Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Measurements of carbon dioxide fluxes and concentrations at multiple scales in Vancouver, Canada Crawford, Benjamin R. 2014

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

Item Metadata


24-ubc_2014_september_crawford_ben.pdf [ 145.51MB ]
JSON: 24-1.0166944.json
JSON-LD: 24-1.0166944-ld.json
RDF/XML (Pretty): 24-1.0166944-rdf.xml
RDF/JSON: 24-1.0166944-rdf.json
Turtle: 24-1.0166944-turtle.txt
N-Triples: 24-1.0166944-rdf-ntriples.txt
Original Record: 24-1.0166944-source.json
Full Text

Full Text

Measurements of carbon dioxide fluxes and concentrationsat multiple scales in Vancouver, CanadabyBenjamin R. CrawfordB.S., Indiana University, 2003M.S., Indiana University, 2007A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinTHE FACULTY OF GRADUATE AND POSTDOCTORALSTUDIES(Geography)The University Of British Columbia(Vancouver)August 2014c￿ Benjamin R. Crawford, 2014AbstractGlobal carbon dioxide (CO2) atmospheric mixing ratios and near-surface air tem-peratures are projected to rise for the foreseeable future. Given that human popu-lations and activity are concentrated in urban areas, knowledge and quantificationof CO2 emissions and uptake processes is important for urban sustainable devel-opment applications and emission reduction efforts. Atmospheric measurementsof CO2 in Vancouver, BC were conducted from 2008-2012 to improve spatial res-olution of emissions monitoring techniques and advance understanding of urbanCO2 atmospheric transport processes. Three datasets representative of three urbanclimate scales were collected.At the neighborhood-scale, four years (2008-2012) of eddy covariance (EC)net CO2 emissions measurements were analyzed with spatial turbulent flux sourcearea models. This allowed high-resolution spatial attribution of the net flux toindividual source/sink processes and reduced spatial bias in EC-measured annualnet emissions totals. Empirical models developed from three years of hourly ECdata (2008-2011) were used to predict net emissions for a fourth year (2011-2012)with errors of 6.7% compared to direct EC measurements on daily timescales.At the micro-scale, high-resolution spatial variations in CO2 mixing ratiosmeasured at 2 m height from a mobile, vehicle-mounted platform were observed.Nighttime CO2 mixing ratio patterns were correlated with potential air tempera-ture, suggesting micro-scale advective processes are an important determinant ofUCL pollutant transport. Micro-scale observations are linked to local-scale EC fluxmeasurements through consideration of the EC storage flux term (FS). Uncertain-ties in hourly FS calculated from a single observation height at EC level are causedby flushing of CO2 from the UCL shortly after sunrise and from CO2 buildup iniithe UCL shortly after sunset.CO2 mixing ratios representative of the city-scale were measured in the urbanboundary layer (UBL) using a tethered-balloon system. The net CO2 flux represen-tative of the city-scale is inferred from hourly changes in UBL CO2 together withUBL height measurements from a ceilometer. The surface flux calculated usingthis method is comparable to local-scale urban EC measurements during the sameperiod, however there is uncertainty in the horizontal advective flux resulting fromsensitivity to parameterization of CO2 mixing ratios in upwind, non-urban areas.iiiPrefaceThe work presented in this dissertation is original research and was conducted withcontributions from several members of the Environmental Prediction in CanadianCities (EPiCC) project. (PIs: T.R. Oke UBC and J. Voogt UWO)A version of Chapter 2 has been published in Theoretical and Applied Clima-tology [Crawford B, Christen A. (2014): ‘Spatial source attribution of measuredurban eddy covariance carbon dioxide fluxes.’, Theoretical and Applied Climatol-ogy, DOI: 10.1007/s00704-014-1124-0.] I was the lead author and responsible forresearch conception, data collection and analyses, and creation of the manuscriptand all figures. A. Christen was involved with research concept formation andmanuscript editing. This work used experimental infrastructure (eddy covarianceequipment and tower) and design by A. Christen and T.R. Oke that were part of theEPiCC project. This work also relied on geospatial data acquisition and analysis byN. Coops, R. Tooke, and N. Goodwin. Additional analyses of building propertiesand building energy modeling was performed as part of M.Sc. research by M. vander Laan. Vegetation analyses and data collection of soil and vegetation chamberrespiration samples was conducted as part of an undergraduate research project byK. Liss. All contributions to this work by others is cited where appropriate.A version of Chapter 3 has been accepted for publication in Atmospheric En-vironment [Crawford B, Christen A. (2014): ‘Spatial variability of carbon dioxidein the urban canopy layer and implications for flux measurements’]. I was the leadinvestigator and responsible for research conception and experimental design, datacollection and analyses, and authorship of the manuscript and all figures. A. Chris-ten also contributed to research conception, experimental design, and editing themanuscript. This work used a mobile measurement and data acquisition systemivdesigned by A. Christen and R. Ketler and modified by myself for the purposes ofthis research.Chapter 4 has not been submitted for peer-review publication, but prelimi-nary results were presented at the 7th International Conference on Urban Climate[Crawford B., Christen A., McKendry I.G., van der Kamp D. (2009): ‘Verticalprofiles of carbon dioxide in the urban boundary layer - measurements and model-ing’, 7th International Conference on Urban Climate, Yokohama, Japan, June 29 -July 3, 2009]. I was lead investigator and responsible for experimental conceptionand design of the measurement system, data acquisition and analysis, and creationof the manuscript and figures. A. Christen contributed to research conception andpreliminary analysis and programming for Figure 4.4. I.G. McKendry supplied thetethered balloon system and D. van der Kamp contributed toward data collectionduring the experiment. D. van der Kamp also was responsible for operation ofa ceilometer during the experiment and used ceilometer observations to estimatethe urban boundary layer height using during data analysis. Observations fromthis experiment were also featured in a journal article in Atmospheric Environment[McKendry I.G., van der Kamp D., Strawbridge K.B., Christen A., Crawford B.(2009): ‘Simultaneous observations of boundary-layer aerosol layers with CL31ceilometer and 1064/532 nm lidar’. Atmospheric Environment, 43 (36), 5847-5852.] In this publication, wind velocities and potential air temperature measuredduring the experiment were part of a case study measuring elevated aerosol layers.A version of Appendix A has been published as a non-peer reviewed techni-cal report: [Crawford B., Christen A., Ketler R. (2013): ’Processing and qualitycontrol procedures of turbulent flux measurements during the Vancouver EPiCCexperiment.’ EPiCC Technical Report No. 1, Technical Report of the Depart-ment of Geography, University of British Columbia., 28pp,Version1.1].vTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Symbols and Acronyms . . . . . . . . . . . . . . . . . . . . . . . xiiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Cities and climate change . . . . . . . . . . . . . . . . . . . . . . 31.2 Urban carbon dynamics . . . . . . . . . . . . . . . . . . . . . . . 61.2.1 Urban ecosystems and urban metabolism . . . . . . . . . 61.2.2 Urban CO2 source and sink processes . . . . . . . . . . . 81.3 Urban atmospheric processes and scales . . . . . . . . . . . . . . 111.4 Measuring urban CO2 emissions . . . . . . . . . . . . . . . . . . 161.4.1 City-wide approaches . . . . . . . . . . . . . . . . . . . . 161.4.2 Neighborhood-scale net emissions . . . . . . . . . . . . . 191.5 Research objectives . . . . . . . . . . . . . . . . . . . . . . . . . 21vi2 Local-scale fluxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.1.1 Spatial complexity of sources and sinks . . . . . . . . . . 242.1.2 Source and sink attribution . . . . . . . . . . . . . . . . . 262.1.3 Study objectives . . . . . . . . . . . . . . . . . . . . . . 262.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.2.1 Eddy-covariance measurements . . . . . . . . . . . . . . 272.2.2 Geospatial data . . . . . . . . . . . . . . . . . . . . . . . 322.2.3 Turbulent flux source area models . . . . . . . . . . . . . 342.2.4 Linking emissions processes with land cover . . . . . . . 362.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . 372.3.1 Source area land cover and location bias . . . . . . . . . . 372.3.2 Statistical model development and downscaling . . . . . . 402.3.3 Modeling net emissions . . . . . . . . . . . . . . . . . . 582.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643 Micro-scale horizontal transects . . . . . . . . . . . . . . . . . . . . 673.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 733.2.1 Study area . . . . . . . . . . . . . . . . . . . . . . . . . . 733.2.2 Tower-based measurements . . . . . . . . . . . . . . . . 763.2.3 Mobile measurements in the UCL . . . . . . . . . . . . . 763.2.4 Data processing and spatial averaging . . . . . . . . . . . 803.2.5 Storage calculations . . . . . . . . . . . . . . . . . . . . 833.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . 853.3.1 Spatial patterns of CO2 mixing ratios in the UCL . . . . . 853.3.2 Storage and venting from the UCL . . . . . . . . . . . . . 923.3.3 Integrating UCL and above-canopy storage . . . . . . . . 953.3.4 Storage analysis using long-term tower measurements . . 973.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1034 City-scale vertical profiles . . . . . . . . . . . . . . . . . . . . . . . . 1054.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105vii4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1104.2.1 UBL CO2 observations . . . . . . . . . . . . . . . . . . . 1104.2.2 CO2 observation network . . . . . . . . . . . . . . . . . . 1134.2.3 Boundary layer CO2 budget . . . . . . . . . . . . . . . . 1174.2.4 UBL measurement source area . . . . . . . . . . . . . . . 1204.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . 1224.3.1 Observed meteorology and CO2 mixing ratios in the UBL 1224.3.2 Observed UBL dynamics and CO2 mixing ratios . . . . . 1274.3.3 Modeled FC using a boundary layer budget . . . . . . . . 1314.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1385 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1425.1 Summary of findings . . . . . . . . . . . . . . . . . . . . . . . . 1425.2 Contributions and implications . . . . . . . . . . . . . . . . . . . 1455.3 Reflections and future work . . . . . . . . . . . . . . . . . . . . . 146Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151A Eddy covariance data processing and quality control procedures . . 169A.1 Instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 169A.2 Sensor calibrations . . . . . . . . . . . . . . . . . . . . . . . . . 173A.3 High frequency quality control . . . . . . . . . . . . . . . . . . . 173A.3.1 Sonic anemometer diagnostic value . . . . . . . . . . . . 174A.3.2 IRGA diagnostic value . . . . . . . . . . . . . . . . . . . 174A.3.3 High frequency spike detection . . . . . . . . . . . . . . 174A.3.4 Flow distortion by the sensor head . . . . . . . . . . . . . 177A.3.5 High frequency statistics check . . . . . . . . . . . . . . 177A.3.6 Sonic-IRGA time lag . . . . . . . . . . . . . . . . . . . . 178A.4 Block average calculation and coordinate rotation . . . . . . . . . 180A.5 Post-processing corrections applied to turbulent carbon dioxide flux 180A.6 Comparison of eddy covariance processing software . . . . . . . . 182viiiList of TablesTable 2.1 Air temperature, precipitation, and data availability during thestudy period. . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Table 2.2 Landcover in the Sunset study area. . . . . . . . . . . . . . . . 32Table 2.3 Model parameters for multiple linear regression hourly trafficemissions model. . . . . . . . . . . . . . . . . . . . . . . . . . 43Table 2.4 Model parameters for hourly segmented regression building space-heating emissions model. . . . . . . . . . . . . . . . . . . . . 49Table 2.5 Emissions and uptake process statistical model summary. . . . 59Table 2.6 Modeled emissions scaled for source area and entire study do-main. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63Table 3.1 Transect run summary. . . . . . . . . . . . . . . . . . . . . . . 78Table 4.1 Hourly UBL profile averages. . . . . . . . . . . . . . . . . . . 125Table 4.2 Boundary layer budget FC and eddy covariance FC . . . . . . . 134Table A.1 Sonic anemometers. . . . . . . . . . . . . . . . . . . . . . . . 171Table A.2 Infrared gas analyzers. . . . . . . . . . . . . . . . . . . . . . . 172Table A.3 Infrared gas analyzer calibrations. . . . . . . . . . . . . . . . . 174Table A.4 Quality control statistical filters. . . . . . . . . . . . . . . . . . 176Table A.5 Flow distortion wind directions. . . . . . . . . . . . . . . . . . 178Table A.6 Sonic-IRGA timelag. . . . . . . . . . . . . . . . . . . . . . . 179Table A.7 Sonic-HMP temperature difference. . . . . . . . . . . . . . . . 182Table A.8 EC software comparison settings. . . . . . . . . . . . . . . . . 183Table A.9 EC software linear regression comparison. . . . . . . . . . . . 184ixList of FiguresFigure 1.1 Urban climate scales. . . . . . . . . . . . . . . . . . . . . . . 14Figure 2.1 Map of the Vancouver Sunset study domain. . . . . . . . . . . 28Figure 2.2 Wind, FC, and landcover by direction. . . . . . . . . . . . . . 31Figure 2.3 Landcover by wind direction and stability. . . . . . . . . . . . 39Figure 2.4 Diurnal course of modeled traffic emissions. . . . . . . . . . . 41Figure 2.5 Building space-heating hourly segmented regression model. . 47Figure 2.6 Modeled space-heating emissions by hour and air temperature. 50Figure 2.7 Daytime FC and solar radiation. . . . . . . . . . . . . . . . . 52Figure 2.8 Nighttime FC and soil temperature. . . . . . . . . . . . . . . . 55Figure 2.9 Comparison of modeled and observed FC. . . . . . . . . . . . 61Figure 2.10 Mapping emissions and uptake processes. . . . . . . . . . . . 65Figure 3.1 Study area and transect map. . . . . . . . . . . . . . . . . . . 74Figure 3.2 Conceptual diagram of FS calculations. . . . . . . . . . . . . 84Figure 3.3 CO2 mixing ratios and ∆c/∆t . . . . . . . . . . . . . . . . . . 86Figure 3.4 Spatial patterns of CO2 and air temperature . . . . . . . . . . 88Figure 3.5 CO2 and air temperature correlations. . . . . . . . . . . . . . 90Figure 3.6 CO2 and air temperature maps with elevation contours. . . . . 91Figure 3.7 Vertical CO2 gradient and σw . . . . . . . . . . . . . . . . . . 93Figure 3.8 Spatial patterns of ∆c/∆t. . . . . . . . . . . . . . . . . . . . . 95Figure 3.9 Storage calculations. . . . . . . . . . . . . . . . . . . . . . . 96Figure 3.10 Diurnal course of ensemble mean CO2 and storage. . . . . . . 99Figure 3.11 Storage, stability, and mixing . . . . . . . . . . . . . . . . . . 102xFigure 4.1 Tethered balloon measurement system. . . . . . . . . . . . . 112Figure 4.2 Map of CO2 measurements in Vancouver. . . . . . . . . . . . 114Figure 4.3 Single box model construct of UBL CO2. . . . . . . . . . . . 119Figure 4.4 Time-height contour plots of UBL CO2 and potential temper-ature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123Figure 4.5 Time-height plot of U-V wind vectors. . . . . . . . . . . . . . 124Figure 4.6 CO2 mixing ratios during July-August 2008. . . . . . . . . . 126Figure 4.7 Height of daytime UBL and nighttime NBL. . . . . . . . . . 128Figure 4.8 Vertical CO2 and potential temperature profiles in the UBL. . 129Figure 4.9 Boundary layer budget FC. . . . . . . . . . . . . . . . . . . . 132Figure 4.10 UBL measurement source areas. . . . . . . . . . . . . . . . . 135Figure 4.11 Modeled surface flux, entrainment, and advection. . . . . . . 137Figure A.1 EC system photographs. . . . . . . . . . . . . . . . . . . . . 170Figure A.2 Sonic anemometer intercomparison. . . . . . . . . . . . . . . 173Figure A.3 Sonic anemometer flow distortion. . . . . . . . . . . . . . . . 177xiList of Symbols and AcronymsSymbol Definition Unitsc Instantaneous carbon dioxide mixing ratio µmol mol−1 (ppm)c￿ Fluctuation of carbon dioxide mixing ratio µmol mol−1 (ppm)c Temporal average carbon dioxide mixingratioµmol mol−1 (ppm)￿c￿ Spatial average carbon dioxide mixing ra-tioµmol mol−1 (ppm)cρ Carbon dioxide partial density g m−3C Photosynthesis model convexity parame-tern/a∆i Turbulent flux source area location bias %EC Eddy covarianceEV Vehicle combustion carbon dioxide emis-sionsµmol m−2 s−1EB Building combustion carbon dioxide emis-sionsµmol m−2 s−1e Vapor pressure hPaFC Turbulent carbon dioxide flux density µmol m−2 s−1FS Eddy covariance storage flux µmol m−2 s−1g Gravitational acceleration m s−2ISL Inertial sub-layerk Von-Karman constantK ↓ Incoming solar radiation W m−2xiiL Obukhov Length mLAIT Tree leaf area index m−2 m−1LAIL Lawn grass leaf area index m−2 m−1λB Building plan-area landcover fraction %λI Impervious plan-area landcover fraction %λL Lawn plan-area landcover fraction %λT Tree plan-area landcover fraction %λM Major road plan-area landcover fraction %λS Secondary road plan-area landcover frac-tion%λR Residential road plan-area landcover frac-tion%M Maximum quantum yield of photosynthe-sisµmol µmol−1MA Molecular mass of dry air kg mol−1Mq Molecular mass of water vapor kg mol−1NBL Nocturnal boundary layerp Barometric pressure hPaPL Lawn gross photosynthesis carbon dioxideuptakeµmol m−2 s−1Pm Photosynthesis at light saturation µmol m−2 s−1PT Tree gross photosynthesis carbon dioxideuptakeµmol m−2 s−1Ptot Total gross photosynthsis from tree andlawn componentsµmol m−2 s−1PPFD Photosynthetic photon flux density µmol m−2 s−1q Atmospheric water vapor density g m−3QE Latent heat flux density W m−2QH Sensible heat flux density W m−2R Universal gas constant J K−1 mol−1RV Gas constant for water vapor J K−1 kg−1RH Human respiration carbon dioxide emis-sionsµmol m−2 s−1xiiiRL Lawn respiration carbon dioxide emis-sionsµmol m−2 s−1RT Tree respiration carbon dioxide emissions µmol m−2 s−1RH Relative humidity %RL Residual layerRSL Roughness sub-layerρA Density of dry air kg m−3ρc Density of carbon dioxide kg m−3ρq Density of water vapor kg m−3σw Vertical wind velocity standard deviation m s−1T ￿a Fluctuation of acoustic air temperaturefrom sonic anemometer◦CT Air temperature from slow-response sen-sor◦CTS Soil temperature ◦CTH Heating threshold air temperature for EB ◦CTKE Turbulent kinetic energy m2 s−2θ Potential air temperature ◦Cθ Temporal average potential air tempera-ture◦C￿θ￿ Spatial average potential air temperature ◦CU Horizontal wind velocity m s−1u∗ Friction velocity m s−1u Instantaneous longitudinal wind compo-nentm s−1u￿ Fluctuation of longitudinal wind compo-nentm s−1UBL Urban boundary layerUCL Urban canopy layerv Instantaneous lateral wind component m s−1v￿ Fluctuation of lateral wind component m s−1w Instantaneous vertical wind component m s−1w￿ Fluctuation of vertical wind component m s−1xivw￿c￿ Covariance of w￿ and c￿ µmol m−2 s−1w￿t ￿ Covariance of w￿ and t ￿ K m−2 s−1w￿q￿ Covariance of w￿ and q￿ g m−2 s−1z Measurement height mz￿ Effective measurement height mzB Building height mzd Displacement height mzi Convective boundary layer height mzT Tree height mz0 Roughness length mxvAcknowledgmentsThis dissertation would not have been possible without the help and support ofmany people. Research funding was provided by the Canadian Foundation for Cli-mate and Atmospheric Sciences (CFCAS) as part of the network ‘EnvironmentalPrediction in Canadian Cities (EPiCC)’ project (PIs: T.R. Oke UBC and J. VoogtUWO) and by an NSERC Discovery Grant ‘Direct measurement of greenhouse gasexchange in urban ecosystems’ (PI: A. Christen, UBC). Research infrastructurewas supported by NSERC, CFI and BCKDF (PIs: Christen, Oke). Acknowledge-ments are also due to the City of Vancouver and Environment Canada for providingadditional data and BC Hydro for granting access to the Sunset tower site. Furthersupport in terms of equipment placement and access came from Cypress Mountainski resort (CO2 sensor), Sir William Osler Elementary School (Oakridge Tower),and Hugh Reynolds (Westham Island Tower).This research also benefitted from scientific, technical, and administrative sup-port of staff and students at UBC. In particular Rick Ketler was invaluable for tech-nical advice and support. N. Coops, N. Goodwin, and R. Tooke provided expertisein terms of remotely sensed and geospatial data. M. van der Laan, S. Krayenhoff,N. Lynch, and C. Emmel from UBC Geography also provided valuable technicaland intellectual support. Further technical and administrative support was providedby E. Heyman, S. Lapsky, K. Kinsley, K. Liss, Z. Nesic, J. Ranada, C. Siemens, andFred Meier (TU-Berlin). UBC Geography professors Ian McKendry and MarwanHassan also provided valuable counsel and technical support.I feel deeply indebted to my PhD advisory committee. Above all, I am ex-tremely grateful for the support, guidance, and patience of my supervisor AndreasChristen. A. Black was also instrumental in developing and clarifying key ideasxviand concepts related to carbon exchange in the urban ecosystem. T. Oke and D.Steyn provided valuable intellectual input and challenged me in ways to make thisa better project. I also would like to especially acknowledge Sue Grimmond (Uni-versity of Reading) who introduced me to the study of urban climates.Finally, I’d like to acknowledge and thank my parents. Nothing would be pos-sible for me if not for them.xviiChapter 1Introduction‘The great aerial ocean which surrounds us, has the wonderfulproperty of allowing the heat-rays from the sun to pass through it with-out its being warmed by them; but when the earth is heated the air getswarmed by contact with it, and also to a considerable extent by theheat radiated from the warm earth because, although pure, dry air al-lows such dark heat-rays to pass freely, yet the aqueous vapour andcarbonic acid in the air intercept and absorb them.’-Alfred Russel Wallace, Man’s Place in the Universe, 1904Alfred RusselWallace, British explorer, scientist, geographer, and co-discovererof the theory of evolution through natural selection, was in his 80th year when hiswords describing what we call ‘the greenhouse effect’ were published (Wallace,1904). By 1904, the existence of the greenhouse effect and its role in warmingthe Earth had already been known by the scientific community for virtually allof Wallace’s life. The French mathematician Joseph Fourier is generally creditedwith the first proposal of the greenhouse effect in publications in 1824 and 1827(Fourier 1824, Fourier 1827). In these publications, he references earlier exper-iments by the Swiss physicist and mountaineer Horace-Benedict de Saussure inwhich the temperature of air trapped between panes of transparent glass was ob-served to rise when exposed to sunlight. In the 1860s, the British physicist JohnTyndall experimentally measured the infrared absorptive powers of various gases inthe atmosphere, including ‘carbonic acid’, or carbon dioxide (CO2) (Tyndall 1859,1Tyndall 1873). By 1896, an attempt to quantify the role of carbon dioxide in thegreenhouse effect was published by Svante Arrhenius (Arrhenius, 1896). By 1917,prominent scientists such as Alexander Graham Bell had expressed concern thatincreased atmospheric CO2 concentrations from unchecked fossil fuel combustioncould have negative consequences for the climate (Bell, 1917).As this is being written in early 2014, we have nearly a century’s worth ofadvances in measurement technology, atmospheric modeling techniques, and sci-entific understanding, yet we share the same concern about the effect of CO2 accu-mulation in the atmosphere from the burning of fossil fuels. Geologic reserves ofcarbon have been transferred to the atmosphere in the form of CO2 through rapidconsumption and combustion of hydrocarbon fuels and the resulting climatic ef-fects have been observed and documented throughout the world (e.g. Stocker et al.2013). The ubiquity of CO2 makes the subject of ‘CO2 emissions’ extremely com-plex and multi-faceted. CO2 emissions are discussed and analyzed in many fieldsof study including the natural sciences, energy applications, economics, and thesocial sciences. CO2 emissions are also a topic of political discussion and conflictat all levels of government, from local municipalities up to international levels.The goal of this dissertation is to contribute to the study of CO2 emissionsby focusing on a particular place: cities. Human populations and activities areincreasingly concentrated in urban areas so an understanding of the various CO2emissions and uptake processes operating in these spaces is important to mitigateadverse climate effects through emissions reductions. In particular, methods tomeasure and quantify emissions are vital to establish metrics to assess and monitorsustainable development efforts. These methods will also be valuable in a contextof increasing regulation and commodification of CO2 emissions (e.g. Kossoy andGuigon 2012).The research presented here is an investigation of the spatial patterns of par-ticular physical processes and phenomena in a specific geographic area at severalspatial scales. Therefore this is an inherently geographic study of urban CO2 emis-sions, even though the research relies on applied atmospheric surface layer theoryand observation techniques. As such, this manuscript is organized according tospatial scales of urban climates. Though this research is necessarily limited inscope, the diffusive properties of the ‘great aerial ocean which surrounds us all’2mean urban CO2 emissions have influence well beyond city limits.1.1 Cities and climate changeAs of early 2014, world population is estimated to be 7.15 billion people (USCB,2014). By 2050, this figure is forecast to grow by 30% (2.15 billion) to 9.3 billionpeople (UN, 2011). During this same time period, the world’s urban populationis expected to rise to 6.3 billion people, an increase of approximately 2.6 billionindividuals. This means that urban areas will absorb all of the world’s populationgrowth over the next four decades, as well as a portion of the rural population. Asa result, the world population is expected to be 67% urban by 2050 and there willactually be fewer people living in rural areas than today. These global figures, how-ever, obscure the diversity of trends in the mass urban migration that is currentlyunderway throughout the world.In the more industrially developed countries of Western Europe, North Amer-ica, Australia, and New Zealand, 78% of the population is already urbanized. Thetotal urban population of these regions is expected to only increase modestly from 1billion in 2011 to 1.1 billion in 2050 (UN, 2011). In contrast are the less developedcountries, particularly in Asia and Africa, where expected growth of urban popula-tions will increase from 2.7 billion in 2011 to 5.1 billion in 2050 (UN, 2011). Thismeans virtually all the expected growth in urban populations will be concentratedin the developing countries. In particular, China and India together are projected toaccount for a third of the total increase in urban population.The world population is not currently distributed evenly among cities of differ-ent sizes, but will become more concentrated in large cities (UN, 2011). Currently,about half the world’s urban population live in cities with fewer than 500,000 andanother 10.1% live in cities with population between 500,000 and 1 million peo-ple. Together, 61% of the urban population lives in cities with fewer than 1 millionpeople. By 2025, this balance is expected to shift. Cities of 1 million or moreinhabitants are expected to account for 47% of the world urban population by 2025and so-called ‘megacities’ with populations greater than 10 million residents willexperience the largest population percentage growth increase (UN, 2011).The growth of megacities has been rapid. In 1970, only Tokyo, Japan and New3York, USA had populations greater than 10 million. Today, the total populationliving in megacities has increased tenfold and there are now 23 megacities world-wide (13 in Asia, four in Latin America, and two each in Africa, Europe, and NorthAmerica). As of 2011, megacities accounted for 9.9% of the world urban popu-lation and 5.2% of overall population, by 2025 these proportions are expected toincrease to 13.6% and 8%, respectively. Nine of the ten fastest growing megacitiesare located in Asia (UN, 2011).The trends of industrialization and population growth driving urbanization arealso behind the global rise in greenhouse gas (GHG) emissions and atmosphericmixing ratios. It is well documented that global mixing ratios of GHGs have in-creased from pre-industrialization levels to present. Globally since 1750, atmo-spheric mixing ratio of carbon dioxide (CO2), methane (CH4), and nitrous oxide(N2O) have increased by 40%, 150%, and 20%, respectively (Stocker et al., 2013).Based on evidence from ice core samples, present concentrations of all three GHGsare the highest they have been during the past 800,000 years and the rates of in-crease are unprecedented in the past 22,000 years (Stocker et al., 2013).CO2 is the most important GHG in terms of radiative forcing and in 2011 isestimated to contribute an additional 1.68 W m−2 to the global climate system rel-ative to 1750 (Stocker et al., 2013). CO2 is also the largest anthropogenic GHGby total emissions and from 1750-2011, 555 GtC have been released to the at-mosphere. Of this historical total, fossil fuel combustion and cement productionaccount for 375 GtC (67.6%) and deforestation and other land use changes accountfor 180 GtC (32.4%) (Stocker et al., 2013).As mean global near surface air temperatures are projected to rise over thecourse of the 21st century under all emissions scenarios (Stocker et al., 2013),urban populations will be among the most vulnerable to climate change impacts.Many of the world’s largest cities are located in coastal areas and residents andinfrastructure are at risk from rising sea levels and storm surges. This risk is com-mon to cities in more developed countries (e.g. New York City, USA; Miami,USA; Rotterdam, Netherlands; Tokyo, Japan) and in rapidly developing countries(e.g. Kolkata, India; Shanghai, China; Guangzhou, China) (OECD, 2010). An-other hazard to human health are intense heat waves exacerbated by the urban heatisland effect. Urban areas can be 3.5-4.5 K warmer than surrounding rural ar-4eas, and this is expected to increase by 1 K per decade up to a magnitude of 10K in large cities (Voogt, 2002). Additionally, large populations of economicallydisadvantaged people with inadequate housing infrastructure located in the mostvulnerable areas are often clustered in urban areas. These populations are often themost exposed to severe weather and climate conditions and have the least capacityto adapt and protect themselves (OECD, 2010).Some studies estimate urban areas account for up to 75-80% of GHG emis-sions (Stern, 2007) as well as 80% of the world’s gross domestic product (UN,2011). Given the concentration of wealth, industry, and population in urban areas,cities have emerged as a powerful force for regulating greenhouse gas pollution.For example, in the United States 1,060 city mayors have signed the ‘U.S. Con-ference of Mayors’ Climate Protection Agreement’ to meet or exceed Kyoto Pro-tocol emission targets (USCOM, 2009). Combined, these cities represent nearly30% of U.S. population and one third of total GHG emissions (Lutsey and Sper-ling, 2008). Internationally, the C40 Cities Climate Leadership Group counts 40large participant cities and 19 affiliate cities whose aims are to reduce emissionsthrough energy-efficiency and clean-energy programs (Rosenzweig et al., 2010).Cities are a logical level of government for enacting GHG emissions because mu-nicipal decisions about transportation infrastructure (e.g. mass transit, highways,bicycle lanes), built form (e.g. building type and density, land use mix), and trafficcongestion can have significant influence on lifestyle choices and emission levels.In order to reduce urban CO2 emissions, a quantitative understanding of urbanecosystem carbon dynamics is needed. Measuring total CO2 emissions and thecontribution from various sources in different cities can help us to understand howand why emissions vary and how they might be reduced (Bulkeley, 2013). A morecomplete understanding of urban CO2 emission processes, patterns, and efficien-cies will enable cities to better monitor and predict emissions. This knowledge canalso form the basis to develop tools to evaluate and implement emission reductionand sustainable growth strategies.51.2 Urban carbon dynamics1.2.1 Urban ecosystems and urban metabolismUrban areas are among the most complex systems ever built by humans, but theyare not self-sufficient. Cities are open biophysical systems dependent on materialand energy flows across a range of temporal and spatial scales. Urban areas can beconsidered ecosystems because they consist of organisms functionally linked witha physical environment, acting within a spatial area (Pickett et al., 1997). Humansare the primary organisms for which the city is a habitat (though other animals alsomake their homes in cities) and we are functionally linked to this physical environ-ment (buildings, soil, vegetation, roads, air) through water, energy, and chemicalcycles. At larger scales, cities are functionally linked to their surroundings throughexchanges of materials and energy and are considered heterotrophic ecosystemsbecause they depend on external sources of energy and materials (Collins et al.,2000).In terms of material and energy flows, the finer the spatial scale under consid-eration, the more ‘open’ that system is. For example, individual urban householdsdepend almost entirely on external inputs of water, energy, and food to function.Larger systems such as entire metropolitan regions may be more ‘closed’, or self-sufficient, in terms of energy generation and waste management, but are open toimports of food and raw materials. At higher levels (regional, national), a systemmay be largely self-sufficient (closed) in terms of food and raw materials produc-tion, but dependent on fossil fuel inputs such as oil. At the global level, the systemis considered closed, with no external inputs or outputs except solar energy.Studies of urban ecosystems can be divided into studies of ecology in cities andecology of cities (Pickett et al., 2001). The former concerns specific componentsof urban ecosystems such as alteration of soil composition and chemistry (Efflandand Pouyat, 1997), vegetation responses (McPherson et al., 1997), or urban cli-mate (e.g. urban heat island, Voogt 2002). The latter is a more systems-orientedapproach that considers entire cities (or neighborhoods) and their biogeochemicalcycles and component interactions, for example.The concept of ‘urban metabolism’ is a systems-based approach used to re-6fer to the flows of matter and energy through urban ecosystems (e.g. Wolman1965, Kennedy et al. 2011). This term implies cities are analogous to living organ-isms and require input and transformation of energy, water, and raw materials andremoval of waste products to survive. The concept is useful for organization andquantification of input-output material flows for urban areas, but has been critiquedas simplistic and misleading because of potential for confusion between processesat organismal scales and ecosystem scales (Golubiewski, 2012). Depending on thescale of analysis, the concept can also result in a ‘black-box’ approach that ignorescomplex feedbacks and interactions within the area of study. Also implicit in urbanmetabolism-based studies is the challenge of boundary definition (or imposition)in complex, continuous systems. This can lead to conceptualization of the cityas a separate entity from the surrounding environment, rather than part of largersocio-economic and ecologic systems.Limitations with urban metabolism-based approaches to quantifying urban CO2emissions are apparent in the distinction between production-based (direct) emis-sions and consumption-based (indirect) emissions. Production-based approachesassign responsibility for emissions to the geographic area in which emissions oc-cur. For example, emissions from fossil fuel combustion for transportation withincity limits would count towards a city’s emissions totals, but emissions from anelectrical plant that supplies power to the city and generates emissions outsidecity boundaries would not. In contrast, consumption-based approaches attribute‘upstream’ emissions (and also ‘downstream’ emissions, for example from wastetreatment) to the city and recognize that urban residents’ energy consumption, de-mand for food and goods, and waste production drive CO2 emissions that occuroutside city boundaries. This acknowledges the role of the city as consumptiondestination at the end of global supply and manufacturing chains and that importedenergy, food, and goods arrive in cities laden with embodied CO2 emissions.More comprehensive understanding of urban carbon cycles could result fromintegrated frameworks that consider both biophysical components and socio-economicfactors and their interactions (Pickett et al. 1997, Pataki et al. 2006). For example,a particular area conducive for integrative study is the concept of spatial hetero-geneity at neighborhood scales (e.g. Pickett et al. 2001). Social geographers, an-thropologists, historians, urban planners, and economists have much to say on the7development and evolution of neighborhoods along socio-economic divisions (e.g.Knox and McCarthy 2005 ). In turn, these social divisions can lead to physical dif-ferentiation of the urban environment which influences physical cycles and fluxesof energy and matter within neighborhoods and across the entire urban landscape.This broader understanding of how the social and physical components of the urbanecosystem interact could be valuable in terms of creating sustainable developmentand urban resilience policies.These types of integrated urban ecological studies have potential to incorpo-rate underlying drivers of CO2 emissions such as population growth, resource dis-tribution, macro-economic trends, and urban organization along socio-economicdivisions. Though the merits of this type of integrative approach are recognized,application of these methods is beyond the scope of this work. This research ismore narrowly targeted towards advancing observational methods to quantify CO2emissions and atmospheric transport within urban areas. Therefore the followingdescription of urban carbon dynamics is focused on physical processes contribut-ing to direct CO2 emissions within urban areas. Thus, any ability to ‘explain’ CO2emissions through this research is acknowledged to be only in a very narrow phys-ical sense.1.2.2 Urban CO2 source and sink processesCarbon cycles through, and within, urban systems across a range of temporal andspatial scales. Within a given city, there are significant storage pools of carbonin soils, vegetation, concrete, asphalt, and building stock (Churkina et al., 2010).These long-term pools may fluctuate on multi-year to decadal timescales as urbanphysical morphology evolves in terms of vegetation (growth or removal) and build-ing character (density and function). On timescales ranging from 1 hour to severalyears, these long-term stores remain roughly constant, but there can be substantialhorizontal flows of carbon across urban boundaries in the form of fossil fuels (nat-ural gas, gasoline, oil, coal), biofuels (wood, bioethanol), food, and waste. CO2 isemitted as a waste product vertically from the system from combustion of fossil fu-els and biofuels, as well as respiration from vegetation, animals, humans, and soils.In addition, vegetation acts to locally remove atmospheric CO2 through photosyn-8thesis (input to the system). Although CO2 is not directly harmful to humans, it iscorrelated with other harmful air pollutants resulting from fossil fuel combustion(e.g. NOx, CO) (Henninger and Kuttler 2010) and CO2 has potential to be used asa tracer gas for urban atmospheric transport studies (Pataki et al., 2005). The focusof this research is on timescales (hourly, daily, seasonal, annual) relevant to theseemissions and uptake processes.In North America, approximately 40% of fossil fuel combustion and associ-ated CO2 emissions are by the transportation and residential sectors (Pataki et al.,2006). Uses for residential fossil fuel combustion include space heating, waterheating, and cooking. Total emissions and the location of emissions vary basedon the mix of fuels available. For example, natural gas combustion for space andwater heating and cooking would result in direct CO2 emissions from an individualresidence, whereas emissions associated with electric heating and cooking appli-ances would be located elsewhere at the point of electrical generation. Wood andcoal combustion for heating and cooking are also a