22 24T0 (0 C)Chapter 5. THE COMPLETE URBAN SURFACE TEMPERATURE 2085.5 SummaryIt is apparent that, for the majority of cases, T1 is very close to the coolest observedoff-nadir Tr. The two notable exceptions are the mid-day flights over the Industrial andResidential areas; these show T1 to be approximately 10 below the coolest off-nadir viewdirection (South). Differences between T1 and the warmest off-nadir direction are equalto, or greater than the range of off-nadir temperatures; differences of approximately 7°Care observed during the early afternoon flights over the Industrial and Residential area.These results suggest that, where available, off-nadir thermal imagery in the directionof the coolest (most shaded) facet yields the closest observable temperature to T. Thisapproximation appears to be least valid at mid-day. The relation between air and surfacetemperatures shows the great majority of pixels have Tr values substantially warmer thanair temperature. An approximate estimate of air temperature can be better derived byusing the lowest fitted distribution to the composite frequency distributions from a viewdirection towards a shaded facet. -Chapter 6MODELLING URBAN SURFACE THERMAL EMISSIONS6.1 IntroductionIn Section 1.6, a review of models for the prediction of thermal emissions over plantcanopies concluded that a geometric projection modelling approach is appropriate whendeveloping a model for urban surfaces because of the similarities between crop row structures and urban canyons. This chapter details the basis of the modelling approach, themodifications necessary to represent an urban surface, and a simple one-dimensional energy balance model to estimate component surface temperatures for use in the geometricalsurface model.6.2 Geometric Models to Estimate Longwave Surface AnisotropyGeometrical models are one of a number of approaches to estimate longwave surfaceansiotropy. Paw U (1992) presents a brief review of TIR model development and Goel(1988) provides a detailed review of geometric models used to calculate shortwave canopyreflectance. In the case of vegetation canopies, a geometric model consists of a groundsurface pius geometrical objects placed to represent individual or combined canopy elements (e.g., single trees, or rows of closely-spaced plants; see Section 1.6.). The modelused here is a slightly modified version of that first described by Sobrino et at. (1990)and Sobrino and Caselles (1990) for use under nighttime conditions over orange groves,and later modified to include daytime conditions (Caselles et at. 1992). It is referred to209Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 210here as the SC model.6.2.1 Extension to Urban SurfacesRow crops are often assumed to be sufficiently long to allow a two-dimensional surface representation using the row cross-section, thereby simplifying modelling. Urbanstreet canyons are broken at regular intervals by intersecting streets. This creates ablock structure which ideally requires three-dimensional surface representation. Similarmodifications have been suggested by McGuire et al. (1989) for forest canopies and byKimes and Kirchner (1982), and cooper and Smith (1985) for short-wavelength surfacereflection. Incorporation of a three-dimensional surface representation for the urban surface greatly increases the complexity of the sensor model and has been omitted fromthe present work. A slight modification to the form of the surface used by Sobrino et al.(1990) and Sobrino and Caselles (1990) is the addition of a second row spacing parameterto better represent the generally non-equal widths of street and alley canyons observedin the study areas (Figure 6.1).6.3 Model DescriptionSensor detected radiance is assumed to be a linear combination of surface componentradiances weighted according to their fractional contribution (f) within the sensor IFOV:L=Zf.L (6.1)where i represents a component surface within the sensor IFOV and L is a representativesurface radiance for that component.A geometrical model has been formulated to estimate ft for simple two-dimensionalurban surfaces given prescribed Sun-surface-sensor geometry. L may be estimated fromChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 211h,I ‘I It’( W < X >Figure 6.1: Two-dimensional model representation of an urban surface showing inputdimensions. All buildings are of equal height H.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 212observed data or a simple one-dimensional energy balance model (Section 6.4). Themodel has been developed using the concepts of Sobrino and Caselles (1990) but usesindependently derived formulations for the angular portions of the IFOV occupied byroof, sunlit and shaded ground, and sunlit and shaded walls (‘yr, 7g, and fw respectively).Tests of the model replicated the results presented by Sobrino and Caselles (1990)The 7 components are determined using the general form:7x = 7xe — ‘Y (6.2)where x is replaced by r, g, or w and e and s refer to the start and end of the planarangle. The component angles (ignoring shading effects) are shown in Figure 6.2.Specific formulae for roof, ground and wall are:X+(n— 1)BL+nAWA+nsWs7r=— H(6.3)X + UBL + flAWA + isWs7reFigure 6.2: Detail of the component angles 7r, 7g, and 7w calculated in the SC model.(6.4)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 213X+(n— 1)BL+nAWA+nsWs= — H(6.5)X + nBL + riA.WA + flsWs7ge= h (6.6)X+riBL+nAWA+T1SWS7W= h8— H (6.7)X+nBL+nAWA+nsWs 68h3 (.)where X is defined as the distance from the sensor sub-point (directly below the sensor)to the beginning of the IFOV (assumed to be at the start of the first roof element)(Figure 6.1). The parameters n, A and s are counters which refer to the number ofthe element viewed; these differ for each ‘y and depend upon whether an alley or a streetcanyon is viewed.The end of the IFOV is not pre-determined; the individual elemental angles aresummed and a check is made following each addition to the sum, so that a slight overestimate is always made because partial elements are not considered. The overestimateis minor when the elemental angles are small and this can be achieved by increasing thesensor height h.The assumption of the IFOV beginning at the first roof has not been generalized. Thisis not considered to be restrictive for cases when the number of street-alley sequenceson the ground are viewed. It does however, restrict exact comparison with observationswhich view only partial or a few series of street alley sequences.Where surfaces are shaded and/or partially or fully obscured due to a large off-nadirviewing angle, additional expressions of are specified to account for these effects. Casesinclude viewing in both up- and down-Sun azimuth directions. The nadir case is similarexcept that the sensor subpoint is assumed to bisect a street canyon so that the IFOV issymmetrical about this point. Calculations are done separately in the up- and down-Sunazimuths to consider the different shading patterns.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 214Input required to specify the Sun-surface-sensor geometry is listed in Table 6.1.Table 6.1: User input required for the SC model.Parameter Symbol UnitsSensor descriptionsensor view angle 0 degreesheight of sensor msensor IFOV IFOV planar degreessensor azimuth çb8 ±900 relative to reference systemSolar Geometryzenith angle Z degreesazimuth angle degreesSurface DescriptionCanyon azimuth degreesBuilding dimensionsheight H mlength BL mbuilding spacing (street) Ws mbuilding spacing (alley) WA mModel output consists of a tabulation of the sunlit and shaded fractional componentsof the major surface types (roof, wall, and ground) viewed. An example showing thevariation in the proportions of these surfaces with view angle for north and south sensorazimuths at solar noon for YD 228 (August 15) over a model surface representing theIndustrial area is presented in Figure 6.3.The fractional components (f) are then used in eq. 6.1 in combination with observedor modelled estimates of L to estimate L6 for the given Sun-surface-sensor input geometry(Section 6.7).Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 215(a) 0.7________0.6 View North, Solor noonFOV=12° 0 Roof> Q5 --a-- Sunlit Ground--G-- ShadedGroundS —--— Sunlit Wall° 0.4=32.0--V- Shaded Wall0.1;—=ó-- - -. -0.0______________ _ __ _ _ __ __0 10 20 30 40 50 60View Angle (Degrees)I I I I I1 0.7__ __ __ __ __ __ __ ____ _ __ _ _ _°g Dccc0.6 View South, Solar NoonFOV=12° 0 Roof> Q5 --h-- SunlitGroundo--9-- ShadedGroundS —. --G— Sunlit Wallo 0.4 WA2--V- Shaded Wall1 0.2View Angle (Degrees)Figure 6.3: Proportions of roof, wall and ground surfaces calculated by the SC model fora 12° FOV sensor over a 2-D urban surface representing the Industrial study area. (a)View to the North; (b) view to the South. Solar geometry used is for YD 228 (August15), solar noon.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 2166.4 An Energy Balance Model to Estimate Surface TemperaturesThe SC model requires estimates of component surface emittance for use in eqn. 6.1.These can be obtained from observational data (e.g., traverse results, image extraction,or mixed distribution modelling), or surface emittance can be modelled. The abilityto model component surface temperatures is important because it allows the model tobe extended to a wide range of times and surface geometries and avoids the difficultyand expense of acquiring high resolution multi-temporal TIR coverage of various urbansurfaces.6.4.1 Urban Surface Temperature ModelsModelling the temperatures of surfaces which comprise the complex urban system isdifficult. The microscale urban climate model of Sievers and Zdunkowski (1986) is designed to simulate air flow over rectangular obstacles but includes an energy budgetto derive the surface temperature of canyon facets. Evaporative, advective and canyonradiative effects are included in the model. URBAN3 (Terjung and O’Rourke, 1980)combines information on urban morphology, building material, and building dimensionto simulate energy budgets and temperatures for individual surfaces. The major modelcomponents are: view factor/obstructions, radiation interception by surfaces, and energybudget temperature calculations. However, since the ambient air is decoupled from thesurface energy budget, feedbacks between air and surface temperatures are not included.The recent model of Mills (1993) combines a semi-empirical canyon windfield model withan urban canyon energy budget model. In the absence of evaporating surfaces, modelledsurface temperatures agree well with observations. The canyon radiation calculations arehandled in the manner of Arnfield (1982) which includes the ability to vary radiativeproperties in strips in the along-canyon direction. In all these models the structure andChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 217code are extremely complex.The present research requires a canopy layer model capable of rapidly providing temperature estimates for building facets with specified orientation and material propertiesfor use with Sc. The complexity of the urban energy balance models mentioned aboverenders them unsuitable. Model complexity can be reduced if it is assumed that component surface temperatures are largely controlled by the orientation and properties ofthe surface. A first order approximation to component surface temperatures can then begenerated using a simple one-dimensional energy balance model. The goal is to correctlymodel the range and relative temperature difference among facets of different orientationand surface material.6.4.2 The Myrup 1-D Surface Energy Balance ModelOne-dimensional energy balance models use the equilibrium surface temperature concept.An equilibrium surface temperature is iteratively solved from the energy balance equationas the solution of a given set of boundary conditions and solar radiation forcing regime.Such models ignore heat and moisture advection.The Myrup (1969) model with modifications by Outcalt (1971) was selected for usein this thesis. It was originally formulated to provide insight into genesis of the urbanheat island but it has a deliberately simple structure which allows it to be adapted toa wide variety of problems (Myrup, 1969; Outcalt, 1971). It includes relatively simpleparameterizations for the calculation of atmospheric diffusivity and incoming solar radiation when compared with other 1-D models (Loudon Ross and Oke, 1984), and thereforemay be a less than ideal choice. Its merits were pragmatic: it is readily available, easilyimplemented, and most of its input can be satisfied by available data. Standard modelinputs and outputs are listed in Tables 6.2 and 6.3. Further details of the model areprovided in Myrup (1969) and Outcalt (1971).Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 218Table 6.2: Myrup model input requirements.Variable UnitsLabel dataTitle character stringTemporal datalatitude degreessolar declination degreessolar radius vectorstart time hoursMeteorological datastation pressure mbmean daily air temperature °Cmean daily vapour pressure mbmean daily wind velocity m sprecipitable water (w) mmdust particles (D) cm3anthropogenic heat W m2Geographical datasubsurface volumetric heat capacity (C) J m K—’subsurface thermal diffusivity (to) m2 ssurface roughness (z0) malbedo (a) fractionmoisture availability factor fractionshadow ratio fractionChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 219Table 6.3: Myrup model output.Parameter UnitsComputed boundary conditionssoil damping depth matmospheric damping depth madiabatic heat transfer coefficient m2 sTime-dependent output (hourly)Solar:Extraterrestrial, Incoming (K 4), and reflected (K t) W m2Direct shortwave, diffuse shortwave W m2Energy balance terms Net radiation W m2sub-surface heat flux W m2turbulent sensible and latent heat fluxes W m2T0 (°C)The success of the Myrup model for estimating surface temperatures has been testedby Outcalt (1971) with respect to needle-ice events and more qualitatively by LoudonRoss and Oke (1984) against remotely sensed surface temperatures and air temperatures. Outcalt (1971) found the amplitude of T0 and the time dependent behaviourwere correctly simulated, but considerable phase-shift discrepancies were noted becauseof heat-transfer processes not included in the model. Loudon Ross and Oke (1984) notedmore serious discrepancies. Surface temperatures were considered unrealistic becausetheir daily range was less than that observed for air temperatures and the peak at solarnoon was too early when compared with observations. However, the authors note thatimprovements result when the model is run with hourly, rather than mean daily, meteorological input data. This provides some coupling between surface and air temperatureswhich is otherwise missing from the model (as evidenced by modelled maximum temperatures which occur at solar noon). Model comparisons against observed temperaturesChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 220from select component surfaces in this study are given in Section Modifications to the Myrup ModelModifications to the basic Myrup model (which assumes an extensive horizontal surface)were made in order to incorporate the surface structure of urban areas. The modificationsinclude:• provision for radiation upon vertical (walls) and inclined (roof) surfaces throughthe calculation of 9, the angle of incidence between the normal to the slope and thesolar beam for a given slope geometry (slope and azimuth) (see e.g., Oke (1987),Iqbal (1983));• inclusion of an urban canyon structure with given canyon azimuth, H:W ratio, andwall and floor surface radiative properties (albedo and emissivity);• incorporation of canyon radiative effects by either: -— coupling to the canyon radiation model of Arnfleld (1982), or;— addition of a simplified scheme for predicting radiation balance at the midpointof the three canyon facets (floor and two walls), including multiple reflections;• replacement of the mean daily boundary conditions with hourly updates.The simplified radiation scheme calculates incident shortwave radiative fluxes forthe midpoint of each facet. This is accomplished by calculating 0 for each surface anddetermining the horizon obstruction of the opposite wall in the direction of the sun.Diffuse radiation is assumed to be isotropic. The calculated irradiances are then inputto the canyon multiple reflection routine of Arnfield (1982). View factors of each facetfor the three facet midpoints are calculated using equations for plane and perpendicularChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 221surfaces for a point derived using Nusselt Sphere techniques (Steyn and Lyons, 1985;Johnson and Watson, 1984). The sky view factor is determined as a residual. Incidentlongwave radiation at the facet midpoints is calculated as a weighted sum (using viewfactors) of the sky and facet longwave irradiances. Sky irradiance is determined usingIdso (1981) or Idso and Jackson (1969). Facet emissions (except for the facet modelled)are calculated from air temperature. At night, this approximation holds well. During thedaytime, air temperature underestimates surface temperature when the facet is directlyirradiated. Multiple reflections are not incorporated.A number of less substantive changes have also been made to faciliate model useincluding:• replacement of the time of modelling (originally specified by means of the Earth’sradius vector (or eccentricity) and the solar declination by equations from Spencer(1971) (as given by Iqbal (1983)) which require only year day (YD) as the baseinput;• conversion between LAT (Local Apparent Time or solar time) and LST or LDT(Local Standard or Daylight time);• addition of the Idso (1981) longwave radiation formulation to that of Idso andJackson (1969); the former requires air temperature and humidity as input, thelatter only air temperature.The incoming solar radiation simulation of the Myrup model is derived from Gates(1962) and requires as input: extraterrestrial solar irradiance (W m2), surface pressure(mb), dust factor D, and precipitable water w (mm). Precipitable water is estimatedby integrating the measured profiles obtained from the composite atmospheric soundingsand dust content is estimated qualitatively from Gates (1962). The sensitivity to D andChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS0—10—20—30—40—50—60—70Figure 6.4: Sensitivity of K.4. to: (a) dust content D and (b) w.222w are shown in Figure 6.4. Increasing values of both parameters leads to lower modelledvalues of K. Substantial temporal variability exists in the estimates of w, and this isshown to qualitatively match with the temporal variations in K.. (Figure 6.5).Observed and modelled incoming global solar radiation for August 15/92 are plottedin Figure 6.6 for two values of D which cover the approximate range of values for pollutedurban atmospheres (Gates, 1962). The model slightly overestimates but the differenceis not large; percentage differences between the maximum modelled and observed valuesare less than 3% for D = 1.5 and less than 2% for D=2.O.In-canyon radiation estimates obtained using the Arnfield (1982) routines and a moresimple set of equations for the mid-point of each canyon facet, are almost identical (Figure 6.7). Comparison of the facet mid-point against the facet average (Figure 6.8) generated from the Arnfield model shows differences when the facet is partially shaded.However, this is not serious because it is not recommended to use average facet netshortwave radiation to estimate facet temperature. It is preferable to combine separateshaded and sunlit estimates. The difference between the facet mid-point and averagefacet irradiance is limited to a second order effect (in the simple model) associated withE5— 1.5crs —=“‘ Bose Values 20 mm / Bose Values—20 0 = 1 .ocm- 25 rsrs \ / 0 = 1.5 cs’‘ W=2Omm—— 30mm ‘. .i W15rnrs—259 12 15 18 21 24 0 3 6 9 12 15 18 21 24Solar Time (Hours) Solar Time (Hours)0 3 6Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 22340t &<Aug.16E 0 AKAug.17 30U wAug.15 ‘wAug.16 H 20V wAug.1710 10___0 S0 0 0—1 00 \0o 0o °cQ:P0U-—20—30 05 7 9 11 13 15 17 19 21Time (Hours, PDT)Figure 6.5: Difference of K. on August 16 and 17 from that of August 15, together withprecipitable water calculated from the composite soundings made during each remotesensing flight.substitution of the entire facet irradiance by the midpoint during the calculation of multiple reflections. The use of the simple mid-point scheme yields execution times almosttwo orders of magnitude faster than using the Arnfield routine (2.75 s versus 218 s).6.5 Model Tests Against ObservationsThe traverse observations of the surface temperature of the road and canyon walls providedata with which model temperature estimates may be compared. This section presentscomparisons of modelled and observed surface temperatures for both individual component surfaces and composite surfaces (e.g., all canyon facets of a given orientation) andinvestigates the model output sensitivity to various model input variables.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS900800700—-. 