significant source of direct CO2emissions in some cities.Climate is also a significant driver of residential fossil fuel consumption anddirect CO2 emissions. Cities that experience cold winters use more fossil fuels forheating during winter, resulting in direct CO2 emissions. During summer, electri-cal consumption associated with space cooling may increase, but associated CO2emissions are likely to be indirect (non-local) because most air-conditioning sys-tems are electric and do not directly combust fossil fuels. Building emissions oper-ate on temporal scales ranging from diurnal cycles based on occupant activity andschedule to seasonal and annual temperature variations. Spatial scales range fromindividual buildings up to city-wide extent.Built urban form in terms of building density and orientation has also beenshown to influence building energy efficiency and demand by up to 40% (Steemers,2003). This is due to sheltering effects from wind or solar input and to shared wallheating efficiencies in dense, residential complexes. Individual building efficiencyand residential habits also affect energy consumption and direct emissions. Simu-lations by residential building energy models for Canada has shown that emissionreductions of 9% could be achieved with upgrades to housing insulation and appli-ance efficiency (Pataki et al., 2006).9For the transportation sector, urban density significantly influences emissionsthrough average vehicle trip lengths and choice of transportation (Newman andKenworthy, 1999). ‘Urban sprawl’ is a global phenomenon characterized by de-creasing built density, segregation of residential and commercial districts, and ex-pansion of transportation networks (Ewing et al., 2003). These types of urbandevelopment increase commuting distances, vehicle miles traveled, and associatedvehicular CO2 emissions. In contrast, dense urban environments reduce commuterdistance and make mass transit options more viable. Fossil fuel-based mass transitsuch as buses reduce direct per capita CO2 emissions, and electric systems (trol-leys, trains, subways) displace emissions to the point of electrical generation. Elec-tric vehicles may also play an increasing role in urban transportation (Hadley andTsvetkova, 2009). Though adoption of this technology has potential to reduce di-rect CO2 emissions and improve local urban air quality, electricity demand willalso increase. The effect of this would be to shift emissions from multiple fossilfuel-based sources that are mobile and highly variable in space and time to sta-tionary point-sources where fossil fuel-based electricity is generated. Currently,vehicle emissions are relevant on temporal scales from diurnal and weekly com-muter cycles. Traffic patterns change seasonally as well with higher traffic densitytypically found in summer. Spatially, traffic emissions are released by mobile com-bustion sources along roads. From a fixed frame of reference (as opposed to vehicleperspective), scale of vehicle emissions range from individual road lane widths tocity-wide extent.Vegetation and soils have also been shown to have a direct effect on net emis-sions through CO2 uptake by vegetation and emissions by soils and vegetation. Inthe United Sates, carbon storage in urban trees is on the order of 700 Mt C witha annual sequestration rates of 22.8 Mt C yr−1 (Nowak and Crane, 2002). Soilsrepresent a carbon storage pool in urban areas and have been shown to have higherfluxes of both CO2 and N2O compared to non-urban soils (Kaye et al., 2005). Veg-etation also can indirectly affect emissions through building solar shading, windshading, and alteration of evaporative cooling rates (Oke et al. 1989, Taha et al.1991). Some indirect effects may increase energy use, such as increased shadingby trees during winter, but models suggest the net effect is a decrease in energy useand associated direct CO2 emissions (Akbari and Konopacki, 2005). Vegetation10and soil processes are generally controlled by environmental factors such as solarirradiance, soil temperature, moisture availability, humidity, and air temperaturethat fluctuate on diurnal-seasonal timescales. Spatially, these biogenic processesoccur on scales ranging from individual leaf stomata up to city-wide urban forests.Depending on the the particular city, industrial sources may be also be a sig-nificant emissions source. For example, Rotterdam, Netherlands has high percapita emissions (29.8 tons of CO2-equivalent (tCO2e)) compared to national val-ues (12.67 tCO2e) in part due to its busy industrial shipping port (Hoornweg et al.,2011). These activities and emissions are typically localized in industrial centers,separate from residential areas.Human respiration is also a source of direct CO2 emissions in urban areas. Thisrespired carbon enters human bodies from consumption of sequestered carbon inagricultural plants (or via meat from dead animals who consumed plant carbonwhile living). Hence this carbon is considered ‘new’ (and renewable) compared tothe carbon stored in fossil fuels and released during combustion. Total respirationemissions from humans depend on population density and variations in activity.Temporally, population density for a city, or area within a city, may change accord-ing to daily and weekly commuter patterns. Spatially, respiration occurs typicallywithin buildings and is released to the atmosphere over hourly timescales throughbuilding ventilation.1.3 Urban atmospheric processes and scalesBefore techniques of measuring urban CO2 emissions are discussed, concepts inurban atmospheric research must be introduced and the scale domain of researchdefined. Boundary-layer meteorological theories and concepts of scale applied tothe urban environment guide measurements and interpretation of results. Scalesof atmospheric processes must be considered along with scales of CO2 source andsink processes and scales of urban physical form. Here, the study domain is definedusing a phenomenological approach.As described earlier, relevant CO2 emissions and uptake processes occur onspatial scales ranging from individual leaf stomata (10−6 m) up to city-wide ex-tents (104 m). Temporally, CO2 emissions and uptake processes vary on hourly,11weekly, and seasonal cycles (103-107 s) according to anthropogenic and environ-mental controls. At smaller scales (10−2-100 m, 100-102 s), CO2 emissions andtransport are determined by photosynthetic biochemical reactions, stomatal con-ductance, and internal combustion processes. These processes are considered tobe ‘relaxed’, in other words their mean contributions are important to the largerscale phenomena of net emissions, but individual fluctuations and sub-processesare ignored. Processes (e.g. continental weathering, socio-economic drivers) op-erating on longer timescales (decadal-century) are assumed to be constant duringthis study.Within a city, the spatial configuration and relative densities of the various CO2emissions and uptake process can vary considerably. To organize research, urbanclimate scales are usually expressed in terms of a hierarchy of repeated urban el-ements (buildings, trees, streets/lots) and urban units (street canyon, block, neigh-borhood) (Oke, 1987). At the micro-scale, the outdoors urban climate for any givenlocation is a function of the proximity, orientation, and climatic (roughness, ther-mal, radiative, moisture) properties of individual urban elements. The micro-scaleurban climate is modified by the individual elements through creation of turbulentvortices and wakes, thermal or pollutant plumes, and patterns of temperature andrainfall created by shade and wind effects.Roads flanked by buildings on either side form urban street canyons. Streetcanyons are defined by their height to width ratio and may have vegetation andsoils present. Urban blocks are formed by road patterns and can be formed ofa perimeter of buildings with vegetation, courtyards, or alleyways in the blockinterior. Other block-level structures can include shopping malls, institutions (e.g.schools, hospitals), factories, or apartment complexes.At larger length scales, urban areas can be differentiated in terms of neighbor-hood characteristics. At the neighborhood, or local, scale urban environments canbe distinguished by variations in building, vegetation, and population density (e.g.high rise apartments versus low density suburbs) or land use (e.g. industrial, com-mercial, residential, parkland). Urban climate at the local is the integrated responseof individual urban units and elements within the neighborhood. Stewart and Oke2012 have developed a classification scheme composed of 17 ‘local climate zones’representative of this scale.12At the largest scale, the entire city is considered as a whole and can be differ-entiated from its surrounding areas. Urban climate phenomena at this scale includethe urban boundary layer and urban plume downwind of a city which are represen-tative of all source and sink activity within the urban area.The spatial and temporal domain of urban CO2 emissions/uptake processes andurban physical form corresponds with meteorological phenomena at the micro-scale, local-scale, and lower range of the meso-scale (e.g. Steyn et al. 1981). Atthese scales, phenomena responsible for dispersion and transport of atmosphericCO2 include turbulence, the daily evolution of the atmospheric boundary layer,and advection by diurnal thermal circulations (e.g. land-sea breezes, slope flows)and regional winds. At urban element - neighborhood scales, transport and mix-ing of CO2 is dominated by turbulent exchange processes. At city-wide scales,atmospheric mixing is determined by atmospheric boundary layer formation andevolution and advection can result from thermal circulations and regional flow pat-terns.The vertical structure of the atmospheric boundary layer is modified in severalunique ways over urban surfaces. Daytime growth of the mixed layer is driven bythermal heating of the surface and convective circulation until the mixed layer oc-cupies the entire boundary layer. This is termed the urban boundary layer (UBL)(Oke, 1987) (Figure 1.1). Height of the UBL may be higher than the mixed layerover non-urban areas because of stronger bouyant forces driven by mean urbanthermal and moisture properties (e.g. low thermal admittance, reduced moisture).Turbulent mixing by boundary layer scale eddies acts to homogenize atmosphericproperties in terms of temperature, water vapor, wind speed, and pollutant concen-tration. This mixing also acts to spatially average these properties on length-scalesapproaching the physical dimensions of the urban environment. Therefore, mea-surements in the UBL are generally representative of the integrated response of theentire urban environment.The lowest 10% of the UBL is called the surface layer and it can be furtherdivided into the urban canopy layer (UCL), roughness sublayer (RSL), and inertialsublayer (ISL) (Figure 1.1). The lower limit of the ISL can be defined as the heightat which momentum flux becomes constant with height. The existence of an ISLis dependent on horizontal homogeneity at length-scales to support growth of an13Figure 1.1: Schematic of urban atmospheric layers and associated scales ofanalysis during a clear-sky, convective daytime situation in light re-gional winds. Adapted from T.R. Oke.internal boundary layer in equilibrium with the surface. At this height, the atmo-sphere is above the influence of individual roughness elements and represents theblended influence of a larger horizontal surface area. Assumptions of horizontallyhomogeneous flow are considered valid for classic surface layer theory applica-tions and measurements in the ISL are representative of the urban climate local, orneighborhood, scale.Below the ISL is the RSL, where horizontal flow is inhomogeneous and mo-mentum exchange is influenced by individual elements such as buildings, trees,and street canyons. The height of the RSL generally extends upwards 2-5 timesthe height of individual roughness elements and is also influenced by their den-sity and configuration (Oke, 1987). Turbulent kinetic energy (TKE) productionwithin the RSL varies according to height. For example, above roof/tree heightTKE is produced from both local thermal and shear sources, while at mean rooflevel TKE is dominated by local shear production (Christen et al., 2009). As a re-sult, turbulent length scales are largest above roof level and turbulent exchange of14heat and momentum is characterized by intermittent ejections from the roof layeror shear layer below (Christen et al., 2007). At roof level, turbulent length scalesare smaller, exchange is more efficient, and exchange is dominated by downwardsweeps (Christen et al., 2007). Measurements above roof/tree height within theRSL are generally representative of the urban climate micro-scale.In the RSL, several assumptions of surface layer theory begin to break down.For example, conditions are influenced by thermal plumes and wakes from indi-vidual roughness elements resulting in inhomogeneous horizontal flow. Further-more, the assumption of local equilibrium between production and dissipation ofturbulence in Monin-Obukhov Similarity Theory does not hold (Christen et al.,2007). Horizontal spatial inhomogeneities can also lead to dissimilar transport be-tween passive scalars compared to similarity theory-based predictions (Roth andOke 1995, Moriwaki and Kanda 2006).Within the RSL is the urban canopy layer (UCL), the height of which is de-termined by the height of canopy elements (buildings, trees) and their density andconfiguration. The height of the UCL can be defined as the effective buildingheight, at which there is an inflection in the shape of the vertical wind profile(Christen et al., 2009). This is related to the zero-plane displacement length ofthe wind profile. Within the UCL, wind speeds and solar radiation are diminishedand local turbulence production is minimal (Christen et al., 2009). Exchange withthe RSL is dominated by intermittent exchange by sweeps from above and ejec-tions out of the UCL (Christen et al. 2007, Salmond et al. 2005). The UCL alsoincludes indoor building air which is exchanged with outdoor air at rates depend-ing on building structure and porosity. Buildings and vegetation in the UCL alsoact to horizontally decouple atmospheric exchange between streets and interiorcourtyards, alleyways, etc. (Weber and Weber, 2008). The urban subsurface layerconsisting of soil, vegetation, and impervious surface is also included in consider-ation of the UCL. Measurements within the UCL are representative of the urbanclimate micro-scale.151.4 Measuring urban CO2 emissionsGiven that human population and activity are clustered in urban areas, it is logicalto infer that a significant fraction of CO2 emissions are attributable to urban areas.Several sources report that urban areas contribute 75-80% of global anthropogenicCO2 emissions (e.g. Stern 2007), while others argue these estimates are overstatedand that only 30-40% of emissions actually occur within urban boundaries (e.g.Satterthwaite 2008). Estimating urban contribution to global emissions is so un-certain that a report from UN-Habitat states: ‘It is impossible to make definitivestatements about the scale of urban emissions. There is no globally accepted stan-dard for assessing the scope of urban GHG emissions - and even if there was, thevast majority of the world’s urban centres have not conducted an inventory of thistype.’ (Habitat, 2011).Despite this assessment, there has been significant progress in the understand-ing and quantification of urban CO2 emissions both at city-wide and sub-city scales.The following sections are a review of urban CO2 emission and CO2 mixing ratioresearch organized by spatial scale. The intention of this review is to provide anaccount of the various measurement techniques that have been developed and de-scribe features of urban CO2 dynamics as they relate to environmental and anthro-pogenic controls.1.4.1 City-wide approachesThough there is no global standard, the most widely used and accepted method-ology for measuring emissions within local government boundaries are inventory-based approaches (e.g. Kennedy et al. 2010). These approaches are not directmeasurements, but use energy and fuel consumption statistics to compile emis-sions totals for local municipal regions. The accuracy of inventory-based estimatesis dependent on the quality and quantity of available data. In many cities, de-tailed data about population and energy consumption are not available, especiallyif a large proportion of residents live in informal or illegal settlements (Bulkeley,2013). Elsewhere, data may be collected by utilities but not be publicly available,or may be only collected on scales larger than city boundaries (e.g. regional ornational), and only on course timescales (monthly, annually). Furthermore, in-16ventories do not account for direct biogenic influences on emissions such as CO2uptake by vegetation, or CO2 emissions from soils.Emissions inventories can be compiled for a number of different GHGs, mosttypically CO2, methane (CH4), and nitrous oxide (N2O). The total of all GHGemissions can be expressed as ‘CO2-equivalent’ (CO2e). This calculation is de-pendent on the specific radiative forcing factor for each individual non-CO2 GHG.These emissions are expressed in terms of the amount of CO2 that would be re-quired to produce the same radiative forcing as the non-CO2 GHG (Gohar andShine, 2007). By volume, CO2 is the largest GHG and accounts for 77% of allGHG emissions globally (EPA, 2014), but CO2-equivalent emissions are not thesame as CO2 emissions.Comparison of inventory-based emissions totals between cities has potential tobe misleading. Estimates of per capita emissions can vary widely depending on thescale at which system boundaries are set, and whether consumption or production-based approaches are used. For example, a single person in Toronto, ON couldbe responsible for 6.42 tons of CO2-equivalent (tCO2e) GHG emissions per year atthe household level living in a dense, central neighborhood; 9.5 tCO2e as a Torontocity resident; 11.6 tCO2e as an inhabitant of the greater Toronto metropolitan re-gion; 16.0 tCO2e as a resident of Ontario; and 22.65 tCO2e as an average citi-zen of Canada (Hoornweg et al., 2011). At the neighborhood scale, these figuresare production-based for transportation and consumption-based for household en-ergy. City and metropolitan emissions are production-based for fossil fuel combus-tion and industrial processes, and consumption-based for electricity generation andwaste treatment. Provincial and national estimates are entirely production-based.Despite the complexities involved, inventory-based emission totals have beenconducted for many cities (e.g. reviews in Kennedy et al. 2009 and Hoornweget al. 2011). In general, it appears that urban residents have lower per capita di-rect emissions compared to national-level values. For example, Tokyo residentshave an average of 4.89 tCO2e GHG emissions per year, whereas the national av-erage for Japan is 10.76 tCO2e (Hoornweg et al., 2011). Urban residents are oftenable to take advantage of transportation efficiencies such as mass-transit optionsand shorter commutes. Building energy demand for heating in cold climates isalso reduced in urban areas due to generally smaller living spaces, shared-wall17heating efficiencies in dense residential apartment complexes, and warmer temper-atures from the urban heat island effect. An exception to this rule may be in Chinahowever, where per capita urban emissions tend to be higher than national values(Hoornweg et al., 2011). This reflects both a concentration of industrial activity andthe relative affluence of urban areas that is manifested in higher fossil fuel-basedenergy production and consumption.Also evident from inventory-based estimates is the large variability betweencities. Average annual per capita emissions are greater than 15 tCO2e for affluentcities in Australia, Canada, Germany, and the United States but are less than 1tCO2e for cities in countries such as Nepal, India, and Bangladesh (Hoornweget al., 2011).Researchers have developed other top-down methods to directly measure city-wide CO2 emissions. Aircraft-based measurements of CO2 concentrations in thedownwind urban plume were used to estimate total urban emissions for Indianapo-lis, IN (Mays et al., 2009). This study found variable emissions with an averageof 19.2 ± 15.4 µmol CO2 m−2 s−1. In London, a similar aircraft-based mea-surement campaign was used to estimate net city-wide CO2 emissions and foundagreement with regional inventories and in situ neighborhood-scale eddy covari-ance (EC) flux measurements (Font et al., 2013). Space-based measurements ofchanges in column-averaged CO2 molar fraction have also been attributed to sur-face emissions in Los Angeles and Mumbai Kort et al. (2012).Measurements of CO2 mixing ratio within cities have been used for severaldecades to characterize spatial distribution, atmospheric transport, and variationof urban CO2. In 1966, CO2 measurements in Cincinnati, New Orleans, and St.Louis, USA found daily variations up to 90 ppmv, 60 ppmv, and 15 ppmv, re-spectively (Clarke and Faoro, 1966). In the suburbs of Sendai, Japan, Tanakaet al. (1985) measured large diurnal variability in CO2 mixing ratio at 30 m aboveground during summer attributable to local CO2 sources and vegetation sinks. Inthe UK, CO2 measurements along a 15 km urban-rural transect in Notthinghamfound CO2 concentrations in urban areas higher than rural surroundings in win-ter, but not during summer (Berry and Colls, 1990). In Phoenix, AZ, an urban CO2dome of elevated canopy-level concentrations was linked to diurnal commuter traf-fic cycles (Idso et al. 1998,Idso et al. 2001). Also in Phoenix, Wentz et al. (2002)18observed significant differences between CO2 concentrations measured above veg-etated golf courses and motorized freeways. In Vancouver, BC, diurnal patterns ofCO2 concentrations were modeled using an urban boundary-layer transport modelin combination with an emissions inventory from an upwind area Reid and Steyn1997.Other studies at city-wide scales have used biogeochemical measurements ofcarbon isotopes to distinguish between atmospheric CO2 originating from fossilfuel or renewable sources. Isotopic measurements of CO2 in Salt Lake City found60% - 80% of winter atmospheric CO2 to be from natural gas combustion Patakiet al. (2007). In Essen, Germany, Henninger and Kuttler (2010) used vehicle tran-sect measurements of pollutants such as CO, N2O, NO, and O3 to estimate theproportion of CO2 contributions from traffic and vegetation.From these studies, it is possible to generalize certain key features of urbanCO2 environments. CO2 mixing ratios are higher and more variable in cities rel-ative to non-urban areas. Per area direct emissions are often higher in cities com-pared to non-urban areas, but per capita direct emissions are usually lower. Thereis a wide range in the magnitude of direct urban emissions between cities and real-time emissions monitoring at city-scales remains difficult. Generally, inventory-based approaches lack comprehensive source and sink information concerning bio-genic components of the urban carbon cycle. Measurement-based approaches in-herently include the net effects of all source and sink processes, but lack informa-tion about the magnitude of specific processes. A feature of city-wide studies isthat they lack spatial detail at sub-city, neighborhood scales.1.4.2 Neighborhood-scale net emissionsWithin cities, there is often large variation in emissions at neighborhood scales.For example, in Toronto, Canada analysis of annual per capita emissions totals bycensus tract reveal emissions ranging from 1.31 tCO2e in a high-density residentialneighborhood up to 13.02 tCO2e in an affluent low-density suburb consisting oflarge single-family homes (VandeWeghe and Kennedy, 2007).Direct measurements of the integrated net ecosystem flux at the neighborhood-scale can be obtained from eddy-covariance (EC) measurements. EC is a top-down19method that continuously measures the net CO2 exchange of entire ecosystems andhas been widely used to measure long-term CO2 exchanges in forest, grassland,and tundra environments (e.g. Baldocchi 2003). The EC method provides contin-uous measurements of net CO2 exchange (FC) at hourly or half-hourly timescales,usually at the local, neighborhood scale (102 - 104 m).To date, over 30 measurement campaigns of urban CO2 flux have been carriedout since the first reported study in Chicago in the summer of 1995 (Grimmondet al., 2002) and a comprehensive review was recently compiled by Velasco andRoth (2010). Most measurements have been conducted in mid-latitude cities inNorth America and Europe and have covered a range of urban neighborhoods fromheavily vegetated, low built-density suburbs (e.g. Baltimore in Crawford et al.2011) to densely built and populated urban centers with heavy traffic loads andsparse vegetation (e.g. Marseille in Grimmond et al. 2004). Although urban veg-etation can act to reduce local net emissions during the summer growing seasonthrough photosynthetic uptake, virtually all measurements have found urban sur-faces to be a net source of CO2. By convention, positive fluxes denote net transferupwards from the surface to the atmosphere, and negative fluxes are downwardsfrom the atmosphere to the surface (i.e. uptake or sequestration).Diurnal patterns of CO2 flux indicate vehicular and household emissions aredominant CO2 sources. Hourly fluxes often show high correlations with trafficcounts (e.g. Edinburgh in Nemitz et al. 2002, Mexico City in Velasco et al. 2005)and observations from several cities showmorning and afternoon flux peaks associ-ated with rush hour traffic (e.g. Basel in Vogt et al. 2006 and Melbourne-Preston inCoutts et al. 2007). Local emissions from fossil fuel combustion for space heating,cooling, and cooking are also large contributors to urban CO2 fluxes, especially inwinter. In Firenze, measured CO2 fluxes were negatively correlated with air tem-peratures (Matese et al., 2009) and in Tokyo home heating contributed to 62% oftotal winter emissions (Moriwaki and Kanda, 2004).Temporal variability of measured fluxes is linked to both cultural practices andphysical processes. For example, results from Lodz, Poland (Pawlak et al., 2011)show a weekly flux minimum on Sundays, the Christian Sabbath, while measure-ments from Cairo (Burri et al., 2009) show minimum fluxes on Friday, the Muslimday of rest. Additional temporal variability was observed in Marseille, where in-20termittent CO2 venting from urban street canyons was linked to nocturnal storageheat releases from the urban fabric (Salmond et al., 2005).Spatial heterogeneity at measurement scales is also a well-documented featureof urban CO2 fluxes. For example, in Essen, Germany annual mean net fluxesmeasured from a vegetated urban park source area (0.8 µmol m−2 s−1) are lowerthan fluxes from an adjacent urban surface (9.3 µmol m−2 s−1) (Kordowski andKuttler, 2010). In Vancouver, net fluxes observed from wind directions with a busyintersection were nearly double those of other wind directions (Walsh, 2005).This first generation of EC CO2 flux measurements in urban areas has estab-lished the ability of the method to quantify net neighborhood-scale CO2 exchangefrom urban ecosystems. Results have provided realistic baseline local emission to-tals for urban neighborhoods and established stationary building sources and mo-bile vehicular sources as dominant terms in the urban CO2 budget. These studieshave also observed CO2 fluxes to be highly variable in time and space, but fewstudies have delved deeper into the underlying spatial and temporal patterns ofindividual CO2 emissions, uptake, and turbulent transport processes.1.5 Research objectivesThe broad objectives of this research are to advance measurement and analysistechniques to map, partition, and quantify CO2 emissions/uptake processes andatmospheric transport in the urban environment. This is accomplished primarilythrough innovative measurements and data analysis representative of three distincturban climate scales in Vancouver, BC, Canada. Together, three distinct datasetsare used provide a comprehensive picture of urban atmospheric CO2 from the pointof emissions (or uptake) at the surface, storage and transport within the urbancanopy layer, and vertical mixing in the urban boundary layer.The dissertation is organized into three Results sections, each with specificresearch objectives focused on a particular urban climate scale of measurementand analysis:i) Chapter 2 - Neighborhood-scale. This scale represents the focus of the workand the primary dataset is a four-year (May 2008 - April 2012) EC flux datasetover the residential ‘Sunset’ neighborhood in Vancouver. Specific objectives of21this section are to a) quantify source area variability and location bias in termsof EC observations at the measurement site, b) develop methods to isolate andmodel individual CO2 source/sink processes in terms of environmental and landcover controls, and c) calculate spatially unbiased neighborhood-scale net CO2emissions.ii)Chapter 3 - Micro-scale. Spatially heterogeneous distribution of CO2 sourcesand sinks at the micro-scale is investigated in this section. The primary datasetis CO2 mixing ratios measured from a mobile, vehicle-mounted platform at 2 mheight in the urban canopy layer (UCL). Research objectives in this section are toobserve micro-scale horizontal variability and rate of change of CO2 mixing ratiosin the UCL. Physical explanations of observed spatial patterns are investigated andtheir impact on local-scale EC measurements is examined through consideration ofthe EC storage flux term. Specifically, storage in the UCL air volume is calculatedfrom UCL mixing ratios and used to test whether storage estimated from a singlemixing ratio measurement at EC height is representative of storage in the entire airvolume below measurement height.iii) Chapter 4 - City-scale. Observations of CO2 mixing ratios in the urbanboundary layer (UBL), up to a height of 400 m, were measured using a tethered-balloon measurement system. Specific research objectives of this section are toobserve the diurnal evolution of CO2 mixing ratios in the UBL, infer CO2 fluxesrepresentative of the city-scale using a boundary layer budget method, and compareresults to neighborhood-scale EC measurements. Observations from a network offive fixed CO2 mixing ratio sensors located in the greater Vancouver region duringJuly-August 2008 are also presented in this section. The objective of these mea-surements is to observe the degree of spatial variability of near-surface CO2 mixingratios found in the greater Vancouver region during growing season conditions.In the Conclusions section (Chapter 5), specific findings from Chapters 2-4are summarized and this work is situated within the broader field of urban CO2emissions research. The contributions and significance of results are discussedas well as the strengths and limitations of the overall research approach. Finally,directions for future research based on the current state of knowledge are discussed.22Chapter 2Local-scale fluxes2.1 IntroductionGlobally, urban areas account for over 52% of world population (UN, 2011) and30-40% of total anthropogenic carbon dioxide (CO2) emissions (Satterthwaite,2008). Cities have become a focus for CO2 emission reduction efforts and localgovernments are increasingly taking action to monitor, report, and reduce CO2emissions from metropolitan areas (Lutsey and Sperling, 2008). One approach toreduce emissions from cities is through energy efficient urban design strategies andthe neighborhood is an important scale at which planning decisions about trans-portation, land-use, and infrastructure are made (Kellett et al., 2012). Emissionsestimates at urban and intra-urban scales of decision-making are needed to informthe planning process. Downscaling of national, state, or regional fuel consumptionstatistics to city and neighborhood scales can be estimated through spatial proxiessuch as land-use, population, or night-lights (e.g. Raupach et al. 2010). How-ever, these proxies are unable to resolve the diversity of intra-urban transportation,building stock, vegetation, and industrial characteristics and neglect biogenic CO2processes from urban soils and vegetation. High-resolution data describing boththe timing and geographic origin of emissions and uptake by vegetation are neededto identify areas with emission reduction potentials, establish emission baselines,validate emissions inventories, and monitor progress towards reduction targets.One method to directly measure intra-urban emissions and uptake of CO2 at23neighborhood scales is tower-based eddy covariance (EC). EC has been establishedas a robust method of measuring CO2 fluxes in cities at neighborhood scales andthere are currently at least 30 urban EC CO2 sites in operation throughout theworld (Velasco and Roth, 2010). EC measures the net mass flux of CO2 betweenan urban ecosystem and the atmosphere at temporal resolutions of usually 30-60minutes and at spatial resolutions of 0.25-10 km2. Previous studies have demon-strated that the net CO2 flux in most cities is dominated by emissions from fossilfuel combustion from traffic and space heating sources (e.g. Nemitz et al. 2002;Grimmond et al. 2002; Moriwaki and Kanda 2004; Vogt et al. 2006). Biogenicemissions (respiration from vegetation, soils, and humans) and uptake from pho-tosynthesis can be relevant in low-density urban ecosystems and urban parks (e.g.Crawford et al. 2011; Peters and McFadden 2012). EC is advantageous for its abil-ity to directly measure the net CO2 exchange from all local emissions and uptakeprocesses at high temporal resolution, however there are two related features ofurban study areas that make interpretation and analysis of CO2 flux measurementsdifficult. These features are the spatial complexity of source and sink configura-tions in urban areas and accurate attribution of individual source and sink processesto the measured net signal.2.1.1 Spatial complexity of sources and sinksFirst is the question of how to properly account for spatial variability of emissionsand uptake at the scale of EC measurements. The EC method uses instrumentslocated at a single point in the inertial sublayer to make measurements that are rep-resentative of the net surface-atmosphere gas exchanges from an upwind surfacearea. This surface area influencing the flux measurement is called the turbulentflux source area and it is constantly changing as a function of wind direction, at-mospheric stability, and lateral dispersion qualities of the flow (Schmid, 1994).Ideally, EC measurements are taken over extensive, homogeneous areas withuniformly distributed CO2 sources and sinks. The spatial scale of source/sinkvariability is assumed to be much smaller than the scale of the source area andso EC sensors will always measure the same type of surface containing identi-cal source/sink distributions no matter the size or orientation of the flux source24area. This is the usual EC assumption for measurements over ecosystems such asforests, crops, or grasslands (Aubinet et al., 2012) and implies that as long as theEC sensors are above the roughness sublayer and measure a sufficiently mixed at-mosphere, the particular sensor location is not important because the vertical massflux is horizontally homogeneous.In urban ecosystems, however, non-uniform distribution of surface CO2 sourcesand sinks is the rule and urban surfaces are often heterogeneous at spatial scalesapproaching or exceeding those of the turbulent flux source areas. As a result,flux source areas are composed of constantly varying source/sink compositionsand measured fluxes will be biased according to the vertical and horizontal place-ment of the EC instruments (Schmid and Lloyd, 1999). This complexity makestraditional measurement objectives of accumulating and comparing CO2 exchangetotals over longer time-scales difficult. How can measurements with potentiallyquite different source area compositions be compared and aggregated? What arethe ‘true’, spatially unbiased CO2 fluxes from a specific urban ecosystem?Issues of spatial variability and scale are fundamental to all urban micro-meteorologicalstudies and researchers using EC CO2 flux measurements have used several ap-proaches to deal with this problem. The most common method has been to stratifyflux measurements by wind direction sectors. For example, in Essen, Germany, Ko-rdowski and Kuttler (2010) use a clear distinction between CO2 fluxes measuredwhen winds are from a wind direction sector with a large urban park versus whenwind directions are from a residential area. In Vancouver, BC, Canada, Christenet al. (2011) address spatial variability by averaging ensemble diurnal courses ofCO2 fluxes measured from four equally sized wind direction quadrants with differ-ent mean CO2 fluxes. Another approach was developed in London, UK, to locatemicro-scale CO2 sources by wind direction through analysis of short-lived plumesof high CO2 concentrations (Kotthaus and Grimmond, 2012). Several studies havealso used source area modeling to calculate the long-term turbulent flux sourcearea for a particular site (e.g. in Mexico City, Mexico by Velasco et al. (2005), inŁo´dz´, Poland by Pawlak et al. (2011), in Helsinki, Finland by Ja¨rvi et al. (2012),in Vancouver, Canada be Christen et al. (2011)). A meta-analysis of 14 urban CO2flux sites demonstrates the potential for ‘natural’ land cover fraction to predict netemissions (Nordbo et al., 2012), but few detailed analyses of how source area com-25position affects measurement variability have been conducted. An exception iswork by Hiller et al. (2011) in Minneapolis, MN, USA which used detailed sourcearea analysis to quantify the influence of traffic emissions on CO2 flux measure-ments over an urban turf-grass field.2.1.2 Source and sink attributionThe second challenge is determining the origin and composition of the net CO2flux in terms of specific source and sink processes. So far, analysis of EC datasetshas been limited to the measured net flux and few methods have been developedto investigate the behavior of individual processes and how their relative composi-tions vary. Many studies have found high correlations between net CO2 fluxes andindependent models, traffic counts, and down-scaled emissions inventories (e.g.Moriwaki and Kanda 2004; Velasco et al. 2005; Matese et al. 2009; Christen et al.2011). Other studies used statistical relations between environmental variablessuch as soil and air temperature, solar radiation, and CO2 flux to infer informationabout specific processes (e.g. Crawford et al. 2011; Bergeron and Strachan 2011).Peters and McFadden (2012) and Velasco et al. (2013) use micro-scale measure-ments of soil and vegetation respiration and CO2 sequestration in conjunction withlocal-scale EC fluxes to calculate the impact of urban vegetation in Minneapolisand Singapore, respectively, but no study has yet attempted a comprehensive de-construction of the net flux measurements into individual component processes.This type of source attribution would be useful for testing specific emissions mod-els (e.g. building energy or transportation models) and would assist the validationof emissions inventories.2.1.3 Study objectivesThe goal of the present study is to address the related issues of spatial variabilityand source/sink attribution using a long-term CO2 flux EC dataset combined withsurface spatial data and source area modeling in a primarily residential area ofVancouver, BC, Canada. Rather than view CO2 source/sink variability and EClocation bias as an obstacle, this paper regards these characteristics of urban CO2flux measurements as advantages that can be leveraged to gain more knowledge26about the spatial and temporal patterns of individual processes contributing to netCO2 flux. The specific objectives are to i) quantify source area variability andlocation bias at the Vancouver measurement site, ii) develop methods to isolateand model individual CO2 source/sink processes in terms of environmental andland cover controls, and iii) calculate spatially unbiased neighborhood-scale netCO2 emissions. These methods will expand existing spatial analysis frameworksused in urban EC studies and potentially be applicable to a range of neighborhoodtypes and mass fluxes of pollutants and greenhouse gases.2.2 Methods2.2.1 Eddy-covariance measurementsNet CO2 fluxes were measured using an EC system mounted at a height of 28.8m a.g.l. on an open, triangular, lattice tower located in the ‘Sunset’ neighborhoodof Vancouver (123.0784◦W, 49.2261◦N, WGS-84). The tower is located on the SEcorner of the grounds of a BC Hydro power substation which extends 130 m tothe north and 100 m to the west. The substation grounds and base of the tower arerecessed by 4 m from the surrounding terrain resulting in an effective measurementheight of z =24.8 m. Outside of the substation, the neighborhood is classified asLCZ-6 ‘open-set lowrise’ (Stewart and Oke, 2012) and has been identified as arepresentative, primarily residential area. The tower has been used in a numberof