6005004003002001 000224Figure 6.6: Observed (at the Sunset Tower site) and modelled incoming global solarirradiance for August 15, 1992.6.5.1 Road surface temperatures6.5.1.1 Sensitivity AnalysisRoad surfaces are the most homogeneous surface type in the study area. Prior to acomparison of model results with observations, a sensitivity analysis to variations in theinput values of: asphalt thermal, radiative and aerodynamic properties, hourly versusdaily data, canyon H:W ratio, and surface slope and azimuth was conducted (Figure 6.9).Base model input values are presented in Table 6.4.The specification of the air temperature is an important determinant of the magnitudeand shape of the surface temperature curve. Using only the mean diurnal air temperature(recorded at Vancouver International Airport) yields maximum modelled T0 at solar noon(which is unrealistic) and the lowest diurnal temperature range of the tests. When the024681012141618202224Time (Hours, PDT)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 225800 Arnfield Model IISimple Canyon N/s CanyonFloorE600E 500 East WoW jr West WaW>4QQCl) Ac300><D 200 cdl100 2 4 6 8 1012141618202224Solar Time (Hours)Figure 6.7: Comparison of mid-facet estimates of net shortwave radiation generated usingthe mid-point model and the Arnfield (1982) canyon radiation routines. Canyon is alignedN/S with H:W of 5:11, a, = 0.2, o = 0.05.mean daily air temperature is replaced by hourly values, the T1, peak is lagged relative tonoon and the range is increased. A further increase in the surface temperature maximumis achieved when traverse air temperatures (Tat) are used in place of the airport data.The sensitivity to asphalt thermal properties was tested using three separate referencevalues for thermal diffusivity (ic) and volumetric heat capacity (C) (Table variation in modelled temperature is relatively small and large changes (an increase in C is most effective) are required to produce a change in phase.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS8007006005QQ>‘4QQ300200U-1 0000 2 4 6 8 1012141618202224Solar Time (Hours)226Figure 6.8: Comparison of mid-facet and facet average estimates of net shortwave radiation generated using the mid-point model and the Arnfield (1982) canyon radiationroutines. Canyon is aligned N/S with: H:W of 5:11, c = 0.2, c = 0.05.Table 6.4: Base input parameters: asphalt sensitivity tests.EastFacet AverageFacetMidpoint N/s Canyonó %\FIaor0West WallModel Input - Value -0.93a 0.05C 1.94x1060.38x10zo 0.0001Slope 0.0Air temperature hourly (Tat augmented with airport dataLongwave Radiation Idso and Jackson (1969)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 22750401)3020105040C)30I—20100 5 6 9 12 15 18 21Time (Hours, POT)Slope -0 3 624 0 3 6 9 12 15 18 21 24Time (Hours, POT)Figure 6.9: Sensitivity analysis of an asphalt road surface temperature to: (a) air temperature specification, (b) thermal properties (k and C, (c) radiative properties (cr, and ),(d) surface roughness (z0), (e) slope aspect, and (f) longwave radiation parameterization.Solid line represents the set of base input (See Table 6.4).(b)(a)C)F—(c)40302010Thermal Properties (K, C)T0withWRT ‘ Air Temperature— —— ‘YVR Hoarly TJiyT40302010Coward (1 98)) —‘— —— Barber (1 957)——. Yoder and Witczak (1975)(d)40Radiative Properties (, )30— a = 0.05, n=0.93a=0.15, =0.93——. a=0.05.n=0.97200 3 6 9 12 15 18 21Time (Hours, POT)(e)1024(f)Aerodynornic(z0)— z, = 0.0001——- z=0.001—— z. = 0.0005•• z = 0.000050 3 6 9 12 ‘ 15 18 21 24Time (Hours, PDT)‘•‘ ldsoondJacksao (1969) Longwave——— ldso(1961)50403020— Horizontal50 Sooth—— 5° North5° WestU9 12 15 18 21 24 - 0 3 6 9 12 15 18 21 24Time (Hours, POT) Time (Hours, POT)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 228Table 6.5: Reference values of asphalt thermal properties used in the Myrup model.Reference C (J m3) K—’ i (m s’)rber (1957) 2.07x106 0.59x106Yoder and Witczak (1975) 1.41x10 1.03x10Goward (1981) 1.94x106 0.38x106An increase in a (from 0.05-0.15) yields a decrease in the modelled surface temperature of approximately 1.5°C; increasing slightly reduces modelled surface temperature.Specification of the surface roughness length, z0 can yield a large sensitivity in T0. Toachieve best agreement with observed data, low values of z0 were required (0.001 or less).These are somewhat arbitrarily specified considering the surface character of asphalt inisolation, rather than as a component of the urban surface (which has much larger valuesof z0).Slight changes in the slope of the road surface are possible due to: (a) road construction (the middle, or crown is higher than the sides), (b) topographic effects. A slopewith westerly aspect shows a phase shift of T0 towards later times, and increases in themaximum modelled temperature. North and south aspect slopes show no phase shift butrespectively reduce or enhance the maximum surface temperature.To summarize, large sensitivities are observed with respect to the manner in which airtemperatures are updated (if at all) within the model, and the aerodynamic roughnessand slope of the surface. IndustrialModelled road surface temperatures (Troad) were generated using: z0 = 1.0x105,k =0.86x106m2 srn’, C = 1.32x105 J m3 K’, = 0.94, a = 0.05. Traverse observationsfrom two blocks were selected for a comparison of modelled and observed Troad. ResultsChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 229are presented in Figure 6.10 for (a) E/W and (b) N/S-aligned blocks, each with H:W ofapproximately 6:22.Observed results for the E/W block are the mean, and ±lu about the mean, from thetraverse measurements along the block. Relatively constant canyon H:W along the blockyields spatially consistent radiation conditions. The maximum road surface temperatureis well approximated by the model, as is the afternoon cooling. The model over-predictsthe rise of temperature in the morning, leading to an overall shift of the heating curveto times earlier in the day.Individual temperature samples are plotted for the N/S block because of significantalong-street variability due to shading by individual buildings, in contrast to the E/Wcanyon. Generally good agreement exists during the morning period. There is evidenceof a period of shading not incorporated in the model around 1200 PDT. Peak and afternoon temperatures are underpredicted by the model, although the general trend of theobservations is replicated.5045403530252015100246 81012141618202224 024681012141618202224Time (Hours, POT) Time (Hors, PDT)Figure 6.10: Modelled and observed road surface temperatures in the Industrial studyarea: (a) E/W street canyon, (b) N/S street canyon.IrdustrisIE/W Street(6th Mon—Cel)30252015(ndustrtIN/S Street(Unnr Sth—6th) 8—MedeIId v0 TrsvPtl4 oS TrsvPtt500 TrooPt06.5.1.4 Downtown5045403530252015Modelling Troad is expected to be more difficult in the Downtown area because the modelhas a limited ability to handle the complex canyon geometry in this area (unequal canyonwall heights and large differences in building heights in the along canyon direction). FourChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 2306.5.1.3 ResidentialA composite two block comparison of an E/W street with few trees in the Residentialarea (Figure 6.11) exhibits results similar to those obtained in the Industrial area: theroad surface warms too quickly and therefore Troad is overpredicted for the period priorto the peak. Afternoon temperatures and the timing of the maximum are generallywell represented by the model. North/south Residential blocks have not been modelledbecause the shading by street trees cannot be accounted for in the model.ResidentialE/W Street49th Ave— Modelled—e—— ObservedC)0t000 2 4 6 81012141618202224Time (Hours, PDT)Figure 6.11: Modelled and observed road surface temperatures in the Residential studyarea: E/W street canyon.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 231street canyons were tested, two for each of the street azimuths in the study area, wherecanyon geometries were relatively constant along the length of the block and wall heightswere approximately equal. Model specifications relating to asphalt remain the same aspreviously. Canyon geometry was estimated from planning maps and field observations.The NE/SW results (Figure 6.12a,b) show a range of agreement between modelledand observed temperatures. Very good agreement was obtained between the observedmean and the modelled surface temperature for the block on Howe St. between Hastingsand Pender (H:W 29:23). The large error bars (representing ±lcr) arise because thereare areas of both shaded and irradiated pavement along the block. The block on HornbySt. between Dunsmuir and Pender (Figure 6.12b) shows poorer agreement between observations and model output. The difference in model output between the two blocks islargely in the width of the main peak. The peak for the Hornby block is wider due tothe more open canyon geometry (H:W 21:23). Observed road temperatures are generallycooler than the modelled values, particularly in the morning. This is similar to the differences noted in the Industrial and Residential areas. A further difficulty with comparingmodelled and observed values in the Downtown area is the required bias of the traverseroute to stay on one side of the canyon where the radiation conditions may differ fromthose in the centre of the canyon.Results from the NW/SE street canyons (Figure 6.12c,d) show similar variation. Inaddition to the mean block road temperature, data are plotted for individual samplingpoints (labelled as Tray Pt) along the block to illustrate the large spatial variabilityof surface temperature along the street. Generally good agreement is obtained for theblock on Hastings between Hornby and Howe, although large portions of the block receivedirect radiation in the afternoon due to some lower buildings. Agreement for the blockon Dunsmuir between Howe and Hornby is much poorer in the early afternoon whenmodelled temperatures are substantially lower than observed. Again, this is probablyChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS(a)U35:i 3002520(c)154035C) 30252015(b)232Figure 6.12: Modelled and observed road surface temperatures in the Downtown studyarea:(a),(b) NE/SW street canyons; (c),(d) NW/SE street canyons.due to variations in the canyon geometry at scales too small to be resolved by the model.6.5.2 Roof TemperaturesRoofs are an important component of the urban surface and form a relatively simplesurface to model in that canyon effects can be neglected (at least in those areas wherebuilding heights are relatively constant). Roof surface temperatures are modelled andcompared both to an individual building (Summer Equipment site) and for the Industrialarea and Residential area as a whole (Figure 6.13).As input a roof emissivity value of 0.92 was used based on the results of Artis and— Modelled Downtown—e-— Observed (Block) Nw/SE StreetA Trv Pt 1 3 ] (CAST: HORN—HOWr)V TrovPtl5 V4035302520(-)F—15O 2 4 6 81012141618202224Time (Hours, POT) Time (Hours, PDT)(d)DowntownNE/Sw Street(HORN: OFJNS—PENI5)Modelled—0— Observed4035C) 302520150246 81012141618202224 024681012141618202224Time (Hours, PDT) Time (Hours, PDT)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 233Figure 6.13: Roof surface temperatures for: (a) the Summer Equipment site and Industrial area, (b) north- and south-pitched roofs in the Residential area.Carnahan (1982). Albedo was set at 0.10 and the thermal properties ic and C were setto 1.55x107m2 s and 3.26x105 J m3 K—’ respectively (based upon a roof construction using asphalt shingles overlaid upon a plywood base with polystyrene insulation(ASHRAE, 1989). It was necessary to modify the model to handle nighttime surfacetemperatures which were initially greatly overestimated. The modifications include:• setting the latent heat flux to zero;• reducing the nighttime sensible heat flux to 0.15 of the calculated value;• replacing screen-level Ta in the atmospheric longwave emission formulation with abulk atmospheric temperature of 10°C;• reducing z0 to very low values.Nighttime observed surface temperatures are significantly below air temperature, andit was desired to reduce the surface energy budget of the model to a balance betweennet radiative loss and heat conduction, however, tests showed that a solution could notbe obtained when the sensible heat flux was reduced to values less than 15% of the(ci) (b)C-)6050403020100—10SurnrserEqcipoent / Indostriol0F-706050403020to0Residentiol Irnoge Extrocted (It 7)Pitched Roofs—e--— North— —0--- Sooth,04e—0-,,’0 2 4 6 8 1 0 1 21 4 1 6 1 8 20 22 24Time (Hours, PDT)0 2 4 6 81012141618202224Time (Hours, PDT)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 234original for the z0 and thermal properties specified. Reducing the sensible heat fluxand removing the latent heat flux (preventing dewfall) have the effect of reducing themodelled nighttime temperatures by approximately 5°C.The observed temperature of the roof at the Summer Equipment building site showsa lag when compared with the modelled temperature in the morning heating curve,similar to that observed for road surfaces (Figure 6.13(a)). The afternoon and eveningcooling period and night time temperatures are well replicated by the model. The imageextracted roof temperature for Flight 2 (Industrial study area) agrees well with themaximum modelled value.Tests of the model to shifts in the hourly input data, slight surface obstructions andusing measured rather than modelled incoming shortwave radiation do not adequatelyaccount for the observed lag. The addition of a wet surface in the morning hours (i.e.,following dewfall) was able to account for some of the lag, but resulted in surface warmingbefore sunrise which was not present in the observations. The lag seems most likely tobe related to the layering of materials with different thermal properties below the surfaceand/or the low effective thermal admittance of building surfaces.In the Residential area, (Figure 6.13b), south-pitched roofs (a 25° pitch is assumed)are underestimated, although with the high surface temperature and low emissivity, largecorrection values are applied to the image temperatures (in excess of 8°C for the atmospheric correction and 5°C for the emissivity correction). Modelled north-pitched rooftemperatures agree well with temperatures extracted from the early afternoon flight(Flight 7).6.5.3 Grass SurfacesGrass surfaces are an important surface component in urban areas, and may be includedas a component of the horizontal surface of SC. Measured grass temperatures and net40 iIIII!ILIIII!I 600Tq(MOd)0 Tg(Obs) 500._._._Q (Obs) 40030300EA’\4\ 20020100c?%O-.10 i I I—1000 2 4 6 81012141618202224Time (Hours, PDT)Figure 6.14: Modelled and measured surface temperatures and net radiation for a grasssurface in Trafalgar Park.Agreement between measured and modelled net radiation and surface temperature isvery good during the daytime (Figure 6.14), although there is a slight overprediction ofnet radiation in the morning and an increasing difference (model overprediction) afterthe time of maximum surface temperature. Nighttime surface temperatures are (5—8°Ccolder than predicted, while the net radiation is slightly larger than observed. TheChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 235radiation are available from a park study (Spronken-Smith, 1994). The site is in a largepark so canyon effects are neglected. Input parameters are set to typical values for shortgrass: o=0.24,€=0.95, ,c = 0.5x106m2 s’, C=1.8x 106 J m3 K’,z0=0.0O1 (e.g., Oke,1987). The fractional surface moisture parameter was set at 0.5, with the considerationthat measured soil moisture was approximately 13%, the underlying soil had a high sandcontent, and the grass surface was not exhibiting visible signs of water stress (i.e., lossof green colouration).C)0V0’F-Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 236discrepancy between modelled and measured surface temperatures may be due to poorthermal coupling of the grass with the underlying soil at night: the tips of the grass(preferentially viewed by an off-nadir IRT) cool radiatively and are much colder than thebase of the grass and soil surface.6.5.4 Facet Temperatures6.5.4.1 An Individual BuildingAs part of a parallel study, wall surface temperatures of a single building in the Industrialarea (Summer Equipment site) were monitored and are available for comparison with themodel. Input data are shown in Table 6.6 and the results are portrayed in Figure 6.15.Table 6.6: Model input: Summer Equipment building facets. Facet azimuth refers to thedirection in which the facet faces (i.e., North = North-facing). Subscripts w and f referto walls and floor respectively.Parameter Units North South East Westcxf fraction 0.05 0.10 0.20 0.20fraction 0.20 0.20 0.20 0.20fraction 0.94 0.95 0.95 0.95fraction 0.98 0.98 0.98 0.98m2 s1 O.86x 106 O.86x106 0.86x 1O_6 O.86x106C J m K—’ 1.32x106 1.32x10 1.32x106 1.32x10m 1.0x10 1.0x106 1.0x10 1.0x106H m 7 5 5 5W m 13 25 12 29BL m 100 100 35 35Canyon geometry was estimated from field observations. Surface thermal and radiative properties are estimates for yellow-painted hollow concrete blocks (ASHRAE, 1989)estimated in the manner of Hutcheon and Handegord (1983).The general shape of the observed time series for each facet is replicated by the model,Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 237with the exception of the west wall, which, perhaps due to local shading effects, showsmuch cooler surface temperatures in the morning. The peak modelled temperature inall cases precedes the observed data, similar to the results obtained for road surfacetemperatures.45403530252015“V4035302520Figure 6.15: Modelled and observed surface temperatures for the Summer Equipmentsite (Industrial area).North wall temperatures are consistently above those observed (which are very closeto air temperature) during the day; this is a function of heating by diffuse shortwaveradiation. Setting canyon albedo to zero results in better agreement between observedand modelled values but is not a realistic model value.4035Z 30252015Modelled Summer EquipmentObserved North WellT024681012141618202224Time (Hours, PDT)45Summer EquipmentSouth Well40— Modelled35 Observed30250 2 4 6 8 101214 161820 2224Time (Hours, PDT)15C)FC-)I-C)I—Summer EquipmentEest WellModelledObservedSummer EquipmentWest WellModelledObservedO 2 4 6 81012141618202224Time (Hours, POT)0 2 4 6 81012141618202224Time (Hours, POT)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 2386.5.4.2 Canyon Facets in the Study AreasThe modified Myrup model was next compared with the average traverse results fromthe three study areas (Figures 6.16, 6.18, 6.20 and Table 6.7). The plotted error barsindicate ±1o from the mean of the traverse data. Where present, the solid symbolsrepresent the values obtained from airborne thermal imagery.Table 6.7: Model input: Industrial, Downtown and Residential Sites.Parameter Units Industrial Downtown Residentialcf fraction 0.05 0.05 0.05fraction 0.25 0.35 0.20fraction 0.94 0.93 0.93fraction 0.95 0.95 0.95m2 s 0.86x 106 0.38x 10_6 0.37x106C J m3 K1 1.32x 106 1.94x 106 0.76x 106m 0.0001 0.0001 0.0005H m 8 35 9W m 22 30 33BL m 100 88 80Shadow Ratio fraction 0 0 0.2-0.5Wet Fraction fraction 0 0 0.11The simulated temperatures in the Industrial area are warmer than the mean traversetemperatures, particularly for the west and south canyon facets. Because of the mixedFOV problem and perhaps due to neglect of shading devices such as awnings, this findingis not too surprising. However, shaded facet temperatures are also warmer than observed.This leads to facet-pair differences which agree in magnitude with those observed. Theearlier peak temperatures from the model are again evident in the difference plot.Where hourly data are unavailable, the model must be run using mean daily inputconditions for air temperature, vapour pressure and windspeed. The effect of using mean6 8 10 12 14 16 18 20 22Time (Hours, PDT)6 8 10 12 14 16 18 20 22Time (PDT)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 239(a)C-)0‘3)00EI—(b) 45403530252015Time (Hours, PDT)C)C00ESI—Symbols: ObservedLines: Modelled(c) 1510‘A d’ A5 ,‘A /.,A*/ •e-4-e\_A\ p5 q• W—E Obs—10 s. A S—N Obs—0-- W—E Mod-15---S-N ModFigure 6.16: Comparison of modelled and traverse facet temperature for the Industrialstudy area. (a) east and west facets, (b) north and south facets, (c) facet pair differences.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 240daily input on modelled canyon facet temperatures and differences was examined for theIndustrial study area (Figure 6.17). Differences between facets show an increase overthose from the hourly modelled data, but it is not excessive, especially since the traversedata are probably biased on the low side for warm facets. Since the use of hourly airtemperatures had a noticeable effect on the phase lag of the temperature curves, it wasanticipated that the use of the mean daily data might also alter the temporal positionof the maximum differences. The effect is relatively minor for the area tested. Thisimproves confidence in the results when daily data are used.Modelled temperatures for Downtown study areas replicate the observed pattern wellbut again modelled values are always higher (Figure 6.18). The observed temperaturesare from the 30° EIRT, so as to obtain surface temperatures closer to the wall midpoint.A low bias in the traverse observations is more probable in this area because of thelarge H:W ratio and the shadows cast by tall buildings. Modelled facet-pair temperaturedifferences agree well with those observed, although the morning peak occurs somewhatearlier in the model.In the Residential area, comparison between modelled and observed temperatures ismade more difficult by the presence of intervening vegetation which is not included in thecanyon representation. The vegetation (in particular the street trees) shade the buildingwalls and form a surface in themselves which is viewed by the EIRT. To accomodatethis effect, the wet fraction and shaded fraction parameters of the Myrup model wereemployed. The wet fraction was set to 0.11 which is the fractional area of all verticalsurfaces made up of trees (vertical plane projection of the model tree canopies). Thisassumes the trees were transpiring. Shadow fraction was calculated from a considerationof shading by street trees, roof overhangs and balconies. Simplistic representations wereemployed for the tree shading: a tree was considered to be a rectangle with a heightC-)0a,D0a)0Ea,I—C)a)0a)0.Ea>I-Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 241(r-1 50 Lines & Symbols: Doily T0, e, u45 Lines: Hourly T0, e, u4035 /P \30 ,‘,/,,25 /----.‘‘ e- --‘ -20156 8 10 12 14 16 18 20 22lime (Hours, PDT)(5 50Lines & Symbols: Daily T0, e, u45 Lines: Hourly T0, e, u40353025h20 - - -_ --156 8 10 12 14 16 18 20 22Time (Hours, PDT)15-A_10 ,.4— --ç ‘/7 a’ ‘5 70 /—5‘k”\‘ - - — — W— (Hourly)1 0 . — - —— S—N (Hourly)a,--d-- W—E(Doily—15 °- — — S—N (Daily(c)C-)06 8 10 12 14 16 18 20 22Time (P01)Figure 6.17: Comparison of modelled canyon facet temperatures obtained using hourlyand daily input data (Industrial area).C.)ci,:30ci)0Eii)I-Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS(a)2424035C)0cc) 30:3025Eci)‘ 2015(b) 40Symbols; ObservedLines: Modelled6 8 1 0 1 2 1 4 1 6 1 8 20 22Time (Hours, PDT)Symbols: ObservedLines: Modelled353025201515105(c)6 8 10 12 14 16 18 20 22Time (Hours, PDT)C-)8.- 0—5—1 0—1 5_i)-&A A *_ •. ‘A4\. ‘•. W—E ObsA S—N Obs—o—• W—E Mod--0-- S—N Mod6 8 10 12 14 16 18 20 22Time (PDT)Figure 6.18: Comparison of modelled and traverse facet temperature for the Downtownstudy area. (a) northeast and southwest facets, (b) northwest and southeast facets, (c)facet pair differences.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 243equivalent to the mean tree height and an area equal to the mean projected tree (vertical plane) area, placed 10 m from the building. Shadow lengths and orientations werecalculated for the daytime period. When shadows intersected the building wall plane theshaded area was calculated. These shadows are maximized at lower zenith angles (earlyand late in the day). Shading by roof overhangs (estimated as 0.45 m) and balconies(2.1 m; 37% of all buildings) is maximized at small zenith angles leading to an overalltemporal shading fraction which exhibits a “W” shape (Figure 6.19).p0.10.0_____________________________________5 7 9 11 13 15 17 19Time (Hours, LAT)Figure 6.19: Estimated temporal variations of shadow ratio for vertical facets in theResidential area.Model results for all facets show temperatures slightly above the mean observed traverse temperature (Figure 6.20). In most cases the differences between facet pairs cancel,yielding very good agreement between the observed and modelled temperature differences. An exception is the west-east difference in the afternoon when, because of warmermodelled west facet temperatures, the difference is greater than that observed. Thereare slight offsets in the timing of the maximum differences, with the modelled differencespreceding the observed in each case.(200D5,0E5,C)0a)z0a)0Ea,I—C-)06 8 10 12 14 16 18 20 22Time (Hours, PDT)Symbols: ObservedLines: Modelled/,4±710121416182022Time (Hours, PDT)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS(a)24450Symbols: ObservedLines: Modelled40353025(b) 4540353025201515105(c)--‘: --- E’4 cc’40—5—10—150’.... W—E ObsA S—N ObsW—E ModS—N Mod6 8 10 12 14 16 18 20 22Time (Hours, PDT)Figure 6.20: Comparison of modelled and traverse facet temperature for the Residentialstudy area. (a) east and west facets, (b) north and south facets, (c) facet pair differences.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 2456.6 SummaryThe modified Myrup model is able to correctly reproduce the form of surface temperaturevariations for canyon facet and road surface temperatures when local, hourly input dataare available.To achieve best results for surfaces exhibiting large diurnal temperature ranges, extremely low values of z0 are required, and, for roof surfaces, the magnitude of the nighttime sensible heat flux has to be limited. Maximum surface temperature for many of thecanyon surfaces (walls, roofs and roadways) was consistently modelled to occur before thetime of the observed maximum. This effect was not present for the moist grass surface.The exact cause of this effect is not known but may be due to one or several of:• layering effects in the substrate or wall materials (especially the low effective thermal admittance of hollow buildings, Goward (1981));• a slight occurrence of surface moisture (e.g., from dewfall) in the morning;• local shading effects.Excellent agreement between measured and modelled K4. would appear to preclude thethird effect. Because the largest phase shifts were observed for a roof and building, andbecause dewfall is not commonly noted on vertical facets (Richards, peTs. comm.) it issuspected that layering effects are the most probable cause for the phase shift.Temperature differences derived from modelled surface temperatures between canyonfacets agree very well with observed results; the phase shift evident in individual facetsis reduced in the difference results. The use of mean daily input conditions overestimatesthe hourly modelled and observed results slightly, but given the low bias in the observedresults, is not considered to be unduly large. In view of the simplistic nature of the MyrupChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 246model and the limited canyon effects incorporated, the results obtained are consideredto form a good basis for the surface temperature input requirements of SC.6.7 Application of the SC Model6.7.1 Application to the Study Areas: Overflight DaysThis section presents results using the SC model with input data selected to represent theconditions present during the times of the remote sensing overflights. Surface structuralparameters H, BL, Wa, and W have been estimated from the surface structural database(Chapter 5). Surface parameters for the study areas are presented in Table 6.8. Exampletwo-dimensional surface profiles derived from the surface structural database for eachof the study areas as well as a graphic representation of the two-dimensional surfacestructure are presented in Figures in the subsequent subsections (Figures 6.24, 6.21,and 6.27).Table 6.8: SC model surface parameters for the Study areas.Parameter Units Industrial Downtown ResidentialCanyon Azimuth (°) 90 45/135 0Sensor Azimuth (°) N,S NW/SE, NE/SW E/WH m 7.3 35 5.5BL m 32.5 30 23.8W m 22.0 22 32.5WA m 12.0 30 18.8m 5000 5000 5000IFOV degrees 12 12 12YD 228 229 230Surface temperatures may be obtained using vehicle traverse data, remotely-sensedChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 247imagery (image-extracted distributions), or modelled temperatures. In the results whichfollow, image-extracted temperatures were used when available, supplemented where required with vehicle traverse temperatures. When a wall or roadway is partially shaded,temperatures for the sunlit and shaded components may be estimated from the mixed-distribution modelling results of Chapter 3. Tests of the use of mean values versusfrequency distributions yielded differences generally less than 0.2°. DowntownSurface structural parameters in the Downtown study area are difficult to estimate because of the wide range of building heights. In the NE portion of the study area, buildingheights are more nearly equal and a symmetrical block structure exists with generallytwo large buildings per block, separated by a narrow alleyway (Figure 6.21) which bettermatches the capabilities of the SC surface representation (Figure 6.21c).The symmetry in the block structure allows all four view directions to be tested.Input conditions for the SC Model were set to match those observed during Flight 4. Atthis time both SE and SW facets were directly irradiated (although the SW facets werejust beginning to warm following direct solar exposure).Model estimates of proportions of the FOV occupied by the component surfaces arepresented in Figure 6.22. As the off-nadir view angle increases, wall areas increase andground area decreases; for angles greater than 40° the fraction of ground viewed is 0for the input surface conditions. When viewing in the NW direction, there is a slightincrease then decrease of shaded wall which is due to projected shadows on the viewedwall which are initially seen at small off-nadir angles and then become obscured as theview angle increases.Modelled temperatures detected by the sensor for view angles from 0-60° (includingthe mean observed 45° off-nadir Tr) are shown in Figure 6.23. In the view directionChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS75160‘— 45cj 30ci)075160‘45-cci’ 30075160‘— 45c, 3000 50 100 150 200Distance (m)0 20 40 60 80 100 120 140 160 180 200Distance (m)248Figure 6.21: Surface profiles for the NE portion of the Downtown study area. (a) and(b) are example profiles extracted from the surface database in the NE/SW and NW/SEdirections respectively. Solid and dashed lines represent two separate profiles. (c) is thesurface profile defined by W5=22, Wa=1O, H=27.5, BL=30 used in the SC model.0 20 40 60 80 100 120 140 160 180 200Distance (m)250 300Downtown Model SurfaceU_j.__Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 249Figure 6.22: Surface proportions viewed: Downtown, Flight 4. View directions are: (a)NW, (b) NE, (c) SE, (d) SW.0 0 0 0 0 0 0View NW —e— RoofF V — 1 — — — Sunlit Ground—--0-- ShadedGroundH = 27.5—0—- Sunlit Wall= —0— Shaded Wall0W30.0>-.- _..:--_..r—r-—--- -: -o io 20 30 40 50 60View Angle (Degrees)(b) :> 0.50U0.4C.2 0,30.2010.0() :> 0.50U0.4C.2> 0.50U..0.4C.2 0 0 0 0 0 0View NE: —0— RoofFOV = 1 2°-‘0-- Sunlit Ground-‘0-- ShadedGroundH = 27.5—0--- Sunlit Wall= —v— Shaded WallW=30.0_.0A— “A..- - -0.. --.0 10 20 30 40 50 60View Angle (Degrees)o o 0 0 0 0 0View SW —0--- RoofFOv = 1 ° - -A- - Sunlit Ground-‘D-- Shaded GroundH 27.5—0— Sunlit Wall= -g —v— Shaded WallW=30,0-Ta,.. —‘a --.0 10 20 30 40 50 60View Angle (Degrees)View SEFOV= 12°H = 27.5S = 22.0A = 10.0W = 30.0‘a• - -- -.—e— Roof— -A—— Sunlit Ground—-0-— Shaded Ground—0-— Sunlit Wall—a—- Shaded Wall—‘v-.—-—•—•v—-—-—’—a>0U0C0C-)0U-•010 20 30 40 50View Angle (Degrees)60Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 250(NW) of the most directly irradiated (and warmest) facet, increasing off-nadir view angles result in constantly increasing apparent temperature registered by the sensor up toapproximately 40° after which the fractional FOV occupied by the wall changes onlyslightly. For the NE and SW view directions, the modelled sensor temperature decreasesslightly with increasing view angle due to the replacement of sunlit ground surfaces withshaded wall surfaces. In the SE view direction, only minor changes in temperature withview angle are obtained. This arises because both the NW canyon facet and the entirecanyon floor are shaded, and approximately equal in temperature.The model overestimates the apparent surface temperature within the sensor FOVfor all view directions. Differences are slightly greater for the shaded facets comparedto the irradiated. Nadir temperatures differ between the two canyon orientations, anartifact of the two-dimensional surface representation. Modelled nadir temperature issignificantly warmer than observed, particularly that associated with the NW and SEview directions (NE/SW street canyons). This occurs because these street canyons haverelatively little shade at this time; the solar azimuth is very nearly aligned with thecanyon axis, and the model does not take into consideration the cross-streets (completelyshaded; Figure 6.22). Despite the large differences in observed and modelled temperature,the model does replicate the order of temperatures between the different view directions(i.e., NW view direction is warmest, followed by nadir, and the rest are very close intemperature; compare with Figure 1.34). The magnitude of the 45° off-nadir temperaturedifferences with view direction are also fairly well represented. IndustrialThe Industrial study area was selected in part because of the relatively simple nature ofthe surface (box-like buildings, lack of vegetation). In several of the blocks, buildings, orsequences of buildings, extend the entire length of the block (Figure 5.4) which providesChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 2513937F- 350(n 33Ccn 312927___25232 1 • I I0 10 20 30 40View Angle (Degrees)Figure 6.23: Variation of the modelled AGEMA scanner temperature for Flight 4 withview angle over the Downtown study area.the long canyon structure assumed by .the SC model. The surface structure across thebuilding row direction (N/S) is well represented by the parameters H, BL, W3, and Wa(Figure 6.24).In the E/W direction, along the building rows, the model is less able to replicate thecombination of small, nonregularly occuring inter-building spacing and unequal buildingheights. The results which follow are restricted to the N/S view azimuths across thecanyon structure.Modelled apparent surface temperatures for the 12° FOV for north and south viewdirections over the model Industrial surface at 1030 PDT are presented in Figure 6.25.The mean apparent surface temperatures obtained from the nadir and 45° off-nadir flightI I I I IIISensor View Direction(Solid Symbols: Observed)o NWSEO NEp SWX Nadir50 60lines over the study area are again included for comparison.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 252Industrial: E/Wo 100 200 300 400 500Distance (m)1 5 Industrial: ‘N/s • I1:_i0 100 200 300 400 500Distance (m)1 Industrial: Modelled N/S -I 10I0 100 200 300 400 500Distance (m)Figure 6.24: Surface profiles for the Industrial study area. (a) and (b) are exampleprofiles extracted from the surface database in the E/W and N/S directions respectively.Solid and dashed lines represent two separate profiles. (c) is the surface profile definedby W8=22, Wa=12, H=7.3, BL=32.5 used in the SC model.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 2533534H0 10 20 30 40 60View Angle (Degrees)Figure 6.25: Variation of the modelled temperature seen by the AGEMA scanner withview angle in north and south view directions for Flight 1 over the Industrial study areafor (solid line) original SC model output, (dashed line) including along-canyon interbuilding spacing and cross-streets (spacing between blocks).In the morning, the temperature contrast between north and south facets is still developing and has yet to reach its maximum value. Off-nadir observations in the northview direction show a slight increase in apparent temperature as the shaded ground surfaces are obscured and replaced by sunlit road and wall surfaces. At angles greater than300 the apparent temperature begins to decreases because the wall surface is cooler thanthe sunlit road. The south view direction shows an almost linear decrease in temperatureas shaded surfaces occupy greater proportions of the FOV.The dashed lines of Figure 6.25 test the effect of incorporating extra “ground” surfacesto represent cross-streets and inter-building spacing. These ground areas were estimatedfor the study area and divided into sunlit and shaded portions based upon projectedshadows for the mean building height (58% sunlit, 42% shaded). The incorporation of0 eSofld Symbols: Observedo ViewNt ViewS50Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 254the extra ground surface has the overall effect of increasing the shaded area and reducingall temperatures. It also reduces the off-nadir differences with view angle and betweenview directions because there is a smaller proportion of wall area. Compared to theobserved temperatures, the original model (which assigns all ground surfaces to roadsurface temperature), yields much warmer apparent surface temperature estimates. Thedifferences are greater for the south view direction (which “sees” more shaded surfaces”.The addition of the extra ground surface provides a better estimate of the nadir temperature and cooler off-nadir temperature, but now under predicts the warmest off-nadirtemperature.Model results for Flight 2 over the Industrial area (Figure 6.26 show a similar patternto those described for Flight 1 with an enhancement of both the off-nadir and directional differences due to the greater wall temperature difference at this time. Agreementbetween observed and modelled values is improved over that of Flight 1 without modification of the model. At this time, shadows are largely confined to the E/W streetsrepresented by the model, and open areas are approaching roof temperature so thatthe neglect of these surface components in the two-dimensional model is of less importance. The difference between observed and modelled temperatures is clearly greaterwhen viewing shaded surfaces, and this may be an indication of the importance of smallscale surface structure, not represented by the model. When viewing the “real” surface in the direction of the irradiated facet, microscale surface structure has no effectupon the sensor temperature because shadows are not viewed. However, when neglectingsmall scale surface structure (e.g., elevator shaft housings, balconies, roof vents in theup-Sun direction), all their shaded areas are neglected and lead to an overestimate of themodelled temperature.Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 255444342F—0CoCCl)0 39ci380 373635_____________________________________________I I • I I I •0 10 20 30 40View Angle (Degrees)Figure 6.26: Variation of the modelled temperature seen by the AGEMA scanner withview angle in north and south view directions for Flight 2 over the Industrial study area. ResidentialThe Residential study area is the most difficult surface structure to replicate in the SCmodel due to the inability of the model to handle trees or sloping roof surfaces. As withthe Industrial area, the model has been applied only to view directions orthogonal to theblock axis; in the study area subset, the block long axis is oriented N/S so only E/Wview directions are considered here. Surface profiles extracted from the surface databaseand the modelled surface profile are presented in Figure 6.27.Note that the presence of garages cannot be included in the model surface; in somecases these buildings form a second canyon bordering the alley, although their occurrenceis more irregular than that of the buildings. The effective width of the building in thesurface model has been increased slightly to account for the width (roof area) of thegarages, but their wall area and the increase in ground shading due to their presence isSolid Symbols: Observed0 ViewNViewS I50 60Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS8—‘ 7E 6‘— 50’ 3) 208— 7E 6—, 50’ 3•5 2E 108,—‘ 7E 6S— 5.2’ 3) 20400400256Figure 6.27: Surface profiles for the Residential study area. (a) and (b) are exampleprofiles extracted from the surface database in the N/S and E/W directions respectively.Solid and dashed lines represent two separate profiles. (c) is the surface profile definedby W3=32.5, Wa=18.8, H=5.5, BL=23.8 used in the SC model.0 40 80 1 20 1 60 200 240Distance (m)0 100 200 300Distance (m)0 100 200 300Distance (m)Chapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 257not accounted for.The SC model was run for conditions representative of Flights 6 (0945 PDT) and 7(1400 PDT) (Figures 6.28 and 6.28). For each run, the canyons were considered to beboth inifinitely long (solid lines in Figures 6.28 and 6.29) and to include additional groundsurfaces representing the inter-building and street spacing along the block (dashed lines).Because of the extensive vegetated surface area in the residential area, the fractional FOVoccupied by sunlit and shaded ground is further subdivided into paved and vegetatedsurfaces, each of which were assigned temperatures derived from the remotely sensedimagery.28-o 27 Solid Symbols: Observed0o ViewEViewW2524I I Io io 20 30 40 50 60View Angle (Degrees)Figure 6.28: Variation of the modelled temperature seen by the AGEMA scanner withview angle in east and west view directions for Flight 6 over the Residential study area.Results for conditions representing Flight 6 (Figure 6.28) show the apparent temperature seen by the sensor decreases with increasing view angle for the east view direction,and to increase by an approximately equal amount for the west view direction. IncludingChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 258the extra open ground has the same effects as discussed for Flight 1. Nadir and eastview direction temperatures are substantially warmer than observed, while the west viewdirection difference is much