previous studies on turbulence, trace-gas exchange, and the urban energy andwater balance (e.g. Cleugh and Oke 1986; Schmid et al. 1991; Grimmond and Oke1991; Roth and Oke 1995; Reid and Steyn 1997; Walsh 2005; Christen et al. 2011)(Figure 2.1). The neighborhood was originally selected as an EC location becauseof its homogeneity in terms of physical morphometry (surface roughness, buildingand vegetation density, distribution, and height), thermal and moisture properties,and relatively flat topography.The EC system consisted of a sonic anemometer (CSAT 3d, Campbell Sci-entific, Logan, UT, USA) and an open-path infrared-gas analyzer (Li-7500, LicorInc., Lincoln, NE, USA) with 0.40 m horizontal separation between sensors. Addi-tional micrometeorological measurements used in this study include incoming so-2780%50%100 m NSunset Neighborhood Study AreaLawn BuildingTree ImperviousMajor RoadSecondary RoadEC TowerTurbulent !ux source area49th Ave.41st Ave57th Ave.Fraser St.Knight St.Victoria Dr.545165154526015453551495240494290493340Memorial South          ParkGordon ParkTecumseh Park70%60%Figure 2.1: Map of the Sunset study neighborhood land cover centered onthe 28.8 m EC flux tower. The spatial extent of the map is 1900 × 1900m and UTM grid coordinates are given for the center and map edges(NAD83, UTM zone 10). The longterm turbulent flux source area (May1 2008 - Apr 30 2011) is displayed as white contour lines showing thesource area weighting significance-level (see text for details). Major andSecondary roads (see text for definitions) are also labeled on the map.28lar radiation measured at the tower (K↓, z =22 m, CNR-1, Kipp and Zonen, Delft,Netherlands), canopy air temperature (Tair, z =2 m, HMP T/RH sensor, Vaisala,Finland) measured in the backyard of a typical residence within 300 m of the tower,and soil temperatures (TS, -5 cm, T-type thermocouple, Omega Engineering, Inc.,Stamford, CT, USA) measured in the yards of 4 selected residences within 1 km ofthe tower.Three-dimensional wind velocities and CO2 concentrations were recorded at20 Hz. Individual datapoints are withheld from further analysis during precipita-tion events, when there is flow distortion from the sonic mounting block, or whenflagged after being passed through quality control filters (spike detection, logicalmaximum/minimum data thresholds) (Crawford et al., 2010). The 20 Hz data arethen block-averaged in 30-minute periods and rotated into a streamline coordinatesystem aligned with the mean wind direction. The vertical CO2 flux (FC with unitsof µmol m−2 s−1) is then calculated as the covariance between vertical wind ve-locity (w￿) and CO2 concentration (c￿) fluctuations from the block-averaged, 2-drotated, 30-minute mean of w￿c￿.FC values are corrected for air volume and density fluctuations (Webb et al.,1980) and horizontal sensor separation (Moore, 1986). Technically, 30-minutecalculations of FC should also include a storage term to account for changes in CO2concentrations within the measurement volume during the flux averaging period.However, analysis at this site has shown that the storage term is usually very smallcompared to FC (<2%) and using a single-point concentration measurement tocalculate storage change may introduce uncertainties larger than the storage termitself (Crawford and Christen, 2012). Therefore fluxes are not corrected for storagechange in this analysis and the storage flux is considered in detail in Chapter 3.For this work, three years of continuous EC observations (May 1, 2008 - Apr30, 2011) at 30-minute resolution were used for analysis and model developmentand a fourth year (May 1 2011 - Apr 30, 2012) was withheld and used to test themodel. During the entire four-year period, the EC system was in operation for70.0% of the time (6.8% failure due to instrument maintenance and quality con-trol, 23.2% due to interference from rain and snow). Average daily air temperaturemeasured at Vancouver International Airport (YVR) during the 4-year period was10.0◦C, which was 0.4 K cooler than the 30-year climatological mean from 1982-29Table 2.1: Regional values of daily mean air temperatures and total precip-itation are measured at Vancouver International Airport climate station(WMO ID:71892) (Environment Canada, 2013).Year Daily mean Total annual Eddy-covarianceair temperature (◦C) precipitation (mm) data availability (%)5/2008-4/2009 9.7 1051 71.7%5/2009-4/2010 11.2 1203 68.7%5/2010-4/2011 10.2 1249 67.1%5/2011-4/2012 10.0 1000 72.9%1982-2012 10.4 1188 -2012. Mean annual total precipitation from 2008-2012 at YVR was 1146 mmcompared to the 1982-2012 mean of 1188 mm. Table 2.1 shows the annual tem-perature and precipitation conditions for each individual year during the study anddata availability of EC observations.Wind directions measured at the tower are subject to seasonal and diurnal pat-terns with winter (DJF) flows primarily from the NE (peak frequency for 10◦ bins is2.2% from 40-50◦) and summer (JJA) winds from the SE (peak frequency is 2.1%from 150-160◦) (Figure 2.2a and 2.2b). The tower location is also subject to di-urnal thermally-driven land-sea circulations, most notably during summer months(Steyn and Faulkner, 1986). The implication for the EC technique is that measure-ments and flux source areas will be spatially biased according to this wind directiondistribution. Though there are virtually no observations when winds are from theNorth (this is a rare wind direction and is filtered due to sensor head flow distor-tion), this is not expected to limit analysis because the distribution of land-coverand CO2 sources and sinks in this wind direction is similar to other wind directionsin the study area. In other words, there are no unusual CO2 emissions processes,for example industrial sources, present in the Northern wind direction sector thatare not represented in other wind direction sectors.306%4%2%DJF MAM JJA SON Night (1800-0600) Day (0600-1800)-20 - 0 0 - 20 20 - 40 40 - 60 60 - 80 80 - 100a) Wind direction seasonal distribution (%) b) Wind direction diurnal distribution (%)  c) Distribution of  30-minute FC magnitude (μmol m-2 s-1) by wind direction d) Mean FC  by wind direction (μmol m-2 s-1)e) Mean source area plan-area landcover (%) by wind direction f ) Mean source area location bias (Δ) by wind directionλTλLλMλSλRλBλIWeekend (S-S) Weekday (M-F)6%4%2%oooooooo6%4%2%oo5040302010321ooooooooooooooooooooooooooooooooooooooooFigure 2.2: a) Wind direction frequency (%) sorted by season, b) Wind direc-tion frequency (%) sorted by time of day, c) Wind direction frequencysorted by flux magnitude (µmol m−2 s−1), d) Mean FC (µmol m−2 s−1)by wind direction sorted by weekend or weekday, e) Mean source areaplan-area land cover fraction, and f) Mean source area location bias.Data are from May 1 2008 - April 20, 2011 when fluxes are available(70% availability) and are binned in 10◦ wind direction segments where0◦ is geographic North. All roses are cumulative (i.e. ‘stacked’) exceptfor d).31Table 2.2: Values derived from remotely sensed products and source areamodeling. The λI land cover element includes road classifications, ‘Do-main’ refers to the entire 1900 × 1900 m geospatial data coverage, and‘Source area-weighted’ refers to the long-term turbulent flux source areafrom May 1, 2008 - Apr 30, 2011.Land cover element Domain average (%) Source area-weighted (%)Building (λB) 29 24.3Tree (λT ) 11 11.8Lawn (λL) 24 20.3Impervious (λI) 35 43.6Major road (λM) 4.3 6.2Secondary road (λS) 1.2 2.6Residential road (λR) 10.0 7.2Building volume (m3 m−2) 1.6 Geospatial dataIn addition to micrometeorological measurements, several remotely sensed spatialdatasets were used in this study (Figure 2.1). In previous studies, a multispectralQuickbird satellite image taken at 2.4 m resolution was merged with airborne Li-DAR data at 1 m resolution to extract surface plan-area land cover (λ ) for a 1900×1900 m area centered on the tower at 1 m resolution (Goodwin et al. 2009; Tookeet al. 2009). A 1900× 1900 m area was chosen because this encompasses virtuallyall of the long-term turbulent flux source area and is the spatial extent of the LiDARdataset. λ is defined as the plan-area land cover of a surface category within somelarger area and is expressed as a fraction in percent. The categories used in thisstudy are: building (λB), tree (λT ), lawn (λL), and impervious ground (λI) and theysum to 100% (Table 2.2). Using this approach, areas underneath overhanging treecanopies are classified as ‘tree’, although they may be impervious ground, lawn, orbuilding in reality.The largest plan-area fraction by category in the 1900 × 1900 m study do-main is λI (35%), which is composed of sidewalks, parking lots, driveways, alley-ways, and roads. Further analysis was conducted to identify and isolate roads withtraffic as separate λ categories. Independent road line GIS layers available from32the City of Vancouver were clipped to the study area and roads are classified as‘Major’ (λM), ‘Secondary’ (λS), or ‘Residential’ (λR) based on mean road widthdetermined from aerial photographs and traffic counts available from the City ofVancouver (City of Vancouver, 2012). Buffers corresponding to mean road widthwere applied to the road line layers and rasterized at the same resolution as theplan-area datasets described earlier. ‘Major’ roads have been attributed a meanwidth of 20 m and correspond to daily traffic loads of >20,000 cars per weekday.The roads in this category are Knight Street, 41st Street, Fraser Street, and VictoriaDrive and compose 4.3% of the study area (395.2 m2). ‘Secondary’ roads havebeen attributed a mean width of 12 m and their daily total traffic loads are on theorder of 10,000-20,000 cars per weekday. Secondary roads are the least commontype (1.2%, 212.6 m2) and consist of 49th Avenue and 57th Avenue. ‘Residential’sidestreets are 8 m wide on average, have the greatest cumulative plan-area cover-age in the study area (602.2 m2, 10.0%), are distributed relatively uniformly aboutthe tower, and have typical daily weekday traffic loads of less than 1000 cars perday.The second largest λ coverage by category is λB (29%) and a detailed build-ing survey classified 91% of buildings as single, detached, residential structures(van der Laan, 2011). The remaining structures include commercial office and re-tail buildings (clustered at major intersections and along commercial strips) andseveral schools and churches. For a 500 m radius around the tower, mean build-ing roof height is 6.5 m and the built density is 12.8 buildings ha−1. In addition,building volumes are extracted by multiplying the λB layer pixels by LiDAR heightabove terrain measurements at 1 m resolution. Mean building volume for 1 m cellsclassified as building is 5.44 m3 m−2 and for the entire 1900 × 1900 m study do-main is 1.57 m3 m−2. Approximately 90% of buildings in the Sunset neighborhoodhave natural gas heating systems that generate local CO2 emissions.Vegetation in the study area consists of a surface layer of irrigated and regu-larly clipped lawn and park grasses (λL=24%), an understory layer of ornamentalshrubs and small trees, and a canopy layer of mature deciduous and coniferoustrees (λT=11%). There are also several public urban parks with expansive grassfields and mature trees within 1000 m radius of the tower (Memorial South Park,Tecumseh Park, and Gordon Park, see Figure 1) that cover 7% of the study area and33contain 20% of the total vegetation cover. For a 400 m radius around the tower, alltrees were manually counted and identified as either coniferous or deciduous fromaerial photography. In total, there were 742 deciduous trees (77%, 14.76 stemsha−1) and 223 coniferous trees (23%, 4.44 stems ha−1) and overall tree densitywas calculated as 17.1 stems ha−1 (Liss et al., 2010). Average height for all treesdetermined from LiDAR data is 9.0 m (n =965), with coniferous trees on aver-age slightly taller (height=11.1 m, n =223) than deciduous trees (height=8.4 m,n =742).Leaf area index (LAI) was calculated for the study area for both canopy-leveltrees (LAIT ) and surface-level lawns (LAIL) and is defined as the one-sided leafarea per unit ground area (m2 m−2). For trees, LAIT is modeled using allometricrelations between tree height and tree crown diameter that are urban-specific fordeciduous trees (Nowak, 1996) and based on forest ecosystems for coniferous trees(Teske and Thistle, 2004). Tree height and crown diameter are extracted from theLiDAR and plan-area land cover classification datasets described earlier. LAIL isdetermined using grass samples collected from representative plots within 1000 mof the tower during summer 2008 (Liss et al., 2010). Mean LAIT for the entirestudy area is 0.39 m2 m−2 and mean LAIL for the entire study area is 1.43 LAIL.Spatially distributed population density data for Sunset was generated by down-scaling census dissemination area data from Statistics Canada (Statistics Canada,2011) to individual building volumes derived from LiDAR data taking into accountbuilding type and use (van der Laan, 2011). For a 1000 m radius around the tower,average density was calculated as 63.1 inhabitants ha−1. This figure is based oncensus data which provides nighttime residential population and does not neces-sarily reflect daily or weekly patterns of movements into, through, and out of thestudy area.2.2.3 Turbulent flux source area modelsThe micrometeorological and spatial datasets are linked through numerical mod-eling of flux source areas with a 2-dimensional crosswind dispersion and gradientdiffusion model (Kormann and Meixner, 2001). This particular model was selectedbecause of its ability to handle a range of atmospheric stability conditions, rela-34tively straightforward implementation, fast calculation, and use in previous studies(e.g. Chen et al. 2009; Christen et al. 2011). As dynamic input, the model uses30-minute mean wind direction (measured at z=24.8 m on the flux tower), surfaceroughness and displacement lengths (z0 and zd , determined by wind direction sec-tor from surface morphometry as described below), lateral dispersion (measuredstandard deviation of the crosswind velocity, σv, for each 30-minute period), andatmospheric stability (Obukhov Length, L, calculated from measurements from theEC system for each 30-minute period).The displacement height (zd) and roughness length (z0) used as inputs are se-lected based on the mean wind direction sector for each 30-minute period. For 10◦sectors centered on the flux tower out to a radius of 400 m, mean building and treeheights (zB and zT ) and plan-area land cover fraction (λB and λT ) were determinedfrom the geospatial datasets and used in empirical relations derived from wind tun-nel experiments (Counehan, 1971). Estimated mean zd is 2.6 m (max zd=3.3 m,min zd=1.9 m) and mean z0 is 1.7 m (max z0=2.2 m, min z0=1.1 m). Thoughit is expected that the scalar CO2 fluxes are highly variable in space, the studyarea shows a reasonably homogeneous surface structure in terms of momentumexchange. This makes it possible to use a relatively simple analytical formationfor the turbulent source area model at this site, though this approach may not bepossible at sites with highly variable and/or isolated roughness elements.The model was run for the entire study period (May 1, 2008 - Apr 30, 2012)at 30-minute resolution and model output is a 2 × 2 m resolution gridded surfacecovering the same 1900 × 1900 m domain as the geospatial data. The sourcearea plan-area coverage (λ f ) are determined by multiplying source area weighting(φ ) by the 2 × 2 m λ for each plan-area category (i) for each grid cell (x,y) andsumming over the entire 1900 × 1900 m model domain:λi f =1900m∑x=0m1900m∑y=0mφ(x,y)λi(x,y) (2.1)This process was also used to calculate source area-specific LAI for both treesand grass. Normally a small portion of the source area falls outside of the 1900× 1900 m domain. When this occurs, the λ for each land cover category for thefraction outside the study area is set to the mean of the entire 1900 × 1900 m area35(Christen et al., 2011). For the long-term integrated flux source area, calculated byaveraging the individual 30-minute source areas, the percentage outside the modeldomain was 12%.2.2.4 Linking emissions processes with land coverThe net CO2 flux from an urban environment is the result of a complex set of bio-genic, anthropogenic, and micrometeorological processes operating across a rangeof temporal and spatial scales (Vogt et al. 2006; Velasco and Roth 2010). Within anurban neighborhood, there are significant stores of carbon in vegetation, soils, andbuilt materials as well as horizontal flows of carbon through the neighborhood inthe form of fuels, food, and waste (Kellett et al., 2012). On multi-year to decadaltimescales, these long-term carbon stores may fluctuate as neighborhood physicalmorphology evolves in terms of vegetation growth and removal and changes tobuilding character and density. On timescales ranging from 1 hour to several years,however, these long-term stores remain roughly constant and local CO2 fluxes tothe atmosphere result from the combustion of fossil fuels (natural gas, gasoline, oil)and biofuels (wood, bioethanol) that have been transported into the study area aswell as from respiration from vegetation, animals, humans, and soils. In addition,vegetation acts to locally remove atmospheric CO2 through photosynthesis.If these three processes (combustion, respiration, photosynthesis) are consid-ered along with their origin at the surface, the local CO2 budget for a residentialarea without industrial sources can be expressed as:FC = EV +EB+RH +(RL+PL)+(RT +PT ) (2.2)where FC is the total measured net CO2 flux, EV is emissions from fossil fuel com-bustion by motor vehicles, EB is emissions from fossil fuel combustion from withinbuildings, RH is human and animal respiration, RL is total respired CO2 enteringthe atmosphere from lawns, RT is respired CO2 from above-ground tree biomass,PL is CO2 uptake by lawn grass photosynthesis, and PT is uptake by tree photosyn-thesis. RL is used here to denote the total emissions from grass lawns and fieldswhich includes heterotrophic respiration from soil, above-ground autotrophic res-piration from grass, and below-ground autotrophic respiration from grass and tree36roots that extend below lawn areas. RT includes the above-ground autotrophic res-piration from tree leaves, boles, and branches. This framework applies to the localnet emissions from a neighborhood, but does not account for non-local emissionsfrom processes such as up-stream fossil fuel combustion for power generation ordown-stream waste decomposition, e.g. in a landfill outside the study domain.With tower-based EC measurements, neither the actual distribution of the sur-face source and sink processes within the turbulent flux source area nor the frac-tional contribution of different processes to the net measured CO2 flux is known.Here we postulate that the fractional composition of the net flux, i.e. the fractionof measured FC attributable to each of the processes in Eq. 2.2, can be inferredthrough analysis of plan-area land cover fractions weighted by individual turbulentflux source areas. To do this, the processes in Eq. 2.2 must be conceptually linkedto a physical area on the surface so observed changes in source area land cover maybe interpreted as controls on the CO2 flux measured at the tower. For this study,EV is associated with impervious road land cover (λM, λS, λR), EB and RH withbuildings (λB), RL and PL with lawn (λL), and RT and PT with tree land cover (λT )(Table 2.5).2.3 Results and discussion2.3.1 Source area land cover and location biasAs a first step in the analysis, spatial patterns of emissions are assessed by compar-ing FC with wind direction (Figure 2.2c). On average, FC is highest when windsare from the SE (90-180◦) quadrant. When winds are from this sector, FC is greaterthan 20 µmol m−2 s−1 66.4% of the time, compared with 35% of the time inthe NW, NE, and SW quadrants combined. Mean FC during weekdays (Monday-Friday) from the SW is 37.8 µmol m−2 s−1, compared to 6.6 from the NW, 15.4from the NE, and 13.4 from the SW. There is also a temporal component to FCvariations as mean weekend (Saturday-Sunday) and holiday fluxes are on average18.6% lower than weekday FC for all wind directions (Figure 2.2d). These resultsare consistent with earlier findings by Walsh (2005), who observed highest fluxesfrom the SE during the first full year of FC measurements at this site in 2001-2002.37Variations in λ by wind direction (Figure 2.2e) are not as obvious as withFC (Figure 2.2c-d). Mean vegetation land cover (λV=λT+λL) ranges from 25%in the 150-160◦ wind sector to 40% in the 230-240◦ sector; λB is highest from90-100◦ (29%) and lowest from 320-330◦ (13%); and land cover fractions of busyroads (λM+λS) are highest from 140-150◦ (14%) and lowest from 310-320◦ (0.8%).Unsurprisingly, the presence of relatively busy roads in the SE is correlated withhigher FC values even though the overall coverage of the λM and λS road classes isrelatively small.To assess variations in λ measured at this tower location relative to the entire1900 × 1900 m study domain, the location-bias of source area λ can be quantifiedaccording to the following equation (Schmid and Lloyd, 1999):∆i =(λ f i− λ¯i)2λ¯i2(2.3)where ∆ is the location bias for a particular land cover class (i), λ f i is the landcover fraction of an individual source area, and λ¯i is the average land cover frac-tion for the entire 1900 × 1900 m study domain. This calculation was run for each30-minute period for (λT+λL), (λM+λS), and λB. Individual periods are sorted andaveraged by wind direction (Figure 2.2f). Impervious non-road (λI) is excludedbecause it is considered inactive in terms of CO2 emissions and (λM+λS) are com-bined for graphical clarity.Overall, average ∆B and ∆V are relatively small (Figure 2.2f), though a peak in∆B is found at 320◦ because of reduced building coverage compared to the neigh-borhood average. According to this measure, the tower-based observations areleast biased, i.e. most similar to the mean land cover of the study domain, whenwind directions are from the SW (180-270◦). The largest ∆ occurs in the SE for(λM+λS). This means that the tower location is most biased towards major andsecondary roads when winds are from these directions even though absolute plan-area coverage of these categories remains relatively small. For turbulent fluxesof heat, moisture, or momentum this bias may not be relevant, but for CO2 fluxmeasurements the impact is significant (Figure 2.2c and 2.2d).Due to the sensitivity of FC to the presence of (λM + λS) and the influence ofatmospheric stability on the flux source area, variations in plan-area land cover38a) λT+λLb) λM+λSc) λBd) FC (μmol m-2 s-1)36045%20%15%0%35%10%-2080Figure 2.3: Source area land cover composition and FC variations by winddirection and stability parameter. Wind directions are in 5◦ bins, z￿/L isin 0.1 bins.fractions with stability and wind direction are examined (Figure 2.3). In Figure2.3, wind directions are plotted along the x-axis and the stability parameter z￿/L(z￿ = z− zd , where z is the effective measurement height and zd is calculated bywind sector) is plotted along the y-axis for λB, (λM+λS), (λT+λL), and FC.Similar patterns to those observed in Figures 2.2c and 2.2e are apparent inFigure 2.3. Source area (λT+λL) is lowest and (λM+λS), λB and FC are highestwhen winds are from the SE (120-180◦). Highest FC is observed when windsare from this sector as well. Land cover variations with z￿/L (y-axis) are mostpronounced when winds are from 120-160◦. Here (λM+λS) decreases and (λT+λL)increases as the stability parameter z￿/L increases. As the atmosphere becomesmore stable and the stability parameter increases, the turbulent source area grows39in size upwind, and the intersection of 49th and Knight represents a decreasingfraction of the source area plan-area land cover. Accordingly, highest FC from thissector is observed during unstable conditions and decreases as stability increases.This signifies that simply using wind direction as a proxy for the turbulent fluxsource area may not be sufficient to account for all observed variations in plan-arealand cover and FC.2.3.2 Statistical model development and downscalingVehicle emissionsEmissions from fossil fuel combustion by motor vehicles (EV ) are expected to be animportant, mobile emissions source in this neighborhood and fluctuate on hourly,daily, and weekly cycles following commuter traffic patterns. According to trafficcounts available from the City of Vancouver, the majority (nearly 80%) of trafficis found along busy, arterial roads (λM + λS) that grid the neighborhood approxi-mately every 800 m (Figure 2.1) while the traffic load on residential sidestreets (λR)is minimal. As shown in Figures 2.2 & 2.3, FC is sensitive to changes in λM+λSand it is this relation that forms the basis for statistically modeling EV .For each hour of the day (h), a multiple linear regression was performed with30-minute FC and source area λM and λS (Figure 2.4, Table 2.3). This process wascarried out separately for each hour (h) for both weekdays (Monday - Friday) andweekends (Saturday - Sunday and statutory holidays):EVh = mShλSh+mMhλMh−bh (2.4)where mSh and mMh are the linear slope coefficients for each hour of the day forsecondary and major roads and bh is an offset accounting for all other source/sinkprocesses (b= RH +RL+RT +PL+PT ) (Table 2.3). The bh intercept is subtractedhere in order to isolate EVh from the other sources/sinks. Although bh will fluc-tuate depending on environmental conditions, the simplifying assumption is madethat for each hour, bh remains constant. This is justified because EV is at leastan order of magnitude greater than the emission terms that contribute to bh whentemperatures are greater than some heating threshold temperature (TH). For this40Tra!c model emissions (µmol m-2 s-1 )24a) Weekday, no intersectionc) Weekend, no intersection d) Weekend, intersectionb) Weekday,   intersectionSecondary road Major road Intercept (b)Hour (LST)Figure 2.4: Diurnal course of traffic model emissions for a) Weekday, no in-tersection, b) Weekday with intersection, c) Weekend no intersection,and d) Weekend with intersection. The solid line is emissions fromMajor roads and the dashed line is emissions from Secondary roadsdetermined from multiple-linear regression. The model intercept (i.e.background emissions) is the dot-dashed line near zero.analysis, individual 30-minute periods were selected when air temperature (Tair) isgreater than TH=14◦C to exclude emissions from space-heating (see section 3.2.2‘Building emissions’). During several nighttime periods when traffic is minimal,regression slopes are negative leading to unrealistic negative nighttime modeledtraffic emissions. During these situations, hourly regression slopes are forced tozero.A further aspect of analysis is then introduced based on the assumption that thepresence of an intersection will increase EV due to slower average vehicle speeds,idling, and acceleration through the intersection. Of particular relevance is the41intersection of Knight Street and 49th Ave 120 m SE of the tower. To test this,30-minute FC measurements were selected when source-area compositions did notcontain both λS (49th Ave) and λM (Knight Street), but predominantly only λS orλM. Practically, there are hardly ever periods that include only λM or λS, so cutoffpercentages of λM < 1.0% and λS < 0.74% were chosen. The cutoff values wereselected to exclude as much of the intersection as possible while still maintaining asufficient number of datapoints for analysis. These points are then subject to linearregression analysis according to the same procedures described earlier.Based on the regression models, hourly traffic emissions can be expressed interms of Major and Secondary road land cover (Figure 2.4). Results show thatemissions from Secondary and Major roads follow daily and weekly patterns. BothmS and mM show a typical 2-peaked curve with early morning and late afternoonpeaks associated with commuter traffic patterns (Sailor and Lu, 2004). Weekdayemissions are on average 31% higher than weekend emissions for Secondary roads,and 44% higher for Major roads. Results also show that the presence of an inter-section increases EV on average by 68% relative to what would be expected fromroad segments with no intersection during weekdays. These differences should betaken into account during spatial allocation of emissions, though it is not evidentexactly how far the effect of the intersection will propagate ‘upstream’ along theroads and cause decreases in mean vehicle velocity.42Table 2.3: Hourly model parameters for the multiple linear regression model used to determine traffic emissions. Hoursindicate the center of the time interval (e.g. 1200 is the period 1130-1230). For each hour, the top row is Weekdayvalues and the bottom row is Weekend values. Parameters excluding the intersection are given in brackets.Hour Major road slope (mM) Secondary road slope (mS) Offset (b) RMSE n(µmol s−1 m−2 road) (µmol s−1 m−2 road) (µmol s−1 m−2) (µmol s−1 m−2) (#)0 19.73 (0.00) 129.87 (0.00) 2.47 (6.42) 5.33 (2.50) 241 (89)87.75 (41.83) 180.35 (31.39) -0.36 (3.42) 5.23 (4.16) 135 (58)1 0.00 (0.00) 56.43 (0.00) 5.44 (7.42) 3.79 (3.62) 208 (77)37.35 (0.00) 105.47 (0.00) 2.17 (5.95) 5.04 (2.74) 119 (53)2 0.00 (0.00) 57.81 (24.48) 3.77 (4.72) 3.22 (2.15) 177 (56)0.00 (0.00) 53.91 (0.00) 5.10 (7.99) 4.11 (4.27) 106 (41)3 0.57 (0.00) 31.57 (0.00) 4.14 (5.78) 3.57 (3.63) 158 (57)11.62 (12.69) 73.68 (38.04) 3.11 (2.92) 3.72 (3.26) 92 (34)4 20.37 (0.00) 56.93 (0.00) 4.18 (7.46) 4.34 (4.43) 132 (44)0.00 (0.00) 43.63 (0.00) 6.16 (11.11) 4.21 (4.50) 83 (28)5 137.28 (13.27) 258.11 (0.00) -3.13 (5.85) 9.35 (7.31) 148 (62)62.87 (61.39) 87.26 (239.80) 1.84 (1.24) 7.98 (9.56) 75 (24)6 400.41 (236.54) 364.71 (124.43) -6.90 (2.25) 14.40 (10.89) 186 (62)92.46 (103.06) 58.42 (48.14) 4.81 (4.56) 10.56 (8.65) 93 (34)Continued on next page43Table 2.3 – continued from previous pageHour Major road slope (mM) Secondary road slope (mS) Offset (b) RMSE n(µmol s−1 m−2 road) (µmol s−1 m−2 road) (µmol s−1 m−2) (µmol s−1 m−2) (#)7 536.49 (337.65) 344.07 (74.60) -6.21 (3.02) 17.11 (10.21) 247 (82)148.42 (118.51) 116.61 (19.22) 0.32 (1.78) 10.48 (9.17) 126 (46)8 505.64 (351.98) 306.99 (218.03) -4.99 (1.26) 17.51 (15.78) 323 (110)255.90 (261.42) 125.38 (209.87) -3.55 (-4.86) 10.62 (10.56) 156 (56)9 481.82 (306.55) 317.20 (344.45) -7.22 (-5.68) 15.85 (11.55) 397 (132)312.53 (172.15) 170.35 (115.95) -3.68 (-0.24) 9.83 (9.39) 198 (76)10 496.26 (312.21) 258.94 (210.64) -5.68 (-1.71) 13.04 (11.72) 452 (187)285.55 (261.12) 228.05 (200.98) -1.00 (0.22) 12.08 (10.64) 219 (90)11 497.98 (403.66) 342.17 (273.90) -7.72 (4.59) 12.99 (8.48) 491 (240)361.45 (246.02) 321.82 (290.60) -6.54 (-4.36) 12.02 (6.76) 229 (103)12 474.80 (320.80) 334.17 (215.42) -6.96 (1.84) 10.88 (8.16) 518 (295)305.42 (198.62) 302.33 (273.63) -3.54 (-1.72) 10.53 (7.90) 261 (128)13 501.07 (361.46) 349.75 (261.41) -7.91 (-3.99) 10.79 (7.12) 571 (344)361.74 (345.61) 254.27 (280.33) -3.14 (-3.80) 8.41 (5.56) 259 (148)14 509.69 (365.29) 380.23 (312.09) -6.91 (-3.66) 10.28 (6.86) 580 (370)338.97 (291.45) 318.16 (299.75) -4.16 (-3.25) 8.98 (5.26) 272 (168)15 487.90 (288.04) 395.96 (301.25) -5.24 (-1.12) 9.63 (5.86) 584 (383)Continued on next page44Table 2.3 – continued from previous pageHour Major road slope (mM) Secondary road slope (mS) Offset (b) RMSE n(µmol s−1 m−2 road) (µmol s−1 m−2 road) (µmol s−1 m−2) (µmol s−1 m−2) (#)359.19 (324.09) 284.90 (245.77) -2.38 (-0.91) 9.98 (6.01) 265 (175)16 433.42 (273.93) 423.22 (349.80) -5.01 (-1.97) 9.92 (6.72) 550 (368)343.17 (285.88) 289.50 (249.80) -2.82 (-1.27) 10.12 (5.95) 254 (167)17 407.72 (255.25) 403.38 (323.15) -3.47 (-0.24) 9.55 (6.40) 545 (353)344.88 (195.77) 304.82 (226.27) -2.60 (0.85) 10.31 (5.01) 245 (157)18 360.52 (198.28) 355.88 (256.74) -3.01 (0.86) 10.57 (4.65) 513 (312)280.04 (152.43) 248.16 (200.84) -0.95 (1.51) 9.69 (3.87) 223 (135)19 265.09 (81.92) 250.51 (113.12) 0.08 (4.58) 9.87 (4.98) 466 (253)241.87 (135.11) 189.65 (110.04) -0.10 (3.43) 9.11 (4.65) 211 (104)20 218.20 (65.43) 373.41 (104.20) -2.53 (4.42) 9.54 (4.94) 419 (206)240.45 (56.47) 339.85 (25.63) -2.88 (5.82) 10.89 (3.33) 193 (85)21 150.86 (5.42) 410.23 (0.00) -0.46 (8.77) 10.89 (5.31) 358 (156)212.81 (54.22) 468.48 (57.99) -7.00 (4.59) 10.43 (4.64) 176 (66)22 96.50 (4.15) 355.71 (55.08) -1.27 (6.50) 8.06 (3.99) 316 (130)136.08 (34.05) 395.39 (72.24) -4.48 (4.23) 8.36 (3.26) 160 (58)23 39.74 (0.00) 227.17 (0.00) 2.34 (6.50) 5.93 (3.72) 273 (118)77.45 (10.41) 221.33 (9.56) 0.75 (6.46) 6.43 (4.22) 143 (54)45Building emissionsApproximately 90% of buildings in the Sunset neighborhood have natural gasspace heating systems that generate local CO2 emissions. The demand for space-heating is expected to be primarily determined by Tair as long as Tair is below aheating temperature threshold (TH). On monthly timescales, there is a linear rela-tion between measured total CO2 fluxes (emissions) and Tair expressed as heatingdegree days (Christen et al., 2011) and for individual selected hours this linear-ity is expected to hold, though the slope of this relation will change from hour tohour depending on occupant activity, building thermal inertia, and type of heatingsystem control.In order to extract the EB signal from FC, for each hour, individual 30-minuteperiods were selected when λM and λS are minimal (< 2%) in order to avoid trafficemissions (EV ). This conditional sampling technique was used rather than statis-tically modeling and removing EV in order to keep the EV and EB models inde-pendent of each other. Next, to account for different building volumes within eachsource area, FC was divided by the source area-specific building volume derivedfrom LiDAR data. This approach effectively places all measured emissions intothe building volumes and expresses emissions as per unit building volume only(µmol m−3 bldg s−1). Building volume is expressed as λBzB where zB is the aver-age source area building height at 1 m resolution.The 30-minute scaled FC values were then binned into 2 K Tair classes andsegmented linear regression was used on bin median FC values to determine thetemperature threshold for each hour TH(h) (Figure 2.5). Segmented linear re-gression is useful for datasets in which there are different processes acting alongnon-overlapping ranges of some variable (Tair in this situation) (Bates and Watts,1988). To apply a segmented regression fit, a ‘breakpoint’ must be identified at thetransition point between processes (i.e. from heating to non-heating) and a linearregression fit is applied to each section before and after the breakpoint. For thisdataset, the breakpoint for each hour was identified by sequentially fitting two lin-ear segments joined at a breakpoint across a user-defined Tair range (Figure 2.5).The arrangement with the minimum root-mean-square-error was deemed the bestfit (Table 2.4).4615:00 LST bp = 13.7oC , 1.7 µmol m-3 s-1 m = -1.45 µmol m-3 s-1 K-1 rmse = 0.72 µmol m-3 s-1 n = 45490%10%75%50%25%mean2 KFigure 2.5: Building space-heating model segmented regression fit for a sin-gle hour (15:00 LST). Linear segments are fit to binned medians and thebreakpoint (bp) is determined as 13.7◦C, 1.7 µmol m−3 bldg s−1. Emis-sions from space-heating occurs to the left (cooler temperatures) of thebreakpoint according to slope m. The root mean square error (rmse)value is given for the fit to the median values, the number of points (n)is given for all bins. This procedure was separately performed for allhours (0-23 LST) (Table 2.4).47For each hour, when Tair < TH , EB should increase linearly with decreasingTair (along a slope, m) and when Tair > TH , EB is expected to remain constant(m =0). Though EB remains constant (at 0) when Tair > TH , other processes arestill active and contributing to measured FC. For example, nighttime FC may rise asTair increases because of associated increases in soil temperature and consequently,soil and vegetation respiration.As with the EV model, there are background emissions (b) composed of RS, RV ,RH , EV from residential roads, and EB from non-space-heating processes, which areassumed to remain constant for each hour. Thus, for each hour (h), when Tair < TH ,EB from space-heating per m−3 of building can be modeled:EBh =mhTair −bhλBzB(2.5)Using this method, the modeled daily course of EB from space-heating peaksat 0900 LST in the morning, declines through the afternoon, then has a secondarypeak in the early evening (Figure 2.6). The TH is also variable through the day,with an average of 12.0◦ C until noon, then increasing to a maximum of 15.3◦ C at1500, before falling to 7.0◦ C at 2200.The net background emissions at the TH breakpoint show fairly steady emis-sions of 6 µmol m−2 s−1 (where m−2 refers to the entire neighborhood average)in the morning, decreasing emissions during the day presumably due to photosyn-thetic activity, and a peak in the evening at 2100 up to 16 µmol m−2 s−1 (Figure2.6). This peak is unlikely from EV along residential sidestreets because there is nocorresponding morning rush-hour peak, and other biogenic process (RS, RV , RH)do not seem likely either because the magnitude is larger than what would be ex-pected from respiration. A hypothesis is that this peak is primarily from emissionsassociated with domestic hot water production, cooking, and possibly wood firesfor home heating. According to a 2006 BC Housing Assessment, 85% of resi-dences in the study domain have wood fireplaces, though there is insufficient data(e.g. usage statistics or surveys) to determine how frequently they are in use. Still,it is expected they are primarily used in the evening hours.48Table 2.4: Model parameters for the segmented regression model used to de-termine hourly emissions due to space-heating. TH is the heating thresh-old above which space-heating systems are active.Hour Slope (mh) TH Background (b) RMSE n(µmol m−3 s−1 K−1) (◦C) (µmol m−2 s−1) (µmol m−3 s−1) (#)0 -1.48 10.3 3.68 0.19 1501 -1.69 12.4 3.60 0.40 1672 -1.57 9.9 3.60 0.79 1913 -1.73 12.8 4.20 1.23 1894 -1.73 12.0 4.64 0.95 2055 -1.50 13.1 3.44 0.59 2206 -1.38 12.4 3.86 0.72 2297 -1.70 12.9 2.46 1.3 2258 -2.55 12.7 0.77 0.97 2479 -2.29 10.2 2.18 0.83 21110 -2.88 12.7 -1.5 0.99 21411 -2.97 12.3 -1.8 0.73 21112 -1.27 13.3 0.59 0.44 20613 -1.45 15.3 0.57 1.3 28314 -1.48 15.0 1.10 1.2 36315 -1.45 13.7 1.70 0.72 45416 -1.34 13.2 2.11 0.64 54417 -1.35 13.5 3.22 0.61 58418 -1.36 13.0 4.39 0.81 58719 -1.29 12.0 5.40 0.59 52620 -1.44 10.6 7.30 0.47 30421 -1.44 7.0 9.77 0.85 24822 -2.70 6.0 6.87 2.1 20323 -1.24 8.0 4.29 1.6 184492010151015Figure 2.6: Contour plot of EB from space-heating based on hourly seg-mented regression models. The temperature threshold (TH) at whichheating systems are activated is shown as the solid line and emissionscontours are shown as dotted lines labeled with emissions values (µmolm−3 bldg s−1).PhotosynthesisDaytime FC measurements when Tair > TH were selected to account for the up-take of atmospheric CO2 by vegetation gross photosynthesis (P). Additionally, EVis modeled (according to Section 3.2.1) and subtracted from FC and the resulting30-minute values are divided by the source area specific LAI. This approach effec-tively attributes the FC signal to m−2 of leaf area.The LAI-scaled FC values are then sorted by air temperature (2 K bins) andplotted against PPFD (100 µmol m−2 s−1 bins) (Figure 2.7). PPFD was not mea-sured directly at the flux tower but is estimated from flux tower measurements of50incoming solar shortwave irradiance (K↓) based on an empirical linear relation be-tween K↓ and PPFD (K↓ vs PPFD, slope=2.01 µmol J−1, offset=11.91 µmol m−2s−1, r2=0.99, n=2992) developed from measurements at the UBC climate stationlocated approximately 12 km west of the flux tower. There, K↓ (Eppley Preci-sion Spectral Pyranometer, Eppley, Inc., Newport, Rhode Island, USA) and PPFD(SQ-100 Quantum Sensor, Apogee Instruments, Logan, UT, USA) were measuredside-by-side and summertime (Jul-Aug 2010) 10-minute means were used for anal-ysis.A photosynthetic light response curve (O¨gren and Evans, 1993) was then fit tothe binned FC medians for each temperature range (Figure 2.7):P =M ·PPFD+Pm−￿(M ·PPFD+Pm)2−4C ·M ·PPFD ·Pm2C(2.6)where M is the maximum quantum yield (0.01 µmol µmol−1 fit using all air tem-peratures),C is a dimensionless empirical convexity parameter (set to 0.83), and Pmis maximum P at light saturation (7.24 µmol m−2 s−1 fit using all air temperatures).This model was used because of its reliance on few input variables, hence it doesnot explicitly include other environmental variables that influence P such as wa-ter vapor pressure deficit, soil temperature (TS), and vegetation type and condition.As implemented here, this model is also biased towards warm temperatures whenvegetation is most active and therefore uncertainty is introduced when applied towinter leaf-off conditions when deciduous trees are not sequestering CO2. To cap-ture sub-annual temporal variability a seasonal model would be preferable, but thiswas not feasible with the conditional sampling techniques used here because duringwinter, CO2 emissions from building heating sources obscure the relatively smallwintertime P signal. The light response curves also include ecosystem respirationwhich is determined as the y-axis intercept. To model only the absolute values ofP, this y-axis intercept value is subtracted from the light response curve.For comparison, up-scaled cuvette measurements from a Li-6400 portable pho-tosynthesis system (Licor Inc, Lincoln, NE, USA) are shown on Figure 2.7. Mea-surements were taken during summer 2009 from individual leaves on the canopyexterior of 22 trees representative of common species encountered in the study do-main (Liss et al., 2010). The system was also used to measure photosynthetic up-5190%10%75%50%25%mean100 µmol m-2 s-114-16 oC16-18 oC18-20 oC20-22 oC22-24 oCModel !t to EC data (all temperatures)Model !t to scaled cuvette measurementsFigure 2.7: FC plotted against photosynthetic photon flux density (PPFD,µmol m−2 s−1) after EV has been modeled and subtracted from FC.PPFD is binned in 100 µmol m−2 s−1 increments and a photosynthe-sis model (Eq. 2.6) is fit to binned medians. The model fit to all airtemperatures is shown as a solid line and the model fit to scaled cuvettemeasurements is the dashed line. Binned FC medians for different tem-perature ranges are shown as colored dots. Mean LAIT for the entirestudy area is 0.39 m2 m−2 and mean LAIL for the entire study area is1.43 LAIL.52take from four representative lawns within the longterm turbulent flux source area.For each measurement, relative humidity (60%), air temperature (25◦C), and CO2mixing ratios (380 ppm) were kept constant within the cuvette while PPFD wasvaried from 0-1500 µmol m−2 s−1 (Liss et al., 2010). Measurements are scaled tom2 of lawn and m2 grass, respectively, and the photosynthesis model (Eq. 2.6) isfit to the average values for lawn and leaf photosynthesis separately. The lawn andleaf photosynthesis models are then area-averaged based on the longterm sourcearea mean values of LAIT and LAIG. The up-scaled gross photosynthesis systemmeasurements show 21% higher values of CO2 uptake than the EC measurements.This is expected given that cuvette measurements assume full exposure to sunlightby the leaf surface, when in reality most tree leaves act to shade each other andreduce total CO2 assimilation.The resulting model fit to the EC measurements does not distinguish betweenphotosynthetic uptake by tree leaves (PleafT) or lawn grass (PleafL). To do this, amean PleafT to PleafL ratio of 0.78 is calculated from the photosynthesis system. Thisis used to spatially partition photosynthetic uptake with the following expression:Ptot = PL+PT = PleafLLAIL+PleafTLAIT (2.7)where Ptot is the total model photosynthetic uptake, PleafL and PleafT are uptake perm2 of lawn and tree leaves respectively, and LAIL and LAIT are leaf area indexvalues for lawns and trees.RespirationTo investigate the ecosystem respiration terms in Eq. 2.2 (RL, RT , and RH), two in-dependent methods are applied. The first method uses nighttime (0100-0500 LST)FC measurements and the second uses the intercept of light-response curves (Eq.2.6) derived from daytime (K↓> 5 Wm−2) FC measurements. Both estimates arecompared against independent CO2 exchange measurements from closed cham-bers and cuvettes within the long-term turbulent flux source area (Liss et al. 2009,2010).For both daytime and nighttime methods, 30-minute periods are selected whenTair > TH (Section 3.2.2) to exclude periods with substantial emissions from build-53ing space-heating. Additionally, EV is modeled (Section 3.2.1) and subtracted frommeasured FC, trading model independence for a greatly increased number of data-points and geographic coverage available for analysis.For the nighttime method, the resulting 30-minute FC values are binned in 1K soil temperature (TS) intervals and an exponential soil respiration model is fit tothe binned FC medians (Figure 2.8). Measured nighttime emissions include below-ground autotrophic and heterotrophic respiration, above-ground autotrophic respi-ration from lawns and trees, and human respiration (RL+RT +RH). It is expectedthat autotrophic and heterotrophic respiration from lawns will dominate due to thegreater plan-area land cover occupied by lawn within the source area (λL=24% vs.