less.A further difficulty in this area is the large number of trees which are not included inthe surface representation. Shading by trees is one possible factor which can account forthe much greater difference between observed and modelled temperatures in the directionof the shaded facet: in this direction both the cool side of the tree canopy as well as theshadow cast by the tree are visible and contrast with sunlit ground surfaces whereas,inthe down-Sun direction, the shaded area can be wholly or partially obscured by thecanopy, depending upon the particular Sun-surface-sensor geometry.4842- Solid Symbols: Observedoo ViewEViewW3836I I I • I IO 10 20 30 40 50 60View Angle (Degrees)Figure 6.29: Variation of the modelled AGEMA scanner temperature with view anglein east and west view directions for Flight 7 over the Residential study area.Off-nadir variations with view angle are only minor for the model results from Flight7. Both view directions show a slight decrease of apparent temperature seen by the sensorChapter 6. MODELLING URBAN SURFACE THERMAL EMISSIONS 259with off-nadir viewing angle. These are not unexpected given that the temperatures of thecanyon walls are approximately equal at this time. The difference between observed andmodelled temperatures are again large but the relative difference between view directionsand between the nadir and 45° off-nadir viewing angles is reproduced well by the model.6.7.2 SummaryA slightly modifed version of the Sobrino and Casselles (1990) model was run using inputdata representative of the across-block surface structure in the three study areas. Resultswere compared with the nadir and 45° off-nadir mean apparent surface temperatures obtained from the helicopter-mounted scanner. Large differences exist between observedand modelled results. These are attributed to the two-dimensional surface representation employed by the model, the simplistic representation of the surface structure (andomission of important elements such as trees from the surface model), and to biases inthe image-extracted component temperatures. The model performs best when the solarazimuth is orthogonal to the canyon azimuth so that shadows are confined to the viewdirection of the sensor. Despite the large modelled-observed differences in temperature,the model does correctly portray the relative difference of facet temperatures with viewdirection and appears to replicate the pattern of temperature change with increasingoff-nadir angle, although this conclusion is limited by the lack of other observed off-nadirangle data.Chapter 7CONCLUSIONSThe overall objective of this thesis was to investigate how the three-dimensional formof the urban surface affects observations of radiative surface temperature made by thermal infra-red remote sensors. The thesis is observationally based and employs a uniquecombination of sensor platforms in an effort to fully define the temperature of the complete urban surface. This approach has allowed the detailed examination of both theanisotropic emissions from selected urban land use areas, and the creation of a databasefrom which estimates of the complete urban surface temperature can be made. Theseobjectives would be difficult and/or prohibitively expensive to obtain using more conventional (i.e. satellite- or aircraft-based remote sensors).7.1 Summary of ConclusionsIndividual chapters provided detailed summaries of findings. In view of the overall objectives outlined in Chapter 1, the main conclusions of this thesis may be described as:• Distributions of vertical facet apparent surface temperature obtained from vehicletraverses demonstrated large variations with facet orientation and strong temporal variation, particularly of the directly irradiated facets. The spatial variationsare conditioned by variations in building properties at the land use scale, and arealso influenced by the amount of vegetation “seen” by the sensor. Temperaturedifferences between facet pairs are greater than 10°C in the Industrial area and260Chapter 7. CONCLUSIONS 261slightly less in the Residential, with the largest difference occurring mid-morningbetween east and west facets. Other times of maximal temperature differences between facet pairs occur within approximately 1 hour after solar noon between northand south facets and in the late afternoon between east and west facets. In theDowntown area, the different street orientation yields one time period with a largefacet temperature difference and a second where the differences are minor, withresults showing a fairly high degree of symmetry. These differences are an important factor in the establishment of anisotropic surface emissions from urban areas.Spectral reflection by low emissivity surfaces may pose a problem in some land-useareas; the effect of such reflections depends upon the sensor location relative to theobserved surface.The analysis of distributions is complicated by the presence of “mixed-pixel” observations because the FOV of the sensor allows portions of both sky and buildingto be seen. Distribution truncation and distribution modelling were investigatedas potential ways of separating facet temperatures from mixed building-sky or skyobservations. Distribution modelling allows separate component populations oftemperatures to be recovered. It is shown to successfully discriminate shaded andsunlit temperature distributions on road surfaces. Results for walls are dependentupon the surface materials and radiation regime: small scale temperature variations (at the building scale, or smaller) of directly irradiated facets within thesensor IFOV leads to a distribution with a long “tail” of warmer temperatureswhich is not easily separable into component Normal populations.• A simple model of the apparent surface temperature distributions created by moving a sensor FOV across a sequence of model buildings with specified temperatureChapter 7. CONCLUSIONS 262distributions which are separated by gaps emitting at a specified sky radiant temperature, was shown to give qualitative agreement with observed distributions.More precise agreement may result with improvements in the representation of thesurface structure.Frequency distributions of apparent surface temperature in each of the study areasobtained from an airborne thermal scanner reveal a characteristic bimodal shapewith the lower peak related to shaded surfaces and the upper peak coinciding withsunlit road surface temperatures. The distribution is most strongly defined inthe morning when differential heating of surfaces yields the greatest variation intemperature between surface types. Distributions from land use areas characterizedby more “simple” surfaces (regular, block-like buildings, and lack of vegetation) leadto more sharply defined temperature distributions.• Mixed distribution modelling techniques were applied to the remotely-sensed apparent surface temperature distributions and frequency distribution differences withthe objective of recovering component surface temperature information. The bestresults were obtained using morning temperature frequency distributions from theIndustrial area, which has the most strongly defined composite temperature frequency distribution. Sunlit road, roof, and shaded surface components were themost successfully extracted components. Vertical facet temperatures, and particularly the most directly irradiated facet, were more difficult to extract. This is notsurprising given the wide range of building materials which exist within the studyareas. Significant differences between traverse and image-extracted estimates ofthe most directly irradiated facet were noted, and are attributed to the observationmethodologies employed, and possible spectral emissivity effects.Chapter 7. CONCLUSIONS 263Strong directional variations (anisotropy) of apparent urban surface temperaturewere observed over each of the study areas. Differences were greatest over theDowntown study area (> 9°C) and coincide with the time of maximum surfacetemperature differentiation between opposing canyon facets. Anisotropy varies withtime and view direction depending upon the orientation of the surface and the solargeometry for that time of the year. Temperature differences due to anisotropy overurban surfaces are shown to be equal to or greater in magnitude than those which resuit from surface emissivity and atmospheric absorption/emission (depending uponthe Sun-sensor-surface geometry). This emphasizes the importance of including aconsideration of anisotropy when using thermal remotely-sensed imagery over urban areas.Complete surface temperatures which attempt to combine temperatures of all themajor surface orientations present in the study areas, have been estimated for thefirst time. Frequency distributions show an enhancement of low temperatures whichyield mean values close to the coolest observed off-nadir apparent surface temperature. This suggests that, where available, off-nadir thermal imagery in the directionof the coolest (most shaded) facet may be the most representative remotely-sensedtemperature to use where a temperature representative of the entire urban surfaceis desired. This approximation appears to be least valid at mid-day.• Geometric projection models are advanced as a possible means of extending theobservational results of anisotropic surface emissions to other times, locations andview angles. A simple two-dimensional model (based upon the work of Caselleset al. (1992) and Sobrino and Caselles (1990), and slightly modified to betterrepresent the study area surface profiles in the across-block view direction), wasapplied to the study areas and compared with observed results. The model is ableChapter 7. CONCLUSIONS 264to replicate the order of temperatures in different view-directions, but shows largedifferences from the airborne scanner results in the direction of shaded facets. Themodel performance is limited by the two-dimensional surface assumption which isnot representative of most urban land use areas.A simple one-dimensional energy balance model was modified to incorporate canyonradiative effects and allow representation of vertical facets for the purpose of estimating temperatures of canyon surfaces in geometric projection models or toallow estimates of complete surface temperature in combination with the buildingdatabases for each of the study areas. The model performed reasonably well for arange of surface types, considering its simplicity. More recent models, incorporatinga layered sub-surface, and better surface-air temperature feedback are advocatedfor future work.7.2 Suggestions for Future WorkThe results of this thesis have indicated the potential for further research in a number ofareas.The most important and direct extension of the current work is the .developmentand application of a geometrical sensor-surface model which would incorporate a three-dimensional urban surface representation as well as solid angle sensor geometry. Whencoupled with a surface energy balance model, the extent of anisotropy over a range ofsurface geometries, locations and times could be investigated.In conjunction with the development of a three-dimensional model, several otherimprovements are suggested including:• Application of more recent and realistic canyon energy balance models (e.g., Mills,1993) to better characterize the component surface temperatures.Chapter 7. CONCLUSIONS 265• Addition of building material data at the same scale as the building height database.This would allow the identification of typical surface material combinations foruse in temperature modelling, as well as providing a link between the observedtemperatures and building properties.• Incorporation of solar geometry/shading routines to the building database to define three-dimensional surficial shading patterns. When coupled with temperatureestimates from energy balance models, this would allow complete surface temperature estimates to be modelled for the surface geometries characteristic of the studyareas.Complete surface temperatures have been estimated for urban surfaces for the firsttime. They form a new temperature which has not been directly confirmed observationally. Pyrgeometers or other wide FOV instruments capable of viewing vertical as wellas horizontal surfaces should be tested over urban surfaces to determine if these provide a better representation of complete urban surface iemperature than do narrow FOVremote sensors. It is also recommended that the feasibility of estimating the completesurface temperature using component surface temperatures from a few selected surfacesbe investigated. Complete surface temperatures also need to be examined in relation tothe surface energy balance.Further work is also suggested on the utility of the mixed distribution modellingtechnique as applied to the identification of component surface temperatures at variousscales. The simple model used to explore the form of surface temperature distributionsmeasured by the traverse vehicle should be further investigated to determine if it can beinverted and used to separate building and mixed-building and sky temperatures.The current work has clearly identified the presence of strong anisotropic emissionsover urban areas at the land use scale. The extent to which it is present in currentChapter 7. CONCLUSIONS 266and past operational remote sensors (primarily satellites) should be investigated andcompared with the present observations. The extension of modelling will also aid in thisanalysis.References[1] AGEMA Infrared Systems 1991: AGEMA Operator’s Manual. AGEMA InfraredSystems AB, Danderyd, Sweden.[2] Arnfield, A.J. 1982: An approach to the estimation of the surface radiativeproperties and radiation budgets of cities. Phys. Geog. 3, 97-122.[3] Artis, D.A. and Carnahan, W.H. 1982: Survey of emissivity variability in thermography of urban areas. Rem. Sens. Environ. 12, 313-329.[4] Arya, P. 1988: Introduction to Micrometeorology. Academic Press: New York.[5] ASHRAE 1989: ASHRAE Handbook. American Society of Heating, Refrigeratingand Air-Conditioning Engineers Inc. Atlanta, Ga.[6] Balick, L.K. and Wilson, S.K. 1980: Appearance of irregular tree canopies innight-time high-resolution thermal infrared imagery. Rem. Sens. Environ. 10,299-305.[7] Balick, L.K. and Hutchison, .B.A. 1986: Directional thermal infrared exitancedistributions from a leafless deciduous forest. IEEE Trans. Geosci. Rem. Sens.GE-24, 693-698.[8] Balick, L.K., Hutchison, B.A., Smith, J.A., and McGuire, M.J. 1987: Directionalthermal infrared exitance distributions of a deciduous forest in summer. IEEETrans. Geosci. Rem. Sens. GE-25, 410-412.[9] Balling, R.C. Jr. and Brazel, S.W. 1988: High-resolution surface temperaturepatterns in a complex terrain. Photogram. Eng. Rem. Sens. 54, 1289-1293.[10] Barber, E.S. 1957: Calculation of maximum pavement temperature from weatherreports HRB Bull. 168, 1-8.[11] Barring, L., Mattsson, J.O., and Lindqvist, 5. 1985: Canyon geometry, streettemperatures and the urban heat island in Malmö, Sweden. Journal of Climatology 5, 433-444.[12] Becker, F. and Raffy, M. 1987: Problems related to the determination of landsurface parameters and fluxes over heterogeneous media from satellite data. Adv.Space Res. 7, 45-57.267References 268[13] Becker, F. and Li, Z.-L. 1990: Towards a local split window method over landsurfaces. mt. J. Rem. Sens. 11, 369-393.[14] Birnie, R.V., Ritchie, P.F.S., Stove, G.C. and Adams, M.J. 1984: Thermal infrared survey of Aberdeen City: data processing, analysis and interpretation.mt. J. Rem. Sens. 5, 47-63.[15] Boissard, P., Guyot, G., and Jackson, R.D. 1990: Factors affecting the radiative temperature of a vegetative canopy. In: Applications of Remote Sensing inAgriculture, Steven, M.D. and Clark, J.A. (eds.) Butterworths, London, 45-72.[16] Byrnes, A.E. and Schott, J.R. 1986: Correction of thermal imagery for atmospheric effects using aircraft measurements and atmospheric modeling techniques. Appi. opt. 25, 2563-2570.[17] Callison, R.D., Blake, P. and Anderson, J.M. 1987: The quantitative use ofAirborne Thematic Mapper thermal infrared data. mt. J. Rem. Sens. 8, 113-126.[18] Campbell, G.S. and Norman, J.M. 1989: The description and measurement ofplant canopy structure. In: Plant Canopies: Their Growth, Form and Function,Russell, G., Marshall, B., and Jarvis, P.G. (eds.) Cambridge University Press,pp.1-19.[19] Carison, T.N., Augustine, J.A. and Boland, F.E. 1977: Potential application ofsatellite temperature measurements in the analysis of land use over urban areas.Bull. Amer. Meteorol. Soc. 58, 1301-1303.[20] Carison, T.N., Dodd, J.K., Benjamin, S.G. and Cooper, J.M. 1981: Satellite estimation of the surface energy balance, moisture availability and thermal inertia.J. Appl. Meteorol. 20, 67-87.[21] Caselles, V., Sobrino, J.A., and Becker, F. 1988: Determination of the effective emissivity and temperature under vertical observation of a citrus orchard.Application to frost nowcasting. h-it. J. Rem. Sens. 9, 715-727.[22] Caselles, V. and Sobrino, J.A. 1989: Determination of frost in orange grovesfrom NOAA-9 AVHRR data. Rem. Sens. Environ. 29, 135-146.[23] Caselles, V., Sobrino, J.A., and Coil, C. 1992: A physical model for interpreting the land surface temperature obtained by remote sensors over incompletecanopies. Remote Sens. Environ. 39, 203-211.[24] Charles-Edwards, D.A. and Thornley, J.H.M. 1973: Light-interception by anisolated plant: A simple model. Ann. Bot. 37, 919-928.References 269[25] Chen, J.-M. and Zhang, R.-H. 1989: Studies on the measurements of crop emissivity and sky temperature. Agric. Forest Meteorol. 49, 23-34.[26] Chen, S.G. Impens, I., Ceulemans, R., and Kockelbergh, F. 1993: Measurementof gap fraction of fractal generated canopies using digitalized image analysis.Agric. Forest Met. 65, 245-259.[27] Choudhury, B.J. 1989: Estimating evaporation and carbon assimilation usinginfrared temperature data: vistas in modeling. In: Theory and Applications ofOptical Remote Sensing. Asrar, G. (ed.), J. Wiley and Sons Ltd, New York. pp.628-690.[28] Cooper, K. and Smith, J.A. 1985: A Monte Carlo reflectance model for soilsurfaces with three-dimensional structure. IEEE Trans. Geosci. Remote Sens.GE-23, 668-673.[29] Cox, C. and Munk, W. 1954: Measurements of the roughness of the sea surfacefrom photographs of the sun glitter. J. Opt. Soc. Amer. 44, 838-850.[30] Cracknell, A.P. 1993: A method for the correction of sea surface temperaturesderived from satellite thermal infrared data in an area of sunglint. Irit. J. Rem.Sens. 14, 3-8.[31] Crosby, D.S. and Glasser, K.S. 1978: Radiance estimates from truncated observations. J. Appi. Met. 27, 1712-1715.[32] Davies, J.A., Robinson, J., and Nunez, M. 1971: Field determinations of surfaceemissivity and temperature for Lake Ontario. J. Appl. Met. 10, 811-819.[33] Desjardins, R., Gray, J. and Bonn, F. 1990: Atmospheric corrections forremotely-sensed thermal data in a cool humid temperate zone. mt. J. Rem.Sens. 11, 1369-1389.[34] Dousset, B. 1989: AVHRR-derived cloudiness and surface temperature patternsover the Los Angeles area and their relationships to land use. Proceedings ofIGARSS-89, 2 132-2137.[35] Dozier, J. 1981: A method of satellite identification of surface temperature fieldsof subpixel resolution. Rem. Sens. Env. 11, 221-229.[36] Duggin, M.J. and Saunders, R.W. 1984: Problems encountered in remote sensingof land and ocean surface features. In: Satellite Sensing of a Cloudy Atmosphere.Henderson-Sellers, A. (ed.). Taylor and Francis: London, pp. 241-287.References 270[37] Dutton, E.G. 1993: An extended comparison between LOWTRAN 7 computedand observed broadband thermal irradiances: Global extreme and intermediatesurface conditions. J. Atmos. Ocean. Tech. 10, 326-336.[38] Eliasson, I. 1991: Urban geometry, surface temperature and air temperature.Energy and Buildings 15-16, 141-145.[39] Eliasson, I. 1992: Infrared thermography and urban temperature patterns. mt.J. Rem. Sens. 13, 869-879.[40] Elliott, W.P. and Gaffen, D.J. 1991: on the utility of radiosonde humidityarchives for climate studies. Bull. Amer. Met. Soc. 72, 1507-1520.[41] Everest Interscience 1991: Model 4000A Infrared Temperature Transducer Operating Manual. Everest Interscience: Fullerton, CA.[42] Forsythe, G.E., Malcom, M.A., and Moler, C.B. 1977: Computer Methods forMathematical Computation, Prentice-Hall Inc., Englewood Cliffs, N.J.[43] Foster, J.L., Ormsby, J.P., and Gurney, R.J. 1981: Satellite observations ofEngland and North-Western Europe. Weather 36, 252-259.[44] Fuchs, M., Kanemasu, E.T., Kerr, J.P., and Tanner, C.B. 1967: Effect of viewingangle on canopy temperature measurements with infrared thermometers. Agron.J. 59, 494-496.-[45] Gallo, K.P., McNab, A.L., Karl, T.R., Brown, J.F., Hood, J.J. and Tarpley, J.D.1993: The use of a vegetation index for assessment of the urban heat islandeffect. mt. J. Rem. Sens. 14, 2223-2230.[46] Garand, L., irassotti, C., Hale, J. and Klein, G.L. 1992: On differences inradiosonde humidity reporting practices and their implications for numericalweather prediction and remote sensing. Bull. Amer. Met. Soc. 73, 1417-1423.[47] Gates, D.M. 1962: Energy Exchange in the Biosphere. Harper and Row: NewYork.[48] Goel, N.S. 1988: Models of vegetation canopy reflectance and their use in estimation of biophysical parameters from reflectance data. Remote Sensing Rev. 4,1-212.[49] Goldreich, Y. 1985: The structure of the ground-level heat island in a centralbusiness district. J. Clim. Appl. Met. 24, 1237-1244.References 271[50] Goward, S.N. 1981: Thermal behaviour of urban landscapes and the urban heatisland. Phys. Geog. 2, 19-33.[51] Griggs, M. 1968: Emissivities of natural surfaces in the 8 to 14 micron spectralregion. J. Geophys. Res. 73, 7545-7551.[52] Grimmond, C.S.B. 1988: An evaporation-interception model for urban areas.Unpubi. Ph.D. Thesis, The University of British Columbia, Vancouver, B.C.[53] Gustavsson, T. and Bogren, J. 1991: Infrared thermography in applied roadclimatological studies. mt. J. Rem. Sens. 12, 1811-1828.[54] Henry, J.A., Dicks, S.E., Wetterqvist, O.F., and Roguski, S.J. 1989: Comparisonof satellite, ground-based, and modeling techniques for analyzing the urban heatisland. Photogram. Eng. Rem. Sens. 55, 69-76.[55] Holbo, H.R. and Luvall, J.C. 1989: Modelling surface temperature distributionsin forest landscapes. Rem. Sens. Environ. 27, 11-24.[56] Hook, S.J., Gablell, A.R., Green, A.A., and Kealy, P.S. 1992: A comparison oftechniques for extracting emissivity information from thermal infrared data forgeologic studies. Rem. Sens. Environ. 42, 123-135.[57] Hoyano, A. 1984: Relationships between the type of residential area and theaspects of surface temperature and solar reflectance (based on digital imageanalysis using airborne multispectral scanner data). Energy and Buildings 7,159-173.[58] Huband, N.D.S. 1985: An Infra-Red Radiometer for Measuring Surface Temperature in the Field. Part II. Calibration and Performance. Agric. For. Meteorol.34, 227-233.[59] Huband, N.D.S. an Monteith, J.L. 1986a: Radiative surface temperature andenergy balance of a wheat canopy. I. Comparison of radiative and aerodynamictemperatures. Boundary-Layer Met. 36, 1-17.[60] Huband, N.D.S. and Monteith, J.L. 1986b: Radiative surface temperature andenergy balance of a wheat canopy. II. Estimating fluxes of sensible and latentheat. Bounday-Layer Met. 36, 107-116.[61) Hutcheon, N.B. and Handegord, G.O.P. 1983: Building Science for a Cold Climate, John Wiley and Sons, Toronto.[62] Idso, S.B. 1981: A set of equations for full spectrum and 8-14 um and 10.5-12.5im thermal radiation from cloudless skies. Water Resources Res. 17, 295-304.References 272[63] Idso, S.B. and Jackson, R.D. 1969: Thermal radiation from the atmosphere. J.Geophys. Res. 74, 5397-5403.[64] Iqbal, M. 1983: An Introduction to Solar Radiation, Academic Press: Toronto.[65] Jackson, R.D. 1988: Surface temperature and the surface energy balance. In:Flow and Transport in the Natural Environment: Advances and Applications.Steffen, W.L. and Denmead, O.T. (eds.). Springer-Verlag: Berlin, 133-153.[66] Jackson, R.D., Reginato, R.J., Pinter P.J. Jr., and Idso, S.B. 1979: Plant canopyinformation extraction from composite scene reflectance of row crops. Appl. optics 18, 3775-3782.[67] Johnson, R.B. and Branstetter, E.E. 1974: Integration of Planck’s equationby the Laguerre-Gauss quadrature method. Journal of the Optical Society ofAmerica 64, 1445-1449.[68] Johnson, G.T. and Watson, I.D. 1984: The determination of view-factors inurban canyons. J. Clim. Appi. Met. 23, 329-335.[69] Jupp, D.L.B., Walker, J. and Penridge, L.K. 1986: Interpretation of vegetationstructure in Landsat MSS imagery: A case study in disturbed semi-arid eucalyptwoodland. Part 2. Model-based analysis. J. Environ. Manag. 23, 35-57.[70] Kahie, A.B. 1977: A simple model of the earths surface for geological mappingby remote sensing. J. Geophys. Res. 82, 1673-1680.[71] Kahle, A.B., Gillespie, A.R., and Goetz, A.F.H. 1976: Thermal inertia mapping:A new geologic mapping tool. Geophys. Res. Lett. 3, 26-28.[72] Khattak, S., Vaughan, R.A., and Cracknell, A.P. 1991: Sunglint and its observation in AVHRR data. Rem. Sens. Environ. 37, 101-116.[73] Kidder, S.Q. and Wu, H.-T. 1987: A multispectral study of the St. Louis areaunder snow-covered conditions using NOAA-7 AVHRR data. Rem. Sens. Env.22, 159-172.[74] Kim, H.H. 1992: Urban heat island. mt. J. Rem. Sens. 13, 2319-2336.[75] Kimes, D.S. 1980: Effects of vegetation canopy structure on remotely sensedcanopy temperatures. Rem. Sens. Environ. 10, 165-174.[76] Kimes, D.S. 1981: Remote sensing of temperature profiles in vegetation canopiesusing multiple view angles and inversion techniques. IEEE Trans. Geosci. Remote Sens. GE-19, 85-90.References 273[77] Kimes, D.S. 1983: Remote sensing of row crop structure and component temperatures using directional radiometric temperatures and inversion techniques.Rem. Sens. Environ. 13, 33-55.[78] Kimes, D.S., Idso, S.B. Pinter, P.J. Jr., Jackson, R.D., and Reginato, R.J. 1980a:Complexities of nadir-looking radiometric temperature measurements of plantcanopies. Appi. Optics 19, 2162-2168.[79] Kimes, D.S., Idso, S.B., Pinter, P.J. Jr., Reginato, R.J., and Jackson, R.D.1980b: View angle effects in the radiometric measurement of plant canopy temperatures. Rem. Sens. Environ. 10, 273-284.[80] Kimes, D.S., Smith, J.A., and Link, L.E. 1981: Thermal JR exitance model of aplant canopy. Appl. Optics 20, 623-632.[81] Kimes, D.S. and Kirchner, J.A. 1982: Radiative transfer model for heterogeneous3-D scenes. Appl. opt. 21, 4119-4129.[82] Kimes, D.S. and Kirchner, J.A. 1983: Directional radiometric measurements ofrow-crop temperatures. mt. J. Rem. Sens. 4, 299-311.[83] Kimes, D.S., Holben, B.N., Tucker, C.J., and Newcomb, W.W. 1984: Optimaldirectional view angles for remote-sensing missions. mt. J. Rem. Sens. 5, 887-908.[84] Kneizys, F.X., Shettle, E.P., Abreu, L.W., Chetwynd, Jr. J.H., Anderson, G.P.,Gallery, W.O., Selby J.E.A., and Clough, S.A. 1988: User’s Guide to LOWTRAN 7, Air Force Geophysics Laboratory Report No. AFGL-TR-88-0177,Hanscom AFB, MA 01731. 137 pp.[85] Kramer, P.J. and Kozlowski, T.T. 1979: Physiology of Woody Plants, AcademicPress: New York.[86] Kustas, W.P., Jackson, R.D. and Asrar, G. 1989: Estimating surface energy-balance components from remotely sensed data. In: Asrar, G. (ed.) Theory andApplications of Optical Remote Sensing. J. Wiley and Sons Ltd.: New York.604-627.[87] Kustas, W.P., Choudhury, B.J., Inoue, Y., Pinter, P.J., Moran, M.S., Jackson,R.D., and Reginato, R.J. 1990: Ground and aircraft infrared observations overa partially-vegetated area. mt. J. Rem. Sens. 11, 409-427.[88] Labed, J. and Stoll, M.P. 1991: Spatial variability of land surface emissivity inthe thermal infrared band: spectral signature and effective surface temperature.Remote Sens. Environ. 38, 1-17.References 274[89] Lacis, A.A. and Oinas, V. 1991: A description of the correlated k distributionmethod for modeling nongray gaseous absorption, thermal emission, and multiplescattering in vertically inhomogeneous atmospheres. J. Geophys. Res. 96, 9027-9063.[90] Lang, A.R.G. 1991: Application of some of Cauchy’s theorems to estimation ofsurface areas of leaves, needles and branches of plants, and light transmittance.Agric. Forest Met. 55, 191-212.[91] Lang, A.R.G. and McMurtrie, R.E. 1992: Total leaf areas of single trees of Eucalypt’us grandis estimated from transmittances of the sun’s beam. Agric. ForestMet. 58, 79-92.[92] Leak, S.M. and Venugopal, G. 1990: Thematic mapper thermal infrared data indiscriminating selected urban features. mt. J. Rem. Sens. 11, 841-857.[93] Lee, H.-Y. 1993: An application of NOAA AVHRR thermal data to the studyof urban heat islands. Atmos. Environ. 27B, 1-13.[94] Lewis, J.E. and Carlson, T.N. 1989: Spatial variations in regional surface energyexchange patterns for Montreal, Quebec. The Canadian Geographer 33, 194-203.[95] Li, X. and Strahier, A.H. 1985: Geometrical-optical modelling of a conifer forestcanopy. IEEE Trans. Geosci. Remote Sens. GE-23, 705-721.[96] Liption, A.E. 1992: Effects of slope and aspect variations on satellite surfacetemperature retrievals and mesoscale analysis in mountainous terrain. J. Appi.Met. 31, 255-264.[97] Lorenz, D. 1966: The effect of the long-wave reflectivity of natural surfaces onsurface temperature measurements using radiometers. J. Appi. Met. 5, 421-430.[98] Loudon Ross, S. and Oke, T.R. 1988: Tests of three urban energy balance models.Boundary-Layer Meteor., 44, 73-96.[99] Matson, M., McClain, E.P., McGinnis, D.F. Jr., and Pritchard, J.A. 1978: Satellite detection of urban heat islands. Month. Weather Rev. 106, 1725-1734.[100] Matson, M. and Legeckis, R.V. 1980: Urban heat island detected by satellite.Bull. Amer. Meteorol. Soc. 61, 212.[101] McGuire, M.J., Balick, L.K., Smith, J.A. and Hutchison, B.A. 1989: Modelingdirectional thermal radiance from a forest canopy. Rem. Sens. Environ. 27, 169-186.References 275[102] Mills, G.M. 1993: Simulation of the energy budget of an urban canyon - 1. Modelstructure and sensitivity test. Atmos. Environ. 27B, 157-170.[103] Monteith, J.L. and Szeicz, G. 1962: Radiative temperature in the heat balanceof natural surfaces. Quart. J.R. Met. Soc. 88, 496-507.[104] Myrup, L.O. 1969: A numerical model of the urban heat island. J. of AppliedMet. 8, 908-918.[105] Nakamura, Y. and Oke, T.R. 1988: Wind, temperature and stability conditionsin an east-west oriented urban canyon. Atmos. Environ. 22, 2691-2700.[106] Nichol, J.E. 1994: A GIS-Based approach to microclimate monitoring in Singapore’s high-rise housing estates. Photogramm. Eng. and Rem. Sens. 60, 1225-1232.[107] Nielsen, D.C., Clawson, K.L., and Blad, B.L. 1984: Effect of solar azimuth andinfrared thermometer view direction on measured canopy temperature. Agron.J. 76, 607-610.[108] Norman, J.M. 1975: Radiative transfer in vegetation. In: Heat and Mass Transfer in the Biosphere. deVries, D.A. and Afgan, N.H. (eds.). Scripta Book Company: Washington D.C.[109] Norman, J.M. and Welles, J.M. 1983: Radiative transfer in an array of canopies.Agron. J. 75, 481-488.[110] Norman, J.M., Divakarla, M., and Goel, N.S. 1995: Algorithms for extractinginformation from remote thermal-JR observations of the Earth’s surface. RemoteSens. Environ. 51, 157-168.[111] Oke, T.R. 1984: Towards a prescrpition for the greater use of climatic principlesin settlement planning. Energy and Build. 7, 1-10.[112] Oke, T.R. 1987: Boundary Layer Climates. Methuen: London.[113] Otterman, J., Brakke, T.W., and Susskind, J. 1992: A model for inferringcanopy and underlying soil temperatures from multi- directional measurements.Boundary-Layer Meteorol. 61, 81-97.[114] Outcalt, S.I. 1971: A numerical surface climate simulator. Geographical Analysis3, 379-393.[115] Parsons, A.J. 1985: Directionality in emitted infrared radiation from grassysurfaces. mt. J. Rem. Sens. 6, 903-910.References 276[116] Paw U, K.T. 1992: Development of models for thermal infrared radiation aboveand within plant canopies.ISPRS. J. Photogramm. Rem. Sens. 47, 189-203.[117] Paw U, K.T., Shaw, R.H., and Myers, T.P. 1985: Evapotranspiration as modelledby a higher-order closure scheme. Proc. Nat. Conf. on Advances in Evapotranspiration. Am. Soc. Agric. Eng., Chicago, IL 43-50.[118] Paw U, K.T., Ustin, S.L., and Zhang, C. 1989: Anisotropy of thermal infraredexitance in sunflower canopies. Agric. For. Meteorol. 48, 45-58.[119] Pease, R.W., Lewis, J.E., and Outcalt, S.I. 1976: Urban terrain climatology andremote sensing. Ann. Assoc. Amer. Geog. 66, 557-569.[120] Press, W.H., Flannery, B.P., Teukolsky, S.A. and Vetterling, W.T. 1986: Numerical Recipes, Cambridge University Press, Cambridge.[121] Price, J.C. 1979: Assessment of the urban heat island effect through the use ofsatellite data. Month. Weather Rev. 107, 1554-1557.[122] Price, J.C. 1982: On the use of satellite data to infer fluxes at meteorologicalscales. J. Appi. Met. 21, 1111-1122.[123] Price, J.C. 1989. Quantitative aspects of remote sensing in the thermal infrared.In: Theory and Applications of Optical Remote Sensing. Asrar, G. (ed.) JohnWiley: New York, 578-03.[124] Quattrochi, D.A. and Ridd, M.K. 1994: Measurement and analysis of thermalenergy responses from discrete urban surfaces using remote sensing data. mt. J.Rem. Sens. 15, 1991-2022.[125] Rao, P.K. 1972. Remote sensing or urban heat islands from an environmentalsatellite. Bull. Amer. Met. Soc. 53, 647-648.[126] Rees, W.G. and James, S.P. 1992: Angular variation of the infrared emissivitiyof ice and water surfaces. mt. J. Rem. Sens. 13, 2873-2886.[127] Roth, M. 1991: Turbulent transfer characteristics on a suburban surface. Unpubl.Ph.D. Thesis, The University of British Columbia, Vancouver, B.C.[128] Roth, M., Oke, T.R., and Emery, W.J. 1989: Satellite-derived urban heat islandsfrom three coastal cities and the utilization of such data in urban climatology.mt. J. Rem. Sens. 10, 1699-1720.[129] Sader, S.A. 1986: Analysis of effective radiant temperatures in a Pacific Northwest forest using Thermal Infrared Multispectral Scanner data. Rem. Sens. Environ. 19, 105-115.References 277[130] Salisbury, J.W. and D’Aria, D.M. 1992: Emissivity of terrestrial materials in the8-14tm atmospheric window. Rem. Sens. Environ. 42, 83-106.[131J Sampson, D.A. and Smith, F.W. 1993: Influence of canopy architecture on lightpenetration in lodgepole pine (Finns contorta var. latifolia) forests. Agric. andForest Met. 64, 63-79.[132] Schmid, H.-P. 1988: Spatial Scales of Sensible Heat Flux Variability Representativeness of Flux Measurements and Surface Layer Structure over SuburbanTerrain. Unpubl. Ph.D. Thesis, The University of British Columbia, Vancouver,B.C.[133] Schmugge, T.J., Becker, F., and Li, Z.-L. 1991: Spectral emissivity variationsobserved in airborne surface temperature measurements. Rem. Sens. Environ.35, 95-104.[134] Sheahen, T.P. 1983: Effects of atmospheric attenuation on temperature measurements made using JR scanning systems. Appi. Optics 22, 1070-1077.[135] Shoshany, M., Aminov, R. and Goidreich, Y. 1994: The extraction of roof topsfrom thermal imagery for analyzing the urban heat island structure. GeocartoInternational 4, 61-69.[136] Sievers, U. and Zdunkowski, W.G. 1986: A microscale urban climate model.Beitr. Phys. Atmosp. 59, 13-40.[137] Sinclair, A.J. 1976: Applications of Probability Graphs in Mineral Exploration.The Association of Exploration Geochemists, Special Vol. 4.[138] Smith, J.A., Ranson, K.J., Nguyen, D., Balick, L., Link, L.E., Fritschen, L., andHutchison, B.A. 1981: Thrma1 vegetation canopy model studies. Rem. Sens.Environ. 11, 311-326.[139] Smith, R.C.G., Prathapar, S.A., Barrs, H.D. and Slavich, P. 1989: Use of athermal scanner image of a water stressed crop to study soil spatial variability.Rem. Sens. Environ. 29, 111-120.[140] Smith, W.L. 1985: Satellites. In: Handbook of Applied Meteorology. Houghton,D.D. (Ed.). John Wiley and Sons: New York, pp. 380-472.[141] Sobrino, J.A., Caselles, V., and Becker, F. 1990: Significance of the remotelysensed thermal infrared measurements obtained over a citrus orchard. ISPRSJourn. of Photogramm. and Rem. Sens. 44, 343-354.References 278[142] Sobrino, J.A., and Caselles, V. 1990: Thermal infrared radiance model for interpreting the directional radiometric temperature of a vegetative surface. RemoteSens. Environ. 33, 193-199.[143] Spencer, J.W. 1971: Fourier series representation of the position of the Sun.Search 2, 172.[144] Spronken-Smith, R.A. 1994: Energetics and Cooling in Urban Parks. Unpubl.Ph.D. Thesis, The University of British Columbia, Vancouver, B.C.[145] Stanley, C.R. 1987: Instruction Manual for PROBPLOT, Association of Exploration Geochemists, Special Volume #14.[146] Steyn, D.G. 1980: Turbulence, Diffusion and the Daytime Mixed Layer Depthover a Coastal City. Unpubi. Ph.D. Thesis, The University of British Columbia,Vancouver, B.C.[147] Steyn, D.G. and Lyons, T.J. 1985: Comments on “The determination of view-factors in urban canyons”. J. Gum. Appi. Met. 24, 383-385.[148] Stoll, M.J. and Brazel, A.J. 1992: Surface-air temperature relationships in theurban environment of Phoenix, Arizona. Phys. Geog. 13, 160-179.[149] Sutherland, R.A. 1986: Broadband and spectral emissivities (2-l8jim) of somenatural soils and vegetation. J. Atmos. and Oceanic Tech. 3, 199-202.[150] Sutherland, R.A. and Bartholic, J.F. 1977: Significance of vegetation in interpreting thermal radiation from a terrestrial surface. J. Appi. Met. 16, 759-763.[151] Taconet, 0., Bernard, R., and Vidal-Madjar, D. 1986: Evapotranspiration overan agricultural region using a surface flux/temperature model based on NOAAAVHRR data. J. AppL Gum. Met. 25, 284-307.[152] Terjung, W.H. and O’Rourke, P.A. 1980a: Simulating the causal elements ofurban heat islands. Boundary-Layer Meteorol. 19, 93-118.[153] Terjung, W.H. and O’Rourke, P.A. 1980b: Influences of physical structures onurban energy budgets. Boundary-Layer Meteorol. 23, 415-424.[154] Van de Griend, A.A., Owe, M., Groen, M., and Stoll, M.P. 1991: Measurement and spatial variation of thermal infrared surface emissivity in a savannaenvironment. Water Resour. Res. 27, 371-379.[155] Verseghy, D.L. and Munro, D.S. 1989: Sensitivity studies on the calculation ofthe radiation balance of urban surfaces: II. Longwave radiation. Boundary-LayerMet. 48, 1-18.References 279[156] Voogt, J.A. 1989: Validation of an Urban Canyon Radiation Model for Nocturnal Long-Wave Radiative Fluxes and the Effect of Surface Geometry on Coolingin Urban Canyons. Unpubi. M.Sc. Thesis, University of British Columbia, Vancouver, B.C.[157] Vukovich, F.M. 1983: An analysis of the ground temperature and reflectivitypattern about St. Louis, Missouri, using HCMM satellite data. J. Appi. Meteorol.22, 560-571.[158] Vukovich, F.M. 1984: A comparison of surface temperature derived from HCMMinfrared measurements with field data. Rem. Sen. Env. 15, 63-76.[159] Willmott, C.J. 1981: On the Validation of Models. Phys. Geogr. 2, 184-194.[160] Wilson, S.B. and Anderson, J.M. 1986: The applicability of LOWTRAN 5 computer code to aerial thermographic correction. mt. J. Rem. Sens. 7, 379-388.[161] Woodcock, C.E. and Strahler, A.H. 1987: The factor of scale in remote sensing.Rem. Sens. Environ. 21, 311-332.[162] Wright, I.R. 1990: A laboratory calibration of infrared thermometers, mt. J.Rem. Sens. 11, 181-186.[163] Yoder, E.J. and Witczak, M.W. 1975: Principles of Pavement Design. JohnWiley: New York.Appendix AAGEMA 880 LWB INFRARED SCANNERA.1 Instrument DescriptionThe thermal infrared imager used in this study is the AGEMA Thermovision 800 BRUTSystem equipped with a Model 880 LWB scanner. The system is owned by the OntarioLaser and Lightwave Research Corporation (OLLRC) and was loaned for the study period. The complete system consists of a scanner, system controller (essentially a speciallyconfigured computer), monitor and keyboard. The system is designed for real-time thermography in industrial and research applications. There are no special system featureswhich adapt or preclude it from use on airborne platforms, although direct DC power-inis an available option which could be used to draw power from aircraft power systems.AGEMA systems have been successfully employed from helicopters in other studies ofurban and roadway climate (Eliasson, 1992; Gustavsson and Borgen, 1991). Airworthiness regulations prohibit modification of any aircraft structure without inspection andre-certification, so the AGEMA system was installed independent of all helicopter systemsincluding a DC power supply and inverter.The scanner uses a cryogenically-cooled (LN) Mercury Cadmium Telluride detectorwhich is sensitive primarily in the 8—12 tm waveband (typical spectral response curvesare shown in Figure A.1).A 12° lens with a stated geometric resolution of 1.2 mrad (measured as slit response at50% contrast) was used with the scanner. Table A.1 presents a summary of the technical280Appendix A. AGEMA 880 LWB INFRARED SCANNER 2811 .00.8(I)C0.60.0Figure A.1: AGEMA 880 LWB system response. Data supplied by Linnander (pers.comm.).specifications of the AGEMA 880 LWB system used.Incident radiation is directed by a set of rotating mirrors though a scalable aperatureunit, a filter unit, and finally onto a point detector. Detector output is converted into a12 bit digital signal by the processor. The instrument gain is software selectable; a gainof 2 allows the resolution to be doubled but limits the maximum signal from the scannerto 0.5 of the FSR. This is not a limitation for the range of surface temperatures presentin urban areas, therefore most images were recorded using a gain of 2.Images are built up from a set of scanned lines with 140 pixels per line. A number ofscanner modes are available (Table A.1). These vary the number and order (interlaced ornon-interlaced) of scanned lines. Images with fewer lines are scanned more quickly; thisallows a greater frame rate to be sustained. Modes which scan less than 140 lines perframe may consist of non-overlapping adjacent lines and were not used. In this study,4 5 6 7 8 9 10 11 12 13 14Wavelength (gm)Appendix A. AGEMA 880 LWB INFRARED SCANNER 282Table A.1: Technical specifications: AGEMA 880 LWB TIR system.Range -30°C to 1300°CSensitivity 0.05°C at 30°CAccuracy ±2% or ±2°CDigitization 12 bit (4096 levels) full scale range (FSR)12 bit (4096 levels) half FSR (GAIN = 2)Scanning Modes 0 - 280 lines © 6.25 Hz 4:1 interlace1 - 140 lines © 6.25 Hz 2:1 interlace2 - 70 lines © 25 Hz non-interlaced3 - 140 lines © 12.5 Hz non-interlaced4 - 280 lines © 6.25 Hz non-interlaced5 - Line scanning modemode 4 (280 lines x 140 pixels) was most commonly used. In turbulent flight conditions,the slower frame rate of mode 4 resulted in some blurring of the images. Under theseconditions scanner mode 3 (140 lines x 140 pixels) was selected at operator discretion.A.2 Instrument TheoryThe AGEMA system uses two special units in the measurement and storage of the thermalimages. Sample values (SV) are the digital output from the 12 bit k/D converter (0to 4095). Thermal values (TV), which use “isotherm units (IU)”, are described asa “... practical arbitrary unit of measurement” (AGEMA Operators Manual, 1991 p8.1). They are a measure of the amount of infrared radiation received and detectedby the scanner. Thermal values are used for level and range settings (which controlthe greyscale of the displayed image) and in the calibration curves of the scanner. Thefollowing equations present relationships between scanner output voltage, sample value,thermal value, incident photon flux, and object temperature.TV = 100 Vscan (A.1)Appendix A. AGEMA 880 LWB INFRARED SCANNER 283where Vscan is the scanner voltage andSV = (A.2)“upswhere iups is an internal constant which varies oniy with the gain. Because the AGEMAutilizes a photon detector rather than an energy detector, the relationship between thethermal value (represented by IU) and received photon radiation (F) is given byTV=CP (A.3)Spectral photon emission at wavelength A and temperature T, P(A, T) is obtained bydividing the Planck functionhc2L(A, T)= hc (A.4)A (exp() — 1)by the energy of a photon(A.5)where h is Planck’s constant (6.6262 x io— J s), k is Boltzmann’s constant (1.281 x 10—23J K—1), and c is the speed of light (2.998 x 108 m s’). Scanner detected photon emissionsmay be obtained usingP(T)= If2. P(A, T) . (A)dA (A.6)fA2 CF(A)dAwhere I (A) is the relative spectral response of the instrument as a function of wavelength(Figure A.1), and€ is the wavelength dependent value of the surface emissivity. In narrow wavebands the wavelength dependence of is often ignored (greybody assumption)and€is removed outside the integral. The calibration function used to relate temperature(T) and thermal value (as measured in IU units) isTV= exp(B) - F (A.7)Appendix A. AGEMA 880 LWB INFRARED SCANNER 284where R, B, and F are the response, spectral, and shape factors respectively. Thesefactors are obtained from a least squares fit of eq. A.7 over a wide temperature range (-30°C— 1300°C) during factory calibration. The factors are instrument, filter and aperturespecific and are fixed in the system memory. Table A.2 presents the scanner constantsfor the system configuration used.Table A.2: Scanner constants: AGEMA THV880 LWB (AGEMA, 1991)Filter NOF no filterAperature 0 fully openLens 12 FOV = 11.55° exactR 16930 response factorB 1470 spectral factorF 7.08 shape factora 0.008 atmospheric constant0 atmospheric constantiups 0.1257545 gain = 1The atmospheric constants a and /3 are used in a model to determine atmospherictransmission T, based upon a user-entered air temperature and object distance. The constants are calculated for standard atmospheric conditions and incorporate the instrumentspectral response (including the effects of any filter selected). This correction methodis not used here, because it applies to a fixed relative humidity (50%) and is designedfor applications in which the path between sensor and object is near-horizontal, with novariation in atmospheric properties. This study uses independent atmospheric corrections determined for measured atmospheric profiles at the time of image acquistion usingthe LOWTRAN 7 model B. Sample calculations show the internal AGEMA corrections candiffer by more than 2 degrees from those estimated using the LOWTRAN 7 model.Appendix A. AGEMA 880 LWB INFRARED SCANNER 285A.3 Instrument CalibrationCalibration of the full-scale instrument range uses a factory-set reference voltage and iscarried out each time the instrument is turned on and each time a sequence of imagesis to be viewed in real time and/or recorded. The AGEMA technical specifications (Table A. 1) state a temperature accuracy of ±2°C or ±2% (no indication of when to applyeither specification) over a wide temperature range (-30° C to 1300°C). An independentcheck of the accuracy of AGEMA temperature measurements was made using the black-body calibration facility of the UBC Soil Science Department. This facility consists of acylindrical blackbody cavity heated or cooled by a stirred water bath (Figure A.2).StirrerBath thermocoupleWater Bath5Thermocouples 4Cavity2AperatureAGEMAScannerFigure A.2: UBC Soil Science blackbody calibration facility.Cavity temperatures between -20°C and 70°C may be obtained using a precooledwater-glycol mixture and a small immersion heater. Cavity temperature is measured byfine-wire (0.003”) welded chromel-constantan thermocouples affixed to the inside of theAppendix A. AGEMA 880 LWB INFRARED SCANNER 286chamber by specially designed mylar/aluminum “sandwiches” which electrically insulatethe thermocouples from the chamber walls and provide good thermal contact with thecavity. The thermocouples are accurate to within 0.01°C; total error, including dataloggerlimitations, is ±0.05°C (R. Adams, pers. comm.). The maximum spatial variation incavity temperature among the four (of five) operating thermocouples was 0.11°C at -15°C, but on average was 0.02°C.The AGEMA scanner was mounted vertically below the cavity port (Figure A.2). Thisposition is less than ideal because the scanner dewar containing the liquid nitrogen (LN)has a reduced holding time when used at angles close to 90°. To avoid loss of detectorsensitivity, the dewar was refilled with LN at short intervals.The configuration of the scanner lens assembly, and the design of the calibration cavity, prevent the scanner lens from being mounted flush with the bottom of the cavity.The scanner lens is large with respect to the size of the cavity port, so the scanned imagewas averaged over a subset (140 x 70) of the complete frame (280 x 140) to reduce possible edge-effects from non-cavity emissions. The presence of non-cavity emissions may beevidenced by a non-symmetrical distribution of temperatures, where the mode is skewedtowards the cavity temperature and a tail of higher (lower) temperatures exists for low(high) cavity temperatures; the “tail” corresponding to the ambient room temperature.Figure A.3 shows that such a distribution does exist for the complete image but that theeffects are substantially reduced over the image subset.Visual inspection of the images revealed that temperature patterns associated withthe image subset tended to be organized into vertical bands and lines, rather than asradial patterns about the image centre. Patterns were similar for different cavity temperatures. They do not appear to be related to the cavity and are thought to be associatedwith internal system variations. Variations of temperature within these patterns were amaximum of about 0.8 degrees at AGEMA apparent temperatures of 339 K and 260 K.Appendix A. AGEMA 880 LWB INFRARED SCANNER 287870EF—. 5EH260 270 280 290 300 310 320330 340350‘mod (K)Figure A.3: Relative position of the image mode for complete and subset areas of thecalibration images.However, the majority of pixels scanned showed relatively little variation; 75% of themost frequently occurring classes from a histogram analysis yielded a temperature rangeof between 0.2 and 0.4 degrees.Results of the calibration test are presented in Figure A.4. The temperature plottedis the image modal temperature based upon a histogram analysis employing 15 classes.There is a consistent overestimation by the AGEMA with respect to the cavity temperaturewhich is especially evident at low temperatures. Differences exceed 2°C below 0°C,decreasing to 0.2—0.5 degrees above 50°C. These results have been reported previously(Isbrucker, peTs. comm.) and are accepted as accurate. When atmospheric correctionsare not required, image temperatures may be corrected using++ Complete ImageImage Subset+++T= 1.08217+1.0137Tag (A.8)70C)60a)5040Ea)i— 30a)200100uJC-)<—10—20C)>‘>0C)C)C)Figure A.4: Comparison of AGEMA scanner temperature with average blackbody cavitytemperature. (a) Calibration plot, (b) Temperature differences.for the range 8 T 700 C.A.4 Radiance-Photon-Temperature RelationsConversion between radiance, photon emission, and TV is required because the atmospheric radiation model used to estimate atmospheric emission and transmission producesradiance values. Blackbody cavity temperatures were converted to photon emissions using eq. A.4 and eq. A.5, and integrated over the spectral band pass of the AGEMA,weighted by the relative sensor response (using the curve supplied by Linnander (pers.comm.) (Figure A.1)). The integral is then normalized by the total integrated sensorresponse. Integrations were performed using an adaptive quadrature method (Forsytheet al., 1977). Results are compared to the AGEMA thermal values where apparent AGEMAmodal temperature is used as T. This implicitly accounts for the temperature differencebetween the cavity and AGEMA (Figure A.3). A plot of the results (Figure A.5) shows aslight deviation from linear.Appendix A. AGEMA 880 LWB INFRARED SCANNER4288(b)320—1—20 —10 0 10 20 30 40 50 60 70Bloc kbody Cavity Temperature (° C)—20 —10 0 10 20 30 40 50 60 70Bloc kbody Cavity Temperature (° C)Appendix A. AGEMA 880 LWB INFRARED SCANNER 289A better approximation over the calibration temperature range is obtained using apower law model. Parameters for the two empirical models are included in Figure A.5.230 I21 0= a + b(PR) a=9.526, b=0.2557 -IU = a (PR)b a=0.4379, b=0.9257c 190 -z- 170D150E 130I— 110 -W 9Q - Temperature Range: 258 — 343 KTemperatures Read (Calibration Data)- Wavelength Range: 4.0 — 1 4mSensorResponse:Yes50 i • I I I I I200 300 400 500 600 700 800Photon Radiance (Photons s1 mm2sr1 xl 012)Figure A.5: Relation between AGEMA Thermal Value and photon emission over thecalibration range.The power law model yields an average deviation from the mean of —0.01°C with amedian value of 0.0962°C. Over the range of temperatures most commonly observed inthe field, (5°C — 50°C) equivalent temperature differences are on the order of 0.3°C withmost differences less than ±0.2 degrees. Larger differences (up to 0.7°C) exist at verylow (below —10°C) or high (above 55°C) temperatures. No improvement is gained fromlinear or power-law models fit to subset ranges of observed temperatures, therefore thepower-law model fit to the entire range of data was selected for operational use.Appendix A. AGEMA 880 LWB INFRARED SCANNER 290A.5 Spectral Response ConsiderationsA.5.1 IntroductionThe spectral response diagram for the AGEMA scanner (Figure A.1) indicates somesystem response is present at wavelengths in the shortwave infrared. The inclusion ofsolar radiation, via surface reflection or atmospheric scattering, into the scanner field ofview increases the total radiance incident upon the detector (by the amount Psolar andresults in an overestimation of the apparent surface radiative temperature (T,.). Becauseof Wien’s law and the respective radiating temperatures of the Earth and Sun, the ratioof reflected solar radiation to the infrared radiation emitted by Earth increases rapidlyat shorter wavelengths (Figure A.6) leading to significant errors in Tr. This sectioninvestigates the degree to which system response to solar radiation affects the temperaturemeasurements made by the AGEMA scanner. The analysis is conducted with the aid ofthe LOWTRAN 7 model in order to estimate the solar and thermal radiances, includingthe processes of atmospheric absorption and scattering, at typicalobservation altitudes.A.5.2 LOWTRAN 7 ParametersThe LOWTRAN 7 model was run using the parameters listed in Table A.3. The surfaceemissivities considered were: a blackbody; a greybody (e = 0.3), representative of highlyreflective urban surfaces, such as uncoated aluminum roofs; a greybody (e = 0.9), typicalof many urban building materials; and, as a limiting case, a surface with = 0.0.The reflective character of the surfaces is assumed to be diffuse; while spectral reflections are possible from smooth surfaces such as glass, observations of this effect are confined to very limited Sun-target-sensor geometries when the reflecting surface is smooth(Duggin and Saunders, 1984). When spectral reflection does occur, errors may be significant, even for thermal infrared wavelengths. Errors in remotely sensed sea surfaceAppendix A. AGEMA 880 LWB INFRARED SCANNER 291C I0 103 \(1)102 0.93 4 5 6 7 8 9 10 11 12 13 14Wavelength (gm)Figure A.6: Ratio of reflected solar radiance to Earth-emitted radiance for threesurface emissivities. Solar radiance is approximated as the blackbody radiance atT = 6000K(appropriately reduced by the inverse square law for the Earth-Sun distance).Earth emissions are calculated assuming a surface temperature of 300 K. Emissions andreflections are related as p,, = 1 —temperatures of 2 K have been attributed to sunglint from waves on the ocean surface(Khattak et al., 1991) using the thermal channels of the AVHRR (band 4 10.3 4um — 11.3m, band 5 11.5 im — 12.5 /Lm).LOWTRAN 7 model runs produce estimates of:• atmospheric radiance which includes:— surface emission— atmospheric emission— atmospheric scattered thermal emission— surface reflected thermal emissionAppendix A. AGEMA 880 LWB INFRARED SCANNER 292Table A.3: LOWTRAN 7 input parameter summary.Model Parameter Settingatmospheric profile mid-latitude summerexecution mode solar radiancescattering multiple scatteringT0 294.2 K (from profile)0.9, 0.3, 0.0boundary layer aerosol rural, visibility 23 kmstratospheric aerosol background stratosphericclouds nonesensor altitude 647 mground altitude 0 msensor angle 00 (nadir angle)zenith angle 37° (solar noon, August 15)latitude 49° 15’ N• path scattered radiance (solar radiation)• ground reflected radiance including:— reflected direct solar— reflected scattered solar• total radiance (sum of the above components)Sensor normalized photon emissions were then calculated from the LOWTRAN radiance resuits using eq. A.6. These were converted to thermal values using the results of section A.3and finally to apparent temperature differences using the internal AGEMA functions.Appendix A. AGEMA 880 LWB INFRARED SCANNER 293A.5.3 ResultsThe results (Table A.4) show that the equivalent temperature error, due to inclusion ofPsolar in the total normalized photon emission reaching the scanner, is small, except oververy low emissivity surfaces. Because the reflected solar component is constant, whilesurface emission increases non-linearly with temperature, the relative error decreases astemperature increases. The surface temperature used in this analysis (294.2 K) is lowwith respect to the observed daytime temperature of many urban surfaces, thus providinga relatively conservative estimate of expected temperature errors. The AGEMA technicalspecifications list a sensitivity of 0.05°C at 30° C but an accuracy of only ±2% or ±2°.This suggests that while Psoja may be detectable, it is small with respect to the statedaccuracy of the system and is comparable to, or smaller than the temperature differencesattributed to system noise in the calibration tests.Table A.4: Sensor normalized photon emissions calculated from LOWTRAN 7 radianceestimates with multiple scattering forp a p b p c d£ atm solar scan r ‘-‘ rPhotons2cmrx 1016 (K) (K)0.0 2.2739 1.4051E-2 2.2878 267.08 0.280.3 2.7645 9.8388E-3 2.7742 276.31 0.180.9 3.7441 1.4069E-3 3.7452 292.16 0.021.0 3.9051 3.4871E-6 3.9051 294.53 0asensor normalized atmospheric photon emission (including surface emission)bsensor normalized sum of ground reflected and path scattered solar emissionctotal sensor normalized emission at sensor altitudedapparent temperature from Pscan (from relations in A.3)echange in T,. due to presence of Psoja,.Appendix A. AGEMA 880 LWB INFRARED SCANNER 294A.6 ConclusionsThe calibration measurements show a substantial systematic error (an overestimation)between the cavity temperature and apparent AGEMA temperature over the range ofcavity temperatures tested. The differences decrease with increasing cavity temperature;over the range 1O°C—50°C AGEMA temperatures are 0.1—1.0°C warmer than those measured by the cavity. Confidence in the AGEMA temperatures is restricted by evidenceof non-cavity emissions within the scanner FOV, and by structurally consistent apparent temperature variations within the images attributed to system noise. The form ofthe relation between thermal value and photon emission was found to be slightly bettermodelled by a power law model rather than the linear model suggested by eq. A.3. Thismodel agrees to within 0.3°C of the observed cavity temperature for temperatures above0°C. The analysis assumes that the sensor spectral response function includes all systemcomponents and is representative of the instrument used. Errors due to inclusion of reflected solar radiation are shown to be small based upon an analysis using the LOWTRAN7 model coupled with specified instrument spectral response.Appendix BATMOSPHERIC CORRECTIONS OF THERMAL IMAGERYB.l IntroductionRemotely-sensed surface temperatures require correction for atmospheric absorption andre-emission which takes place along the path length of atmosphere between the sensorand surface. For the short path-lengths viewed using airborne thermal imagery theprimary influence of the atmosphere is absorption and re-emission of thermal radiationby water vapour, with absorption and re-emission by ozone, and the uniformly mixedgases of carbon dioxide, nitric oxide, carbon monoxide and methane also having an effect(Desjardins et al.; 1990; Wilson and Anderson, 1986). Atmospheric absorption and reemission serves to reduce the range of apparent surface temperatures measured by athermal remote sensor, because surface emissions are replaced by atmospheric emissions.The atmospheric correction technique adopted in this study is the single infrared channel method (Becker and Li, 1990). This method uses an atmospheric radiation transfermodel to estimate atmospheric emission and transmission using a description of the atmosphere obtained from vertical remote soundings, radiosonde ascents or climatologicaldata, and a model of the sensor-detected radiance which combines the atmospheric, surface, and reflected sky radiance components.295Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 296B.2 MethodologyThe radiance L observed by a sensor at altitude h3 and pointing at an angle 0 to thenormal may be represented as (Byrnes and Schott, 1986)L(h, 0) = r(h, 0) . L, + r(h, 0) (1—Lsiy + La (B.1)where r(h,0) is the atmospheric-path transmission to the sensor, is the surface emissivity, L0 is the surface blackbody radiance at temperature T, L8k is the hemisphericincoming sky radiance at the ground surface, and La is the atmospheric emission in thepath between the sensor and the ground surface. Note that all the quantities in eq. B.1are spectral quantities.The actual radiance received by the sensor is an integration over the sensor bandpassweighted by the sensor responseL — f L(., h, 0) )d) B 2(see Section A.2). As in Section A.2, the wavelength dependence of is ignored, therebyinvoking the assumption that surface materials behave as greybodies. Because the AGEMAthermal scanner used has a detector with a linear response to photon radiation ratherthan radiance, the above equations are converted to photon emittances by dividing thespectral quantities by the energy per photon (eq. A.5).Using the relations determined between AGEMA IU units and photon emission, andthe internal AGEMA calibration function, model estimates of photon emission at sensorheight and view angle can be converted to apparent temperature estimates. Results areexpressed by way of look-up tables for specified surface apparent temperature and sensorapparent temperature from which a low degree polynomial can be used to speed theapplication to image data. The accuracy of the polynomial approximation is generallyAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 297within ±0.03°C for the temperature range 0—55°C, and to within ±0.06°C outside thisrange.No additional correction due to angle within the sensor FOV is incorporated. Fornadir imagery, the ±6° swath across the image yields temperature differences of less than0.02°C compared with the 0° look-up table. When scanning at the 45° off-nadir angle,the differences are greater, ranging from —0.4—0.3°C for the 51° upper limit of the FOV,and —0.2—0.3°C for the 39° FOV limit. However, actual differences may be larger, dueto uncertainties in the absolute angle of the scanner (likely ±10° from the specified 45°angle); apparent temperature differences calculated between a 40° and 50° off-nadir anglerange from —0.4°C at 270 K to 0.6°C at 330 K.B.3 LOWTRAN 7 Atmospheric Radiation ModelThe atmospheric radiation transfer model used is the LOWTRAN 7 Atmospheric Transmittance/Radiance model (Kneizys et al., 1988). This is a narrow-band radiation modelwhich has been used extensively in other studies employing aircraft-mounted thermalscanners (Hook et al., 1992; Schmugge et al., 1991; Desjardins et al., 1990; Byrnes andSchott,1986; Callison et al., 1987; Wilson and Anderson, 