λT=11%) and a directly measured leaf respiration to lawn respiration (Rleaf : RL)ratio of 0.13 based on chamber and cuvette measurements (see below for ratiocalculation details). It is assumed that TS is an important environmental controlfor nighttime respiration and use of the standard soil temperature-dependent soilrespiration model by Lloyd and Taylor (1994) is appropriate:RL = Rref exp￿E0￿1Tref−T0−1TS−T0￿￿(2.8)where Rref (2.09 µmol m−2 s−1) is a reference RL at a reference TS (Tref=285.15K), E0 is a temperature sensitivity parameter (357.4 K), and T0 is a minimum limitset to 227.13 K according to Lloyd and Taylor (1994).The second independent method uses daytime FC values in which the EV signalhas been statistically removed during periods when Tair > TH . These values arethen sorted by TS (2 K bins) and PPFD (100 µmol m−2 s−1 bins) (Figure 2.8). Alight-response curve based on PPFD (Eq. 2.6) is then fit to the binned FC mediansfor each TS range. The model fits show increasing CO2 uptake (negative FC) asPPFD increases and have a positive value (net emissions) at the intercept of they-axis. The y-intercept values of the fitted light response curves at PPFD=0 areinterpreted as mean CO2 respiration (RL +RT +RH) for each TS range (e.g. Falgeet al. 2002). The respiration model (Eq. 2.8) is then fit to the daytime y-interceptpoints for comparison with the nighttime model fit (Figure 2.8).The respiration estimates from the nighttime model are greater than the day-54Fit based on nighttime EC dataFit based on scaled chamber data90%10%75%50%25%mean1 KFit based on daytime EC data (y-intercept photosynthesis model points)Daytime EC y-intercept photosynthesis model pointsFigure 2.8: Soil respiration model curves (Eq. 2.8) based on fits to night-time FC binned medians (solid curve), y-intercept points (when PPFD=0µmol m−2 s−1) using a photosynthesis model (Eq. 2.6) fit to day-time EC measurements (square points), soil respiration model fit tothe y-intercept points (dashed line), and independent plot-scale cham-ber measurements (dotted curve, individual chamber measurements arenot shown).time model estimates by on average over 250% so an independent dataset of lawnchamber and leaf cuvette measurements is used as reference. Measurements ofRL were conducted during 2008-2009 and are comprised of over 200 individualmeasurements on representative urban lawns within the long-term flux source areafootprint. Measurements were taken with a Li-800 infrared gas analyzer (Licor,Inc., Lincoln, NE, USA) using a custom-built opaque-domed chamber system.Observations covered a range of TS (0-30◦C)) and soil moisture conditions (Lisset al., 2009). The respiration model (Eq. 2.8) fit to these points is multiplied55by (λL+λT ) to scale chamber measurements to the land cover proportions of thelongterm turbulent flux source area. The λT area is included to account for ar-eas of lawn underneath overhanging tree canopies which are assumed to be lawns.This causes an overestimation of lawn land-cover because tree canopies also over-hang buildings, sidewalks, and roads, but this bias is at least partially offset byexpected increases in RL below tree canopies resulting from higher below-groundautotrophic respiration from tree roots. Tree canopies also can influence RL bymodulating soil temperatures through shading and reduced reduced sky-view fac-tors resulting in generally cooler daytime soil temperatures and warmer nighttimesoil temperatures compared to more open areas of lawn.To estimate the contribution of RT , leaf respiration was determined from cu-vette observations during summer 2009 from a portable photosynthesis measure-ment system. As described in Section 3.2.3, relative humidity, air temperature, andCO2 mixing ratios were kept constant within the cuvette while PPFD was variedfrom 0-1500 µmol m−2 s−1. The mean value of leaf respiration when PPFD=0µmol m−2 for all vegetation samples was determined as 0.7 ± 0.2 µmol m−2 leafs−1 at Tair=25◦C.During the May 2008 - Apr 2011 study period, mean TS is 21.6◦C when Tair >24.5◦C and Tair < 25.5◦C. According to the soil respiration model fit based on soilchamber measurements, RL when TS=21.6 ◦C is 5.4 µmol m−2 s−1. Therefore atthis TS there is a Rleaf to RL ratio of 0.13. Assuming this ratio remains constant,Rleaf is estimated as 0.13RL and is scaled up to RT using LAIT . This is a rough es-timation and assumes that Rleaf is also temperature dependent and follows an expo-nential curve. Though controlled experiments have shown that Rleaf does increasewith increasing temperature (Villar et al., 1995), the approach used here neglectsother variables influencing Rleaf and stomatal conductance such as leaf transpira-tion, photosynthetic rates, humidity, and CO2 mixing ratio differences between theair and leaf interior (Campbell and Norman, 1998).The (RL +RT ) model curve derived from scaled chamber and photosynthesissystem measurements is nearly identical to the respiration model curve fit to thedaytime FC measurements (Figure 2.8). Although daytime respiration is not ex-pected to exactly equal nighttime respiration due to differences in temperature sen-sitivity and possible light inhibition of plant respiration (Reichstein et al., 2005),56these effects are not expected to be large enough to explain the magnitude of night-day differences observed here. A speculative explanation for the night-day dif-ferences is human respiration (RH). Most neighborhood residents will be homeovernight and RH will occur indoors. According to building engineering codes,a minimum ventilation rate for residential areas is 0.35 m3 m−3 hr−1 (ASHRAE,2004). This means that on average the entire volume of indoor air will be ventedover the course of 2.86 hours and indoor CO2 emissions will be measured by theEC system. Based on published per capita CO2 emission rates of 201.54 µmolcap−1 s−1 (Moriwaki and Kanda 2004; Christen et al. 2011) and mean source areapopulation density of 0.0064 persons m−2, source area-scaled RH is expected to beapproximately 1.2 µmol m−2 s−1. This value is reasonably close to the night-daydifference of 1.4 µmol m−2 s−1 for the average TS (11.3◦C) observed over the May1, 2008 - Apr 30, 2011 study period.Due to the agreement between soil respiration model fits between scaled cham-ber measurements and daytime FC photosynthesis model y-intercept points, thedaytime model fit is chosen for modeling (RL+RT ) for both day and night periods.RH is included as an additional constant CO2 source (1.4 µmol m−2 s−1) duringnighttime periods only, though this assumes virtually all residents leave the neigh-borhood during the day to attend work or school and does not fully explain theapparent night-day difference dependence on TS.So far, respiration models based on FC observations have been expressed assource-area scaled values. To downscale respiration and express it in terms oflawn and leaf area, the measured Rleaf to Rlawn ratio from chamber observations isused. For each time-step, total modeled source area-scaled respiration (Rtot) can beexpressed as:Rtot = Rlawn(λL+λT )+Rlea f LAIT (2.9)where Rleaf and Rlawn are emission factors scaled per m−2 of leaf and lawn respec-tively. Given that Rlea f : Rlawn=0.13 for this site, this equation is solved for bothRlea f and Rlawn which are then used to distribute total respiration between λL (RL)and LAIT (RT ). This method calculates RT indirectly based on RL and does not ac-count for variables such as leaf temperature or leaf moisture content, or respirationfrom tree boles and branches.572.3.3 Modeling net emissionsEmissions totalsThe spatial attribution analysis of FC returns spatio-temporal models separated intoindividual emission or uptake processes based on environmental variables and ex-pressed in m−2 land cover (Table 2.5). Three years of FC measurements (May 1,2008- Apr 30, 2011, Year 1-3) were used as the dataset for statistical analysis andmodel development (Section 3.2), and a fourth year (May 1, 2011-Apr 30, 2012,Year 4) was withheld from analysis to be used for model testing. Each of the pro-cesses listed in Table 2.5 is modeled at 30-minute resolution for Year 4 and twoseparate spatial scalings are applied. The first is turbulent flux source area-scalingbased on plan-area land cover and LAI extracted from individual turbulent fluxsource area models at each 30-minute timestep. This scaling reflects the locationbias of the EC flux tower and can be directly compared to FC measurements fromthe EC system during Year 4. The second scaling uses the entire 1900 × 1900 mdomain and is interpreted as the spatially un-biased ecosystem averaged flux for alarger urban subset area.58Table 2.5: Individual CO2 emissions and uptake process sub-models with associated environmental variables and landcover elements used for model development and spatial scaling. Sub-model development details are found in thetext.Symbol Description Environmental variable Land cover elementEV Vehicle traffic emissions Hour of day, day of week λM, λSEB Space and water heating, cooking, wood burning Hour of day, Tair (◦C) Building volumeRL Lawn respiration (soil and grass and roots) Tsoil (◦C) λLRT Respiration from tree leaves Tsoil (◦C) LAITPL Photosynthetic uptake by lawn grass PPFD (µmol m−2 s−1) LAILPT Photosynthetic uptake by tree leaves PPFD (µmol m−2 s−1) LAITRH Human respiration Night-day respiration difference Building volume(applied to nighttime periods only)59At 30-minute resolution, there is general agreement between individual FC ob-servations and source area-scaled modeled fluxes, through there is also substantialscatter (the overall RMSE is 12.41 µmol m−2 s−1). This scatter is not unexpectedgiven that the models were based upon binned-median FC values instead of individ-ual 30-minute datapoints in part because high variability is a noted characteristicof this site and other urban FC datasets (e.g. Velasco and Roth 2010).The model performs better when predicting the annual ensemble mean diurnalcourse of CO2 fluxes (Figure 2.9a). During daytime (0500-1500 LST), the modelis higher than observed ensemble means by an average of 4.0 µmol m−2 s−1, andovernight (2200-0400 LST) is lower by an average of 2.5 µmol m−2 s−1. Themodeled source area-scaled daily total emissions are 22.17 g C day−1 comparedto an observed daily total of 20.77 g C day−1 (6.7% difference). The primaryemissions source is traffic emissions accounting for 18.10 g C m−2 day−1 (81.6%of the daily total), followed by building emissions (3.65 g C m−2 day−1, 16.4%).The daily respiration total (lawn, tree, and human) is 1.60 g C m−2 day−1 (7.1%)and photosynthesis acts to remove 1.18 g Cm−2 day−1 (-5.3%). The annual diurnalensemble means include both weekend and weekday periods (Table 2.6).Annual mean daily total domain-scaled modeled emissions are 16.93 g C day−1which is 24% lower than the source-area scaled model total (Figure 2.9b). Theprimary emissions source is traffic accounting for 12.14 g C m−2 day−1 (71.8% ofthe daily total), followed by building emissions (4.27 g C m−2 day−1, 25.2%). Thedaily respiration total (lawn, tree, and human) is 1.85 g C m−2 day−1 (10.8%) andphotosynthesis acts to remove 1.34 g C m−2 day−1 (-7.9%).Monthly emissions totals are also calculated using source area- and domain-scalings (Figures 2.9c and 2.9d). During calculation of long-term (monthly-annual)exchange totals, measurement gaps in the EC system from rainy periods or systemmaintenance and calibration must be accounted for. To account for gaps here,monthly totals are calculated using ensemble hourly means for each month. Forthe source area-scaled monthly totals (Figure 2.9c), the diurnal course of hourlyensemble means for both weekends and weekdays and for observed and modeledFC is calculated for each month. The daily total exchange is then determined forweekdays and weekends separately and scaled up to monthly totals by the numberof weekends and weekdays that occur in the month.60Buildings Tra!cRespirationPhotosynthesisa) Source area-scaled b) Domain-scaledc) Source area-scaled d) Domain-scaledFigure 2.9: Modeled (bar) and observed (line) CO2 emissions and uptakefor a) hourly source area-scaled area, b) hourly domain-scaled area, c)monthly source area-scaled area, and d) monthly domain-scaled area.The source area-scaled modeled monthly totals agree well with observationsduring the warmer months (Apr-Sep average difference is 6.1%), but overestimateemissions during cool months (Oct-Mar) by an average of 36% (Figure 2.9c). Peakphotosynthesis occurs in Jul (-0.06 kg C m−2 month−1, -14.9% of the monthlytotal) and largest respiration totals are in Aug (0.07 kg C m−2 month−1, 15.7%of the monthly total). The discrepancy between model and observations duringOct-Mar is in part due to traffic model bias towards warm-weather (summer) trafficpatterns. This bias is a result of selecting warm periods to create statistical relationsfor EV to avoid emissions associated with building space-heating. From trafficcounts, it is expected that overall traffic is decreased by 10% in winter relative tosummer (Christen et al., 2011), meaning the traffic model overestimates wintertime61EV .Domain-scaled modeled building emissions range from 5.3% (0.02 kg C m−2month−1) of the monthly total in Aug to 47.7% in Jan (0.35 kg C m−2 month−1)(Figure 2.9d). A potential source of EB model error is bias towards residentialbuildings. The EB model developed here was formulated using periods selectedspecifically to predominantly include residential buildings, which account for over90% of buildings in the study domain. Within 200 m of the tower however, there isa cluster of commercial buildings including a fast food restaurant, a school at the in-tersection of Knight Street and 49th Avenue, and a residential apartment complex.Different occupant activities, building functions, energy consumption profiles, andenergy efficiencies associated with these buildings are likely to result in differentCO2 emissions compared to residential, detached homes.In year 4, source area-scaled modeled annual total FC is 9.76 kg C m−2 y−1,compared to a measured annual total of 9.41 C m−2 y−1 (Table 2.6). Domain-scaled annual total emissions are 6.42 kg C m−2 y−1 (-34% difference from thesource area-scaled total). The largest domain-scaled source is from traffic (4.42 kgC m−2 y−1, 68.8%), followed by building emissions (1.79 kg C m−2 y−1, 27.9%),respiration (0.67 kg C m−2 y−1, 10.5%), and photosynthesis (-0.46 kg C m−2 y−1,-7.2%).The domain-scaled annual emissions total agrees well with independent neighborhood-scale modeling of annual CO2 emissions based on component-specific bottom-upmodels (e.g. building energy models, traffic counts) within the same 1900 × 1900study domain (Kellett et al., 2012). That study calculated an annual total local flux(emissions from traffic, buildings, respiration from humans, soil, and vegetation,and offsets from photosynthesis) of 5.73 kg C m−2 y−1 (-11% difference with thepresent study). Their model also suggests of all emissions, 47% originate from ve-hicles (compared to 68.8% in the present study), 40% are from buildings (presentstudy: 27.9%), and 12% from human, soil, and vegetation respiration (presentstudy: 10.5%). Photosynthesis accounts for -9% emissions offset in the bottom-upmodel and -7.2% in the present study. Although agreement between the total an-nual FC and the contribution of respiration and photosynthesis calculated by the twostudies is good, there is less agreement regarding the distribution between trafficand building emissions. This discrepancy suggests an over-estimation of the traffic62Table 2.6: Individual sub-model daily and annual emission totals with spatialdomain- and turbulent flux source area-scalings applied. Relative propor-tions of total net emissions are given in parentheses.Daily totals (g C m−2 day−1) Annual totals (kg C m−2 yr−1)Emissions component Source area Domain Source area DomainPL+PT -1.18 (-5.3%) -1.34 (7.9%) -0.40 (-4.0%) -0.46 (-7.2%)RL+RT 0.87 (3.9%) 1.01 (5.9%) 0.30(3.1%) 0.35 (5.5%)RH 0.73 (3.2%) 0.84 (4.9%) 0.46 (4.7%) 0.32 (5.0%)EV 18.10 (81.6%) 12.14 (71.8%) 6.86 (70.2%) 4.42 (68.8%)EB 3.65 (16.4%) 4.27 (25.2%) 2.55 (26.1%) 1.79 (27.9%)Total 22.17 16.93 9.76 6.42contribution (in part due to seasonal bias discussed earlier) and under-estimationof building emissions (in part due to residential building bias discussed earlier) inthe statistical models of the present study.Emissions mappingModeled CO2 emissions can also be downscaled spatially and visualized as high-resolution emissions maps (Figure 2.10). Maps of annual emission totals are con-structed by summing the modelled unscaled year-long 30-minute timeseries foreach emissions component and then multiplying by the appropriate land-cover andLAI layers (Table 2.5) in grid cells at 20 m resolution. The 20 m resolution is arbi-trary but is used because it approximates width of Major roads and is fine enough todistinguish between vegetation and buildings on individual residential lots in thisstudy area.Highest emissions occur at intersections of Major and Secondary roads withhighest mapped emissions of 118.4 kg C m−2 y−2 at the intersection of Knightand 41st (intersection of two Major roads). The statistical EV model as developedhere only considers Major and Secondary roads, but for mapping purposes, 24% ofmodeled emissions are re-distributed evenly to residential side-streets. This local-trip share factor is based on a regional trip diary survey (Ministry of Transportation,2004). Spatial patterns of building emissions vary with building volume and high-63est emissions originate from the schools, churches, and commercial buildings inthe neighborhood with largest volumes. Respiration (human, lawn, and tree) andphotosynthesis are included on one map and park areas with high vegetation den-sities stand out as net CO2 sinks on an annual basis. Highest modeled net uptakeis -1.6 kg C m−2 y−2 in Memorial Park and the mean (median) net uptake for veg-etated areas is -0.11 (-0.07) kg C m−2 y−2. The maximum net uptake observedin Memorial Park is a very high value compared to forest ecosystems and is theresult of LAI values up to 10.5 m−2 m−2 representing mature trees in the park. Thephotosynthesis is scaled to this LAI value and is likely an overestimation becausethe model assumes all leaves are equally exposed to light and does not account forlight extinction through the canopy.2.4 ConclusionsThe ‘Vancouver-Sunset’ flux tower is characterized by a high location bias in termsof CO2 fluxes due to the proximity of a busy road intersection on one side of thetower. The presence of this intersection makes spatially un-biased interpretationand comparison of hourly flux measurements difficult. Therefore, spatial analysismethods were developed to exploit this location bias and develop empirical modelsfor individual emissions processes in this urban setting. The models rely on re-lations between measured CO2 fluxes, observed environmental variables, and fluxsource area averaged plan-area land cover compositions and LAI. The results pre-sented here are specific to this site, but the methods and principles can likely beapplied to other locations with relatively uniform roughness and thermal characteras well as measurements of other greenhouse gases or pollutant fluxes.Modeling techniques developed here were able to take advantage of the patchy,variable urban surface to isolate individual emissions processes, as well as pro-vide ecosystem-wide, spatial averages of the net flux and high-resolution emis-sion maps. Modeled net emissions correspond well with measurements on daily(domain-scaled annual daily emissions totals are 16.93 g C day−1, within 6.9% ofmeasurements) and monthly timescales (within 5.7% of measurements). Modeleddomain-scaled annual flux totals of 6.42 kg C m−2 y−1 converge with results fromindependent bottom-up models (within 11%). Annually, modeled traffic emissions64b) Tra!c emissions545260154516515453551TowerE 41st AveE 49th AveKnight St.Victoria Dr.Fraser St.E 54th AveMemorial S. ParkGordon ParkTecumseh ParkTowerE 41st AveE 49th AveKnight St.Victoria Dr.Fraser St.E 54th AveMemorial S. ParkGordon ParkTecumseh Park545260154516515453551494290493340495240TowerE 41st AveE 49th AveKnight St.Victoria Dr.Fraser St.E 54th AveMemorial S. ParkGordon ParkTecumseh Park494290493340495240TowerE 41st AveE 49th AveKnight St.Victoria Dr.Fraser St.E 54th AveMemorial S. ParkGordon ParkTecumseh Parka) Space and water heating emissionsc) Respiration and photosynthesis d) Net annual emissionsN250 m N250 mN250 m N250 mFigure 2.10: Maps of total annual emissions at 20 m resolution for a) Spaceand water heating from buildings, b) Vehicle emissions , c) Respiration(human, soil, and vegetation) and photosynthesis, and d) Total annualemissions from all processes.are calculated to account for 68.8% of total net emissions, building sources con-tribute 27.9%, respiration from soil and vegetation is 5.5%, respiration from hu-mans 5.0%, and photosynthesis offsets are -7.2%.There remain uncertainties with using flux source area models in urban en-vironments, even in areas of uniform roughness. For example, uncertainties areintroduced due to below-canopy channeling effects along preferential street axesdepending on mean wind direction and from the variable heights within the canopythat pollutants are released. Vehicle emissions and respiration processes occur near65or at ground level while space-heating emissions are injected at roof level and veg-etation photosynthesis can occur both at the ground (grass) and above roof height(trees). Also, emissions in urban areas are usually associated with heated plumesand sufficient mixing by mean wind and turbulence is needed to avoid individualmicro-scale plume buoyancy effects. This aspect should also be considered whenselecting measurement height of the EC system.Additional uncertainties are introduced to statistical models from selective sam-pling bias of tower-based measurements during model development. For example,summer periods with traffic counts higher than winter were used to model year-round traffic emissions. Also, turbulent flux source areas were selected with pre-dominantly residential homes, instead of commercial buildings, to model build-ing emissions. The model also is inherently limited to CO2 sources and sinksfound within this neighborhood and plan-area landcover types included in the geo-spatial datasets. Therefore, there will be uncertainty resulting from exclusion ormis-classification of any relevant processes not represented by a specific landcovertype, for example emissions from underground sewage networks. These uncertain-ties imply that a combination of empirical models based on direct measurementsand inventory-based bottom-up modelling approaches should be used to quantify,monitor, and develop models of neighborhood net CO2 emissions. Despite un-certainties, spatially-averaged measurements have potential to provide validationfor bottom-up models (e.g. Christen et al. 2011) and are a means for real-timeemissions monitoring. In turn, bottom-up models can be calibrated against mea-surements and used to model additional neighborhoods or future scenarios.This study demonstrates that high resolution source area modeling in urbanareas is a promising technique to leverage more information out of urban long-termflux datasets. The location of the ‘Vancouver-Sunset’ flux tower is particularlysuited to this type of analysis because of homogenous surroundings in terms ofroughness and thermal properties, relatively low-density open canopy structure,and the heterogeneous distribution of CO2 sources and sinks and their positionsrelative to the tower.66Chapter 3Micro-scale horizontal transects3.1 IntroductionEddy covariance (EC) has been established as a robust technique to directly mea-sure net carbon dioxide (CO2) surface-atmosphere exchange, or flux density, inurban areas in recent decades. The first published flux measurements of CO2 bymeans of EC were from Chicago in 1996 (Grimmond et al., 2002), and there arecurrently at least 30 urban CO2 flux sites in operation worldwide (Velasco andRoth, 2010). While measurements of the net vertical turbulent CO2 flux have beenreported for a range of urban sites worldwide, relatively little attention in thesestudies has been given to determination of non-turbulent advective fluxes and howchanging mean CO2 mixing ratios within the air volume between ground and mea-surement level affect measured fluxes (i.e. storage change terms). These termshave long been recognized as sources of uncertainty in EC measurements aboveforest ecosystems on timescales of typical flux-averaging periods (30-60 minutes)(e.g. Goulden et al. 1996) and air-column storage has been shown to account forup to 60% of individual hourly net ecosystem change (NEE) values (Yang et al.,1999). Given the application of EC methods in urban neighborhoods to measureCO2 emissions and validate emission inventories and models at high spatial andtemporal resolution, a better quantitative understanding of storage and advectivefluxes in urban ecosystems and evaluation of measurement techniques is needed.The focus of this work is on storage changes of CO2 within and above the urban67canopy layer (UCL) using measurements observed during a case-study in a resi-dential neighborhood in Vancouver, BC, Canada.The EC technique relies on a mass balance framework for a measurement vol-ume that extends from the surface up to a height in the surface layer above theinertial roughness sublayer (e.g.Finnigan et al. 2003). Following Reynold’s av-eraging and neglecting molecular diffusion, the conservation of mass for a scalarquantity such as CO2 (c), at a point within this volume can be expressed as:SC =￿∂ (u￿c￿)∂x +∂ (v￿c￿)∂y +∂ (w￿c￿)∂ z￿￿ ￿￿ ￿I+∂c∂ t￿￿￿￿II+￿u∂c∂x + v∂c∂y +w∂c∂ z￿￿ ￿￿ ￿III(3.1)where Sc is the net source term (emissions or uptake), term I is the mean turbulentflux divergence transported by u, v, and w wind velocity components in the x, y,and z directions, respectively, term II is the storage term, and term III representsadvection by mean wind components. Typically, there is an assumption of well-mixed and horizontally homogenous flow, so that horizontal mean gradients ( δcδx ,δcδy ) and horizontal flux divergence (δ (u￿c￿)δx ,δ (v￿c￿)δy ) are negligible. Furthermore, themeasurement volume coordinate system is rotated to align with the mean flow foreach flux averaging period so that the mean vertical wind velocity (w), and hencevertical advective flux, are also equal to zero.After these simplifying assumptions and integrating for a finite volume thatextends from the surface up to a measurement height z, the source term for CO2can be evaluated as:SC = ρaw￿c￿z+￿ z0∂cρ∂ t dz (3.2)where c is the CO2 molar mixing ratio (µmol mol−1 or ppm), ρa is the dry molar airdensity (mol m−3), and cρ is CO2 partial density (µmol m−3). In practice, the pointmeasurement of the vertical covariance w￿c￿ at height z is relatively straightforwardto measure and this time-averaged quantity is assumed to represent an area-averageat the top of a measurement volume, parallel to the surface. Over periods longerthan a day, the integrated storage term (second term on RHS) is negligible com-pared to the vertical flux (i.e. no long-term accumulation or depletion of CO2),so the measured vertical flux is interpreted as the net source (i.e. net ecosystem68exchange, NEE, in natural ecosystems). On hourly timescales, however, there isoften storage change within the measurement volume so that the measured w￿c￿does not reflect the ‘true’ SC surface source or sink activity.To measure the storage term (FS), researchers at forest or agricultural sites oftenuse vertical profile systems to measure and integrate changes in mean CO2 partialdensity (∆cρ ) at several heights (z) across the flux averaging period (t) (e.g. Aubinetet al. (2005)):FS =1tn∑i=0∆cρzi (3.3)A vertical profile is used because of different rates of CO2 buildup (or depletion) atdifferent heights within and above a canopy. For example, during stable conditionsthere is potential for a near-surface inversion where in-canopy and above-canopyflows become de-coupled leading to ∆cρ∆t divergence. When a profile system is un-available, researchers have used the change in CO2 concentration at a single height(usually the same height where w￿c￿ is measured) to calculate FS, with the assump-tion that ∆cρ∆t is constant throughout the depth of the measurement volume (e.g.Hollinger et al. 1994, Black et al. 1996). This assumption has been tested in aDouglas-fir forest ecosystem and was found to produce equivalent FS compared tocalculations using four measurement levels at different heights, except for differ-ences of about 10% during mid-morning and mid-afternoon (Morgenstern et al.,2004).Using CO2 concentration measurements to calculate FS is potentially problem-atic for several practical and theoretical reasons. In theory, the storage term couldbe calculated from the instantaneous concentration differences between the endand beginning of the flux averaging time, but these instantaneous measurementsare easily biased by an errant gust or eddy penetration into a canopy where mea-surements are taken. To resolve this, average CO2 concentrations centered on thebeginning and end of the flux averaging time are typically used instead, but thismethod can underestimate storage by over 50% by acting as an averaging filter dueto the choice of averaging time period and instrument switching rate (Finnigan,2006).Additional complications arise from the different source areas influencing CO2concentration sensors at different heights in a profile system. Sensors lower down69in the profile will be more influenced by the immediate micro-scale tower surround-ings while sensors higher up are representative of a larger concentration source areaupwind of the tower. Integrating measurements from sensors at different heightsrepresentative of different source areas presents a potential scale conflict.Furthermore, CO2 flux densities and CO2 concentrations are the result of dif-ferent atmospheric processes and different source areas even when measured at thesame point. Absolute concentrations at any given moment are the result of localprocesses (biogenic and anthropognic emissions, vegetation uptake) modified byboundary layer-scale advection and entrainment processes, superimposed on a re-gional and global background signal resulting from processes acting on seasonal(e.g. Northern hemisphere summer vegetation uptake), decadal - multi-decadal(e.g. anthropogenic emission trends), and geologic (e.g. continental weathering,climate variation) time-scales. In contrast, CO2 flux measurements ideally filter outall longer-scale influences by averaging over a relatively short time period (typi-cally 30-minutes) representing boundary layer-scale eddy circulations. At typi-cal 30-minute flux-averaging timescales, concentration scalar source areas can belarger than turbulent flux source areas by an order of magnitude (Schmid 1994).Despite these issues, researchers in forest ecosystems have found FS to be animportant factor on flux-averaging timescales, particularly at night during stableconditions. A comparison of six forest flux sites in Europe found that 20-80% ofcarbon released by nocturnal respiration sources is stored in the air volume belowmeasurement height, depending on the slope surrounding the measurement site(Aubinet et al., 2005). Slope can be important because of removal of stored car-bon due to horizontal advection by slope drainage flows (e.g. Froelich and Schmid2006, Feigenwinter et al. 2008, Leitch 2010). FS has also been found to vary ac-cording to friction velocity (u∗), as a variable representing turbulent mixing, withnegligible storage at higher u∗ values. As a result, researchers in forest ecosystemsoften use a visually determined u∗ threshold to filter out periods of low turbulenceand high storage that can be gap-filled using a variety of methods (e.g. Falge et al.2002). This procedure however has potential to introduce bias to long-term esti-mates of NEE (SC) through subjective choice of u∗ threshold and potential double-counting of CO2 stored in the measurement volume during gap-filling when thatCO2 is eventually vented from the canopy and measured by the EC system (Gu et70al., 2005).In an urban environment, the physical form of the canopy and the distributionof heat sources is expected to alter the atmospheric processes affecting FS. Thepresence of anthropogenic heat sources and relatively large heat storage releasesduring night are likely to contribute to more frequent periods of instability and indenser urban areas result in a shallow, weakly convective mixed layer overnight(Christen and Vogt, 2004). Similarly, the increased roughness of the urban surfaceis hypothesized to result in greater mechanically produced turbulence and mixingcompared to porous vegetation canopies. The relatively open urban canopy struc-ture compared to forests could result in greater vertical coupling between canopyand above-canopy conditions, but at the same time urban form can cause horizon-tal de-coupling (e.g. between different streets and between backyards and streets(Weber and Weber, 2008). Additionally, the greater overall strength and spatialvariability of surface emission sources (e.g. vehicles, buildings) will affect themagnitude and relative strength of FS relative to net CO2 flux (FC).Several studies have documented observational evidence of CO2 storage andventing (i.e. negative storage) from the urban canopy layer (UCL). In Tokyo, Japan,vertical profile measurements of CO2 concentrations within a street canyon showan increase in UCL CO2 of over 40 ppm relative to above-canopy measurementsduring stable, wintertime conditions. Similarly, in Basel, Switzerland, UCL accu-mulation of CO2 and differences between a street canyon and above-canopy mea-surements up to 15 ppm are observed during late afternoon and overnight periodsduring summer (Vogt et al., 2006). Venting between the UCL and the surface layerabove the canopy is also observed in Marseille, France (Salmond et al., 2005).Here, discontinuous overnight ‘bursts’ of CO2 from within a canyon are shown tobe related to intermittent convective plumes of relatively warm canopy-layer airbuoyantly escaping the UCL. Together, these results indicate that the mechanismsfor CO2 buildup and venting are dynamic and vary according to a combination ofsite-specific atmospheric conditions (stability, energy balance, wind speed, winddirection), built form (canyon depth and orientation), and timing and magnitude oflocal emissions processes.Despite these findings, consideration of storage in the urban CO2 flux litera-ture is relatively rare. In Edinburgh, Scotland, an hourly storage flux correction71term is calculated using the single-level method and is found to modify the hourlymeasured flux by 11%, on average (Nemitz et al., 2002). In Basel, Switzerland, thestorage flux is calculated from a vertical profile system measuring ∆cρ at ten differ-ent heights in an individual street canyon. These measurements found storage to beparticularly relevant during the morning when the onset of thermal mixing ventedCO2 from within the canyon and modifies the measured flux by -23% (Feigenwin-ter et al., 2012). In Baltimore, USA (Crawford et al., 2011), and Beijing, China(Liu et al., 2012), the storage term is acknowledged but assumed to be negligiblebased on frequent overnight urban instability and a primary focus on monthly andannual exchange totals. Several other studies report buildup of CO2 concentra-tions on diurnal cycles or recognize potential for measurement uncertainty due tostorage, but do not explicitly attempt a storage correction (Grimmond et al., 2002;Coutts et al., 2007; Velasco et al., 2005; Helfter et al., 2011; Pawlak et al., 2011).In a review of urban CO2 flux literature in 2010, there is no mention of storage inguidelines given for processing urban EC flux data (Velasco and Roth, 2010).Despite the theoretical challenges in calculating FS from concentration mea-surements, an examination of a practical, working approximation is needed forurban areas nonetheless. Given the potential for extreme micro-scale horizontalspatial heterogeneity in terms of CO2 concentrations, use of a tower-based verticalprofile system is likely impractical for most urban measurement locations becauseof source area differences between sensors and resulting scale conflict when in-tegrating profile measurements vertically. A solution that is technically easier toimplement and avoids this scale conflict is use of concentration changes from asingle CO2 concentration sensor located above the roughness sublayer, at the sameheight as (and usually available from) the EC system. The underlying assumptionwith this method is that of vertically consistent changes in CO2 concentration fromthe measurement height down to the surface. This assumption is potentially validduring neutral and unstable conditions when canopy-layer and above-canopy con-ditions are well-coupled (also when FS is likely negligible), but could lead to anunderestimation of FS (and net emissions) if the concentration change below mea-surement height is greater than at measurement height, and an overestimation ifthere is greater buildup above-canopy relative to the canopy. However, this methodis also subject to scale-related uncertainties because of the differences in source72areas between scalar concentrations and turbulent fluxes.The objectives of this work are to measure the micro-scale variability and rateof change of UCL CO2 mixing ratios (c) in a residential neighborhood in Van-couver, BC, Canada. These measurements are used to calculate the magnitude ofFS on hourly timescales to test the assumption that a single measurement heightis representative of the entire air volume below measurement height. This is ac-complished through use of vehicle-mounted CO2 measurement system to capturehorizontally spatially-averaged concentrations within the UCL in a 2.5 km2 urbanarea representative of the longterm turbulent flux source area of an EC tower.Mobile measurements of CO2 concentrations and mixing ratios have been con-ducted previously in several cities to observe urban-scale concentration patterns,and statistical analysis has shown that the method can yield reliable, reproducible,and representative results (Henninger and Kuttler, 2007). In Phoenix, AZ, vehi-cle mounted CO2 sensors were driven on transects across the metropolitan areato characterize the ‘urban CO2 dome’ (Idso et al. 1998, Idso et al. 2001). In Es-sen, Germany, urban-rural differences were analyzed by measuring along the sameroute over a range of seasons, time of day, days of week, and meteorological con-ditions (Henninger and Kuttler, 2010).3.2 Methods3.2.1 Study areaThe observation program was conducted in a 1600 × 1600 m urban area in the‘Sunset’ neighborhood of Vancouver, BC, Canada (centerpoint coordinates are123.0784◦ W, 49.2261◦ N, WGS-84). The neighborhood is classified as LCZ-6‘open-set lowrise’ (Stewart and Oke, 2012) and is a primarily residential area withan average population density of 63.1 inhabitants ha−1. Over 90% of buildings aresingle, detached, residential structures with a mean building density of 12.8 build-ings ha−1 and mean building height of 5.1 m (van der Laan, 2011). There are alsoseveral expansive public parks with grass fields and mature trees within the studyarea and overall tree density is 17.1 stems ha−1 within a 500 m radius of the tower(Liss et al., 2010). Mean treetop height for the entire study area is 7.7 m.73Figure 3.1: Map of the Sunset neighborhood study area. Plan area landcoverhas been determined from remotely sensed datasets. The transect routestarts and ends in Memorial Park in the NW portion of the study areaand the route passes through 64 200 x 200 m grid cells. The cumulativeturbulent flux source area for the 7-8 Sep. 2013 observation period isalso shown on the map.74The neighborhood is organized into blocks delineated by relatively large, busyroads approximately every 800 m. Based on weekday traffic counts (City of Van-couver, 2012), the busiest of these arterial roads (Knight Street and 41st Street)transport daily traffic loads upwards of 20,000 cars through the study area and sec-ondary arterial roads (49th Ave and 57th Ave) carry 10,000-20,000 cars per day.The block interiors are accessed via networks of smaller, residential side-streetsand back lanes with weekday daily loads of less than 1,000 vehicles per day.The 1600 × 1600 m area of the transect is a subset of the 1900 m × 1900 mdomain of previous modeling and spatial analysis studies conducted in the neigh-borhood (Figure 3.1). The 1900 m × 1900 area encompasses over 80% of thetower’s long-term turbulent flux source area and corresponds to coverage by satel-lite imagery and a Light Detection and Ranging (LiDAR) dataset (Crawford andChristen, 2014). These remotely sensed datasets have been used to extract plan-area landcover and 3-d built and vegetation characteristics at 1 m resolution (Tookeet al. 2009, Goodwin et al. 2009). Plan-area landcover fractions in the 1900 m ×1900 m study area are 29% building, 11.8% tree, 20.3 % lawn, and 43.6% imper-vious.The tower and surrounding neighborhood have been the site of a number of pre-vious CO2-related measurement campaigns and modeling studies (Reid and Steyn,1997; Walsh, 2005; Christen et al., 2011; Kellett et al., 2012). Modeling resultsand EC observations report spatially-averaged annual emissions of 5.73 - 6.42 kgC m−2 year−1 for the 1900 m × 1900 m study area. The net annual emissionsare dominated fossil fuel emissions from motor vehicle and home heating sources(87-96% combined), with an additional 10-12% from human, soil, and vegeta-tion respiration, with -7 to -9% offset by vegetation photosynthesis (Kellett et al.(2012), Crawford and Christen (2014)). The urban surface has been found to be anet CO2 source for all hours and all seasons on average despite CO2 sequestrationby vegetation. Average measured daily emissions range from 16.0 g C m−2 day−1in August to 22.1 g C m−2 day−1 in December (Christen et al., 2011).753.2.2 Tower-based measurementsThe study area is centered on an urban flux tower located within the grounds of apower substation (tower coordinates are 123.0784◦ W, 49.2261◦ N, WGS-84). Thetower is located on the SE corner of the substation grounds that extend 100 m tothe west and 130 m to the north of the tower. The tower is an open, triangular,lattice tower and the base is recessed by 4 m from the surrounding terrain. Netmixing ratios, molar densities, and fluxes of CO2 (FC) have been continuouslymeasured at the tower sinceMay, 2008 at an effective height of 24.8 m a.g.l. using asonic anemometer (CSAT 3-d, Campbell Scientific, Logan, UT, USA) and an open-path infrared gas analyzer (IRGA, Li-7500, Licor Inc, Lincoln, NE, USA). TheIRGA was calibrated in-house every 6 months during the study period according tostandardized procedures (Li-Cor, 2002) against reference tanks from EnvironmentCanada.Three-dimensional wind velocities and CO2 concentrations were recorded at20 Hz and subject to several quality control procedures (e.g. filtered for interfer-ence from precipitation, realistic max/min thresholds). Fluxes are then calculatedfrom block-averaged means of 30-minute periods and a 2-d coordinate rotation isperformed to align the coordinate system with the mean wind direction. Fluxes areadditionally corrected for density effects (Webb et al., 1980) and sensor separationbetween the sonic anemometer and IRGA (Moore, 1986). Full EC quality con-trol and data processing details are described in Crawford and Christen (2014) andCrawford et al. (2010).3.2.3 Mobile measurements in the UCLCO2 mixing ratios were measured from a mobile, vehicle-mounted platform over acontinuous 26-hour period from 0500 LST Wednesday September 7, 2011 to 0700LST Thursday September 8, 2011. Synoptically, the region was under the influ-ence of a high pressure system and clear-sky, calm conditions were selected forgreatest overnight actual stability. Based on FC observations and spatial modelingresults presented in Chapter 2, mobile measurements were conducted mid-week toensure enhanced traffic emissions and therefore greatest FS. EC tower-based mea-surements at 24.8 m a.g.l. recorded a mean air temperature of 22.0◦C (minimum76= 16.0◦C at 7 Sep 0640 LST, maximum = 27.4◦C at 7 Sep 1640 LST) and meanwind speed of 2.0 m s−2 during the study period.The transect route was designed to measure along arterial roads, residentialside-streets, alleys, and urban parks and to be completed in approximately onehour so that assumptions of atmospheric stationarity could be plausibly applied.Transects were driven every second hour, except during sunrise when they weredriven every hour to better resolve CO2 venting from the UCL. Measurementstimed to correspond with the 30-minute flux-averaging period of the tower. Intotal, 15 individual transect runs were completed (Table 3.1).77Table 3.1: Summary of transect runs from Sep 7 - 8, 2011. Times are in LST (PST) and runs started at least 5-minutesbefore the hour to establish baseline air temperature values. a UCL 200 × 200 m spatial average, b Mean valuemeasured by the EC system at 24.8 m a.g.l., c UCL linear correlation coefficients between 20 × 20 m grid cells oftemperature and CO2 differences from run mean.Run Start Time End Time CO2 (ppm)a θ (C)a W.S. (m/s)b W.D. (◦)b u∗ (m/s)b z￿/Lb θ vs. Cc1 4:50 5:50 486.70 14.16 1.40 23.0 0.26 0.66 0.242 5:55 6:50 504.13 14.15 1.51 37.6 0.31 -0.57 0.213 6:55 8:05 474.84 15.96 0.59 322.8 0.17 -0.81 0.674 8:55 10:00 448.48 20.63 1.49 170.2 0.21 -1.87 0.345 10:55 11:55 424.35 24.37 2.30 242.0 0.37 -1.72 0.586 12:55 14:00 405.42 26.59 2.50 294.7 0.51 -1.04 0.607 14:55 16:10 402.04 28.33 3.02 296.6 0.26 -0.73 0.598 16:55 18:05 407.68 27.89 2.56 304.6 0.25 -0.57 0.659 18:55 20:10 438.48 22.88 2.39 340.2 0.20 0.20 0.1410 20:55 22:05 461.09 20.84 2.77 38.0 0.19 -0.33 -0.1911 22:55 23:55 460.64 18.83 2.25 52.2 0.34 0.98 -0.4912 0:55 1:55 457.30 17.34 2.07 26.2 0.14 0.16 -0.7613 3:00 4:00 469.49 15.66 1.44 7.5 0.21 -214.82 -0.5114 4:55 6:00 482.89 15.37 1.01 262.6 0.15 -16.98 0.3115 6:00 7:05 505.80 11.80 0.85 230.4 0.13 1.82 0.2278The total length of each individual transect is 22.3 km, of which 14.8 km(66.4%) are along residential sidestreets and alleyways, 5.9 km (26.5%) along arte-rial roads, and 1.6 km (7.2%) along roads that go through, or immediately adjacentto, parks. The transect route is limited to measurements above roadways and doesnot sample directly above lawns or other vegetated surfaces. This bias is partiallymitigated by segments that traverse through parks and from expected horizontalmixing of air above roadways and adjacent vegetated areas. This horizontal mixingin the UCL can result from advection by mean winds, traffic-generated mechanicalturbulence, and micro-scale thermal circulations between roads and lawns. There-fore, it is expected that observed above-road mixing ratios are representative of thetotal above-road and above-lawn outdoor air volume in the UCL.Ambient CO2 mixing ratios were measured with a Li-820 closed-path IRGA(Licor Inc., Lincoln, Nebraska, USA) from an intake port at 2.1 m height. Theintake port was mounted on the outside of the measurement (Toyota Tacoma truck)vehicle 2.0 m ahead of the rear exhaust pipe to avoid contamination from it’s ownemissions and facing backwards to reduce static pressure changes in the sampletube from vehicle acceleration. Air was drawn to the IRGA in the truck interiorthrough a 1 m length tube (3 mm internal diameter) and filter (1.0 µm PTFE Acro50, Pall Corporation, Washington, New York, USA) by a pump (KNF unmp015b,KNF, Freiburg, Germany) at a flow rate of 500 cc min−1. CO2 mixing ratios weresampled every 1 s and recorded on a datalogger (21x, Campbell Scientific, Logan,Utah, USA). Air temperature was also sampled at 1 s with an actively ventilatedfine-wire thermocouple (T-type) located at the intake port and recorded on the 21xdatalogger. Latitude, longitude, elevation, and ground speed data were determinedevery 1 s with a GPS receiver (Garmin 76s, Garmin, Inc., Olathe, Kansas, USA)and roof antenna (GA 25MCX, Garmin, Inc.). Data from the GPS and 21x datalog-ger were merged with a custom program using LabView 8.6 (National Instruments,Texas, USA) which allowed on-line visualization of incoming data. All data arereported using Local Standard Time (LST is equivalent to Pacific Standard Time,not Daylight Savings Time).793.2.4 Data processing and spatial averagingThe 1-s CO2 and air temperature samples were subject to several quality controlprocedures before analysis. Raw data were withheld from analysis if individualpoints were greater than 5 standard deviations from the transect mean. This thresh-old was determined visually to eliminate infrequent, physically implausible spikesin CO2 mixing ratio. Individual points were also withheld during occasions that theGPS receiver either could not receive a location signal due to overhead obstructions(e.g. dense tree canopy) or recorded a location error in the GPS internal diagnosticvalue (e.g. from satellite errors or reflected multipath signals). Raw data were alsowithheld when vehicle speed was less than 5 km hr−1 to avoid contamination bythe vehicle’s own exhaust or from nearby vehicles while idling in traffic.A 5-minute average measured at the beginning and end of each transect sur-vey at the same location in Memorial Park was used to obtain baseline values forair temperature and a linear trend calculated between the beginning and end aver-age was removed from the temperature signal during the traverse. The purpose ofthis procedure is to remove background temporal trends to more clearly assess thespatial patterns along the transect. This procedure assumes both a linear rate oftemperature change and a uniform rate of change across the entire study area. Toassess the magnitude of uncertainty introduced from the assumption of linear tem-perature change, measurements from a stationary UCL temperature sensor installedat 2 m height in a representative backyard in the study area were used. At this sta-tion, 5-minute air temperatures were linearly interpolated between observations atthe beginning and end of each transect period. The interpolated air temperaturesdiffer by 1.2% on average from measurements at this station. Next, the rate of tem-perature change between the Memorial Park location and stationary UCL stationare compared. On average, the linear rates of temperature change differ by 0.66K hr−1. To control for adiabatic influences on air temperature from topographicvariations in the study area, de-trended air temperatures are converted to potentialtemperature (θ ) at 0 m a.s.l. using the dry adiabatic lapse rate of 0.0098 K m−1(Stull, 1988) and elevations recorded by the GPS.The de-trending procedure is not done for CO2 because of potential uncertain-ties and practical difficulties in determining a background CO2 trend. The above-80canopy CO2 measurements by the EC system may be unsuitable because they areexpected to be de-coupled from the UCL during stable conditions and be generallyrepresentative of a different source area than UCL measurements. Furthermore,this would utilize the very assumption of vertically constant ∆c/∆t that this workaims to test. Determining average trends from the UCL measurements is difficultdue to the extreme micro-scale differences in CO2 source and sink strength andbehavior. For example, a trend measured in an urban park is likely not applicableto an area containing a busy intersection because of different source and sink pro-cesses (vegetation photosynthesis and biogenic respiration vs. traffic emissions).Therefore, the operating assumption is that the magnitude of CO2 mixing ratiospatial variability in the UCL will render any background temporal trends over thetime of the each individual traverse negligible. For ∆c/∆t calculations betweenindividual transects, the de-trending is also unnecessary because the ∆t betweentransects is roughly equal at any individual point along the route.For this study, spatial averages of CO2 (￿c￿) and potential temperature (￿θ￿)are preferable to temporal averages due to variable transit times across the transectroute throughout the day. To accomplish this, first a 20 m × 20 m grid approxi-mating street width was overlaid on the study area and sample points falling withineach grid-cell are averaged. This resolution is used for analysis and visualizationof micro-scale variations in air temperature and CO2. In a second step, the 20 m× 20 m grid cells containing CO2 or air temperature values are averaged to make200 m × 200 m grids. This resolution ensures an equal number of data pointsfor each transect run and is used to determine the average CO2 concentration forthe entire canopy-layer during a particular run and for comparison between runsdespite variable vehicle speeds. The transect route was designed to pass througheach of the 200 m× 200 m grid cells in the 1600× 1600 m study area at least onceand measurements over both arterial road and residential side-streets were requiredwithin grid cells that contained both road types.As noted earlier, a methodological challenge with FS calculations is to inte-grate turbulent flux and scalar mixing ratio measurements that are representative ofdifferent source areas. In an urban environment with a heterogeneous distributionof CO2 sources and sinks at the scale of turbulent flux source areas, changes inwind direction and stability that alter source areas and therefore measured fluxes81and mixing ratios should also be taken into account. To incorporate the affects ofchanging wind directions and atmospheric stability on tower-based observations,the turbulent flux source area influencing EC tower measurements was calculatedfor each transect period. It is recognized that the turbulent flux source area is notthe same as the scalar concentration source area, however the use of the source areamodel provides a means to estimate the effects of changing flow conditions on thetower-based instruments. The flux source area model used here is a 2-d crosswinddispersion and gradient diffusion model by Kormann and Meixner (2001). Modelinputs for each transect run are mean wind direction (measured at the tower), sur-face roughness and displacement lengths (determined morphometrically), lateraldispersion (measured standard deviation of crosswind velocity), and atmosphericstability (Obukhov length) determined at EC system height. The Obukhov length(L) is calculated as:L =−u3∗TvkgQH(3.4)where k is von Karman’s constant (0.4), g is gravitation acceleration (9.81 m s−2),and u∗ (friction velocity, m s−1), TV (virtual temperature, K), and QH (sensible heatflux, W m−2) are measured at the tower by the EC system (Crawford and Christen,2014).The turbulent flux source areas are calculated at the same temporal resolutionas the measured CO2 flux. For each 60-min transect run, two equally-weighted30-minute turbulent flux source areas are calculated at 2 m × 2 m resolution andeach pixel is weighted by its contribution to the measured tower signal. The 200m × 200 m spatial averages of UCL CO2 mixing ratio were then overlaid andmultiplied by the source area weights (φ ). Average CO2 mixing ratio weighted bythe turbulent flux source area is determined by summing the result over the 1600× 1600 m study area domain. For weighted flux source area pixels that fall outsideof the study area, the mean of the entire 1600 × 1600 m domain is used so that thesummation of φ is equal to unity.823.2.5 Storage calculationsFS for a specific volume is calculated as:FS =∆cρ∆t z (3.5)where z is the measurement volume (m−3 m−2) and ∆cρ∆t is the measured changein CO2 molar density (µmol m−3) over time t (s). For the mean UCL ∆cρ , molardensity of CO2 is calculated by dividing the UCL mean mixing ratio (µmol mol−1)by molecular weight of air (m =28.97 g mol−1) and multiplying by dry air density(g m−3). Density is calculated using regional air pressure (p (Pa)) measured atVancouver International airport corrected for mean transect elevation, measuredmean air temperature (T (K)), and the gas constant r=8.314 J mol−1 K−1 (ρ = pmTr ).Five variations of z are used in this study to calculate the storage flux (Figure3.2). The first method (FS1) uses a point measurement of ∆cρ at the EC towertop which is applied over the entire measurement volume from the ground up tothe measurement height (z). This method assumes a constant ∆cρ throughout theentire measurement volume. The exact measurement volume is determined by sub-tracting the mean building volume (1.6 m−3 m−2) and mean tree volume (0.8 m−3m−2) per unit area in the study area from the air volume up to the EC measurementheight (24.8 m−3 m−2). The building volume is not included because the indoor airvolume is not subject to the same turbulent mixing processes as the outdoor air vol-ume. CO2 emissions released indoors (e.g. from human respiration, or natural gascombustion during cooking) are expected to remain in storage within the buildinguntil they are released to the UCL through ventilation from the buildings. Us-ing a building code minimum ventilation rate of 0.35 m−3 s−1 (ASHRAE, 2004),the entire volume of indoor air, including CO2 is expected to be exchanged withoutdoor air every 2.86 hours. Analysis of LiDAR data indicates 27.7% of singlefamily dwellings in the study area are built after 1990, 44.7% between 1965-1990,and 27.6% before 1965 (van der Laan, 2011). Though actual ventilation rates areexpected to vary with building vintage, the assumption is that exchange of theindoor-air volume takes place on timescales longer than the flux averaging period.To match the transect periods of this observation period, ∆cρ at the tower is cal-83Figure 3.2: This schematic illustrates the FS calculation methods compared inthis study. Curved lines represent hypothetical vertical CO2 gradients,points represent CO2 measurements, and shaded areas represent storagein the measurement volume. Tree and building volume are not includedin the measurement volume for the FS calculations.culated as the difference between mean CO2 concentrations measured at the towerduring each transect and ∆t is the duration (s) between the temporal centerpoint ofeach transect run.The second method (FS2) is a two-layer approach which adds together resultsfrom Eq. 6 calculated for two layers (z1 and z2) and using two different∆cρ∆t values.The first layer volume (z1) is defined as the surface up to the mean treetop height,excluding tree and building volume, and uses the spatially averaged ∆cρ∆t measuredin the entire 1600 × 1600 m transect area. Trees are the highest canopy element inthis neighborhood and mean tree height in the study area is 7.7 m (z1=7.7-(1.6+0.8)m−3 m−2). The z2 layer volume extends from the top of the mean tree height up tothe measurement height of the tower EC system (z2=24.8-7.7 m−3 m−2). For thislayer, ∆cρ∆t is calculated from the CO2 molar density measured at a single locationby the EC system.The third method (FS3) uses∆cρ∆t measured in the UCL and attributes this changeto the entire single-layer volume used for FS1. This method assumes a constant ∆cρthroughout the entire measurement volume equal to that measured in the UCL.The fourth method (FS4) is equivalent to FS2, except the turbulent flux source84area-weighted ∆cρ values are used instead of spatially averaged UCL values overthe entire 1600× 1600 m study area. The fifth method (FS5) is equivalent to FS3, butuses turbulent flux source area-weighted ∆cρ values instead of spatially averagedUCL values over the entire 1600 × 1600 m study area.3.3 Results and discussion3.3.1 Spatial patterns of CO2 mixing ratios in the UCLThroughout the study period, CO2 mixing ratios measured in the UCL are greaterthan those measured on the tower (Figure 3.3a). The spatially averaged (mean of200 m grid cells) mixing ratio in the UCL is 48.8 ppm (11.9%) higher than thetower-top measurement during the entire study period on average. This is consis-tent with flux measurements at this site that show the surface to be a net source ofCO2 during both day and night, during all seasons (Crawford and Christen, 2014).During the entire observation period, mixing ratios for the grid cell including theintersection of Knight St. and 49th (grid cell E5) were on average 19.8 ppm (4.3%)higher than the UCL average. The mixing ratios measured in the Memorial Parkgrid cell (B2) are 10.2 ppm (-2.2 %) lower than the UCL average. On average, thesource area-weighted CO2 mixing ratio differs from the UCL average by only 2.9ppm (0.6%).For all hours, the maximum individual grid cell is greater than the minimumindividual grid cell by 16% (71.4 ppm) on average, and the mean spatial standarddeviation (between all 200 × 200 m grid cells) for all runs is 15.0 ppm. Observedspatial variability in the UCL is also greater than temporal variability measured atthe EC tower. The average temporal standard deviation of tower-based 5-minuteaverages during transect periods is 7.4 ppm and mean absolute change in CO2mixing ratio across each transect period is 12.3 ppm. This result is in line withthe expectation that above-canopy measurements are representative of a blendedaverage of micro-scale UCL conditions.Spatial patterns of CO2 mixing ratio vary through the study period. Duringthe daytime (0700-1800 h), mixing ratios in the Knight/49th grid cell (E5) are27.0 ppm (6.2%) higher than the UCL average and 36.7 ppm (8.7%) higher than85Figure 3.3: a) Diurnal course of UCL CO2 mixing ratios at 200 m grid res-olution during the study period (Sep. 7-8, 2011). b) Diurnal course of∆c/∆t during the transect study period. ‘Memorial Park’ refers to gridcell B2 and ‘49th/Knight’ refers to E5.86in Memorial Park (B2) (Figure 3.4). During this time, the 200 m grid cell withhighest average CO2 mixing ratio contains the intersection of Knight St. and 41stAve (E1). The daytime mean mixing ratio for this cell is 460.4 ppm and is onaverage 33 ppm (7.7%) higher than the UCL run mean. The grid cell with thelowest average mixing ratio during daytime (415.3 ppm) is found near MemorialPark (B2) and is on average -11.8 ppm (-2.7%) below the UCL run mean. Thisimplies that during the day, CO2 mixing ratios in the UCL are largely a functionof proximity to vehicle traffic sources during weekday, summertime conditions.This is consistent with results presented in Chapter 2, in which FC observationsand spatial emissions models show highest emissions originating from roads, inparticular intersections.Overnight, measured CO2 mixing ratios reflect diminished traffic volume andgreater relative magnitude of emission processes such as soil and vegetation res-piration. From 2300-0400 h, mixing ratios in the Knight/49th (E5) grid cell fallbelow the UCL mean (-6.7 ppm, -1.5%) and from 2300-0200 h, mixing ratios mea-sured in Memorial Park (B2) surpass those measured in the Knight/49th (E5) gridcell (15.6 ppm, 3.5%) and the UCL mean (6.6 ppm, 1.4%) (Figure 3.3). Spatially,highest overnight (2100-0400 h) concentrations are found in the vicinity of Tecum-seh Park (H3) and are on average 493.8 ppm, 32 ppm above the nighttime UCLmean (Figure 3.4). This is consistent with results from FC observations and spatialemissions model results presented in Chapter 2 in which summertime overnightrespiration from soil and vegetation is an important source of CO2 and emissionsfrom vehicular traffic are at a minimum. During this night, air temperatures remaingreater than specific hourly heating thresholds determined in Chapter 2, meaninghome space heating systems are inactive and do not produce CO2 emissions.CO2 mixing ratios are correlated with potential air temperature during bothdaytime and nighttime (Figure 3.5). During daytime (0700-1800 h), there is a pos-itive correlation between 200 m grid cells (linear correlation coefficient R = 0.88).Reasons for this include co-located CO2 and anthropogenic heat emissions fromvehicle engine combustion along arterial roads, and large sensible heat flux overimpervious arterial roads relative to surrounding lawn and park surfaces due toreductions in surface water availability, transpiration, reduced shading from veg-etation. In this study area, residential side-streets are typically lined with mature87Figure 3.4: a) Mean UCL CO2 mixing ratio at 200 x 200 m resolution (c) dif-ference from individual run spatial average (￿c￿) during daytime (0700-1800), b) Mean 200 m potential air temperature (θ ) difference fromindividual run spatial averages (￿θ￿) during daytime, c) Average c−￿c￿during nighttime (2100-0400 h), d) Average θ −￿θ￿ during nighttime(2100-0400 h). Elevation contours (10 m) derived from LiDAR surfacereturns are overlaid on each map.88trees whereas arterial roads are not.During the overnight runs (2100-0400 h), there is a negative linear correlationbetween potential temperatures and CO2 mixing ratios at 200 m resolution (R =-0.57). To explain this feature, a more detailed inspection of individual runs at20 m spatial resolution is performed (Figure 3.6). Strongest overnight negativecorrelation at 20 m is observed during run 12, from 0100-0200 h (R = -0.76).During this run, the atmosphere was stably stratified (z￿/L = 0.16, mean u∗ = 0.1m s−1 measured at the EC tower). Spatially, 20 m resolution patterns of CO2mixing ratios and potential temperature are found to conform to the micro-scaletopography of the study area during this run, with higher mixing ratios and lowertemperatures concentrated in low-lying areas. Tecumseh Park (H3) has the highestCO2 mixing ratios (80 ppm above UCL mean) and lowest air temperature (1.5K below UCL mean) during this run and the park lies in a slight depression (-5m) relative to the surroundings. There is also a pool of cool, CO2-rich air in ashallow valley descending southeastwards from Memorial Park (B-C,3-7, D7-8).This region’s temperature is 0.2-1.5 K below the UCL mean and CO2 is 5-25 ppmabove the UCL mean. This suggests advection processes influenced by micro-scaletopography, i.e. cold-air pooling and drainage, can be a determinant of nocturnalUCL pollutant concentrations during de-coupled, stable conditions.Based on the linear correlation coefficient between 20 m resolution air temper-ature and CO2 mixing ratios for each run (Table 3.1), the cold air drainage featureappears to develop between run 8 (1700-1800 h) and run 10 (2100-2200 h) when Rdrops from 0.65 to -0.19. The negative correlation is strongest during run 12 (R=-0.76, 0100-200 h) and persists until run 14 (0500-0600 h), which spanned sunriseat 0549 h.Cold air drainage at a different scale has been documented in an individualstreet canyon byMoriwaki et al. (2006) using vertical profile measurements of CO2mixing ratios in Tokyo, Japan. In this instance, the cold air subsidence occurredovernight and flow was from relatively cold rooftops down to ground level. Thesecurrents transported CO2 released from rooftop vents above the canopy down toground level within the canyon where it was accumulated overnight.The magnitude of observed micro-scale spatial differences in the UCL dur-ing this observation period are comparable to results measured at larger, regional89Figure 3.5: Daytime and nighttime linear correlations between differencesfrom the transect spatial average for CO2 (c− ￿c￿) and air tempera-ture (θ −￿θ￿). The best fit linear equation, R2 value, and correlationcoefficient (R) for each period are given. Labeled individual points cor-respond to individual 200 m grid cells. Tecumseh Park refers to gridcell H3 and 41st/Knight refers to E190Figure 3.6: a) Nocturnal c−￿c￿ and b) θ −￿θ￿ overlaid on 5 m ground ele-vation contours (m a.s.l.) derived from LiDAR surface returns dataset.91urban-rural scales. In Essen, Germany, transect measurements measured averageurban-rural mixing ratio differences of 6.5% in summer and 11.3% in winter (Hen-ninger and Kuttler, 2010). Urban standard deviations during the Essen transectswere 17.6 ppm during summer and 21.3 ppm during winter. In Phoenix, Arizona,USA, measured urban mixing ratios during weekdays were on average 43% higherthan rural areas during a 2-week wintertime study period (Idso et al., 2001). InBaltimore, MD, 5-years of transects show urban mixing ratios higher than a ruralreference by 66 ppm (George et al., 2007) and mixing ratios in Portland, Oregonare 5-6 ppm above an up-wind rural reference station (Rice and Bostrom, 2011).3.3.2 Storage and venting from the UCLAs shown in Figure 3.3a and discussed in the previous section, spatially-averagedUCL CO2 mixing ratios are higher than those measured on the flux tower (24.8m). This is consistent with stationary profile measurements that show decreasingmixing ratios with height in Basel, Switzerland (Vogt et al., 2006). In Basel, thestrength of the vertical gradient (though not absolute mixing ratio values) is de-termined by traffic volume and is strongest within the UCL in a street-layer upto a height of approximately 20% of the street canyon depth. In Tokyo, Japan(Moriwaki et al., 2006), however, measured vertical profiles are homogeneous withheight during well-mixed conditions, because street-level traffic is negligible attheir site and the primary CO2 source is from roof-top building vents in the upperUCL. From this injection point, CO2 is vertically dispersed throughout the UCLby turbulent mixing during unstable conditions, but is transported to the surface bycold-air subsidence from relatively cold rooftops during stable conditions leadingto negative gradients overnight (Moriwaki et al., 2006).During the present study, the vertical gradient between the tower measurementsand source area-weighted CO2 mixing ratios in the UCL is correlated with verti-cal wind velocity variations (σw, standard deviation of vertical wind fluctuations)measured above the canopy at 24.8 m (Figure 3.7). When the atmosphere is sta-ble and σw decreases, mixing is suppressed, UCL-tower gradients are strongest(strongest negative gradient is -4.1 ppm m−1), and there is potential for storagebuildup of CO2. As stability, mixing, and σw increase, tower measurements be-92Figure 3.7: The vertical CO2 mixing ratio gradient (ppm m−1) as a functionof σw measured on the EC tower at 24.8 m a.g.l.. The dotted line is anexponential function of the form y = a0exp(a1x), where a0 =−4.93 anda1 = 3.51.come more tightly coupled with the UCL measurements and the magnitude of thenegative gradient is reduced (weakest negative gradient is -0.5 ppm m−1). Thisimplies there must be venting of built-up CO2 from the UCL during the transitionfrom stable to unstable conditions after sunrise.There is evidence of venting during the morning of Sep. 7. A timeseriesof run-to-run ∆c shows divergence between tower and UCL measurements from0700-0800, shortly after sunrise (0537 h) (Figure 3.3b). Initially at 0700 h, themean UCL CO2 mixing ratio has increased by 17.4 ppm hr−1 from early morningtraffic emissions injected into the stable surface layer, consistent with results fromChapter 2. By the next run at 0800 h, UCL mixing ratios have decreased to -13.593ppm hr−1 and mixing ratios at the tower top have risen to 13.1 ppm hr−1. By 1000h, both levels show a decrease in CO2 mixing ratios as the traffic morning rushhour wanes, relatively cleaner air from the residual layer is brought to the surfaceduring the growth of the daytime mixed layer. At the same time, photosynthesisincreases and vegetation takes up a fraction of emitted CO2, consistent with FCobservations and photosynthesis model results presented in Chapter 2. By 1200 h,UCL and above-canopy air appear well-coupled and remain so for the duration ofthe day. After sunset from 2000-2200 h, the UCL shows an initial increase in CO2relative to the tower (1.78 ppm hr−1 vs 0.62 ppm hr−1) and a small buildup againin the early morning (0400-0700 h).Spatially, patterns of UCL ∆c/∆t vary throughout the day at 200 m grid res-olution (Figure 3.8). During the initial storage buildup after sunset (1700-2200),∆c/∆t is distributed fairly uniformly across the study area and is always positive,ranging from 0.1 - 1.8 ppm hr−1 (Figure 3.8 a). Overnight after the UCL has cooledand the atmosphere has become stably stratified, spatial patterns of ∆c/∆t can beexplained by cold air drainage following the underlying topography of the studyarea (Figure 3.8 b). Grid cells with relatively high elevations (e.g. F5-6,G5-6,H6-8) show negative ∆c/∆t, even though there is no active photosynthesis to removeCO2, and grid cells with lower elevations show an increase in CO2 mixing ratio. Inparticular, depressions such as the Tecumseh Park area (G3-4) and the shallow val-ley extending southwards from Memorial Park (C4-8, D7-8, E7-8) show increasesin CO2.After sunrise and warming of the surface, venting of CO2 from the UCL isstrongest (-8 to -12 ppm hr−1) in the low-lying areas where cool, CO2-rich air haspooled overnight (G2-3, C4-8, D5-8, E7-8) (Figure 3.8 c). All other grid cells alsoshow negative ∆c/∆t and values range from -1.9 - 12 ppm hr−1. After the flushof built-up CO2, daytime (1100-1600) ∆c/∆t are all negative (0 to -2.5 ppm hr−1)with lowest values in the SW and highest values in the NE, generally conformingto topography.The measured ∆cρ/∆t is then converted to mass storage change in the UCLvolume (excluding tree and building volume) using Equation 6. The magnitudeof UCL FS is small using either the mean of the entire UCL, or the source area-weighted mean. Mean canopy-layer FS using the entire UCL is -0.04 µmol m−294Figure 3.8: a) Mean 200 m UCL ∆c/∆t (ppm hr−1) from 1700-2200 h, b)2300-0000 h, c) 1100-1600 h, and d) 0700-1100 h. Elevation contours(10 m) derived from LiDar surface returns are overlaid on each map.s−1 (maximum = +1.1 µmol m−2 s−1, minimum = -0.82 µmol m−2 s−1). Usingthe source area-weighted values, mean canopy-layer FS is -0.08 µmol m−2 s−1 andthe range is from -0.57 to +0.16 µmol m−2 s−1. The mean absolute differencebetween the two methods for all hours is 0.15 µmol m−2 s− Integrating UCL and above-canopy storageVehicle transect measurements illustrate horizontal CO2 gradients as well as CO2storage in, and venting from, the UCL. The objective of this section is to incorpo-rate these measurements into calculations of the hourly storage flux term (FS) for95Figure 3.9: Storage (FS) calculated during the Sep. 7-8 study period usingfive variants of measurement volume (Figure 3.2).the entire measurement volume to investigate their affect on local-scale EC obser-vations. Five variations of FS are compared to test assumptions regarding constant∆c/∆t with height and assess the uncertainty using the FS1 method (Figure 3.9).Divergence between FS calculations is greatest during the UCL venting at 0800h, when FS3 and FS5 show negative values (-3.4 and -2.4 µmol m−2 s−1 respec-tively), and FS1,2,4 are positive (+1.74 to +3.37 µmol m−2 s−1). This spread isdue to negative measured ∆c/∆t in the UCL and positive ∆c/∆t measured abovethe canopy on the tower during this time. During the afternoon (1300-1900 h), allmethods track closely and steadily increase from -1.0 to 0.0 µmol m−2 s−1. Af-ter sunset (1900-2300 h), FS3 and FS5 show a greater buildup relative to the othermethods (e.g. at 2000 h, FS5 = +0.67 and FS1= +0.15 µmol m−2 s−1). During theremainder of the observation period calculated FS is near 0 µmol m−2 s−1 and all96methods are within 0.2 µmol m−2 s−1 of each other.For the entire observation period, the median standard deviation (σ ) of FS1−5is 89% of FS1. Maximum values occur overnight during canopy buildup (621%,σ=0.4 µmol m−2 s−1) and during morning canopy venting (148%, σ=1.25 µmolm−2 s−1) and minimum values are during mid-day (9%, σ=0.11 µmol m−2 s−1).Although this is only for one day, these results suggest that results obtained usingthe single-layer method at the EC measurement level (i.e. FS1) have potential forvery large uncertainties, especially during individual venting or storage builduphours.Although there are relatively large differences between methods at individualhours, the choice of storage calculation does not have a large effect on net FCduring this study period. Median absolute FS/FC for the entire observation periodis greatest for FS5 (15.8%) and FS3 (6.7%), the methods which assume UCL ∆cρ/∆tfor the entire measurement volume. Median FS/FC is least for FS2 (3.3%). MedianFS/FC for FS1 is 4.1% and for FS4 is 5.7%.For each hour, the maximum error that the choice of FS method has on FC iscalculated as the difference between the maximum and minimum of all methodsdivided by FC ((max(FS1−5)-min(FS1−5)) / FC). Choice of FS method has the great-est effect on FC (123%) at 0700 h during UCL venting when the absolute differencebetween storage methods (3.69 µmol m−2 s−1) is larger than the measured towerflux (3.00 µmol m−2 s−1). The median value for all hours during the 24-hourobservation period is 5.2%.3.3.4 Storage analysis using long-term tower measurementsObservations from the transect observations can be placed in context by comparingagainst FS calculated using the single level method (FS1) from the long-term CO2molar density dataset observed at this tower continuously from May 2008 - April2012. Ensemble mean hourly CO2 mixing ratio (measured at 24.8 m) and FS (cal-culated according to Eq. 6) are determined using 30-minute data from all seasons,wind directions, and days of the week. ∆cρ/∆t at the tower top is calculated asthe difference between mean molar densities centered at :00 minutes (:45 - :15)and :30 (:15-:45) minutes. Hourly ensemble values are centered on each hour, for97example the 1200 h value indicates the ensemble average of measurements from1130 - 1230 h. For this calculation, z is the volume from measurement height (24.8m) down to the surface and excluding the building and tree volume, as describedearlier.The diurnal course of hourly ensemble means over the entire period showsteady CO2 mixing ratios (approximately 395 ppm) from midnight until 0500 h(Figure 3.10a). From 0600 - 0800 h, CO2 rises to a maximum of 402 ppm beforedropping to a daily minimum of 388 ppm at 1500 h, after which levels rise steadilyup to 395 ppm at 2300 h. Mean wintertime (DJF) weekday mixing ratios are onaverage 22.4 ppm (6%) higher throughout the 24-hour daily period than summer-time. Daily wintertime mixing ratios peak at 0900 h (418 ppm), drop steeply to anafternoon minimum at 1500 