1986; Sheahen, 1983).The LOWTRAN model is used with a composite atmospheric profile (Section B.4) consisting of measured and climatological values of pressure, temperature and humidity. Theconcentrations of ozone, methane, nitrous oxide and carbon monoxide are obtained fromthe mid-latitude summer atmosphere model (as supplied by LOWTRAN) and other atmospheric trace gas concentrations are given by the U.S. Standard Atmosphere (Kneizyset aL, 1988). The model produces estimates of T and La tabulated at 5 cm1 intervals for use in (eqs. B.1, B.2, A.5). Emitted surface radiance, L0, is calculated usingthe Planck function. The hemispheric incoming sky radiance L8k is estimated followingAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 298Dutton (1993) using the quadrature method of Lacis and Oinas (1991) whereby L5k isapproximated by a weighted sum of the radiances obtained at zenith angles of 0, 60, and84.26°.B.4 Composite Atmospheric ProfilesThe use of temporally and spatially relevant profiles has been found to be important forsuccessful modelling of the atmospheric radiative fluxes using LOWTRAN and thereforethe accurate retrieval of surface radiative temperature (Desjardins et al., 1990; Wilsonand Anderson, 1986).In order to meet this requirement, three sources of data are used: local radiosondelaunches coincident with the time of the over-flight to characterize the lower atmosphere,climatological soundings from the nearest upper-air station to supplement local soundings, and standard climatological values for the highest levels of the atmosphere, abovethe range of radiosondes. Two composite atmospheric profiles were constructed for eachlocal sounding. The composite profiles are used to estimate incident sky radiance at thesurface, and for path emission and transmission for satellite data. The components ofthe composite profile are described in the following subsections.B.4.1 Local Radiosonde ProfilesProfiles of pressure, dry and wet-bulb temperature in the lower atmosphere were obtained from radiosondes launched (using an AIRSONDETM system) near the study siteduring the time of helicopter over-flights. Altitudes and dewpoint temperature Td required for LOWTRAN were derived from measured variables using the online-software ofthe AIRSONDETM system (Tetens saturation vapour pressure formula) with the modification that saturation vapour pressure with respect to the ice phase was not used untilAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 299evidence of wet bulb freezing was obtained. The height of ascent varied between launches,some launches ended prematurely due to instrument failure and/or data transmission orreception problems (Table B.1).Table B.1: Local radiosonde (Airsonde) launch times and maximum altitude. est. indicates maximum altitude was estimated based upon assumed ascent rate and comparisonof temperature and humidity profiles with other flights.Flight Date Launch ZmaxTime(PDT) (m)1 15/08/92 0949 22982 15/08/92 1337 66133 15/08/92 1738 32714 16/08/92 1123 41425 16/08/92 1616 54916 17/08/92 0953 1768(est.)7 17/08/92 1354 3811-8 17/08/92 1712 3538(est.)9a 24/08/92 1544 failed9b 24/08/92 1646 5710(est.)10 24/08/92 2029 415111 24/08/92 2357 failed12 25/08/92 0530 6725Of the 13 launches, 2 failed completely (Flights 9 and 11), and 3 suffered problemsthought to be associated with the pressure sensor (Flights 6, 8, and 9b). For thoseflights with suspect pressure data, heights were estimated using assumed ascent rates,and by comparing the structure of the temperature and humidity profiles with those of“good” flights on the same day, or from a similar launch time on another day. PressuresAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 300were then calculated using the height and temperature data. Of the failed launches,Flight 9a was replaced by a second launch (Flight 9b), and Flight 11 was estimated bytemporal interpolation using Flights 10 and 12. Plots of the finalized atmospheric profiles(which include upper atmospheric data from climatological soundings (Section B.4.2))are provided in Figure B.1.B.4.2 Climatological Radiosonde ProfilesTemperature and humidity at altitudes above those measured by the local radiosondeprofiles were obtained from reported pressure, temperature and dewpoint depressions(DPD) at significant and mandatory levels at one of the 2 nearest 12 hour radiosondelaunch sites (Figure B.2): Port Hardy, B.C., (station YZT) or Quillayute WA, (stationUIL). Table B.2 summarizes the radiosonde data collected.No temporal interpolation of the data is attempted because, for the most part, it isused only for altitudes well above the boundary layer, where most of the diurnal variationtakes place. The soundings are plotted in Figure B.1 which allows comparison betweenthe two stations and the local sounding.Humidity data are much sparser for station UIL due in part to differences in radiosonde reporting practices between the U.S. and Canada (Table B.4.2 and Garand etaL, 1992; Elliott and Gaffen, 1991). Reported DPD for the uppermost levels at stationYZT are subject to a 1% relative humidity limit; this practice sets all layers with a humidity < 1% as 1% and calculates the DPD based upon reported air temperature. Thisintroduces a moist bias into the profile. The US practice of replacing low RH valueswith a fixed DPD of 30 K results in a dry bias for those levels so assigned. For thermalimagery obtained from the helicopter, these biases will have a minor effect, because itaffects only the estimation of L0, which is reflected to the sensor in only small amountsfrom most surfaces. For satellite data, the effects are more important, since the layersAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY101E03I10010’E01II0301Figure B.l: Atmospheric profiles of temperature and humidity as measured by localAIRSONDE launches, and the closest (temporally) profile obtained from the upper airreporting stations Port Hardy B.C. (YZT) and Quillayute WA (UIL) for each flightconducted: (a-c) Flights 1—3 respectively.S30— 00/10/0210092 A1,.ood.1—0— 00/l5/921224Z OIL 900/I5/0212002 000101EI10010-I(a)(b)(C)10’EI10010-I10’E3:10010_I— 00/15/021900221I4002eI 200/15/9210240 OIl. 90 09/15/9212000 47170 —50 —30 —10 10Air Temperature (° C)cma.— 09/15/0220370 2i,soodo2 I70-EE:.1.Air Temperature (0 C)-91.1 —/13 —51.)—30 —13 10Dewpoint Temperature (0 C)Dewpoint Temperature (0 C)la-I10’EIio10-I49’rDiA-p— 09/15/9200370 oi’.ode3 7:09/19/9212390 OIL 4 a107/10Air Temperature (° C)r99.0)-00.— 00/15/9200377 211.0,14.3—0— 09/19/9212392 UIL0 09/l6/92000079Z1 ,—90 —70 —50 —30 —10 10Dewpoint Temperature (0 C)E6Z11)1 001 0’E0)I1 00Figure B.1 (Continued). Atmospheric profiles for: (d) Flight 4, (e) Flight 5.Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY(d)3021 01d’°°— ci08/16/921 822 Z Arsod 4-— 08/1 6/92 1 239 Z LJIL 40- 08/17/920000Z YZT10—’ • I I I—70 —50 —30 —10 10Air Temperature (0 C)1 OlE0ii)I1001 0_i0:.08/1 6/92 1822 Z Airsonde 408/16/921239Z1J1L0 08/17/9200002 YZTI I I I I—90 —70 —50 —30 —10 10Dewpoint Temperature (0 C)1 0’(e)20)I1 0’10°ci0ci0.00U08/1 6/92 231 6 Z Airoonde 5-‘‘ 08/17/920024Z UIL13- 08/17/920000ZYZTaio’ I 1,111 I. 1,1,1.1.1—90 —70 —50 —30 —10 10Dewpoint Temperature (° C)Air Temperature (0 C)Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 303Figure B.1 (Continued). Atmospheric profiles for: (f) Flight 6, (g) Flight 7, (h) Flight 8.F01eI0’100I‘001908/l7/92I00322k0o608/17/9212242 OILo 00/17/0212092070(0(9)(h)10—70 —50 —30 —10 10 30Air Temperature (° C)‘i00/I1/92205424iooo4o7 A00/10/9200332 OIL1 —l_03010’F01I10010-I101F.041I10010’10’F.0010’I10010_IQ440%— 00/17/9210532 3In4o000/11/9212242 00.Dewpoint Temperature (° C)O.R300— 00/17/9220542 O,oo,,2o700/00/9200332 OIL0’• 00/10/9200002 027 0—90 —70 —50 —30 —10 10Dewpoint Temperature (° C)@89-00•.-o-1O4.n9C,&— 00/10/9200122 Ai,,a,.4o0-4— 00/10/9200332 OIL- 0 00/10/9200002 022 0—90 —70 —50 —30 —10 10Dewpoint Temperature (0 C)70 —50 —30 —10 10Air Temperature (° C)SIIN—00/l8/92001220i,oa,,0o0 A‘ 00/10/9200352 OIL10 ,0:/10T IAir Temperature (0 C)(1)(j)(k)(I)I-’.aqn8101p—40”IC) 10-I-I50.10‘—08/25/920329241184112810•—09/25/9200242OIL3.>D08/29/9200000029Ie9—ie_’I.,1:11E—70—50—30—1010QAirTemperature(0C)Cd)9d D CD I-0<‘)t0r.o’•::..0fra•a050_4Q99.,.10110’00%10’10’ecç,-iIIS’‘‘oo°oo°o>o‘1000C\,00—09/25/920039ZM,,41120(4.08/25/9209554208403811—08/25/921220Z0i,,,2o12—08/24/922444021,840389q00/24/9200242211——09/25/92022422112.0—09/25/920224208.2,6006/25/9200240OIL-‘e08//0000eo8//0e06/2/202010a08/2/000001Ci)10—90—70—50—30—1000—90—70—50—30—0010—90—70—50—30—1010—90—70—50—30—1010CDewpoiotTemperoture(0C)DewpointTemperature(0C)DeapointTemperature(°C)DewpointTemperature(4C)I,‘zAirTemperature(0C)AirTemperature(0C)0 0 I C’., CAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 305Figure B.2: Location of the Vancouver study area and upper air reporting stations PortHardy, B.C. (YZT) and Quillayute, WA (UIL).are now part of the atmospheric path between the surface and sensor. In the absenceof raw sounding data (which are accurate down to a RH of 1% (Garand et al., 1992))the recovery of the actual humidity is not possible. This leaves only the possibilitiesof layer omission, replacement or interpolation. In this study, humidity data from themid-latitude summer atmosphere are used to replace missing data, or data below the 1%RH limit.B.4.3 Climatological DataWhen atmospheric data are required for levels above the maximum altitude of the radiosondes, pressure and temperature data are taken from the mid-latitude summer atmosphere as specified in the LOWTRAN 7 model. This climatological profile matches observedPort Hardy -AQuillayute - UIL0 100 200kmAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 306Table B.2: Summary of radiosonde data collected.Flt Airsonde YZT UIL Reported Levelsa/ ZmaxbDate Timec Date Time Date Time YZT UIL(Ta) UIL(Td)1 15 1648 15 1200 15 1224 43/33.3 35/23.3 20/9.72 15 2037 16 0000 15 1224 42/35.2 35/23.3 20/9.73 16 0037 16 0000 16 1239 42/35.2 29/25.4 19/9.64 16 1822 17 0000 16 1239 42/35.3 29/25.4 19/9.65 16 2316 17 0000 17 0024 42/35.3 34/26.1 19/9.76 17 1653 17 1200 17 1224 33/37.2 23/22.5 13/9.67 17 2054 18 0000 18 0033 50/35.3 36/31.7 18/9.38 17 0012 18 0000 18 0033 50/35.3 36/31.7 18/9.39 24 2346 25 0000 25 0024 51/34.1 42/26.6 27/9.310 24 0328 25 0000 25 0024 51/34.1 42/26.6 27/9.311 24 0656 25 1200 25 1224 40/35.1 32/22.5 18/9.412 24 1230 25 1200 25 1224 40/35.1 32/22.5 18/9.4aflumber -bkmCUT timeprofiles of air temperature well (Figure B.3a). The local soundings are shown to be somewhat drier than the mid-latitude summer profile, especially at mid-levels (Figure B.3b),but there is some variation between soundings. Generally good agreement exists at upperlevels when relative humidities are greater than 1%. The 1% RH limitation for stationYZT is clearly evident (Figure B.3c).B.5 Airborne Imagery vs. Surface-Based MeasurementsCoincident with the time of the over-flights, ground-based observations of the temperature of selected surfaces were conducted at a nearby site. The surfaces sampled wereAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 307(a) (b)100€100— Midlotiio00 — MidIotitide Srnr,erAiniopheeYZTO8/15/921200Z 0 ° YZTO8/15/92120000 YZ1 08/1 6/92 0000 Z 0 YZT 08/1 6/92 000072T 08/1 7/92 0000 Z YZI 08/1 7/92 0000 Z10—I I I. I — 10_i—110 —70 —30 10 —160 —120 —80 —40 0Air Temperature (° C) Dewpoint Temperature (° C)(c)1000000— UidIoIiIud SooerAiroo0phrYZ108/15/921200ZYZTO8/16/920000Z0 YZTO8/17/920000Z1 IIII •I[•I• I,!IIIrII100 101 102Relative Humidity (Z)Figure B.3: Comparison of the LOWTRAN 7 Midlatitude summer atmosphere model withmeasured profiles from station YZT; (a) air temperature, (b) dewpoint temperature, (c)relative humidity.Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 308Table B.3: Summary of reporting differences for radiosonde measurements. Compiledfrom Garand et al. (1992) and Elliot and Gaffen (1991).U.S. Practice (1) when RH < 20%, DPD set to 30 K(2) when Ta < —40°C, no humidity reported(3) when RH > 90%, use old sensor algorithm forsignal to RH conversionCanadian (1) when RH <9%, set RH =9% and convert to DPDPractice (2) when Ta < —65°C, no humidity reported(3) when RH > 90%, use new sensor algorithm for signal to RHconversionchosen to give as wide a temperature range as possible (Table B.4).Ground-based sampling was carried out using a 60° FOV Everest IRT (sensitive toradiation in the 8—14 um waveband) held approximately 0.5—0.75 m above the targetsurface, and at an off-nadir angle of approximately 30—45°, to avoid viewing the operator.Calculated surface areas viewed for these sensor positions range from approximately0.5 m to 3.1 m. The large FOV of the Everest allows better spatial sampling and matchingwith the IFOV of the AGEMA scanner at higher flight altitudes. However, the off-nadirsampling is not ideal for rough surfaces such as grass, which may be susceptible tovariations in emission with view angle and direction (Parsons, 1985) and the large viewedarea may encompass several AGEMA pixels, at the lower scanning altitudes used for Flights1—8. Samples were made at 1 s intervals for 6 s per surface and recorded on a CS 21Xdatalogger. The fast sampling rate and relatively few samples per site allow all surfacesto be sampled within 5—10 mm of the time of the overpass. Comparisons are madeusing apparent surface temperatures uncorrected for emissivity. This requires that thesurface emissivities be constant over the spectral range of both instruments (surfaces areAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 309Table B.4: Ground calibration sites and surface types sampledFlights Calibration Site Surface Types Samples1 3 Johnathan Rogers Park grass (3)concrete (3)asphalt (3)4— 5 Portside Park grass (3)sand (3)asphalt (3)water (1)6 — 8 Memorial Park grass (3)asphalt a (3)asphalt b (3)water (3)9 — 1 Trafalgar Park grass (3)concrete (3)greybodies), and for sky radiance to be considered as blackbody emission. If radiancereceived by the Everest IRT on the ground is represented byLevLo+(1)Lsky (B.3)then the radiance at the AGEMA scanner at a given flight altitude can be writtenLag TLev+La. (B.4)Look-up tables (LUT) for AGEMA apparent temperature were constructed at 1°C intervals of surface (Everest IRT) apparent temperature using eq. B.4 (converted to photonemissions), with Lev given by an integration of Planck’s law for the surface apparenttemperature, and -r and La modelled using LOWTRAN 7. Plots of the LUT are presentedAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 310in Figure B.4.The temperature corrections are equivalent to the temperature difference which mustbe added to, or subtracted from, the apparent AGEMA surface temperature in orderto obtain the true surface apparent temperature. Note the large range of correctionrequired across the range of temperatures possible in urban areas. Cool, irrigated lawnsmay require a reduction in temperature while very hot surfaces such as roads or rooftopsrequire a large positive correction.The calibration site is located on an AGEMA image and a subset of pixels centredon this point is extracted. The pixel apparent temperatures are converted to surfaceapparent temperatures (corrected for atmospheric emission and transmission) via theLUT. A window of pixels is used because the Everest FOV may cover several pixelsand precise location of the ground calibration point on the image is difficult in somecases. This is important for those surfaces which show significant spatial variation insurface temperature but are devoid of highly contrasting thermal features which couldbe used to more precisely fix the ground measurement site. This is often the case with thegrass surfaces which show local spatial variations of uncorrected apparent temperatureexceeding 2 degrees. The use of the low emissivity target as a control point helps toovercome this problem but is not completely satisfactory; image blurring due to platformand operator motion results in a smearing of the relatively cool target temperature intothe nearby surfaces which makes comparison of grass temperatures difficult for highaltitude flights.A check was made to compare the results of this method with that of using eq. B.1.Measured values of € were obtained for the sites (Table B.5). Generally good agreementexists between the measured values and those reported in the literature (Table B.6),keeping in mind expected variations in of sand (due to quartz content) and grass (due tomoisture content, species and surface structure). Sky radiance in the 8—14 ,um wavebandAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY(a)311Figure B.4: Look-up table plots of the atmospheric correction required to account foratmospheric absorption and transmission (but not surface reflection) on each day flightswere conducted: (a) Flights 1—3, (b) Flights 4—5, (c) Flights 6—8.86420—2—4—6—8C-)C0UV0C-)V:30VaEVI—(b)84C)e 2V.i-200)0 —4E0)—-6yo a io 20 30 40 50Surface Temperature (0 C)0 0 0 10 20 30 40 50Surface Temperature (0 C)8()a, 6C.2 4U2I :I::—10(C)0 0 10 20 30 40 50Surface Temperature (0 C)6C-)C00V0C)V0VaSVF—0 0 10 20 30 40 50Surface Temperature (0 C)1 —— FlightS P0001. -20—2—4—6—80 10 20304050Surface Temperature(0C)6C02V60 0 0 10 20 30 40 50Surface Temperature (° C)60Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 312(d) 1210Flqht 9S FIht1O6 —— FIi9htllFhghtl240020DQ0—20E —40‘ -60O10 20 30 40 50 60Surface Temperature (° C)Figure B.4 (Continued). (d) Flights 9—12.was modelled using LOWTRAN 7 using the composite atmospheric profiles. Ground-sampled apparent temperatures were converted to Lev, and L0 was obtained from (B.3)using measured surface emissivities and modelled L8k9.Differences between the methods are generally less than 0.1°C. For Flights 1—8, twosets of ground calibration checks were made: one at the start of the mission, at thealtitude used for nadir image acquisition (referred to as pass 1), and a second, uponcompletion of the lower, 45° off-nadir fiightlines, (pass 2). Both calibration checks aredone at nadir, or near-nadir, sensor angles. Flights 9—12 have a single calibration setand use a limited set of surface types. Results are presented in Figure B.5 where thefollowing plotting conventions have been used:• vertical solid lines represent the temperature range in the small pixel subset extracted about the estimated location of the sampling point;• vertical dashed lines represent the range of temperatures viewed over a much largerpixel subset containing the sampling point to indicate the temperature variationspossible for that surface type;Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 313• horizontal error bars represent ±lu of the ground sampled surface apparent temperature (Tr).B.5.J. DiscussionSeveral consistent observations arise from inspection of Figure B.5:1. agreement between corrected scanner temperature and ground-sampled temperature is better for the second pass of Flights 1—8; the lower scanning altitude resultsin a shorter path length with a slightly reduced atmospheric correction and alsoallows more precise location of ground sample points on the image;2. sea surface temperatures obtained for Flights 1—5 do not agree well. The EverestIRT values are consistently lower and show substantially more variation than thecorrected AGEMA temperatures;3. grass (and to some extent sand) surfaces exhibit a large range of temperaturesaround the calibration sites, both in the image data and from ground-sampleddata;4. flights conducted after sunset (10—12) show significantly less variation in apparentgrass temperature by both Everest IRT and AGEMA (the extension towards coolertemperatures of the AGEMA temperatures is more a function of smearing of the lowemissivity target into the pixel subset extracted for the ground sample point, andnot indicative of true temperature variation of the grass surface;5. asphalt and road surfaces exhibit a relatively narrow range of temperature (except the shaded portion of the asphalt surface in Flights 6—8) and generally showgood agreement despite the large correction applied, and a relatively low surfaceemissivity.(a)Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 314-oaUa)0C)C)I—wC)EIRT T, (° C) EIRT Tr(°C)28 30 32 34 36 38 40 42 44LIRT Tr (° C)a)C)a)0C)C)I—uJC)(b)-;‘ 40 --- 3538o ))) 0C) C)5 34 3’ 25o o- 32w wC) 15(c)40- 35C, Ua) a,30o o3 25o 0I—’20LU LUC) C)< 15Figure B.5: Comparison of ground-sampled Tr with corrected AGEMA scanner Tr. Seetext for plotting conventions. Surface types sampled are: a - asphalt (road surface for allbut Flights 6—8), c - concrete, g - grass, s - sand, r - road surface (asphalt) (Flights 6—8only), w - water. Figures (a - c) are Flights 1—3 respectively.15 25 35 45EIRT Tr(°C)29 31 33 35 37 39 41EIRT Tr(°C) CIRT Tr(°C)Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 315(d)____________ ____________/aEIRT Tr(°C) EIRT Tr(°C)(e)___J5Poss2AI:488ma/25252729313335373941 4345 10 20 30 40EIRT Tr(°C) EIRT Tr(°C)Figure B.5 (Continued). (d) Flight 4, (e) Flight 5.>C) m-1C)C)0aC)a0C) Cl>C)C) 0CDC)aaC)m>C)C) 0aC)aaC-)C))C)C) 0aC)aa(D0)’l1èJ13—9sq(q-,j)(pnuizuo)ci(D)‘i±iI]C) P19-Coccoaaoe.L2(D0)’liei(D)1!dI(Do)’lIdi]L()o>C)C))-C°C-) 0aC)oa(5)OLc()A?JDVJ4TI‘IVI41JHLLOSNOLLOHHOOOIHIIJSOI4TJVE1xTpUOddy(D)1idi]9IAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 317(i)____________(j)____________/LIRT T1(° C) EIRT Tr (° C)(k) (I)____1AIt1524rn/1/EIRT Tr (° C) EIRT Tr(° C)Figure B.5 (Continued). (i-i) Flights 9—12.Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 318Table B.5: Surface emissivities for ground calibration surfaces.Location Surface a n Date MethodJonathan grass 0.973 0.009 9 08/30/92Rogers 0.988 0.004 12 08/03/93 1park 0.993 0.004 12 08/03/93concrete 0.956 0.008 8 08/30/92 10.960 0.003 19 08/03/93 10.956 0.003 19 08/03/93 2asphalt 0.932 0.004 7 08/30/92 1(road) 0.932 0.005 19 08/03/93 10.926 0.006 19 08/03/93 2Portside sand 0.957 0.004 20 08/30/93 1Park 0.961 0.005 20 08/30/93 2asphalt 0.959 0.004 19 08/30/93 10.956 0.005 19 08/30/93 2grass 0.972 0.008 20 08/30/93 1(dry) 0.980 0.007 20 08/30/93 2grass 0.988 0.003 18 08/30/93 1(green) 0.992 0.005 18 08/30/93 2Memorial grass 0.979 0.007 8 08/30/92 1Park 0.985 0.004 24 08/04/93 10.992 0.004 24 08/04/93 2asphalt 0.944 0.004 7 08/30/92 1(road) 0.937 0.003 23 08/04/93 10.928 0.006 23 08/04/93 2asphalt 0.956 0.004 24 08/04/93 1(playing sfc) 0.953 0.005 24 08/04/93 2aMethod 1: Davies et al. (1971).6Method 2: Chen and Zhang (1989)Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 319Table B.6: Reported values of for surface types tested. Except where noted spectralrange is 8—14 ,um.Surface Referencegrass (dead) 0.966 Sutherland (1986)grass (live) 0.990very short grass 0.979 Labed and Stoll (1991)tufts of grass 0.981grass (almost 0.958 Van de Griend et al. (1991)complete cover)lawn: dense and 0.973 Lorenz (1966); 7—15 pmwell keptBig blue stem 0.972 Salisbury and D’Aria (1992); 11.3—11.6 pmIndian grass 0.971Switch grass 0.969Rye grass (senescent) 0.903sand 0.893—0.915 Sutherland (1986)0.938 Lorenz (1966); 7—15 pmsand (dry) 0.84—0.90 Arya (1988)concrete 0.942 Lorenz (1966); 7—15 pm0.71-0.90 Oke (1987)asphalt 0.955 Lorenz (1966); 7—15 pm(old road surface) 0.95 Oke (1987)water 0.972 Davies et al.(1971)0.990 Salisbury and D’Aria (1992); 11.3—11.6 pm0.925—0.969 Rees and James (1992); e decreases asobserving angle (relative to normal) increasesAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 320B.5.1.1 Sea Surface TemperaturesPerhaps the most troubling of these observations is the disagreement between AGEMAand Everest apparent sea surface temperatures for Flights 1—5. Presuming the correctionmethod, atmospheric profile, and scanner calibrations to be correct, there are two possibleexplanations for the disagreement.Firstly, the off-nadir sampling by the surface-based Everest IRT with it’s large FOVmay lead to a contribution of direct atmospheric radiance within the FOV. The totalFOV of the Everest IRT is actually larger than 60°, with the 60° being defined as theFOV which contributes 80% of the total power distribution (Everest Interscience, 1991).Figure B.6 illustrates the potential for a reduction in Tr for off-nadir instrument angles,due to atmospheric radiance in the Everest FOV. This diagram was constructed as follows. Surface emissions are predicted for given SSTs using the Planck function with aconstant emissivity of 0.98. Atmospheric radiance is calculated for a range of angles usingLOWTRAN 7 with the composite atmospheric profile constructed for Flight 1. Radianceat the sensor is tabulated at 2° intervals with the source of radiance determined as eitherthe sky or sea surface. Surface emissions are reduced by the prescribed and a reflectedcomponent of atmospheric radiance added. The radiance is weighted with the sensorFOV response, and the total radiance is then converted to an apparent temperature.The results in Figure B.6 indicate that for off-nadir angles greater than 30°, sharp decreases in apparent surface temperature are possible. The slight warming trend evidentat 35° for lower SSTs is due to the relatively high atmospheric emission for angles closeto the horizontal (i.e., close to blackbody emission at air temperature).The importance of these effects is enhanced if the angular emissivity variations ofRees and James (1992) are considered. These authors found a decrease in emissivity of0.05 with depression angles between 30 and 60°, for shallower angles further decreasesU0ci,Da0)E0)Fci)>000C0)0a0Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 32119171513119753e=O.98Atmosphere: Fight 177’ e-4-.‘7,.-/0U/. ‘B—/ ‘I ,0WO 10 20 30 40 50 60 70 80 90Viewing Angle (Degrees from horizontal)Figure B.6: Modelled variations in apparent radiative surface temperature for a 600 FOVEverest IRT viewing a water surface. -were observed (down to 0.4) but confidence in these results was limited. These resultswere for calm water conditions and so are not directly applicable here, however, thegeneral trend, leads to the type of disagreement found here. The relatively large standarddeviation obtained from the Everest IRT water samples may be a result of variations inthe angle at which the instrument is held between samples. Secondly, the apparentdiscrepancy between AGEMA and Everest IRT SSTs might be due to specular reflectionof solar radiance into the sensor FOV. This is commonly known as sunglint. If the sensoris pointing towards the specular point of the Sun, specular reflection of solar radianceinto the sensor FOV will result in increased apparent surface temperatures. When thesea surface is calm, the range of angles at which this is possible is low, but as windspeeds increase and surface roughness increases, the range of observation angles affectedAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 322Table B.7: Solar position and AGEMA scanner and Everest azimuth angle direction:Flights 1—5. Image azimuths approximate only. Scanner angle is nominally 00 (nadir)but may be up to 200 off-nadir. Z - solar zenith angle, qS - solar azimuth angle. Time islocal time (PDT).Flight Time Z q Image Azimuths Everst Az Time1 1103 44.9 131.2 203 292.5 11142 1424 38.4 207.5 180 “ 14383 1744 65.0 261.5 180 - 225 “ 17584 1208 38.6 153.2 225 - 248 337.5 12175 1608 50.5 239.7 315 “ 1634increases up to 30° from the specular point (Duggin and Saunders, 1984).Recent observations of sunglint in NOAA polar orbiting satellites indicate SST differences of up to 2 K in the thermal infrared channels 4 and 5 (Khattak et al., 1991).Greater errors may be anticipated for the AGEMA sensor which also responds to someshorter wavelength infrared radiation. In lieu of attempting to model the degree to whichspecularly reflected solar radiation may be present in the AGEMA (see e.g., Cox and Munk(1954), Khattak et al. (1991), Eitner (1992), Cracknell (1993)) which requires knowledge of slope distribution functions of waves, or a second visible channel, Table B.7 ispresented to determine the extent to which the view angles are aligned with the solarspecular point.Table B.7 indicates there is potential for several of the flights to be affected whenconsidering the spread of sunglint about the specular point with waves on the sea surface, but there is no consistent pattern of alignment for all flights. The Everest alignmentalso shows some potential for sunglint observation in Flights 3 and 5, however, significant differences between it and the AGEMA apparent temperature remain (Figure B.5).Appendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 323Inspection of the AGEMA images taken over False Creek (Flights 1—3), and Burrard Inlet(Flights 4—5), fail to show evidence of sunglint in terms of areas of increased surfaceapparent temperature or increased local variation of temperature. Unfortunately, noground-based observations of water temperature were made during the flights after sunset; the available remotely-sensed images taken over Lost Lagoon, (a small freshwaterlake) and English Bay (tidal, saltwater) do not indicate an increase in the range, or variation of surface apparent temperatures, taken during the afternoon flight, when comparedto the post-sunset flights.In conclusion, the available evidence appears to indicate a discrepancy in the EverestIRT value rather than the AGEMA temperature.B.5.1.2 Grass TemperaturesThe apparent scatter between airborne- and surface-sampled grass temperatures mayarise as the result of a number of factors. The relatively large range of temperaturesobtained for both the surface and airborne measurements indicate there is significantspatial variation over the grassy surfaces which makes the correct identification of theground sampling point critical. These variations may be a function of microscale differences in moisture content, slope, emissivity and surface structure. Differences in the lookangle and IFOV of the two instruments used may also contribute, especially consideringgrass, as with other vegetation canopies (see e.g., Paw U, 1992), has been observed tobe a non-Lambertian emmitter (Parsons, 1985). This behaviour occurs due to the combination of vertical (grass blades) and horizontal (turf layer) components which are seenin differing proportions by sensors at different viewing positions.Although the agreement in the mean between the points is not good, the rangesspanned by the two instruments tend to be similar for most flights, and decreases significantly for the night time flights. This suggests local differences in sensor position andAppendix B. ATMOSPHERIC CORRECTIONS OF THERMAL IMAGERY 324orientation, coupled with the relatively large, true variation in surface temperatures, areresponsible for the apparent scatter.Appendix CEVEREST INTERSCIENCE MODEL 4000A IRTC.1 Instrument Description.The EIRT is a small, light, self-contained, micro-computer based infrared temperaturetransducer incorporating a mechanical chopper. Instrument specifications are listed inTable C.1.Table C.1: Everest Interscience Model 4000A Infrared Transducer Specifications. (Everest Interscience, 1991)FOV: 15° (80% of total signal)Spectral response: 8 - 14 mScale Range: -30C—100°CPrecision: ± 0.1°CAccuracy: + 0.5°CResolution: 0.1°CEmissivity: 0.1—0.999 software selectable (normally 0.98)C.2 Instrument OperationThe EIRT optics collect incident radiation and focus it on a detector. The stated FOVis for 80% of the total signal; the radiant power distribution received from a surface isshown in Figure C.1.The detector produces an electrical signal proportional to the radiation received.This is conditioned, using internal circuitry, to produce a linear voltage-temperature325Appendix C. EVEREST INTERSCIENCE MODEL 4000A IRT1 .00.9Co 0.800.70.600.5(Di) 0.4cv. 0.3o 02G.)° 0.10.0—1 .0 —0.6 —0.2 0.2 0.6Relative Position Across FOV1 .0326Figure C.1: Radiant power distribution received by EIRTInterscience, 1991).from a surface. (Everestrelation. The output analog voltage signal was recorded using a Campbell Scientific CS21X datalogger. A block diagram of the EIRT components is presented in Figure C.2.The Model 4000A EIRT uses a mechanical chopper of known temperature to self-calibrate during instrument operation. . The chopper operates on a 500 ms period. Forthe first 250 ms of the period the chopper obscures the optical aperature. During theremaining 250 ms, the aperature is open and the temperature value is updated, basedupon the last 50 ms of the time during which the aperature is open (C. Everest pers.comm.). The use of the mechanical chopper significantly improves the performance of theEIRT under conditions of changing ambient temperature, removing the need to monitorinstrument body temperature (Wright, 1990). During operation, the instruments wereshielded and insulated to minimize heating and cooling gradients imposed by alternatesun and shade while traversing, and to reduce the effects of diurnal temperature changes.IT j I I100Z80ZAppendix C. EVEREST INTERSCIENCE MODEL 4000A IRT 327DCOMFUER___________o IDETECTOR CONVERTER [JIR SENSOR TR8NSr’JrfERFigure C.2: Block diagram of EIRT. (Everest Interscience, 1991).The software selectable emissivity is used to amplify the detector output signal (by a factor of ) so that the output is analogous to that of a blackbody at the same temperature;was set at 0.98 on the EIRTs used in this study. Conversion to blackbody apparenttemperature is possible using eqs. C.1- C.4, supplied by the manufacturer:TriTriTdet (C.1)E1 = —4.022402 i03 + 1.47937473 102 T1 + 9.84784552 .— 1.16050526• i0 T— 1.08464168• 10—8+ 4.39174381 10k’ (C.2)E2=•E1 (C.3)Tr2 = 0.3794+68.1983 E2 — 21.9711 E + 15.1732 E— 11.8715•E+5.2213•E (C.4)T2 Tdet+Tr2 (C.5)where Tm is the relative temperature at emissivity setting,T is the target temperature recorded with , E, is the adjusted voltage, and Tdet is the detector temperature.Appendix C. EVEREST INTERSCIENCE MODEL 4000A IRT 328Figure C.3 illustrates the sensitivity of recorded target temperature to a change in from0.98 to 1.0 for two detector temperatures, and the difference in temperature incurredby errors in estimating a detector temperature. The results suggest that if equivalentblackbody temperature is desired, the accuracy in estimating the detector temperature isless important than not including the emissivity setting at all, except for surfaces muchcolder than the detector.C.3 Instrument CalibrationThe EIRT are delivered having undergone factory calibration checks for two surface temperatures (26 and 76°C) at an air temperature of approximately 25°C. Checks on theinstrument calibrations were conducted using the calibration facilities of the University ofCalifornia (Davis) (1991) and the UBC Soil Science Department (1992, 1993). Differencesin the calibration procedures lead to differences in the derived relations between cavitytemperature (Tcav) and EIRT temperature (Tev) (Table C.2). The 1991 and 1993 calibrations utilized short exposure to a blackbody cavity of relatively constant temperature,whereas the second calibration used a long term exposure to a slowly changing blackbodycavity temperature. Greater differences between the cavity and EIRT temperatures wereobtained for the 1992 calibration, especially at low cavity temperatures, which may bedue to thermal coupling between the cavity and EIRT creating temperature gradientswithin the EIRT. These are known to cause errors in EIRT output temperature (Wright,1990) on the order of 0.5°C for a body temperature gradient of -2°C/b mm. Measurement of external EIRT surface temperature changes (probably larger than internaltemperature gradients) during some parts of the 1992 calibration, yielded surface coolinggradients at the start of the calibration between .-1.5--2.5°C over 10 minute periods. Someinstances exceeded 4°C/b mm at the beginning of the calibration (calibrations beganAppendix C. EVEREST INTERSCIENCE MODEL 4000A IRT 329at the low temperature end of the EIRT scale). During traverse operations, instrumenttemperature is assumed to be near air temperature and the results of the 1991 and 1993calibrations are probably most relevant, because instrument temperature is more constant in those calibrations. Calibration equations are based upon an emissivity settingof 0.98. Conversion of EIRT recorded temperature to apparent blackbody temperaturerequires estimation of Tdet. This can be accomplished by using air temperature (available for the 1992 and 1993 calibrations) although this introduces an error if Tdet # Ta(see Figure C.3). Regression relations between Tcav and Ta,, (f = 0.98) and Tcav and Tev(€ = 1.0) yield results generally to within +0.1°C.Table C.2: EIRT calibration results using Tev = a + bTcav; Eev = 0.98.Instrument a b Temp. range(°C)2094-1 0.203 1.001 10-542094-2 0.124 0.996 12-502094-3 1.632 0.988 12-54- 2094-4 0.571 0.993 11-542094-5 5.051 0.937 15-552094-6 1.835 0.941 12-52Results of the most recent calibration indicate that one EIRT (SN 2094-5) incorporates a substantial offset (Table C.2). The remaining EIRT show some differences froma perfect fit, although only 2 were best fit by models which lie outside the expectedaccuracy range of the EIRT (±0.5°C). In this study, instruments are corrected usingthe most recent calibration results to obtain equivalent blackbody temperature, with aninstrument emissivity setting of 0.98.Appendix C. EVEREST INTERSCIENCE MODEL 4000A IRT 3300.8 I0.6 Tdet 25° C0.40.2 ..—0.4 T0.98—T=1.oErr(Td, =±1)—0.6 Err(Td,=±2)Err(Td,=±5)—0.80 10 20 30 40 50Target Temperature (° C)(b) 0.7’ ITdet—0.620 C0.50.4 •.... .0.3o U.L—0.20 3 Err(Tdet=±1)Err(Tdt=±2)—0.4 Err(Tt=±5)—0.5 I0 10 20 30 40 50Target Temperature (° C)Figure C.3: Temperature differences due to conversion of Tev from 6ev = 0.98 to 1.00(symbols) and differences due to using ±1,2,5°C in place of Tdet (lines).Appendix DTIR REMOTE SENSING COVERAGE OF THE STUDY AREASThis appendix presents maps illustrating the location of the study sites within the Vancouver area (Figure D.l) and the approximate bounds of coverage obtained for eachscanner orientation on each flight (Figures D.2 — D.9). The figures include compositecoverage maps for orthogonal scanner azimuths (North-South, East-West), a completecomposite for all four scanner azimuths, and a total composite which includes the nadircoverage.Gaps in the coverage of some flights are a result of difficulties by the scanner operatorin maintaining the required scanner orientation due to motion of the helicopter anddifficulties in locating the correct target on the ground. The area of coverage for Flights6-8 (Residential study area) was restricted by air traffic control because of noise concernsof residents.331Appendix D. TIR REMOTE SENSING COVERAGE OF THE STUDY AREAS 332Downtown8rdwy Avo.aI iN —1’ipuquJt1WUI1W E1UUJluwiiii__II___Iil I I IFigure D.1: Study areas in Vancouver, B.C. Clockwise, from top right: Industrial (FalseCreek South) Flights 1—3, Residential (Sunset) Flights 6—8, Downtown Flights 4—5.Industrial(False Creek South)IIo=Ffl IEfl flIEtEEJ IEIJ UIEIIi IEfl ‘DLJ LJ cZtU= EnLtIE U——49th Av..ACalibration SiteFlux Measurement TowerResidential(Sunset)Appendix D. TIR REMOTE SENSING COVERAGE OF THE STUDY AREAS 333_SEEZEI000 0 1cocci Eli rzzzoco LOUJui1 11tIC] 0 00LUflcc ci ‘aooouoUESCANNER ORISNTATION4rNoOh40 Southp05 CompofleaAAS 200wColOration SiteFlux Measoremeot TowerSCANNER ORIENTATION4 East4EWest20W Compositeo CatEnation SiteA Flux Measurement TowerAS 20CmSCANNER ORIENTATION9 Nadir4rComposireTotal Compositeo Celibrotios SiteA Elao Meaxonement TowerAo 200wFigure D.2: Scanned areas (by direction) and composites for Flight 1 (Industrial area:August 15, 1992; 1030 PDT). (a) North and south scanner azimuths, (b) east and westscanner azimuths, (c) nadir and total composites. Scanner angle is approximately 450off-nadir for each azimuth.I IC L1Z7Zfl rjI.iNAppendix D. TIR REMOTE SENSING COVERAGE OF THE STUDY AREAS 334SCdNNER ORIENTATION45North4S•South• N(S0AAo 2000rSCANNER ORIENTATION45Eow4S WooS•o CaItoelIooSeA Rux Measurement TowerAo 2050,SCANNER ORIENTATIONNadir45 Cunrpou,teTotel Cotopositeo Culibrutron SiteA Flux Measurement TowerA0 2000,Figure D.3: Scanned areas (by direction) and composites for Flight 2 (Industrial area:August 15, 1992; 1400 PDT).U/1 ci’LCallbretion SitsFlux Measurement TowerEJ \1EEI DEli UEEli LII__________Z1J EIUILCjDCDcoc1”)-CDUUUEUUUU\IT]Ti.WJL.IILI1111[iF-inE1ii ‘UUIIULIULIUUULILIIIU[IUoornuornI°0-4zTi1111:11lUU’ODüLULi[jjLJUUULJ1.S,’?7cUUUUUUULiL.OE1DWWOOWinnLJ° 1I C) H C/) H4-. CDz-1Zc1Appendix D. TIR REMOTE SENSING COVERAGE OF THE STUDY AREAS 336SCANNER ORIENTATIONLi 4545 SoNhweetNEISW Corttpoeite0 CeItbraIin SrteASCANNER ORIENTATIONLi 4545 SoUIheetNWISE CompociteO Calibratton SiteA—SCANNER ORIENTATIONNedir45 Co,eposlloTold Con,posito0 Colibrotioe SiteAFigure D.5: Scanned areas (by direction) and composites for Flight 4 (Downtown area:August 16, 1992; 1130 PDT).SCANNER ORIENTATIONPOtade Park45Northeast45 Southwest•NE/SW CompositeO Calibration SiteASCANNER ORIENTATION45 Northwest•4E SoutheastNW/SE ComposheO Calibration SiteASCANNER ORIENTATIONPo.tstde ParkNadir4E Composite•Total CompositeO CalibratIon SiteAFigure D.6: Scanned areas (by direction) and composites for Flight 5 (Downtown area:August 16, 1992; 1630 PDT).Appendix D. TIR REMOTE SENSING COVERAGE OF THE STUDY AREAS 337Poieid. ParkFAppendix D. TIR REMOTE SENSING COVERAGE OF THE STUDY AREAS 338SCANNER ORIENTAtiON45 North45 SouthN/S CompositeCalibration SiteFlux Measurement TowerSCANNER ORIENTATION45 East46 WestE/W ComponiteCatibratioo SiteElsa Measorement Tower—SCANNER ORIENTATIONTotal CompositeCalibration SiteFlax Meanorewent TowerFigure D.7: Scanned areas (by direction) andAugust 17, 1992; 0945 PDT).composites for Flight 6 (Residential area:ANadir45 Composite0AASCANNER ORIENTATIONNadir45 Composite•Calibration SiteFlux Measurement TowerFigure D.8: Scanned areas (by direction) and composites for Flight 7 (Residential area:August 17, 1992; 1400 PDT).Appendix D. TIR REMOTE SENSING COVERAGE OF THE STUDY AREAS 339SCANNER ORIENTATION1 45North45 South•N/S CompositeO Calibration SiteFkix Measurement TowerASCANNER ORIENTATION45 East•45 West•E/W Composite0 Calibration SitePlus Measurement TowerAJuIE1I1I‘lu—IUo,jwAppendix D. TIR REMOTE SENSING COVERAGE OF THE STUDY AREAS 340SCANNER ORIENTATION45 North45 SouthN/S CompositeCalibretios SiteFlux Measurement Tower—SCANNER ORIENTATION45 East45 WestE/W CompositeCalibration SiteFlux Measurement Tower—SCANNER ORIENTATION45 CompositeTotal CompositeCalibration SeFlax Measurement TowerFigure D.9: Scanned areas (by direction) and composites for Flight 8 (Residential area:August 17, 1992; 1715 PDT).Nadir


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items