h (400 ppm) before a prominent secondary afternoonpeak (over 410 ppm at 1900 h). During summer, the daily peak is at 0600 h (399ppm) and then fall gradually, relative to winter, throughout the afternoon (380 ppmat 1600 h) before rising steadily to 390 ppm at 2300 h.During all seasons, observed mixing ratios fall during the day even though theurban surface remains a net source of CO2 (Crawford and Christen, 2014). Thisdiurnal pattern is consistent with results reported at this site by Reid and Steyn(1997) and at other urban locations (e.g. Grimmond et al. 2002,Vogt et al. 2006).Using a series of box-model formulations, Reid and Steyn (1997) were able repro-duce the observed diurnal pattern of CO2 mixing ratios at this site by simulatingupwind summertime CO2 surface fluxes along with regional advection and ventingand entrainment exchanges with the free atmosphere at the inversion base duringthe daytime growth of the boundary layer. That study indicates that during day-time, dilution from advection and boundary-layer entrainment processes are theprimary factors determining CO2 concentrations on hourly timescales at this site.The dilution by entrainment of reduced-CO2 air from above would yield a positiveFC measured by the EC system, even if net surface emissions remain unchanged.This positive flux is not attributable to surface source/sink processes and wouldbe corrected by FS calculations based on ∆cρ/∆t. In terms of horizontal advec-tion however, a potential scale mismatch arises between scalar (∆cρ ) and flux (FC)source areas if there is horizontal divergence in the measurement layer (i.e. 0-24.4m).98Figure 3.10: a) Hourly ensemble means of CO2 mixing ratio, b) Hourly en-semble means of FS. Data are from May 2008 - April 2012 and mea-sured by the EC measurement system at 24.8 m a.g.l. The y-axis onthe RHS is for FS/FC (dashed line).99FS is calculated according to Eq. 6 and hourly ensemble mean FS valuesrange between -1.7 and +0.8 µmol m−2 s−1 during the diurnal course (Fig 3.10b).Overnight, (2200 - 0400 h), FS fluctuates about 0 µmol m−2 s−1, and increasesfrom 0400-0600 h (up to 0.8 µmol m−2 s−1), indicating an accumulation of CO2just before sunrise. From 0600 - 0900, FS decreases and becomes negative (mini-mum -1.7 µmol m−2 s−1 at 0900 h), indicating depletion of CO2 through the morn-ing after sunrise. From 09:00 - 17:00, FS steadily increases and becomes positiveat 1600 h and remains slightly positive until 2200 h.Throughout the day, mean absolute FS is on average 2.8% of hourly ensembleFC measurements. The largest positive hourly value is at 0500 h (6.0%) and thelargest negative is at 0900 h (-4.2%). This pattern is consistent with observed stor-age buildup (positive storage) before sunrise and venting of stored CO2 (negativestorage) after sunrise observed during the 24-hour transect observational period.At this site, FS is smaller relative to FC compared to the reported value of 11% inEdinburgh, Scotland (Nemitz et al., 2002).There are also seasonal differences in FS magnitude and timing related to chang-ing sunrise and sunset times. During winter (DJF), peak positive FS is at 0800 hduring morning rush hour before sunrise (December 21 sunrise is approximately0810 h) and peak negative FS is at 1100 h. FS remains negative throughout the dayand increases after sunset from 1600-2000 h (December 21 sunset is approximately1615 h). This is in contrast to summer (JJA) when morning peak storage occurs at0500 (June 21 sunrise is approximately 0510 h) and peak negative storage is threehours earlier than in winter. The magnitude of early morning positive summer stor-age is less than winter due to reduced emissions from building heating and thereis also an overnight storage increase (2000-2200 h) after sunset (June 21 sunsetis approximately 2020 h). Daylight savings time (DST) also impacts the timingof emissions by altering anthropogenic emissions patterns. From March-October,DST is observed and clocks are shifted 1 hour forward, though all times reportedhere are in Standard Time.The diurnal evolution of FS here is similar to diurnal cycles reported in forestecosystems (Yang et al. (1999), Aubinet et al. (2005)). In these studies, summer-time FS increases significantly in the first part of the night and becomes negative inthe morning after sunrise. A difference in the diurnal pattern at this urban location100is the prominent increase in FS before sunrise from anthropogenic traffic emissionsthat are not present in the forest ecosystems. At six European forest flux sites, FSis found to decrease overnight due to cooling soil temperatures which inhibit soilrespiration and also likely advective transport out of the study volume (Aubinetet al., 2005). The mean FS at these sites ranges from 1600-0400 h ranges from 0.2- 2 µmol m−2 s−1 and mean FS at this urban site for the same season and hours is0.5 µmol m−2 s−1.The magnitude of FS in the present study is correlated with several variablesrepresenting atmospheric stability and mixing (Figure 3.11). FS decreases as hor-izontal wind speed, vertical wind velocity variance, friction velocity, turbulent ki-netic energy, and z￿/L increase. This is also consistent with FS behavior in forestecosystems where researchers have often visually inspected friction velocity rela-tions with FC to determine a low turbulence threshold to withhold or flag FC mea-surements as suspect (Massman and Lee, 2002). Results shown here indicate thatother measures, such as σw (Figure 3.7), may work equally well, or better, thoughthis threshold method has been questioned as being too subjective and with poten-tial to introduce bias to annual emission and uptake estimates (Gu et al., 2005). Inboth summer and winter, there is a notable increase of FS following sunset due tothe collapse of the daytime mixed layer, and continued injection of evening CO2emissions (primarily building and traffic sources) into the developing stable layer.After this initial increase, FS decreases and remains slightly positive during sum-mer and is slightly negative during winter for the remainder of the night. Duringthese overnight periods (000-0400 h), atmospheric stability parameters indicatemore stable conditions on average during winter (mean z￿/L = +0.37) than dur-ing summer (mean z￿/L = +0.27) which would seem to indicate greater storagebuildup in winter. Horizontal wind speeds, however, are higher overnight duringwinter (mean wind speed = 2.7 m s−1) than summer (mean wind speed = 2.3 ms−1), as are the vertical wind velocity variance (DJF σw = 0.32 m s−1 vs JJA σw =0.29 m s−1) and friction velocity (DJF u∗ = 0.24 vs. JJA u∗ = 0.23). This indicatespotentially greater mixing from mechanically produced turbulence during winterwhich contribute to turbulent transport and negative FS values.FS may also remain positive overnight during summer at this location becausethe urban surface is potentially a stronger source of CO2 overnight during sum-101Figure 3.11: Binned median 30-minute values of FS compared to stability andmixing parameters measured by the EC system at 24.8 m a.g.l..mer compared to winter. During winter, statistical analysis of measured FC fromthis site and building energy modeling in the study area indicate that the bulkof CO2 emissions related to home heating occur during the late afternoon andevening (1600-2400) instead of overnight (Crawford and Christen 2014, van derLaan 2011). Soil respiration is also minimal during winter because of low soil tem-peratures. During summer, however, warmer soil temperatures lead to increases insoil respiration emissions overnight (Christen et al. 2011, Crawford and Christen2014).Negative nocturnal storage (i.e. venting) was also observed above a streetcanyon in Marseille, France by Salmond et al. (2005). That study takes place insummer and observed intermittent ‘bursts’ of CO2 are attributed to buoyant ther-mal plumes from within the canyon on turbulent timescales of less than 8-seconds.102In forest ecosystems, nocturnal reductions in FS are observed at six European for-est sites during summer as well (Aubinet et al., 2005). This phenomena cannotbe explained by decreased source intensity nor increased turbulent transport, soadvective transport is stated as the likely cause. However, advection estimates atthese sites are not considered accurate enough to completely support this hypothe-sis (Aubinet et al., 2005).3.4 ConclusionsVehicle transect measurements of UCL CO2 mixing ratio over a 26-hour periodwere conducted in the long-term turbulent flux source area of an urban EC systemin Vancouver, BC. During daytime, highest CO2 mixing ratios (+33 ppm, 7.7%> spatial mean) in the UCL are found above busy roads and spatial patterns arelargely a function of proximity to traffic sources. This is consistent with FC obser-vations and spatial emissions modeling results presented in Chapter 2 which foundhighest emissions originating from arterial roadways and intersections in particu-lar. Overnight, UCL CO2 and potential air temperature are negatively correlatedand spatial distribution of CO2 is influenced by micro-scale cold-air drainage andpooling. Based on FC observations and spatial modeling presented in Chapter 2,respiration from soils and vegetation is expected to be an important source of noc-turnal CO2. This is the first study to measure spatial patterns of UCL CO2 mixingratios at the micro-scale and the degree of horizontal spatial variability observedhere was unexpected. Given that CO2 can be regarded as a tracer gas for fossilfuel emissions of CO and NOx (Henninger and Kuttler, 2010), findings reportedhere have health implications for urban populations in terms of exposure to air pol-lution. These findings also highlight the difficulty in making representative pointmeasurements of scalar quantities such as air pollutants or temperature in the UCL.UCL measurements are also used to improve vertical resolution of ∆c/∆t andestimate FS in the measurement volume below EC measurement height. Five vari-ations of FS calculations are used to test the assumption of vertically consistent∆c/∆t implicit in single-layer FS calculations from CO2 measurements at the EClevel (FS1). Results here show this assumption to be untrue during certain hours,particularly morning venting, when the UCL ∆c/∆t was -13.5 ppm hr−1 and above-103canopy ∆c/∆t was +13.1 ppm hr−1.This results in large uncertainties associated with calculated values of FS us-ing FS1. Median uncertainty using FS1 during this observation period was ±89%,with values ranging from ±9% during midday to ±621% during the initial noc-turnal CO2 buildup. The magnitude of this uncertainty is mitigated because of thehigh levels of anthropogenic emissions at this site. Overall, FS is relatively minorcompared to FC and the median uncertainty to FC resulting from the choice of FScalculation method is 5.2%.From analysis of long-term measurements at the EC tower (May 2008 - April2012), ensemble hourly means of FS (using FS1) for all seasons range from -1.7to +0.8 µmol m−2 s−1. On average, hourly ensemble absolute means of FS are2.8% of FC, but for individual ensemble mean hours can be up to 6.0% of FC. Thediurnal course and FS magnitude here are comparable to forest sites, except forthe notable storage buildup before sunrise due to anthropogenic traffic emissionsand subsequent canopy layer venting. FS is also found to vary with atmosphericstability and mixing, with reductions in FS as instability and mixing increase. Theseatmospheric properties interact in subtle ways with underlying seasonal and diurnalsurface emission and uptake patterns to influence FS.In summary, there are several practical and theoretical difficulties with con-ducting continuous measurements of FS in an urban area. In particular, the practicalquestion of how to integrate CO2 mixing ratio and flux measurements with differ-ent source areas remains unsolved. Evidence from this site suggests FS is likelyto be minimal for long-term measurements and on hourly scales in relatively lowdensity urban areas with low canyon aspect ratios. In high density areas with highcanyon aspect ratios, potential for significant hourly FS is greater. In general, un-certainties to net emissions at all sites from FS are largest during the night-day andday-night transition periods. It is recommended that, at the very least, all urbanCO2 flux studies estimate the magnitude and relative importance of FS using thesingle-level calculation method at EC height, even though this method has largeuncertainties and on average underestimates the absolute value of FS.104Chapter 4City-scale vertical profiles4.1 IntroductionAlthough a majority of global anthropogenic greenhouse gas emissions, includingcarbon dioxide (CO2), originate from urban areas, measurements of CO2 mixingratios in the urban boundary layer (UBL) are relatively rare. In contrast, mea-surements of local-scale neighborhood net emissions have become more preva-lent in recent years from the application of eddy-covariance (EC) methods to ur-ban ecosystems (e.g. review by Velasco and Roth (2010)). Additionally, mobilevehicle-mounted measurements of CO2 have revealed near-surface spatial patternsranging from regional-scale urban-rural differences (e.g. Idso et al. (2001), Hen-ninger and Kuttler (2007)) to micro-scale variations based on proximity to CO2sources and topography (Crawford and Christen, 2014). One advantage of mea-surements in the UBL is that CO2 mixing ratios are representative of a larger areathan local-scale EC measurements because of the spatial averaging properties ofatmospheric turbulence. This means there is potential to infer regional-scale netemissions from mixing ratio timeseries that could be used to monitor emissionsand validate municipal emissions inventories and models. Furthermore, measure-ments in the UBL are useful because CO2 is a tracer gas that can be used to studyurban atmospheric transport mechanisms (Pataki et al., 2005) and is correlated withother air pollutants such as nitric oxide (NO), nitrogen dioxide (NO2), and carbonmonoxide (CO) (Henninger and Kuttler, 2010).105Two general approaches have been commonly used to infer net emissions fromentire urban areas from concentration or mixing ratio measurements. The first usesmixing ratio observations together with inverse atmospheric transport modelingto derive upwind source areas and surface fluxes. Examples include the Stochas-tic Time-Inverted Lagrangian Transport (STILT) model (Lin et al., 2003) coupledwith wind fields generated by a weather prediction model such as the Weatherand Research Forecasting (WRF) model (Skamarock et al., 2005). An examplestudy using this approach was conducted in Salt Lake City, where mixing ratioobservations at three sites were used with a WRF-STILT model configuration todetect changes in anthropogenic CO2 emissions of greater than 15% on monthlytimescales (McKain et al., 2012). In Los Angeles, a similar inverse modeling ap-proach was used to determine that at least eight surface-based observation sitesare needed to resolve anthropogenic CO2 emissions at 8-week temporal resolutionat 10 km spatial scales (Kort et al., 2013). In Heidelberg, Germany, atmospherictransport modeling was combined with a 5 min by 5 min resolution CO2 emissionsmodel to predict observations from a stationary network of CO2, CO, and radonsensors (Vogel et al., 2013). Another example of this approach was conductedin Paris, France, where a network of CO2 mixing ratio measurements are usedto model FC on 6-hour timescales. This study concluded that inversion modelingtechniques that rely on CO2 upwind-downwind gradients perform better than thosethat simply use raw CO2 mixing ratios (Bre´on et al., 2014).The second common approach is formulation of a box-model based on massconservation principles. Several studies have used aircraft-based measurements ofCO2 mixing ratios in the urban plume to calculate integrated urban CO2 net fluxusing a box-model formulation. In Indianapolis, USA, measurement flight pathswere conducted through the depth of the ABL downwind of the city. A spatiallyinterpolated vertical plane of CO2 mixing ratio values normal to the prevailingwind direction downwind of the city was then used to attribute net emissions toan upwind urban spatial area. Fluxes calculated using this method showed hightemporal variability (mean 19.2 ± 15.4 µmol m−2 s−1), but mean values were notstatistically significantly different from an independent spatial surface emissionsmodel (Mays et al., 2009). In London, UK, a measurement campaign used aircraft-based observations to estimate urban CO2 fluxes and found daytime averages of10646 - 104 µmol m−2 s−1. These values were statistically similar to synchronousmeasurements from local-scale urban EC flux towers and emissions inventories(Font et al., 2013). In Vancouver, BC, Canada, Reid and Steyn (1997) used a seriesof eight box models describing UBL growth combined with a surface emissionsmodel to predict CO2 mixing ratios at a measurement location located 28 m abovethe urban surface. A similar approach was developed in Salt Lake City, USA usinga total of 162 boxes and a near-surface CO2 monitoring network comprised ofseven stations (Strong et al., 2011).Box-models rely on atmospheric boundary layer budget techniques and gen-erally need only mixing ratio measurements in the boundary layer to representregional fluxes, in addition to measurements or models of boundary layer depth(Cleugh and Grimmond, 2001). Assumptions include a well-mixed boundary layerthat has fully adjusted to the underlying urban surface. Following scaling argu-ments by Raupach (1995) and Cleugh and Grimmond (2001), the downwind dis-tance (X) required for boundary layer adjustment during convective conditions isrelated to the convective time scale (t∗), horizontal wind velocity (U), and convec-tive velocity scale (w∗):X =ziUw∗(4.1)where zi is the ABL height and w∗ is:w∗=￿gFHvziθv￿1/3(4.2)where FHv is the virtual heat flux and θv is virtual temperature.The convective time scale for a surface signal to reach the top of the boundarylayer is:t∗ =XU=ziw∗(4.3)It then follows that a convective boundary layer will be fully mixed over lengthsgreater X and times longer than t∗. If the scale of horizontal heterogeneity ofsurface fluxes is much less than X , boundary layer concentrations will be represen-tative of the average of the surface source/sink processes underlying the UBL.107Boundary layer budget methods have been used primarily over agricultural andforested landscapes to infer regional CO2 fluxes on hourly-daily timescales. Day-time measurements of near-surface CO2 mixing ratios over wheat fields in Aus-tralia were used to infer regional fluxes that compared well with local-scale CO2flux measurements by Raupach et al. (1992) and Denmead et al. (1996). In Ger-many, aircraft measurements of convective boundary layer CO2 mixing ratios wereused to compute regional fluxes with 10-20% error estimates over landscapes ofbeech forests and agricultural fields (Laubach and Fritsch, 2002). In Iowa, USA,aircraft measurements during growing season over mixed soybean and corn crop-land were found to be consistent with tower-based local-scale fluxes (Martins et al.,2009). In Sacramento, CA, USA, Cleugh and Grimmond (2001) used a boundarylayer budget method to calculate regional heat and water vapor fluxes. In this study,regional heat fluxes derived using budget methods combined with radiosonde mea-surements of potential temperature were found to agree with spatially-weightedsurface flux measurements from irrigated agricultural, non-irrigated agricultural,and urban landcover regions.Studies using boundary layer budget methods in urban areas must take intoaccount the unique ways that urban areas modify the overlying boundary layerstructure. In idealized daytime conditions with clear skies and a regional back-ground wind, an internal boundary layer grows downwind with distance from theupwind urban edge until it occupies the entire planetary boundary layer. This in-ternal boundary layer is the urban boundary layer (UBL) and strong heating ofthe urban surface relative to surrounding non-urban areas (e.g. from urban heatisland effect and modification of surface energy balance towards higher Bowen ra-tio) drives buoyant convection that propels the UBL higher than the surroundingnon-urban boundary layer (Oke, 1987). Within this daytime UBL, strong mixingfrom convection and turbulence is expected to result in uniform vertical profiles ofwind speed, wind direction, water vapor, potential temperature, and suspended airpollutants (e.g. Oke and East 1971).Overnight, urban areas modify the overlying atmosphere through thermal andmechanical processes. In non-urban areas, cooling of the urban surface after sunsetcreates a layer of stable air that acts to suppress vertical mixing and transport.Over urban areas, a shallow, slightly unstable roughness sub-layer (RSL) has been108observed due to increased mixing from storage heat releases, anthropogenic heatemissions, and greater urban surface roughness (e.g. Oke and East 1971, Uno et al.1988). Above the nocturnal RSL is an elevated temperature inversion above whichis a stable nocturnal boundary layer (NBL) and residual mixed layer (RL) (Unoet al., 1988). The RL is a near-neutrally stable layer of residual turbulence, heat,and pollutants from the previous day capped by a temperature inversion. Nocturnalprofiles of potential air temperature are expected to be roughly constant from thesurface through the RSL, increase in the NBL, then become roughly uniform withheight in the RL (e.g. Oke and East 1971). Vertical profiles of pollutants with activenocturnal surface sources, such as CO2, are expected to decrease with height abovethe nighttime RSL due to near-surface buildup below the temperature inversion andlack of vertical mixing (e.g. Oke and East 1971).To date, neither boundary layer budget approaches nor inverse modeling tech-niques are able to monitor emissions near real-time nor resolve hourly-scale urban-scale fluxes. Improved resolution and capability of CO2 emissions monitoring isimportant to municipal governments to identify and monitor individual sources/sinksand assess progress towards meeting emissions reduction targets. Column-basedmeasurements in urban areas are likely to yield more information about urban-scalefluxes than a near-surface measurement because they represent a larger source areaof influence (McKain et al., 2012). The objectives of this work are to present aunique dataset of CO2 mixing ratios observed in the UBL column over Vancouver,Canada and attempt to model urban-scale CO2 fluxes at hourly resolution using aboundary layer budget approach. This approach is chosen because of its relativesimplicity in terms of data input requirements and for its potential to resolve fluxesat hourly temporal scales. Results from the boundary layer budget model are thencompared to local-scale EC measurements from a network of three flux towers inoperation within the greater Vancouver regional area.1094.2 Methods4.2.1 UBL CO2 observationsStudy location and periodDuring August 14 - 15, 2008, vertical profiles of CO2 mixing ratios in the UBLwere observed with a measurement system mounted on a tethered balloon. Mea-surements took place in Memorial Park cemetery (49.2362◦ N, 123.0392◦ W)in Vancouver, BC, Canada (Figure 4.2). The cemetery is centrally located inmetropolitan Vancouver with a mean elevation of 80 m a.s.l. and is located ap-proximately 5.0 km from downtown at a heading of 340◦, 10 km away from theStrait of Georgia (270◦), and 7.0 km away from YVR International Airport (230◦).The cemetery is approximately 0.5 km x 1 km in area and is bounded by arterialroads to the east (Fraser St.) and south (49th Ave) with daily vehicle loads upwardsof 20,000 vehicles per day. Landcover in the cemetery is primarily clipped and ir-rigated grass. The cemetery is interspersed with individual trees up to 25 m height,though no trees were within 100 m of the observation site.Synoptically, the region was under the influence of a high-pressure systemwhich resulted in mostly clear skies and warm temperatures during the August 14-15, 2008 experimental period. At the nearby Vancouver-Sunset flux tower (Section4.2.2), mean air temperature recorded during the 24-hour experiment at 24.8 mheight a.g.l. was 24.9◦C, with minimum of 21.0◦ C at 0800 h and maximum of30.2◦ C at 1800 h. Clear-sky, summertime conditions were selected in order tosample a representative diurnal cycle of morning-afternoon convective boundarylayer development, evening boundary layer collapse, and nocturnal developmentof a stable surface layer and residual layer aloft. Weekday conditions (Thursday-Friday) were selected to be representative of mid-week CO2 emission loads. Sunseton August 14th was at 1930 LST and sunrise on August 15th was 0505 LST.InstrumentationUBL CO2 mixing ratios were measured with a GMM220 CO2 sensor (Vaisala,Inc., Vantaa, Finland) and recorded on a CR1000 data logger (Campbell Scien-110tific, Inc. Logan, UT, USA). Air temperature and relative humidity were measuredwith a HMP50 T/RH sensor (Vaisala, Finland), and air pressure, along-wind veloc-ity, cross-wind velocity, and wind direction were measured using a Kestrel 4500Weather Tracker (Nielsen-Kellerma Co., Boothwyn, PA, USA). All instrumentswere sampled at 5 s intervals. The measurement system was suspended 0.5 m be-low a 5 m3 volume meteorological balloon inflated with helium (Figure 4.1). Winddirection measured by the Kestrel 4500 was determined by the balloon’s orien-tation (always into mean horizontal wind direction). The balloon’s altitude wascontrolled by an electric winch and the balloon was authorized to measure up to aheight of 400 m a.g.l. by Navigation Canada air traffic control. During the 24-hourperiod from 1100 August 14 - 1100 August 15, 48 vertical profiles were measuredin total. Each profile was completed in 30 minutes (i.e. 2 profiles per hour, 1 goingup, 1 coming back down) and were timed to coincide with 30-minute CO2 fluxaveraging periods at three EC towers in operation in the region (see Section 4.2.2).Voltage signals from the GMM220 CO2 sensor recorded by the CR1000 loggerwere converted to CO2 mixing ratios using results from an in-house calibrationagainst an Li-7000 closed-path infrared gas analyzer (Li-Cor Biosciences, Lincoln,NE, USA). During the calibration, the GMM220 and Li-7000 intake valve wereplaced in a sealed chamber in which the CO2 mixing ratio was varied from 440 -970 ppmv. Mixing ratios were kept constant at six different values and ten-minuteaverages from the two sensors were compared. Resulting CO2 mixing ratios werethen corrected for air temperature and atmospheric pressure using manufacturersupplied algorithms (Vaisala, 2014).Data processingHourly vertical profiles of CO2 mixing ratio, air temperature, wind direction, rel-ative humidity and wind velocities used for subsequent analysis were obtained byaveraging 5 s samples (c) from two individual, consecutive 30-minute profiles (1going up, 1 coming back down) in 10 m vertical increments:￿cz￿=∑cznz(4.4)111Figure 4.1: Instrumentation to measure and record meteorological data is sus-pended below the tethered balloon during the 24-hour observation pe-riod (photograph by Andreas Christen).112where ￿cz￿ is the mean CO2 mixing ratio of nz individual 5 s samples (c) at 10 mheight increment z. Using this method, the balloon passes through each 10 m heightincrement twice per hour, though the time interval between passes is not constantwith height. Regions near the top of the profile have a shorter interval betweenmeasurements ( 5 minutes), while regions near the surface have longer intervals(up to 60 minutes). Profile-averaged CO2 mixing ratio values (￿c￿) reported in thiswork are the mean of the 10 m vertical increment averages for a height range z1(m) to z2 (m):￿c￿=∑z2z1￿cz￿(z1− z2)/10m(4.5)This procedure (Eqs. 4.4-4.5) is also applied to potential air temperature (θ ),and wind velocity (U) to retrieve profile-averaged values of ￿θ￿ and ￿U￿, respec-tively. Potential temperature at 0 m a.s.l. was determined from balloon systemmeasurements of air temperature (t) using the dry adiabatic lapse rate of 0.0098 Km−1 (Stull, 1988) and measurement height (z).4.2.2 CO2 observation networkDuring the tethered balloon experiment, there were three micrometeorological tow-ers within 20 km of the balloon launch site equipped with eddy-covariance instru-mentation to measure local-scale CO2, energy, and water fluxes. The Vancouver-Sunset tower was located 1.9 km distance to the SE (143◦) in a residential neigh-borhood, Vancouver-Oakridge tower was located in a residential neighborhood 3.0km to the W (250◦), and Westham Island tower was established as a rural referencestation located 18.5 km away to the SW (200◦) (Figure 4.2). Additionally, CO2mixing ratios were measured in the urban canopy layer of the high-density cen-tral business district of downtown Vancouver (Robson Square) and at a forestedoffseason ski resort on Cypress Mountain at 1100 m elevation above sea level.The Sunset flux tower is located within the grounds of a BC Hydro substation(tower coordinates are 123.0784◦ W, 49.2261◦ N, WGS-84). The tower is locatedon the SE corner of the substation grounds that extend 100 m to the west and 130m to the north of the tower. The tower is an open, triangular, lattice tower and thebase is recessed by 4 m from the surrounding terrain. Net mixing ratios, molardensities, and fluxes of CO2 (FC) were continuously measured at the tower from113Figure 4.2: Map of CO2 measurements in Vancouver during July-August2008. Elevation refers to the base elevation above sea level of the mea-surement tower, tripod, or tethersonde and height refers to the instru-ment height above ground. The basemap is a false-color (Bands 745)image from USGS Landsat 5 TM taken September 7, 2011 at 18:50LST. 114May 2008 - April 2012 at an effective height of 24.8 m a.g.l. The Sunset neigh-borhood is classified as LCZ-6 ‘open-set lowrise’ (Stewart and Oke, 2012) and isa primarily residential area with an average population density of 63.1 inhabitantsha−1. Over 90% of buildings are single, detached, residential structures with amean building density of 12.8 buildings ha−1 and mean building height of 5.1 m(van der Laan, 2011). Plan-area landcover fractions in a 1900 m × 1900 m areacentered on the tower are 29% building, 11.8% tree, 20.3 % lawn, and 43.6% im-pervious. This tower and surrounding neighborhood have been the site of a numberof previous CO2-related measurement campaigns and modeling studies (Reid andSteyn (1997), Walsh (2005), Christen et al. (2011), Kellett et al. (2012), Craw-ford and Christen (2013)). A ceilometer (CL31 Vaisala) was also in operation atthe Sunset tower location during the balloon experiment (McKendry et al., 2009)and is used to estimate daytime convective boundary layer heights through remotesensing of vertical aerosol structure (van der Kamp and McKendry, 2010). Theceilometer measured backscatter profiles at 15 s resolution in 5 m vertical incre-ments up to 7.5 km. For analysis of daytime UBL height, 10 min averages wereused for heights above 50 m (van der Kamp and McKendry, 2010).The Oakridge tower (123.1329◦ W, 49.2306◦ N) was in operation from June-August 2008. EC instrumentation was mounted at 29 m on a guyed hydraulicmast located on the grounds of an elementary school in the Oakridge neighborhood(LCZ-9 ‘open-set lowrise’). The tower is 0.5 km from arterial roads and a parkcomposed of grass recreational fields (approximately 250 m x 250 m) is locatedimmediately east of the tower. Based on analysis of satellite imagery, landcoverfractions in a 1000 m radius about the tower are 24% building, 23% impervious,and 56% vegetation (both trees and lawn) (Tooke et al., 2009). Population densitywithin 1000 m radius is 27.6 inh ha−1, built density is 8.0 bldg ha−1, and meanroof height is 5.8 m.A rural reference EC tower was located on Westham Island (123.1768◦ W,49.0863◦ N), 18 km south of the Sunset and Oakridge urban sites. The tripod towerwas installed in a flat, unmanaged, non-irrigated grass field in a region character-ized by intensive agriculture. The local-scale turbulent flux source area is classifiedas Zone 11 ‘low plant cover’ and grass heights ranged from 10 cm in winter to 1.75m in summer. On average, grass was 1.60 m high during August 2008 (Liss et al.,1152010). EC instrumentation was mounted at 1.8 m a.g.l. and the tower was located300 m horizontally away from the Strait of Georgia.All three flux sites measured CO2 fluxes using a sonic anemometer (CSAT 3-d, Campbell Scientific, Logan, UT, USA) and an open-path infrared gas analyzer(IRGA, Li-7500, Licor Inc, Lincoln, NE, USA). Each IRGA was calibrated in-house every 6 months according to standardized procedures (Li-Cor, 2002) againstreference tanks from Environment Canada. At all sites, three-dimensional windvelocities and CO2 mixing ratios were recorded at 20 Hz and subject to severalquality control procedures (e.g. filtered for interference from precipitation, realis-tic max/min thresholds). Fluxes were then calculated from block-averaged meansof 30-minute periods and a 2-d coordinate rotation was performed to align the co-ordinate system with the mean wind direction. Full EC quality control and dataprocessing details are described in Appendix A.Additional measurements of CO2 mixing ratios were measured during July-August 2008 at the University of British Columbia (UBC) campus (-123.2490◦W, 49.2554◦ N), in downtown Vancouver (-123.1217◦ W, 49.2825◦ N), and at Cy-press Mountain (-123.2120◦ W, 49.3915◦ N) (Figure 4.2). The UBC and Cypressstations both used Vaisala GMM220 sensors that sampled every 5 s and recorded10 minute averages during July - August 2008. The UBC sensor was mounted at2 m a.g.l. and the immediate surroundings are an irrigated and clipped grass fieldand experimental agriculture plot and orchard. The station is approximately 0.5 kmaway from the Strait Georgia to the west. This sensor was removed from UBC andused for the balloon-based system on August 14-15. The Cypress Mountain CO2sensor was located on a ski run of Cypress Mountain ski resort at 1100 m elevation.The sensor was mounted at 2 m a.g.l. above rocky soil with sparse vegetation withmature coniferous trees within a 30 m radius. The downtown measurement loca-tion at Robson Square is located in the densely built Vancouver central businessdistrict. Immediate surroundings are hotel, office, and apartment buildings of 30-40 stories and busy roads with high traffic and pedestrian volume (intersection ofHornby and Robson Streets). CO2 mixing ratio observations were conducted witha closed-path Li-800 GasHound (LiCor, Inc.) co-located with a Greater VancouverRegional District (GVRD) air quality monitoring station with an intake valve at 5m a.g.l..1164.2.3 Boundary layer CO2 budgetSurface CO2 fluxes are inferred from UBL measurements using a boundary layerbudget approach applied to a simple, single-box model construct (Figure 4.3). Thismodel assumes a column of air being advected along the surface at the mean windspeed and CO2 emissions from the surface are evenly mixed throughout depth ziduring time t. Using this method, the hourly surface flux (FC) is calculated as(Denmead et al. 1996, Font et al. 2013):FC = zi∆￿c￿∆t − (cb−￿c￿)∆zi∆t − zi￿U￿∆cx∆Sx(4.6)where zi is the depth of the convective UBL (daytime) or stable NBL (nighttime)layer, ￿c￿ is the profile-averaged CO2 molar density measured by the balloon sys-tem through depth zi (Eq. 4.5), and t is the time (s) over which FC is calculated.The second term on the RHS describes the entrainment flux from the overlying freeatmosphere where cb is the background CO2 molar density above the UBL. A nega-tive entrainment flux signifies a reduction in UBL CO2 content via dilution from aninflux of relatively low-CO2 density air from the free atmosphere, whereas a posi-tive entrainment flux results from a decrease in zi. In this formulation, subsidencevelocity is assumed negligible. The third term on the RHS describes the horizontaladvective flux, where ￿U￿ is the profile-averaged horizontal wind speed and ∆cx∆Sx isa horizontal CO2 molar density spatial gradient. This gradient is determined as thedifference between the profile-averaged CO2 molar density and background CO2molar density upwind of the urban area. A negative advection term indicates an in-flux of relatively low-CO2 content air from the upwind direction and reduction ofUBL CO2 content, whereas a positive advection flux is the reverse. The horizontaldistance Sx is calculated for each hour as (Cleugh and Grimmond, 2001):Sx = ￿U￿t (4.7)This assumes constant wind speed during time t and that the travel distance of anair parcel during this time is equal to the length scale of the averaging area. Dur-ing unstable conditions (hourly z￿/L <0, measured at Sunset EC tower), ￿U￿ isaveraged for the entire depth of the UBL measurements (0-400 m) and during sta-117ble conditions (hourly z￿/L >0), ￿U￿ is averaged for heights below 50 m from theballoon-based measurements. This assumes atmospheric conditions in the night-time RSL and stable NBL are de-coupled from the overlying RL.During daytime, the height of the UBL (zi) is determined at hourly resolu-tion by measurements from the ceilometer at the Sunset EC tower (van der Kampand McKendry, 2010). Overnight, the NBL height zi cannot be determined fromceilometer measurements because non-physical, semi-periodic backscatter signalfluctuations below 50 m obscure definition of NBL depth (van der Kamp andMcKendry, 2010). Instead, the height of the NBL was determined using threedifferent methods. First, zi is defined at which potential temperatures are 90%of the potential air temperature in the residual layer (defined as the measured av-erage temperature from 350-400 m a.g.l.). This is an ad hoc method based onobserved nighttime temperature profiles (Figure 4.4, Figure 4.8) that defines zi ata height where the potential temperature profile stops increasing with height andbecomes roughly uniform. The second method relies on the assumption that me-chanical forces dominate near-surface mixing compared to buoyancy and meanwind speeds at 10 m (measured by the tethered balloon system) are used to deter-mine zi based on empirical relations described by Benkley and Schulman (1979)where zi = 125U10. This model was specifically designed for 10 m wind speedsbecause this is the World Meteorological Organization standard wind speed mea-surement height. This height is also appropriate for this study because of poten-tial de-coupling between the stable NBL and overlying RL conditions. The thirdmethod describes the growth of the stable NBL in terms of cumulative surfacecooling with time (Stull, 2000):zi = 5￿aURL3/4t1/2￿(4.8)where a is a constant set to 0.15 m1/4 s1/4 (Stull, 2000), URL is the mean windspeed in the RL (defined here as the mean of balloon-system measurements from350-400 m), and t is time since sunset in seconds.During overnight, stable conditions when there is no convective boundary layergrowth, the entrainment term is ignored. During daytime, the background CO2mixing ratio (cb) above the convective boundary layer is set to three different val-118Figure 4.3: Conceptual diagram of the single box model used to calculatecity-scale FC based on balloon measurements.119ues: 375 ppm (average 24-hour value measured during July-August, 2008 at thenon-urban Cypress Mountain, Westham Island, and UBC CO2 monitoring sites(Figure 4.6), 382 ppm (average measured during the experiment at Estevan Point,a global CO2 monitoring station on the west coast of Vancouver Island), and adynamic value set to the hourly difference between the Westham Island and UBLprofile-averaged CO2 mixing ratio during the experiment. These values are con-verted to molar density by multiplying the mixing ratio (µmol mol−1) by air den-sity (g m−3) and the inverse of the molecular weight of dry air (28.96 g mol−1).These same three cb values are also used to determine the horizontal∆cx∆Sx gradi-ent contained in the advection term of Eq. 4.6. The advection term is added to FCduring individual hours only when wind directions are from 200◦ - 270◦. Duringthese conditions, marine air with relatively low-CO2 mixing ratio is expected to in-fluence the model domain during the hourly FC calculation intervals. When windsare from 270◦-200◦, the horizontal gradient is influenced by extensive upwind ur-ban landscape and so ∆cx∆Sx is assumed to be negligible.The result is three variations for each hourly FC value calculated using Eq.4.6: three variations using different values of background CO2 mixing ratio duringdaytime unstable hours and three variations using different values of NBL heightduring nighttime stable conditions.4.2.4 UBL measurement source areaTo aid analysis of observed UBL profiles and interpretation of inferred regional FC,2-d surface source areas are estimated for balloon-based measurements. The sourceareas are not used analytically in this context (e.g. for quantification of sourcearea landcover composition), instead their purpose is to roughly define the upwindsurface area contributing to the UBL observations as an aid for interpretation ofcalculated FC. As a result, several simplifications are used during their calculation.The along-wind dimension (Sx) of the source area is calculated using Eq. 4.7described earlier. The lateral spread of the 2-d source area is estimated throughcalculation of the standard deviation of lateral dispersion distance (σy) for an indi-vidual plume during unstable conditions (Stull, 2000):σy = 0.5Xz (4.9)120where X is the normalized downwind distance:X =xw∗zU(4.10)For stable periods a formulation based on Taylor’s Lagrangrian statistical tur-bulence theory is used (Stull, 2000):σ2y = 2σ2v t2L￿xUtL−1+ exp￿−xUtL￿￿(4.11)where x is downwind distance from the source and tL is the Lagrangian time scale,set to 60 s for dispersion in the boundary layer (Stull, 2000). These methods as-sume Gaussian concentration distribution about the central plume axis as well asconsistent and uniform wind speed U (measured at Sunset Tower) and cross-windstandard deviation (σv) (measured at Sunset Tower). Averages from Sunset towerare used because they are based on continuous measurements (20 Hz) instead of alimited sample from the balloon-based system. Effects on horizontal dispersion bycanopy layer surface roughness elements (buildings and vegetation) and by street-canyon channeling are ignored.For each 60-minute profile measurement period, σy is calculated for downwinddistance Sx for a single plume originating at the balloon measurement location.The plume is triangle-shaped with along-wind dimensions of Sx and a base lengthof 2σy. The plume is then oriented facing into the mean wind direction (upwind)measured during the 60-minute profile and originating at the balloon measurementsite. This plume inversion approximates the upwind area from which individualsurface plumes would be advected across the balloon measurement location andtherefore measured by the balloon system. In reality, if horizontal distribution isGaussian, measurements will be more representative of CO2 emissions releasednear the source area centerline, however source area weighting in terms of contri-bution to measurements is not included in this simplified source area model. Thisapproach also neglects effects on the source area of turning wind directions withheight.1214.3 Results and discussion4.3.1 Observed meteorology and CO2 mixing ratios in the UBLDuring the August 14-15 experiment, measured CO2 mixing ratios up to 400 ma.g.l. range from 372 - 456 ppm. Lowest mixing ratios (372-384 ppm) are observedduring late afternoon from 1600-1800 h LST (LST=PST) and highest (420-456ppm) are observed overnight and early morning (2000-0700) below 50 m a.g.l.(Figure 4.4, Table A.1). Observed profile-averaged CO2 mixing ratios for the 0-400 m depth are 387.5 ppm during the initial profile from 1100-1200 h, and fallto 376.2 ppm from 1700-1800. Overnight, the profile-averaged CO2 mixing ratioincreases to 406.1 ppm at 0300-0400 h.During the experiment, measured potential air temperatures range from 34.0◦Cat 400 m at 1900 h to 19.0◦C at 10 m at 0500 h (Figure 4.4). During daytime(1100-1800 h), potential air temperatures are generally uniform with height and0-400 m profile-average potential temperature increases from 22.5◦C at 1100 h to29.3◦C at 1800 h. After sunset (1800-1900), the air temperature near the surface(<50 m) rapidly cools from 29.9◦C to 25.9◦C.Observed wind directions and velocities conform to a pattern of diurnally re-versing land-sea breeze thermal circulations typical in this area during summer-time (Figure 4.5, Table A.1). During the afternoon (1100-1800 h), wind directionsabove 10 m are from west and southwest and profile-averaged wind speeds are 2.6m s−1. Maximum daytime wind speeds are 4.2 m s−1 at 140 m a.g.l. at 1600 h.After sunset, the thermal circulation reverses and observed wind directions shift tothe northeast. From 1900-2000, there is directional wind shear with height, witheasterly winds observed at 100-250 m, and westerly winds from 250-400 m. By2200 flow at all heights up to 400 m was from the east, with a maximum velocityof 3.9 m s−1 at 140 m. Overnight, winds are light (< 1.5 m s−1), with increasesup to 2.5 m s−1 below 100 m at 0500 h near sunrise. After sunrise, onset of theseabreeze front is observed at 0700-0800 h as wind directions shift back towardsthe southwest and velocity increases to 4.5 m s−1.122Figure 4.4: Time-height contour plots of UBL a) CO2 mixing ratio, and b)potential temperature.123Figure 4.5: Time-height wind vector plots. Length and color of arrow indi-cate horizontal wind speed, and direction of arrow indicates wind direc-tion is coming from using meteorological convention (e.g. 0◦=North,90◦=East).The diurnal course of the profile-averaged CO2 mixing ratio from 0-400 m iscompared to CO2 mixing ratios measured at four locations in the Greater Vancou-ver Regional District (Figure 4.6). Highest CO2 mixing ratios throughout the studyperiod are observed at the downtown Robson Square site (mean 440.2 ppm, max481.2 ppm, min 414.5 ppm) and lowest values are at Westham Island (mean 375.8ppm, max 393.6 ppm, min 366.1 ppm). During the afternoon (1100-1800 h), meanUBL values measured by the balloon-based system are within 0.9 ppm (0.2%)of measurements at Sunset tower and within 4.5 ppm (1.2%) of measurements atOakridge tower, measuring at 24.8 and 29 m, respectively. Overnight (1800-0800h), 0-400 m average is lower than Sunset tower by -23.6 ppm (-5.5%) and by -8.0124Table 4.1: Profile averages from UBL measurements (0-400 m a.g.l.) duringthe 24-hour observation period.Hour ￿c￿ (ppm) ￿θ￿ (◦C) ￿U￿ (m s−1) Wind direction (◦)11:00 387.50 22.52 1.79 257.6212:00 387.36 23.12 2.42 260.3613:00 385.65 24.03 2.95 227.7914:00 383.17 25.16 2.76 228.1615:00 384.04 26.45 2.97 219.2316:00 381.06 27.24 3.48 219.5117:00 376.52 28.05 3.18 232.0618:00 377.22 29.26 2.53 219.9119:00 384.11 28.08 2.59 354.1920:00 389.97 27.42 2.89 0.0021:00 393.24 27.03 2.25 40.1122:00 397.06 26.58 2.56 101.4223:00 400.72 25.89 2.40 102.9300:00 399.97 25.12 1.98 98.9201:00 398.69 24.49 1.30 31.4102:00 402.66 23.74 1.10 338.1503:00 406.14 23.37 1.08 81.8304:00 405.12 23.09 1.63 91.9605:00 405.71 23.22 1.41 59.7006:00 404.17 22.45 1.55 325.7607:00 402.75 22.82 2.50 152.9308:00 396.96 23.33 2.85 194.3809:00 399.27 23.18 2.24 170.2210:00 396.11 24.27 1.77 172.7424-hr mean 393.55 25.00 2.26 174.80ppm (-1.9%) at Oakridge, on average. This result suggests that urban tower-basedmeasurements are representative of the entire UBL depth during unstable, well-mixed conditions. At night however, the tower-based measurements diverge fromthe profile-average, suggesting de-coupling of urban surface layer conditions fromthe overlying RL (see also Chapter 3).Ensemble mean CO2 mixing ratios measured by the CO2 observation networkduring July-August 2008 indicate CO2 content during the 24-hour experimental125Figure 4.6: Hourly mean CO2 mixing ratios measured by the observation net-work (Figure 4.2) during the balloon campaign on August 14-15, 2008(solid lines) and ensemble means during July-August 2008 (dashedlines). Cypress and UBC station values are not available during theballoon observation period. The balloon measurement (dark blue) is acolumn-average value from 0-400 m (￿c￿).period was unusually high overnight and during early morning (Figure 4.6). At theSunset EC tower, mean CO2 mixing ratio from 2000-0600 h during the experimentwas 429.5 ppm, compared to the July-August ensemble mean of 387.4 ppm for thesame hours (8.3% difference). Hourly CO2 mixing ratios from 2000-0500 h at Sun-set tower during the experiment were in the top 10th percentile of all July-August2008 values for each hour. At the residential Oakridge neighborhood EC tower, ob-served mixing ratios during the experiment were 3.8% higher than the July-Augustaverage and at downtown Robson Square, mixing ratios were 9.4% above aver-126age. At the same time, nighttime mixing ratios were -2.2% below average at thenon-urban Westham Island site, though typically nighttime Westham Island CO2mixing ratios are higher than those at Sunset tower by 6.6 ppm (1.7%). During thedaytime, all sites are within 2.2% of their respective July-August ensemble means.The likely reasons for the above-average CO2 mixing ratios during this nightare enhanced thermal stability and reduced mixing and advection frommean winds.This night was characterized by more negative sensible heat flux (QH) and lowerwind speeds than average at the Sunset flux tower. Overnight from 0000-0400 h,mean QH measured at the Sunset tower was -4.94 W m−2, compared with a July-August 2008 ensemble average of 0.675 W m−2. For the same hours, wind speed atSunset tower was 1.96 m s−1 during the experimental period, compared with a July-August 2008 ensemble average of 2.4 m s−1. More negative QH implies greaterheat loss from the surface leading to more stably stratified conditions than usualand lower wind speeds imply reduced vertical mixing and horizontal advection,resulting in above average CO2 mixing ratios during the experimental period. Thisimplies that observations obtained during this period are representative of a spe-cific set of synoptic and regional atmospheric conditions (e.g. summertime high-pressure system, large daytime solar heat input, weak synoptic horizontal pressuregradient). The advantage of observations during this set of conditions is that im-portant processes such as development of the nocturnal UBL, daytime convectiveUBL, and onset of thermal land-sea breeze circulations are more clearly developed.4.3.2 Observed UBL dynamics and CO2 mixing ratiosObserved UBL mixing ratios are strongly influenced by the daytime evolution ofthe convective UBL and overnight formation of a stable NBL. Initial backscattermeasurements from the Sunset ceilometer show a UBL height of 540 m at 1100h (Figure 4.7). This is the maximum UBL height measured during the observa-tion period, and from 1100-1300 h the UBL height fluctuates about 500 m beforeslowly falling to 400 m by 1700 h. From 1700-1900 h, just before sunset, the UBLcollapses rapidly due to reduced surface heating and less vigorous vertical mixing(Figure 4.7). The observed pattern of convective UBL development and magnitudeof UBL heights are consistent with summertime measurements from the ceilometer127Figure 4.7: Nighttime estimates of the NBL height using three differentmethods and daytime measurements from the ceilometer located at theSunset EC tower. The legend refers to the methods used to calculate theNBL height described in the text. Daytime UBL heights are based onmeasurements from the ceilometer.conducted at this site from 2006-2008 (van der Kamp and McKendry, 2010).Vertical profile measurements during this time show a well-mixed UBL withnearly uniform potential temperatures and CO2 mixing ratios with height (Figure4.4). During late afternoon (1500-1800 h), profile-averaged CO2 mixing ratio upto 400 m a.g.l. is 380.2 ppm and the mean vertical gradient for 20 m verticalincrements is -0.01 ppm m−1 (Figure 4.8). Profile-averaged potential air temper-ature during the same period is 30.0 ◦C, with mean vertical gradient of 0.0085 Km−1. During this time, measurements by the EC system at Sunset tower indicatethermally stable conditions (mean z￿/L=-0.70).128Figure 4.8: a) CO2 mixing ratio and b) potential air temperature in the UBLduring afternoon unstable conditions (light blue) and overnight stableconditions (dark blue).After sunset (1930 h), three methods were used to estimate the height of thestable nocturnal boundary layer during the night of August 14-15, 2008 (Figure4.7). The cumulative cooling method (Eq. 4.8) estimates a NBL height of 104 mat 2000, approximately 30 minutes after sunset. This estimate rises to 212 m atmidnight before falling to 111 m by 0100. This peak is explained by the increasein RL wind speeds from 2200-2300 h (Figure 4.5). After 0100, the NBL heightrises up to 168 m by 0500 h, just before sunrise. The empirical estimate based on10 m wind speeds shows an initial rapid increase in NBL height from 15 m to 161m from 2000-2200 before falling back to 15 m by 0200. From 0200-0500 h, NBLheight rises to 132 m. This estimate is directly scaled to observed fluctuations innocturnal wind speeds at 10 m measured by the tethered balloon system (Figure4.5). The NBL height estimates using the ad hoc method based on potential airtemperature profiles rises rapidly from 2000-2300 from 10 m to 130 m. The rest129of the night is a primarily steady upward growth to 140 m at 0500 h. The meanNBL height calculated from all methods and all hours between sunset and sunriseis 115.0 m.Vertical profile measurements overnight (0000-0400 h) show profile-averagedpotential air temperature temperature has cooled to 25.7 ◦C, with a mean verticalgradient of 0.0211 K m−1 and maximum gradient of 0.10 K m−1 from 20 - 40 m(Figure 4.8). During these same hours, profile-averaged CO2 has risen to 403.8ppm and mean vertical gradient is -0.16 ppm m−1 with a steepest negative gradientof -1.5 ppm m−1 from 20 - 40 m. Above the NBL, potential temperature profilesindicate presence of a neutrally stable RL with roughly uniform CO2 mixing ratioswith height.In contrast to measurements of NBL structure over other urban areas (e.g.Montreal, Canada (Oke and East, 1971) and Sapporo, Japan (Uno et al., 1988)),there is no observed shallow thermally unstable layer at this site, though thesestudies are representative of different neighborhood types. Instead, the nocturnalpotential temperature inversion begins very near to the surface, presumably dueto minimal storage and anthropogenic heat releases from the micro-scale park set-ting of the measurements and surrounding low-density residential neighborhood.There is also a clear increase in CO2 mixing ratios in the NBL, which is evidenceof vertical mixing of CO2 injected into the stable NBL from nocturnal canopy layersources (e.g. human, soil, and vegetation respiration, fossil fuel combustion fromtraffic and cooking). Because there is no indication of buoyant thermal turbulenceproduction, this mixing must instead be dominated by mechanical processes (i.e.turbulence generated by wind shear and from flow over buildings and trees).Shortly after sunrise (0505 h), ceilometer backscatter measurements show therapid rise of the convective UBL up to a height of 350 m by 1000 h. Duringthis time, CO2 mixing ratios below 100 m show a decrease from 421.3 ppm to402.4 ppm. This indicates vertical flushing of accumulated CO2 in the NBL duringgrowth of the convective UBL, in addition to entrainment of relatively low-CO2content air from the overlying residual layer.The pattern of overnight CO2 buildup and morning flushing of accumulatedCO2 in the NBL observed by the balloon-based system is consistent with obser-vations of CO2 mixing ratios and potential air temperature measured in the UCL130(Chapter 3). In the UCL, there was an observed increase in CO2 mixing ratiosin the hour after sunset, followed by micro-scale horizontal advection along topo-graphic gradients (i.e. cold-air pooling) during the night. Just after sunset, CO2mixing ratios in the UCL were observed to decrease in a spatially uniform manner(Chapter 3).4.3.3 Modeled FC using a boundary layer budgetUrban-scale CO2 fluxes were modeled using a box-model boundary layer budgetapproach and UBL mixing ratio measurements using Eq. 4.6 (Figure 4.9). Fornighttime hours, three variations of the stable NBL height were used, and for day-time hours, three variations of background CO2 mixing ratios for entrainment andadvection were used. FC was modeled for the period 1200 August 14 - 1200 August15, 2008 and the diurnal course of FC is plotted to be centered at noon in Figure4.9.The mean flux of all three variations for the 24-hour experimental period was10.2 µmol m−2 s−1, compared to local-scale EC measurements of 6.8 µmol m−2s−1 observed at the Sunset tower, 1.1 µmol m−2 s−1 measured in the Oakridgeneighborhood, and -3.7 µmol m−2 s−1 measured at Westham Island (Figure 4.9,Table A.2). The July-August 24-hour directionally-averaged ensemble EC meanmeasured at the Sunset tower was 15.5 µmol m−2 s−1; 2.2 µmol m−2 s−1 atOakridge tower; and -1.6 µmol m−2 s−1 at Westham Island. For the Sunset ECmeasurements, directionally-averaged fluxes representative of the entire Sunsetneighborhood are calculated by the equal-weighted average of hourly ensemblemean FC from four wind direction quadrants (0-90◦, 90-180◦, etc.) (Christen et al.,2011). This procedure is performed because the Sunset tower is located approxi-mately 50 m NW of a busy intersection and the EC system is affected by locationbias in term of CO2 emissions (Chapter 2). Using the spatial modeling techniquesdeveloped in Chapter 2, the spatially-averaged mean FC for the Sunset neighbor-hood during the experimental period is 14.4 µmol m−2 s−1.The diurnal course of FC calculated by the boundary layer method shows posi-tive values overnight (mean 3.4 µmol m−2 s−1 from 2100-0500 h), followed by anincrease in emissions up to 49.8 µmol m−2 s−1 at 0900 h, after which FC values131Figure 4.9: Boundary-layer budget calculations of FC during 14-15 August2008. The mean of all model variations is shown as the black line,error bars are maximum and minimum of individual model runs withvarying background CO2 and stable boundary layer height inputs. Alsoshown are observations from Sunset, Oakridge, and Westham Islandflux towers during the balloon experiment (dashed line), and for July-August 2008 (shaded areas, dotted lines).decline throughout the afternoon. During the late afternoon, the calculated net fluxbecomes negative from 1600-1900 h with a minimum of -17.0 µmol m−2 s−1 at1700 h. Though it is not expected that values modeled using the boundary layerbudget approach will correspond precisely with EC measurements because theyrepresent source areas of different spatial scales and source/sink configurations, theoverall diurnal pattern of the boundary layer budget is plausible and qualitativelyagrees with urban Oakridge and Sunset EC measurements, with the exception ofthe late afternoon period (1600-1900 h).132Overnight (2100-0500 h), the mean boundary layer budget FC of 3.4 µmolm−2 s−1 is within 7.5% of the nighttime EC average measured at Oakridge towerfrom July-August 2008 (3.15 µmol m−2 s−1). This modeled value is less than thedirectionally-averaged July-September 2008 mean of 9.3 µmol m−2 s−1 measuredby EC over Sunset neighborhood and the mean of 8.4 µmol m−2 s−1 observed atWestham Island during the same period (Figure 4.9). The mean hourly overnightuncertainty estimate of ± 2.4 µmol m−2 s−1 is based on three different input valuesof stable NBL height to Eq. 4.6. Differences between NBL height estimates areshown in Figure 4.7.For individual hours from 0100-0200 h and 0400-0500 h, FC values modeledby Eq. 4.6 show uptake (negative FC) by the surface when observations of profile-averaged ∆c∆t decrease. Given that actual CO2 uptake by the surface at night is notpossible because there is no active vegetation photosynthesis, these negative values∆c∆t are the result of either entrainment of advective processes not captured by themodel. During these hours, average estimate of the stable NBL height decrease(Figure 4.7) because of reduced wind speeds, both in the NBL and overlying RL(Figure 4.5). This suggests a reduction in vertical mixing from mechanical tur-bulence, and so near-surface values of ∆c∆t are expected to be positive (i.e. CO2buildup). Visual analysis of source areas influencing UBL measurements for 0000-0200 h, based on mean NBL wind speed and direction, indicate influence by ur-banized regions to the east and northeast (Figure 4.10). It is possible that the modelassumption of negligible horizontal gradients of CO2 mixing ratios in this upwinddirection is not valid. Another possible process contributing to observed negative∆c∆t that the model does not capture is relatively low-CO2 mixing ratio air in the RLbeing mixed downward from shear-produced turbulence from higher wind speedsaloft.After sunrise, the model predicts an initial gradual rise in FC followed by asharp peak up to 49.8 µmol m−2 s−1 at 0900 h. The timing and shape of thispeak is similar to the spatially-averaged ensemble mean measured at the SunsetEC system, but the magnitude is larger (28.2 µmol m−2 s−1 at 0900 h at Sun-set tower). The UBL source area estimate for this time extends southwards and133Table 4.2: Modeled boundary layer budget FC compared with values mea-sured during the 24-hour observation period by eddy covariance. ST isSunset flux tower, OR is Oakridge flux tower, and WI is Westham Islandflux tower. The overbar signifies the ensemble mean flux from July -August 2008 for each tower and ￿ST ￿ is the spatially modeled flux fromSunset neighborhood using methods developed in Chapter 2. Units forall values are µmol m−2 s−1.Hour FC ST OR WI ST OR WI ￿ST ￿00:00 -4.54 2.16 -0.78 -24.89 9.57 3.43 7.55 4.2701:00 -2.74 2.71 0.02 NaN 6.06 3.45 7.33 4.7102:00 7.13 5.57 -2.14 NaN 4.45 2.80 7.89 3.4503:00 9.25 9.78 -2.03 13.53 4.51 2.32 8.17 3.2004:00 -4.86 0.53 -4.78 11.63 5.28 2.04 8.33 3.3105:00 0.18 8.12 -12.93 9.28 6.11 2.96 7.29 4.5906:00 -0.09 1.53 -2.19 4.03 11.43 4.10 4.52 14.5507:00 6.59 12.86 12.11 4.77 19.80 5.76 0.37 18.9508:00 7.01 -6.76 10.82 -5.67 20.55 3.22 -5.56 34.9309:00 49.76 5.14 -7.18 -12.07 27.86 -0.31 -9.47 19.4210:00 42.24 19.23 -4.66 -14.17 17.32 0.26 -12.54 20.5711:00 34.77 9.57 0.20 -14.26 28.37 -0.93 -14.58 26.4112:00 27.30 -0.44 -3.29 -13.55 17.26 -0.11 -15.08 16.3513:00 44.35 3.98 2.88 -14.27 19.02 -1.10 -14.55 24.9614:00 12.68 3.86 -5.33 -12.63 26.63 -0.59 -13.80 21.4415:00 24.83 1.03 10.53 -12.19 18.98 -0.33 -12.35 18.5916:00 -0.23 10.80 7.47 -11.07 19.92 -0.12 -10.55 20.7117:00 -16.99 5.80 4.34 -7.85 21.02 1.31 -7.02 13.1718:00 -10.43 2.47 8.59 -2.50 18.82 2.82 -1.89 24.1519:00 -7.06 3.94 6.29 0.34 15.59 3.23 1.36 6.1920:00 2.95 3.43 -1.62 0.79 15.01 4.41 4.83 16.3421:00 6.24 32.75 -1.29 NaN 14.85 4.02 7.15 14.4522:00 6.95 22.68 6.19 29.55 12.69 5.32 7.15 7.5023:00 9.66 2.19 4.95 -6.30 11.91 4.95 7.28 4.3524-hour mean 10.21 6.79 1.09 -3.69 15.54 2.21 -1.59 14.44134Figure 4.10: Individual 60-minute source areas for UBL CO2 mixing ratiomeasurements from 0000-0100 h, 0100-0200 h, 0900-1000 h, and1700-1800 h. The source area downwind dimension is defined by Sxand the base is 2σy. See text for calculation details. Stations in theCO2 monitoring network in operation during the balloon-based mea-surements are also shown. The basemap is a false-color (Bands 745)image from USGS Landsat 5 TM taken September 7, 2011 at 18:50LST.includes predominantly urban residential landcover in south-central Vancouver,heavily trafficked commuter highways leading into Vancouver, industrial zonesalong the Fraser River, and a mixture of urban and agricultural landcover in Rich-mond (Figure 4.10). Higher FC compared to the residential Sunset EC systemcould result from industrial sources and relatively higher traffic density within theupwind source area.On average, the magnitude of entrainment from 0700-1100 h is -31.5 µmol135m−2 s−1 (Figure 4.11) as the height of the convective UBL rises rapidly after sun-rise to 540 m by 1100 h (Figure 4.7). Mean uncertainty due to the different back-ground CO2 mixing ratios used in the entrainment term of Eq. 4.6 is ± 4.0 µmolm−2 s−1 (Figure 4.11). Overall, the situation is simplified because advection is as-sumed negligible during these wind directions. Additional model uncertainties aregenerated because the single-box model construct used in this study assumes a spa-tially homogeneous rise in convective UBL height across the entire model domain.Though this is acknowledged to be a simplification, the assumption is supportedby modeled mixed layer depths for eight individual boxes each with an along-wind length of 1900 m (cumulative length is 15.2 km) extending westward fromthe Sunset tower to the shoreline of the Strait of Georgia (Reid and Steyn, 1997).This study modeled the mixed layer depth in June over each individual box usingparameterized surface sensible heat flux, inversion intensity, mixed layer tempera-ture, and subsidence velocity (Reid and Steyn 1997, Steyn and Oke 1982). Aftersunrise, the mixed layer depths over all boxes are modeled to rise at the same rateup to nearly 600 m until 1100 am.From 0900-1600 h, mean modeled FC steadily decreases from 49.8 µmol m−2s−1 to -0.2 µmol m−2 s−1. As the diurnal seabreeze circulation is established andwind directions veer towards the SW, advection begins to play an important role(Figure 4.11). From 1100-1600 h, the magnitude of the advection term is -30.2µmol m−2 s−1 on average, enough to offset FC and entrainment combined. Theaverage hourly uncertainty range of modeled advection resulting from the inputof different upwind background CO2 mixing ratios is ± 30.6 µmol m−2 s−1 (Fig-ure 4.11). As the background CO2 mixing ratio decreases the horizontal upwindCO2 gradient becomes steeper and the magnitude of advection decreases (i.e. morenegative). Further uncertainty in the model could result from spatially inhomoge-neous changes in the UBL depth after 1100 h. After the seabreeze is established inmid-morning, higher UBL heights are expected further downwind of the land/seaedge in part because the cooler marine air being advected onshore acts to suppressUBL depth (Reid and Steyn 1997). Evidence for spatially variable UBL depth alsocomes from differences in sensible heat flux (QH) observed at Sunset flux towerand Oakridge flux tower. At Sunset tower, mean QH is 297.45 W m−2 and atOakridge tower mean QH is 189.29 W m−2 from 1100-1600 h. QH is an important136Figure 4.11: Modeled surface FC, entrainment, and advection during the 24-hour experimental period. Uncertainty ranges for entrainment and ad-vection are plotted as error bars. The solid line is the observed profile-averaged UBL ∆c∆t by the balloon system and is equal to the sum of theFC, entrainment, and advection terms (Eq. 4.6).driver of boundary layer development used in several models (e.g. Steyn and Oke1982, Batchvarova and Gryning 1991, Batchvarova and Gryning 1994) and higherQH is associated with more vigorous boundary layer growth.During the late afternoon from 1600-2000 h, mean modeled FC is negative, in-dicating uptake by the surface. The magnitude of uptake (-17.0 µmol m−2 s−1 at1700 h) is more negative than observed at Westham Island rural station (-6.8 µmolm−2 s−1 at 1700 h) and is not plausible given the afternoon traffic rush hour ob-served by the Sunset EC tower at this time and indicated by traffic counts (Chapter2). During these afternoon hours, ceilometer observations indicate that the UBL137height falls from 513 m at 1300 h down to 200 m at 1900 h from reductions insurface heating and less vigorous vertical mixing (Figure 4.7). This drop in zi andUBL volume leads to an exchange with the overlying atmosphere that increasesthe CO2 mixing ratio in the UBL, i.e. positive entrainment flux. (Figure 4.11). TheReid and Steyn (1997) modeling study conducted at this same location ignoredentrainment during periods when the UBL height fell.Another factor contributing towards modeled negative FC during the late af-ternoon is the choice of upwind CO2 mixing ratio input during calculation of theadvection term. Mean advection from 1300-1900 h trends towards zero as theprofile-averaged CO2 mixing ratios decrease to the daily minimum levels (376.5ppm at 1700 h) which acts to reduce the upwind horizontal gradient ∆cx∆Sx . The mea-sured profile-averaged value actually falls below one of the upwind input valuesof 382 ppm which results in a positive advective flux (Figure 4.11). ObservedCO2 mixing ratios below background values during late afternoon was also a fea-ture of the modeling study conducted by Reid and Steyn (1997). Profile-averagedwind speeds are strongest during this time and visual analysis indicates that sourceareas extend upwind past YVR airport into the Strait of Georgia (Figure 4.10), ev-idence that advection of marine air is likely responsible for observed reductions inprofile-averaged CO2 mixing ratios. A shortcoming of the model for these hours isinaccurate parameterization of upwind CO2 mixing ratios in the marine air volume.4.4 ConclusionsMeasurements of CO2 mixing ratio, potential air temperature, wind speed, andwind direction were conducted in the urban boundary layer of Vancouver, Canadaover a continuous 24-hour period in August, 2008. Observations were used tomodel integrated urban-scale CO2 fluxes at hourly timescales using a boundarylayer budget calculation and box-model construct. Model results were compared toobservations from three local-scale EC towers in operation during the experimentin the greater Vancouver region. Observations are representative of a synopticallycalm situation characterized by weak horizontal pressure gradients (low regionalwind speeds) and clear skies (high solar radiation input during day, strong radiativecooling overnight). Though this particular set of atmospheric conditions is not138typical for this area (i.e. not representative of winter or during passage of synoptic-scale systems), these conditions allow for near-ideal conditions to observe specificprocesses operating at the city-scale such as formation of a stable nocturnal UBL,development of a daytime convective UBL, and onset of thermally driven land-seabreezes.Overnight, measured vertical profiles of potential air temperature show devel-opment of a stable NBL with overlying neutrally stable RL beginning just aftersunset. Measured vertical profiles of CO2 mixing ratio during this same time showa clear buildup of CO2 in the stable NBL. Three estimates of NBL height wereused and average NBL height of all three methods is 115 m. Vertical potentialtemperature profiles and EC measurements from nearby Sunset tower indicate ver-tical mixing CO2 in the NBL is dominated by mechanical processes. The boundarylayer budget model predicts positive hourly FC on average overnight (average 3.4µmol m−2 s−1 from 2100-0500 h) and generally performs well (within 7.5% ofOakridge residential flux tower measurements), though individual hours show un-realistic negative FC due to advective and entrainment effects not captured by themodel.After sunrise from 0600-1100 h, there is rapid growth in UBL height andprofile-averaged CO2 mixing ratios decrease due to flushing of accumulated overnightCO2 in the NBL and entrainment from above. Modeled FC during this time risesto a peak at 0900 h (49.8 µmol m−2 s−1). The timing and shape of the increase inFC is similar to EC measurements at Sunset tower, though the magnitude of mea-surements is not as large (28.2 µmol m−2 s−1). Entrainment of low-CO2 mixingratio air from above is an important process during this time and average hourlymodeled uncertainty resulting from different background CO2 mixing ratio inputsto the model is ±4.0 µmol m−2 s−1.Throughout the afternoon (1100-1800 h), observed profile-averaged CO2 mix-ing ratios continue to decrease even though the surface is expected to be a netsource of CO2. Measurements show a well-mixed UBL with roughly verticallyuniform profiles of potential air temperature and CO2 mixing ratios. During thisperiod, advection from upwind low-CO2 mixing ratio marine air is an importantprocess. Though modeled FC magnitude is realistic from 1100-1600 h, there islarge uncertainty due to advection (hourly average ±30.6 µmol m−2 s−1) resulting139from different background CO2 mixing ratio model inputs.During late afternoon (1800-2000 h), the model performs poorly and predictsunrealistic uptake (negative FC). This is largely due to profile-averaged measure-ments in the UBL falling near (and even below) the input values for backgroundCO2 mixing ratio used to calculate advection. Model performance during this pe-riod would be improved with more precise of upwind CO2 mixing ratio measure-ments or parameterization. Performance would also be likely enhanced with useof a model made from a larger number of smaller boxes to simulate spatial hetero-geneity of UBL growth and decay, similar to the model used by Reid and Steyn(1997) to predict CO2 mixing ratios at Sunset tower.In summary, the model performs reasonably well overnight, during the initialgrowth of the UBL, and in the early afternoon. Phenomena such as the growthof the nocturnal stable boundary layer and growth of the convective UBL are spa-tially simplified in this single-box formulation, but uncertainty ranges with regardsto entrainment are acceptable and modeled FC is realistic. The largest model un-certainties result from advection, especially during the late afternoon when modelperformance is poor. Improved measurements or parameterization of upwind CO2mixing ratios used to calculate horizontal spatial gradients, as well as increasedmodel spatial resolution, would likely improve modeling of advection, and alsoFC.Additional measurements, such as observation of carbon isotopes 14 C and 12C,could also help distinguish between advection of marine air with CO2 originatingprimarily from biogenic sources and urbanized air with CO2 primarily from fossil-fuel sources (e.g. Pataki et al. 2007). Measurements of gases such as carbonylsulfide (COS) or methyl iodide (CH3I) could also potentially be used as tracers formarine air (e.g. Bell et al. 2002, Kettle et al. 2002).Given the complex coastal meteorology at this site (e.g. sea breeze circulation,spatial heterogeneity of UBL dynamics), a comparison of model performance us-ing a high resolution inverse atmospheric transport modeling approach would bebeneficial. Such a comparison would be especially interesting during the afternoonperiod when several meteorological processes (entrainment, advection) are inter-acting simultaneously with surface CO2 sources/sinks to influence CO2 mixingratios in the UBL. More sophisticated treatment of the upwind source area influ-140encing UBL measurements using an inverse modeling approach would also likelyyield insights into urban-scale CO2 fluxes.This work also has implications for monitoring of urban-scale CO2 emissionsusing CO2 mixing ratio measurements. Tower-based CO2 mixing ratio observa-tions in the surface layer at Sunset tower are representative of the entire well-mixedUBL during convective situations and below the stable NBL height overnight dur-ing this 24-hr observation period. This implies future research, using either a box-model approach or inverse atmospheric modeling, could take advantage of long-term (multi-year) mixing ratio measurements to estimate urban-scale CO2 fluxes.Repeated measurements averaged over many days in varying atmospheric condi-tions would result in more robust emissions estimates.141Chapter 5Conclusions5.1 Summary of findingsThe broad objectives of this research are to advance measurement and analysistechniques to map, partition, and quantify CO2 emissions/uptake processes and at-mospheric transport in the urban environment. This is accomplished through novelmeasurements and innovative data analysis representative of three urban climatescales in Vancouver, Canada. Together, these datasets and their analysis provide acoherent view of urban CO2 from emissions/uptake at the surface, to storage andtransport within the urban canopy layer, and finally to mixing and transport in theurban boundary layer.First, specific results concerning urban emissions/uptake processes and theirmeasurement are summarized:1. Eddy covariance (EC) carbon dioxide flux (FC) measurements from Sunsettower are linearly related to the plan-area landcover proportion of busy, arte-rial roads in the turbulent flux source area. Linear coefficients vary by hourdepending on diurnal traffic density patterns.2. When traffic emissions are controlled for, hourly FC has a negative linear cor-relation with air temperature. Linear coefficients vary by hour and are validwhen temperatures fall below a statistically defined temperature threshold.This relation is attributed to hourly emissions due to building space-heating.1423. During summer nights, FC measurements are shown to increase exponen-tially with soil temperature observations. This relation is also expectedto hold during winter, however conditional sampling techniques used limitanalysis to summer. The difference between observed EC measurements andsoil chamber measurements (average 1.4 µmol m−2 s−1) is speculated to bethe result of human respiration.4. When traffic emissions are controlled for, observed FC during daytime sum-mer hours decreases as PPFD increases. This is attributed to vegetation pho-tosynthesis.5. Based on these relations, modeled ecosystem-wide annual net emissions forthe Sunset neighborhood are calculated as 6.42 kg C m−2 y−1 with trafficemissions estimated to account for 68.8% of total net emissions. Buildingsources contribute 27.9%, respiration from soil and vegetation is responsiblefor 5.5%, respiration from humans is 5.0%, and photosynthesis offsets are-7.2% of the annual net total.6. On average, the EC storage flux term is minor, only 2.8% of the magnitudeof FC. During a 24-hour observation period, using a two-layer method tocalculate the storage flux changes FC by 5.2% on average. Largest uncer-tainties from hourly storage in the UCL are during the transition from day-night when there is observed UCL buildup and from day-night when there isobserved venting of CO2.Next, findings related to CO2 mixing ratios and atmospheric transport in theurban environment of the Sunset neighborhood are summarized. Specific resultsare organized by description of a typical 24-hour diurnal cycle during calm summeranti-cyclonic conditions in Vancouver, beginning at sunset:1. Beginning at sunset, there is formation of a thermally stable NBL layer withan average height of 115 m over the course of the night. Just after sunset,there is a spatially uniform increase in UCL CO2 mixing ratios up to 1.5 ppmhr−1.1432. Vertical potential temperature profiles gradients overnight up to 400 m areon average 0.0211 K m−1 with a maximum gradient of 0.10 K m−1 from20 - 40 m. Vertical CO2 mixing ratio gradients are on average -0.16 ppmm−1 with a steepest negative gradient of -1.5 ppm m−1 from 20 - 40 m. Thisindicates lack of a shallow, weakly unstable mixed layer at this site overnightand indicates observed vertical mixing in the stable nocturnal boundary layeris the result of mechanical processes.3. Overnight, patterns of CO2 mixing ratio in the UCL are negatively correlatedwith potential air temperature (R=-0.57). This is evidence that micro-scaleadvection processes (i.e. cold air drainage and pooling in low elevation ar-eas) is an important process regulating pollutant transport and exposure inthe UCL.4. Just after sunrise, built-up CO2 is flushed from the UCL and NBL as the con-vective UBL begins to form and entrained air from the RL is mixed down-ward. In the UCL, venting of the UCL is observed throughout the study areawith magnitudes up to -11 ppm hr−1. Mean entrainment flux in the UBLduring this time is calculated as -31.5 ± 4.0 µmol m−2 s−1.5. During daytime, highest CO2 mixing ratios in the UCL (+33 ppm, 7.7%above spatial mean) are found above busy roads and spatial patterns arelargely a function of proximity to traffic sources.6. Throughout the afternoon, CO2 mixing ratios in the UBL decrease eventhough the surface remains a net source of CO2. Advection of low-CO2marine air is an important process with mean magnitude of -30.2 µmol m−2s−1, though the uncertainty range is large (±30.6 µmol m−2 s−1).7. Throughout the 24-hour period, CO2 mixing ratios in the UCL are consis-tently higher than observations on the EC tower in the ISL. The magnitudeof UCL-ISL gradients decreases exponentially with increases to the standarddeviation of vertical wind velocity measured by the EC tower.1445.2 Contributions and implicationsThe key contributions of this research towards advancing understanding of urbanCO2 source/sink processes and atmospheric transport are:1. Development of methods to spatially attribute EC measurements of net FC toindividual source/sink processes. This is an innovative application of turbu-lent flux source area models to partition EC measurements and develop spa-tially unbiased urban exchange totals on hourly to annual timescales. Thisis a significant advancement that has implications for improved direct moni-toring of urban emissions at neighborhood scales and for evaluation and de-velopment of independent emissions inventories and models (e.g. Christenet al. 2011,Kellett et al. 2012).2. Investigation of the hourly EC storage flux term and recommendations for itsmeasurement and calculation. The storage term has largely been neglectedin urban EC studies and its in-depth treatment here establishes measurementguidelines and serves as a basis for future work. This contribution furtherrefines the EC measurement technique in urban areas for increased accuracyof net FC measurements on hourly timescales.3. Observation of micro-scale spatial CO2 mixing ratio patterns. Novel ap-plication of vehicle transect measurement techniques at the urban climatemicro-scale revealed striking and unexpected spatial patterns of CO2 mixingratios in the UCL. Because CO2 can be used as a tracer gas for other pollu-tants, these observations have implications for pollutant transport and humanexposure and health risk in urban areas.4. High-resolution observations of UBL CO2 and thermal dynamics. A uniquedataset of vertical profiles of CO2 mixing ratio and potential air tempera-ture in the UBL quantifies several key features of CO2 transport, including:development of a thermally stable NBL over a medium density residentialneighborhood (LCZ ‘open-set lowrise’) and associated CO2 buildup, vent-ing of near-surface CO2 during growth of the convective UBL, and the in-fluence of entrainment and advection on UBL CO2 mixing ratios. This work145has implications for monitoring and modeling urban-scale emissions fromCO2 mixing ratio measurements through boundary-layer budget or inverseatmospheric modeling approaches. Findings also have implications for de-velopment and evaluation of high-resolution atmospheric models in complexcoastal urban areas (e.g. Leroyer et al. 2014).5.3 Reflections and future workThis research used primarily observational techniques and data analysis methodsto investigate CO2 source/sink processes and atmospheric transport in Vancouver,Canada. These approaches yielded unique findings and significant contributions,but there are several inherent limitations. Specific methodological shortcomingsand potential improvements are discussed in detail in Chapters 2-4, so this space isdevoted to a more general reflection on the overall research approach.A significant limitation is the geographical scope of CO2 flux and mixing ratioobservations. At this site, the EC technique is able to measure direct emissionsfrom within turbulent flux source areas with limited spatial range (lengthscales <2 km). Methods to calculate urban-scale emissions using changes in hourly UBLCO2 mixing ratios have source areas with lengthscales of ∼10 km at this site. Thismeans that important upstream processes outside of city limits such as fossil fuelcombustion for electrical generation and imbedded emissions related to transportand production of food and material goods are not captured by measurements (e.g.Christen et al. 2011). This is also true for downstream processes such as emissionsassociated with waste management and disposal. This production-based approach,rather than a consumption-based approach, is a limited perspective because specificbehaviors and consumption patterns of urban residents control CO2 emissions inlocations outside of city limits.This work also views CO2 emissions and uptake as functions of individualphysical source/sink processes operating independently and simultaneously togetherin an urban area. Interactions and feedbacks between individual processes are ac-knowledged, and in some cases implicitly embedded in observations, but are notexplicitly quantified. For example, higher CO2 content in the urban atmospherecould potentially increase photosynthetic uptake by vegetation (Long et al., 2004).146Changes to vegetation cover can also modify the surface energy balance and alterbuilding energy demand and emissions through shading from sun and wind (Ak-bari et al., 2001). Increased UCL air temperatures and soil temperatures from theurban heat island also impact FC through alteration of building energy demand andemissions related to space heating and modification of biogenic respiration andphotosynthesis rates. Development of dynamic urban emissions models for plan-ning or research applications should investigate and attempt to incorporate theseaffects.Research here is also limited to investigation of physical CO2 source/sink pro-cesses and the underlying social, economic, and technological context of emis-sions are neglected. In part, this is justifiable because these forces vary on longertimescales (roughly decadal-multidecadal) and are assumed constant on the hourly-annual timescales of analysis here. At the same time, a more holistic explanationof neighborhood and urban-scale CO2 emissions must take into account the socio-economic fabric of the city as well as its physical expressions.Finally, as with all observational studies, there is the question of the repre-sentiveness of measurements. Are results here specific to the time and place ofmeasurements, or do they represent more fundamental properties underlying allurban ecosystems? Specific findings, such as ecosystem-average annual emissionstotals or specific parameters fit to soil respiration and photosynthesis models, areunique to the study site, but the hope is that the measurement techniques and an-alytical methods developed here are universal and can be exported to any urbanarea. For example, EC measurements have been applied to different neighborhoodtypes in urban areas the world over and it is reasonable to expect that analysis us-ing turbulent flux source area modeling could yield insights, given the study areasatisfies certain morphological considerations (e.g. spatially homogenous rough-ness). Other phenomena observed here, such as nocturnal advection in the UCLfollowing micro-scale topography, are likely not unique to this research site andobservations can be repeated elsewhere.Despite limitations of measurement-based approaches, there are several poten-tially productive avenues of research that could be followed to build upon findingsin this work:1471. Additional EC CO2 flux measurements in representative neighborhoods. Ex-panding the range of study areas in terms of transportation regimes, popula-tion densities, building densities and structures, fuel sources, vegetation, andclimate conditions is needed to understand the full range of CO2 source/sinkprocesses, configurations, and behavior. To date, EC measurements are bi-ased towards neighborhoods in Northern Hemisphere cities in countries withdeveloped economies. Since future urban growth is projected to be greatestin tropical cities in countries with developing economies, future measure-ments should aim to include neighborhoods in these cities.2. Additional urban EC measurements of CH4 and N2O fluxes. These gases areincluded in many municipal greenhouse gas inventories, but there are rela-tively few measurements of local-scale fluxes in urban areas (e.g.Famulariet al. 2010, Christen et al. 2013). These gases are important to considerbecause of their high radiative forcing potential relative to CO2.3. Further investigation of specific CO2 source/sink process behavior and in-teractions between processes. For example, how does urban vegetation pho-tosynthetic uptake vary with changing ambient CO2 mixing ratios? Whatis the effect of various traffic management strategies (e.g. roundabouts vs.intersections, speed limits, congestion taxes, etc.) on CO2 emissions? Howmuch are emissions from underground sewage or transportation networks?This could have implications for urban planning and design scenario model-ing.4. Establishment of urbanCO2 mixing ratio measurement networks. This wouldbe helpful to monitor urban-scale emissions and refine urban-scale atmo-spheric transport models. Measurements are especially needed in upwindnon-urban areas and above the urban boundary layer to establish horizontaland vertical gradients used to calculate advection and entrainment. Columnmeasurements in the UBL representative of urban-scale conditions are alsovaluable because they represent larger spatial areas than near-surface mea-surements. Emerging technologies such as flying drones and space-basedobservations could be used as innovative measurement platforms.1485. More detailed investigation of micro-scale advection and transport in theUCL. This especially refers to observed nocturnal observations of cold-airdrainage and pooling along micro-scale topographical gradients in Chapter3. What is the frequency of this phenomena? What are antecedent conditionsfor its development? What are spatial patterns of other pollutants? Howdoes micro-scale advection affect EC measurements? High-resolution mea-surements or use of path-averaging measurement systems (Christen, 2014)combined with computational fluid dynamic modeling could be used to in-vestigate these questions.6. Further study on the impact of surface spatial heterogeneity onMonin-Obukhovsimilarity theory predictions. This could be accomplished by comparison ofnormalized scalar standard deviations (e.g. CO2, H2O, heat) under varyingspatial source/sink configurations (e.g. from contrasting flux source areas,during different seasons). Scalar dissimilarity has been observed to evolveseasonally in forest ecosystems (Williams et al., 2007) and this type of studywould build on research in urban ecosystems by Roth and Oke (1995) andMoriwaki and Kanda (2006).7. Collaboration with urban geographers and planners for more holistic treat-ment of emissions and scenario modeling. For example, what socio-economicvariables are correlated with neighborhood scale emissions? How can green-house gas models and inventories be incorporated with neighborhood andurban planning development scenarios?As humans continue to concentrate in urban areas in the upcoming decades, de-veloping a transparent and effective measurement framework for monitoring urbanCO2 (and other greenhouse gases) emissions will be important. If cities and mu-nicipal governments continue to lead on greenhouse gas reduction strategies, theframework should have standardized features and be applicable across many citytypes so that meaningful comparisons between cities can be made. Application todeveloping mega-cities is especially important because these areas are projectedto absorb the bulk of the world’s urban population growth. The work presentedin this dissertation (and ideas for future research) can be viewed as steps towards149establishing such a framework.Specifically, the multi-scale approach used in this work is recommended as animportant feature of an urban greenhouse gas measurement framework. City-scalemeasurements, either from a fixed CO2 sensor network or space-based measure-ments of the urban air column, could be combined with atmospheric models andused to establish emissions baselines, monitor overall progress towards reductiontargets, and evaluate emissions inventories (e.g. Christen 2014). More detailed in-formation that discerns differences at the neighborhood-scale, either through down-scaled inventories or models, could be used to identify emissions hot-spots or de-velop and test mitigation strategies. Direct neighborhood-scale measurements arepossible through EC, but it is likely impractical in the near future to implement anetwork of EC towers in a city due to cost and logistical constraints. Furthermore,EC is not applicable for densely built neighborhoods with very tall buildings be-cause measurements need to be 2-3 times the height of buildings and remain withinthe UBL (Feigenwinter et al., 2012). Therefore, a strategy of using down-scaled in-ventories or emissions models that have been evaluated against EC measurementscould be a viable strategy to track neighborhood-scale emissions across an urbanregion. Comparison to measurements could be accomplished through short-term insitu observation programs or from development of emissions factors based on rep-resentative EC measurements in other cities. At the micro-scale, a network of CO2mixing ratio measurements could be used in conjunction with city-scale measure-ments and high-resolution flow models to constrain neighborhood-scale emissionsestimates and investigate point sources of emissions such as power plants, indus-trial facilities, or busy transportation hubs.Such a multi-scale monitoring approach combining measurements, inventories,and models would provide a coherent picture of a city’s emissions in near real-time, provide linkages between measured concentration fields and emissions, andhelp to identify representative sites for fixed monitoring stations. This would alsoallow on-the-fly assessment of emissions management strategies and allow moreflexibility in terms of policy responses. Such responses are important because theongoing urbanization of our species represents an opportunity to develop along amore sustainable trajectory.150BibliographyAkbari, H. and Konopacki, S. 2005. Calculating energy-saving potentials ofheat-island reduction strategies. Energy Policy, 33(6):721–756. → pages 10Akbari, H., Pomerantz, M., and Taha, H. 2001. Cool surfaces and shade trees toreduce energy use and improve air quality in urban areas. Solar energy,70(3):295–310. → pages 147Arrhenius, S. 1896. On the influence of carbonic acid in the air upon thetemperature of the ground. The London, Edinburgh, and Dublin PhilosophicalMagazine and Journal of Science, 41(251):237–276. → pages 2ASHRAE 2004. Ventilation for acceptable indoor air quality. Technical report,American Society of Heating, Refrigeration, and Air-conditioning Engineers,Inc. → pages 57, 83Aubinet, M., Berbigier, P., Bernhofer, C., Cescatti, A., Feigenwinter, C., Granier,A., Gru¨nwald, T., Havrankova, K., Heinesch, B., Longdoz, B., et al. 2005.Comparing CO2 storage and advection conditions at night at differentCARBOEUROFLUX sites. Boundary-Layer Meteorology, 116(1):63–93. →pages 69, 70, 100, 101, 103Aubinet, M., Vesala, T., and Papale, D., editors 2012. Eddy covariance: apractical guide to measurement and data analsysis. Springer AtmosphericSciences. → pages 25Baldocchi, D. 2003. Assessing the eddy covariance technique for evaluatingcarbon dioxide exchange rates of ecosystems: past, present and future. GlobalChange Biology, 9(4):479–492. → pages 20Batchvarova, E. and Gryning, S.-E. 1991. Applied model for the growth of thedaytime mixed layer. Boundary-Layer Meteorology, 56(3):261–274. → pages137151Batchvarova, E. and Gryning, S.-E. 1994. An applied model for the height of thedaytime mixed layer and the entrainment zone. Boundary-Layer Meteorology,71(3):311–323. → pages 137Bates, D. and Watts, D. 1988. Nonlinear regression analysis and its applications,volume 2. Wiley Online Library. → pages 46Bell, A. G. B. 1917. In Dictionary of Canadian Biography Online, volume XV.University of Toronto and Laval.htt p : // = 7894. → pages 2Bell, N., Hsu, L., Jacob, D., Schultz, M., Blake, D., Butler, J., King, D., Lobert, J.,and Maier-Reimer, E. 2002. Methyl iodide: Atmospheric budget and use as atracer of marine convection in global models. Journal of GeophysicalResearch: Atmospheres (1984–2012), 107(D17):ACH–8. → pages 140Benkley, C. W. and Schulman, L. L. 1979. Estimating hourly mixing depths fromhistorical meteorological data. Journal of Applied Meteorology,18(6):772–780. → pages 118Bergeron, O. and Strachan, I. B. 2011. CO2 sources and sinks in urban andsuburban areas of a northern mid-latitude city. Atmospheric Environment,45(8):1564–1573. → pages 26Berry, R. and Colls, J. 1990. Atmospheric carbon dioxide and sulphur dioxide onan urban/rural transect I. Continuous measurements at the transect ends.Atmospheric Environment. Part A. General Topics, 24(10):2681–2688. →pages 18Black, T., Hartog, G., Neumann, H., Blanken, P., Yang, P., Russell, C., Nesic, Z.,Lee, X., Chen, S., Staebler, R., et al. 1996. Annual cycles of water vapour andcarbon dioxide fluxes in and above a boreal aspen forest. Global ChangeBiology, 2(3):219–229. → pages 69Bre´on, F., Broquet, G., Puygrenier, V., Chevallier, F., Xueref-Re´my, I., Ramonet,M., Dieudonne´, E., Lopez, M., Schmidt, M., Perrussel, O., et al. 2014. Anattempt at estimating Paris area CO2 emissions from atmospheric concentrationmeasurements. Atmospheric Chemistry and Physics Discussions,14(7):9647–9703. → pages 106Bulkeley, H. 2013. Cities and climate change. Routledge. → pages 5, 16152Burri, S., Frey, C., Parlow, E., and Vogt, R. 2009. CO2 fluxes and concentrationsover an urban surface in Cairo/Egypt. In Proceedings of the SeventhInternational Conference on Urban Climate. → pages 20Campbell, G. S. and Norman, J. M. 1998. Introduction to environmentalbiophysics. Springer Verlag. → pages 56Chen, B., Black, T., Coops, N., Hilker, T., Trofymow, J., and Morgenstern, K.2009. Assessing tower flux footprint climatology and scaling between remotelysensed and eddy covariance measurements. Boundary-Layer Meteorology,130(2):137–167. → pages 35Christen, A. 2014. Atmospheric measurement techniques to quantify greenhousegas emissions from cities. Urban Climate. → pages 149, 150Christen, A., Coops, N., Crawford, B., Kellett, R., Liss, K., Olchovski, I., Tooke,T., van der Laan, M., and Voogt, J. 2011. Validation of modeled carbon-dioxideemissions from an urban neighborhood with direct eddy-covariancemeasurements. Atmospheric Environment, 45(33):6057–6069. → pages 25, 26,27, 35, 36, 46, 57, 61, 66, 75, 102, 115, 131, 145, 146Christen, A., Johnson, M., Molodovskaya, M., Ketler, R., Nesic, Z., Crawford, B.,Giometto, M., and van der Laan, M. 2013. Integral emission factors formethane determined using urban flux measurements and local-scale inversemodels. In EGU General Assembly Conference Abstracts, volume 15, page6143. → pages 148Christen, A., Rotach, M. W., and Vogt, R. 2009. The budget of turbulent kineticenergy in the urban roughness sublayer. Boundary-layer meteorology,131(2):193–222. → pages 14, 15Christen, A., van Gorsel, E., and Vogt, R. 2007. Coherent structures in urbanroughness sublayer turbulence. International journal of Climatology,27(14):1955–1968. → pages 15Christen, A., van Gorsel, E., Vogt, R., Andretta, M., and Rotach 2001. UltrasonicAnemometer Instrumentation at Steep Slopes: Wind Tunnel Study - FIeldIntercomparion - Measurements. MAP Newsletter, 15:164–167. → pages 177Christen, A. and Vogt, R. 2004. Energy and radiation balance of a CentralEuropean city. International Journal of Climatology, 24(11):1395–1421. →pages 71153Churkina, G., Brown, D. G., and Keoleian, G. 2010. Carbon stored in humansettlements: the conterminous United States. Global Change Biology,16(1):135–143. → pages 8City of Vancouver 2012. Vanmap., Accessed May 29, 2013.→ pages 33, 75Clarke, J. F. and Faoro, R. B. 1966. An Evaluation of CO2 Measurements as anIndicator of Air Pollution. Journal of the Air Pollution Control Association,16(4):212–218. → pages 18Cleugh, H. and Grimmond, C. 2001. Modelling regional scale surface energyexchanges and CBL growth in a heterogeneous, urban-rural landscape.Boundary-layer meteorology, 98(1):1–31. → pages 107, 108, 117Cleugh, H. and Oke, T. 1986. Suburban-rural energy balance comparisons insummer for Vancouver, BC. Boundary-Layer Meteorology, 36(4):351–369. →pages 27Collins, J. P., Kinzig, A., Grimm, N. B., Fagan, W. F., Hope, D., Wu, J., andBorer, E. T. 2000. A new urban ecology. American Scientist, 88(5):416–425.→ pages 6Counehan, J. 1971. Wind tunnel determination of the roughness length as afunction of the fetch and the roughness density of three-dimensional roughnesselements. Atmospheric Environment, 5(8):637–642. → pages 35Coutts, A. M., Beringer, J., and Tapper, N. J. 2007. Characteristics influencing thevariability of urban CO2 fluxes in Melbourne, Australia. AtmosphericEnvironment, 41(1):51–62. → pages 20, 72Crawford, B. and Christen, A. 2014. Spatial source attribution of measured urbaneddy covariance co2 fluxes. Theoretical and Applied Climatology, pages 1–23.→ pages 75, 76, 82, 85, 98, 102Crawford, B. and Christen, A. August 6-10, 2012. Quantifying the CO2 storageflux term in urban eddy-covariance observations. In 8th InternationalConference on Urban Climates, Dublin, Ireland. → pages 29Crawford, B., Christen, A., and Ketler, R. 2010. Eddy covariance data processingand quality control procedures, EPiCC Technical Report No.1. 11 p. → pages 29, 76, 173154Crawford, B., Grimmond, C., and Christen, A. 2011. Five years of carbon dioxidefluxes measurements in a highly vegetated suburban area. AtmosphericEnvironment, 45(4):896–905. → pages 20, 24, 26, 72CSI 2011. CSAT3 Three Dimensional Sonic Anemometer. Campbell ScientificIncorporated, revision 10/11 edition. → pages 174Denmead, O., Raupach, M., Dunin, F., Cleugh, H., and Leuning, R. 1996.Boundary layer budgets for regional estimates of scalar fluxes. Global ChangeBiology, 2(3):255–264. → pages 108, 117Effland, W. R. and Pouyat, R. V. 1997. The genesis, classification, and mappingof soils in urban areas. Urban Ecosystems, 1(4):217–228. → pages 6Environment Canada 2013. e.html, Accessed May 29,2013. → pages 30EPA 2014. →pages 17Ewing, R., Pendall, R., and Chen, D. 2003. Measuring sprawl and itstransportation impacts. Transportation Research Record: Journal of theTransportation Research Board, 1831(1):175–183. → pages 10Falge, E., Baldocchi, D., Tenhunen, J., Aubinet, M., Bakwin, P., Berbigier, P.,Bernhofer, C., Burba, G., Clement, R., Davis, K. J., et al. 2002. Seasonality ofecosystem respiration and gross primary production as derived fromFLUXNET measurements. Agricultural and Forest Meteorology,113(1):53–74. → pages 54, 70Famulari, D., Nemitz, E., Di Marco, C., Phillips, G. J., Thomas, R., House, E., andFowler, D. 2010. Eddy-covariance measurements of nitrous oxide fluxes abovea city. Agricultural and Forest Meteorology, 150(6):786–793. → pages 148Feigenwinter, C., Bernhofer, C., Eichelmann, U., Heinesch, B., Hertel, M.,Janous, D., Kolle, O., Lagergren, F., Lindroth, A., Minerbi, S., et al. 2008.Comparison of horizontal and vertical advective CO2 fluxes at three forest sites.Agricultural and Forest Meteorology, 148(1):12–24. → pages 70Feigenwinter, C., Vogt, R., and Christen, A. 2012. Eddy covariance measurementsover urban areas. In Eddy Covariance, pages 377–397. Springer. → pages 72,150155Finnigan, J. 2006. The storage term in eddy flux calculations. Agricultural andForest Meteorology, 136(3):108–113. → pages 69Finnigan, J., Clement, R., Malhi, Y., Leuning, R., and Cleugh, H. 2003. Are-evaluation of long-term flux measurement techniques part I: averaging andcoordinate rotation. Boundary-Layer Meteorology, 107(1):1–48. → pages 68,180Font, A., Grimmond, C., Morguı´, J.-A., Kotthaus, S., Priestman, M., and Barratt,B. 2013. Cross-validation of inferred daytime airborne CO2 urban-regionalscale surface fluxes with eddy-covariance observations and emissionsinventories in Greater London. Atmospheric Chemistry and PhysicsDiscussions, 13(5):13465–13493. → pages 18, 107, 117Fourier 1827. Memoire sur la temperature du globe terrestre et des espacesplanetaires. In Memoirs of the Royal Academy of Sciences of the Institut deFrance, pages 569–604. → pages 1Fourier, J. 1824. In Annales de chimie et de physique, pages 236–281. → pages 1Froelich, N. and Schmid, H. 2006. Flow divergence and density flows above andbelow a deciduous forest: Part II. Below-canopy thermotopographic flows.Agricultural and forest meteorology, 138(1):29–43. → pages 70George, K., Ziska, L. H., Bunce, J. A., and Quebedeaux, B. 2007. Elevatedatmospheric CO2 concentration and temperature across an urban-rural transect.Atmospheric Environment, 41(35):7654–7665. → pages 92Gohar, L. and Shine, K. 2007. Equivalent CO2 and its use in understanding theclimate effects of increased greenhouse gas concentrations. Weather,62(11):307–311. → pages 17Golubiewski, N. 2012. Is there a metabolism of an urban ecosystem? anecological critique. Ambio, 41(7):751–764. → pages 7Goodwin, N., Coops, N., Tooke, T., Christen, A., and Voogt, J. 2009.Characterizing urban surface cover and structure with airborne lidar technology.Canadian Journal of Remote Sensing, 35(3):297–309. → pages 32, 75Goulden, M., Munger, J., Fan, S., Daube, B., Wofsy, S., et al. 1996. Measurementsof carbon sequestration by long-term eddy covariance: methods and a criticalevaluation of accuracy. Global Change Biology, 2(3):169–182. → pages 67156Grimmond, C., King, T., Cropley, F., Nowak, D., and Souch, C. 2002. Local-scalefluxes of carbon dioxide in urban environments: methodological challenges andresults from Chicago. Environmental Pollution, 116:243–254. → pages 20, 24,67, 72, 98Grimmond, C., Salmond, J., Oke, T. R., Offerle, B., and Lemonsu, A. 2004. Fluxand turbulence measurements at a densely built-up site in marseille: Heat, mass(water and carbon dioxide), and momentum. Journal of Geophysical Research:Atmospheres (1984–2012), 109(D24). → pages 20Grimmond, C. S. B. and Oke, T. R. 1991. An evapotranspiration-interceptionmodel for urban areas. Water Resources Research, 27(7):1739–1755. → pages27Gu, L., Falge, E. M., Boden, T., Baldocchi, D. D., Black, T., Saleska, S. R., Suni,T., Verma, S. B., Vesala, T., Wofsy, S. C., et al. 2005. Objective thresholddetermination for nighttime eddy flux filtering. Agricultural and ForestMeteorology, 128(3):179–197. → pages 70, 101Habitat, U. 2011. Cities and Climate ChangeGlobal Report on HumanSettlements 2011. London: Earthscan. → pages 16Hadley, S. W. and Tsvetkova, A. A. 2009. Potential impacts of plug-in hybridelectric vehicles on regional power generation. The Electricity Journal,22(10):56–68. → pages 10Helfter, C., Famulari, D., Phillips, G. J., Barlow, J. F., Wood, C. R., Grimmond, C.S. B., and Nemitz, E. 2011. Controls of carbon dioxide concentrations andfluxes above central London. Atmospheric Chemistry and Physics,11(5):1913–1928. → pages 72Henninger, S. and Kuttler, W. 2007. Methodology for mobile measurements ofcarbon dioxide within the urban canopy layer. Climate Research, 34(2):161. →pages 73, 105Henninger, S. and Kuttler, W. 2010. Near surface carbon dioxide within the urbanarea of Essen, Germany. Physics and Chemistry of the Earth, Parts A/B/C,35(1-2):76–84. → pages 9, 19, 73, 92, 103, 105Hiller, R. V., McFadden, J. P., and Kljun, N. 2011. Interpreting CO2 fluxes over asuburban lawn: The influence of traffic emissions. Boundary-layermeteorology, 138(2):215–230. → pages 26157Hollinger, D., Kelliher, F., Byers, J., Hunt, J., McSeveny, T., and Weir, P. 1994.Carbon dioxide exchange between an undisturbed old-growth temperate forestand the atmosphere. Ecology, pages 134–150. → pages 69Hoornweg, D., Sugar, L., and Gomez, C. L. T. 2011. Cities and greenhouse gasemissions: moving forward. Environment and Urbanization, 23(1):207–227.→ pages 11, 17, 18Idso, C., Idso, S., and Balling Jr, R. 1998. The urban CO2 dome of Phoenix,Arizona. Physical Geography, 19(2):95–108. → pages 18, 73Idso, C., Idso, S., et al. 2001. An intensive two-week study of an urban CO2 domein Phoenix, Arizona, USA. Atmospheric Environment, 35(6):995–1000. →pages 18, 73, 92, 105Ja¨rvi, L., Nordbo, A., Junninen, H., Riikonen, A., Moilanen, J., Nikinmaa, E., andVesala, T. 2012. Seasonal and annual variation of carbon dioxide surface fluxesin Helsinki, Finland, in 2006–2010. Atmos. Chem. Phys, 12:8475–8489. →pages 25Kaye, J., McCulley, R., and Burke, I. 2005. Carbon fluxes, nitrogen cycling, andsoil microbial communities in adjacent urban, native and agriculturalecosystems. Global Change Biology, 11(4):575–587. → pages 10Kellett, R., Christen, A., Coops, N. C., van der Laan, M., Crawford, B., Tooke,T. R., and Olchovski, I. 2012. A systems approach to carbon cycling andemissions modeling at an urban neighborhood scale. Landscape and UrbanPlanning. → pages 23, 36, 62, 75, 115, 145Kennedy, C., Pincetl, S., and Bunje, P. 2011. The study of urban metabolism andits applications to urban planning and design. Environmental pollution,159(8):1965–1973. → pages 7Kennedy, C., Steinberger, J., Gasson, B., Hansen, Y., Hillman, T., Havranek, M.,Pataki, D., Phdungsilp, A., Ramaswami, A., and Mendez, G. V. 2009.Greenhouse gas emissions from global cities. Environmental Science &Technology, 43(19):7297–7302. → pages 17Kennedy, C., Steinberger, J., Gasson, B., Hansen, Y., Hillman, T., Havra´nek, M.,Pataki, D., Phdungsilp, A., Ramaswami, A., and Mendez, G. V. 2010.Methodology for inventorying greenhouse gas emissions from global cities.Energy Policy, 38(9):4828–4837. → pages 16158Kettle, A., Kuhn, U., Von Hobe, M., Kesselmeier, J., and Andreae, M. 2002.Global budget of atmospheric carbonyl sulfide: Temporal and spatial variationsof the dominant sources and sinks. Journal of Geophysical Research:Atmospheres (1984–2012), 107(D22):ACH–25. → pages 140Knox, P. L. and McCarthy, L. 2005. Urbanization: An introduction to urbangeography. → pages 8Kordowski, K. and Kuttler, W. 2010. Carbon dioxide fluxes over an urban parkarea. Atmospheric Environment, 44(23):2722–2730. → pages 21, 25Kormann, R. and Meixner, F. 2001. An analytical footprint model for non-neutralstratification. Boundary-Layer Meteorology, 99(2):207–224. → pages 34, 82Kort, E. A., Angevine, W. M., Duren, R., and Miller, C. E. 2013. Surfaceobservations for monitoring urban fossil fuel CO2 emissions: Minimum sitelocation requirements for the Los Angeles megacity. Journal of GeophysicalResearch: Atmospheres, 118(3):1577–1584. → pages 106Kort, E. A., Frankenberg, C., Miller, C. E., and Oda, T. 2012. Space-basedobservations of megacity carbon dioxide. Geophysical Research Letters,39(17). → pages 18Kossoy, A. and Guigon, P. 2012. State and trends of the carbon market 2012. →pages 2Kotthaus, S. and Grimmond, C. 2012. Identification of micro-scale anthropogenicCO2, heat and moisture sources–processing eddy covariance fluxes for a denseurban environment. Atmospheric Environment, 57(301):e316. → pages 25Laubach, J. and Fritsch, H. 2002. Convective boundary layer budgets derivedfrom aircraft data. Agricultural and Forest Meteorology, 111(4):237–263. →pages 108Leitch, A. 2010. Summertime horizontal and vertical advective carbon dioxidefluxes emasured in a closed-canopy douglas-fir forest on a slope. Master’sthesis, University of British Columbia. → pages 70Leroyer, S., Be´lair, S., Husain, S. Z., and Mailhot, J. 2014. Sub-KilometerNumerical Weather Prediction in an Urban Coastal Area: A Case Study overthe Vancouver Metropolitan Area. Journal of Applied Meteorology andClimatology, (2014). → pages 146159Li-COR. LI-7500 Instruction Manual. V4.pdf. → pages174Li-Cor 2002. Li-820 instruction manual. Technical report, Li-Cor Biosciences,Inc. → pages 76, 116Lin, J., Gerbig, C., Wofsy, S., Andrews, A., Daube, B., Davis, K., and Grainger,C. 2003. A near-field tool for simulating the upstream influence of atmosphericobservations: The Stochastic Time-Inverted Lagrangian Transport (STILT)model. Journal of Geophysical Research: Atmospheres (1984–2012),108(D16). → pages 106Liss, K., Crawford, B., Christen, A., Siemens, C., and Jassal, R. 2009. Ecosystemrespiration of suburban lawns and its response to varying management andirrigation regimes. In American Meteorological Society Annual Meeting,Phoenix, AZ. → pages 53, 55Liss, K., Tooke, R., N., C., and Christen, A. 2010. Vegetation characteristics atthe Vancouver EPiCC experimental sites, EPiCC Technical Report No.3. →pages 34, 51, 53, 73, 115Liu, H., Feng, J., Ja¨rvi, L., and Vesala, T. 2012. Four-year (2006–2009) eddycovariance measurements of CO2 flux over an urban area in Beijing.Atmospheric Chemistry and Physics, 12(17):7881–7892. → pages 72Lloyd, J. and Taylor, J. 1994. On the temperature dependence of soil respiration.Functional Ecology, pages 315–323. → pages 54Long, S. P., Ainsworth, E. A., Rogers, A., and Ort, D. R. 2004. Risingatmospheric carbon dioxide: plants face the future. Annu. Rev. Plant Biol.,55:591–628. → pages 146Lutsey, N. and Sperling, D. 2008. America’s bottom-up climate change mitigationpolicy. Energy Policy, 36(2):673–685. → pages 5, 23Martins, D. K., Sweeney, C., Stirm, B. H., and Shepson, P. B. 2009. Regionalsurface flux of CO2 inferred from changes in the advected CO2 column density.Agricultural and Forest Meteorology, 149(10):1674–1685. → pages 108Massman, W. and Lee, X. 2002. Eddy covariance flux corrections anduncertainties in long-term studies of carbon and energy exchanges.Agricultural and Forest Meteorology, 113(1-4):121–144. → pages 101160Matese, A., Gioli, B., Vaccari, F., Zaldei, A., and Miglietta, F. 2009. Carbondioxide emissions of the city center of Firenze, Italy: measurement, evaluation,and source partitioning. Journal of Applied Meteorology and Climatology,48(9):1940–1947. → pages 20, 26Mays, K. L., Shepson, P. B., Stirm, B. H., Karion, A., Sweeney, C., and Gurney,K. R. 2009. Aircraft-based measurements of the carbon footprint ofIndianapolis. Environmental science & technology, 43(20):7816–7823. →pages 18, 106McKain, K., Wofsy, S. C., Nehrkorn, T., Eluszkiewicz, J., Ehleringer, J. R., andStephens, B. B. 2012. Assessment of ground-based atmospheric observationsfor verification of greenhouse gas emissions from an urban region. Proceedingsof the National Academy of Sciences, 109(22):8423–8428. → pages 106, 109McKendry, I., Van der Kamp, D., Strawbridge, K., Christen, A., and Crawford, B.2009. Simultaneous observations of boundary-layer aerosol layers with CL31ceilometer and 1064/532 nm lidar. Atmospheric Environment,43(36):5847–5852. → pages 115McMillen, R. 1988. An eddy correlation technique with extended applicability tonon-simple terrain. Boundary-Layer Meteorology, 43(3):231–245. → pages180McPherson, E. G., Nowak, D., Heisler, G., Grimmond, S., Souch, C., Grant, R.,and Rowntree, R. 1997. Quantifying urban forest structure, function, and value:the Chicago Urban Forest Climate Project. Urban ecosystems, 1(1):49–61. →pages 6Ministry of Transportation 2004. Greater Vancouver Trip Diary Survey.Technical report, British Columbia Ministry of Transportation. → pages 63Moore, C. 1986. Frequency response corrections for eddy correlation systems.Boundary-Layer Meteorology, 37(1):17–35. → pages 29, 76, 181Morgenstern, K., Andrew Black, T., Humphreys, E. R., Griffis, T. J., Drewitt,G. B., Cai, T., Nesic, Z., Spittlehouse, D. L., and Livingston, N. J. 2004.Sensitivity and uncertainty of the carbon balance of a Pacific NorthwestDouglas-fir forest during an El Nin˜o/La Nin˜a cycle. Agricultural and ForestMeteorology, 123(3):201–219. → pages 69Moriwaki, R. and Kanda, M. 2004. Seasonal and diurnal fluxes of radiation, heat,water vapor, and carbon dioxide over a suburban area. Journal of AppliedMeteorology, 43(11):1700–1710. → pages 20, 24, 26, 57161Moriwaki, R. and Kanda, M. 2006. Local and global similarity in turbulenttransfer of heat, water vapour, and CO2 in the dynamic convective sublayerover a suburban area. Boundary-layer meteorology, 120(1):163–179. → pages15, 149Moriwaki, R., Kanda, M., and Nitta, H. 2006. Carbon dioxide build-up within asuburban canopy layer in winter night. Atmospheric Environment,40(8):1394–1407. → pages 89, 92Nemitz, E., Hargreaves, K. J., McDonald, A. G., Dorsey, J. R., and Fowler, D.2002. Micrometeorological measurements of the urban heat budget and CO2emissions on a city scale. Environmental science & technology,36(14):3139–3146. → pages 20, 24, 72, 100Newman, P. and Kenworthy, J. 1999. Sustainability and cities: overcomingautomobile dependence. Island Press. → pages 10Nordbo, A., Leena, J., Haapanala, S., Wood, C. R., and Vesala, T. 2012. Fractionof natural area as main predictor of net CO2 emissions from cities. GeophysicalResearch Letters, 39:L20802. → pages 25Nowak, D. J. 1996. Notes: estimating leaf area and leaf biomass of open-growndeciduous urban trees. Forest Science, 42(4):504–507. → pages 34Nowak, D. J. and Crane, D. E. 2002. Carbon storage and sequestration by urbantrees in the USA. Environmental pollution, 116(3):381–389. → pages 10OECD 2010. Cities and Climate Change. OECD Publishing, → pages 4, 5O¨gren, E. and Evans, J. 1993. Photosynthetic light-response curves. Planta,189(2):182–190. → pages 51Oke, T. 1987. Boundary layer climates, volume 632. Routledge. → pages 12, 13,14, 108Oke, T. and East, C. 1971. The urban boundary layer in Montreal.Boundary-Layer Meteorology, 1(4):411–437. → pages 108, 109, 130Oke, T. R., Crowther, J., McNaughton, K., Monteith, J., and Gardiner, B. 1989.The micrometeorology of the urban forest [and discussion]. PhilosophicalTransactions of the Royal Society of London. B, Biological Sciences,324(1223):335–349. → pages 10162Pataki, D., Alig, R., Fung, A., Golubiewski, N., Kennedy, C., McPherson, E.,Nowak, D., Pouyat, R., and Romero Lankao, P. 2006. Urban ecosystems andthe North American carbon cycle. Global Change Biology, 12(11):2092–2102.→ pages 7, 9Pataki, D., Tyler, B., Peterson, R., Nair, A., Steenburgh, W., and Pardyjak, E.2005. Can carbon dioxide be used as a tracer of urban atmospheric transport?Journal of Geophysical Research: Atmospheres (1984–2012), 110(D15). →pages 9, 105Pataki, D., Xu, T., Luo, Y., and Ehleringer, J. 2007. Inferring biogenic andanthropogenic carbon dioxide sources across an urban to rural gradient.Oecologia, 152(2):307–322. → pages 19, 140Pawlak, W., Fortuniak, K., and Siedlecki, M. 2011. Carbon dioxide flux in thecentre of Ło´dz´, Poland - analysis of a 2-year eddy covariance measurement dataset. International Journal of Climatology, 31(2):232–243. → pages 20, 25, 72Peters, E. B. and McFadden, J. P. 2012. Continuous measurements of net CO2exchange by vegetation and soils in a suburban landscape. J. Geophys. Res,117:G03005. → pages 24, 26Pickett, S., Cadenasso, M., Grove, J., Nilon, C., Pouyat, R., Zipperer, W., andCostanza, R. 2001. Ecological, physical, and socioeconomic components ofmetropolitan areas. Annu. Rev. Ecol. Syst, 32:127–57. → pages 6, 7Pickett, S. T., Burch Jr, W. R., Dalton, S. E., Foresman, T. W., Grove, J. M., andRowntree, R. 1997. A conceptual framework for the study of humanecosystems in urban areas. Urban Ecosystems, 1(4):185–199. → pages 6, 7Raupach, M. 1995. Vegetation-atmosphere interaction and surface conductance atleaf, canopy and regional scales. Agricultural and Forest Meteorology,73(3):151–179. → pages 107Raupach, M., Denmead, O., and Dunin, F. 1992. Challenges in linkingatmospheric CO2 concentrations to fluxes at local and regional scales.Australian Journal of Botany, 40(5):697–716. → pages 108Raupach, M., Rayner, P., and Paget, M. 2010. Regional variations in spatialstructure of nightlights, population density and fossil-fuel CO2 emissions.Energy Policy, 38(9):4756–4764. → pages 23163Reichstein, M., Falge, E., Baldocchi, D., Papale, D., Aubinet, M., Berbigier, P.,Bernhofer, C., Buchmann, N., Gilmanov, T., Granier, A., et al. 2005. On theseparation of net ecosystem exchange into assimilation and ecosystemrespiration: review and improved algorithm. Global Change Biology,11(9):1424–1439. → pages 56Reid, K. and Steyn, D. 1997. Diurnal variations of boundary-layer carbon dioxidein a coastal city - Observations and comparison with model results.Atmospheric Environment, 31(18):3101–3114. → pages 19, 27, 75, 98, 107,115, 136, 138, 140Rice, A. and Bostrom, G. 2011. Measurements of carbon dioxide in an Oregonmetropolitan region. Atmospheric Environment, 45(5):1138–1144. → pages 92Rosenzweig, C., Solecki, W., Hammer, S. A., and Mehrotra, S. 2010. Cities leadthe way in climate-change action. Nature, 467(7318):909–911. → pages 5Roth, M. and Oke, T. R. 1995. Relative efficiencies of turbulent transfer of heat,mass, and momentum over a patchy urban surface. Journal of the AtmosphericSciences, 52:1863–1874. → pages 15, 27, 149Sailor, D. and Lu, L. 2004. A top-down methodology for developing diurnal andseasonal anthropogenic heating profiles for urban areas. AtmosphericEnvironment, 38(17):2737–2748. → pages 42Salmond, J., Oke, T. R., Grimmond, C., Roberts, S., and Offerle, B. 2005. Ventingof heat and carbon dioxide from urban canyons at night. Journal of AppliedMeteorology, 44(8):1180–1194. → pages 15, 21, 71, 102Satterthwaite, D. 2008. Cities’ contribution to global warming: notes on theallocation of greenhouse gas emissions. Environment and Urbanization,20(2):539–549. → pages 16, 23Schmid, H. 1994. Source areas for scalars and scalar fluxes. Boundary-LayerMeteorology, 67(3):293–318. → pages 24Schmid, H., Cleugh, H., Grimmond, C., and Oke, T. 1991. Spatial variability ofenergy fluxes in suburban terrain. Boundary-Layer Meteorology,54(3):249–276. → pages 27Schmid, H. and Lloyd, C. 1999. Spatial representativeness and the location biasof flux footprints over inhomogeneous areas. Agricultural and ForestMeteorology, 93(3):195–209. → pages 25, 38164Skamarock, W. C., Klemp, J. B., Dudhia, J., Gill, D. O., Barker, D. M., Wang, W.,and Powers, J. G. 2005. A description of the advanced research WRF version 2.Technical report, DTIC Document. → pages 106Statistics Canada 2011. Census tract dissemination., Accessed May 29, 2013. → pages34Steemers, K. 2003. Energy and the city: density, buildings and transport. Energyand buildings, 35(1):3–14. → pages 9Stern, N. 2007. The economics of climate change: the Stern review. cambridgeUniversity press. → pages 5, 16Stewart, I. D. and Oke, T. R. 2012. Local climate zones for urban temperaturestudies. Bulletin of the American Meteorological Society, 93(12):1879–1900.→ pages 12, 27, 73, 115Steyn, D. and Faulkner, D. 1986. The climatology of sea-breezes in the LowerFraser Valley, BC. Climatol. Bull, 20(3):21–39. → pages 30Steyn, D. and Oke, T. 1982. The depth of the daytime mixed layer at two coastalsites: A model and its validation. Boundary-Layer Meteorology,24(2):161–180. → pages 136, 137Steyn, D., Oke, T., Hay, J., and Knox, J. 1981. On scales in meteorology andclimatology. Clim Bull, 39:1–8. → pages 13Stocker, T. F., Dahe, Q., and Plattner, G.-K. 2013. Climate change 2013: Thephysical science basis. Working Group I Contribution to the Fifth AssessmentReport of the Intergovernmental Panel on Climate Change. Summary forPolicymakers (IPCC, 2013). → pages 2, 4Strong, C., Stwertka, C., Bowling, D., Stephens, B., and Ehleringer, J. 2011.Urban carbon dioxide cycles within the salt lake valley: A multiple-box modelvalidated by observations. Journal of Geophysical Research: Atmospheres(1984–2012), 116(D15). → pages 107Stull, R. 1988. An introduction to boundary layer meteorology, volume 13.Springer. → pages 80, 113Stull, R. B. 2000. Meteorology for scientists and engineers: a technicalcompanion book with Ahrens’ Meteorology Today. Brooks/Cole. → pages 118,120, 121165Taha, H., Akbari, H., and Rosenfeld, A. 1991. Heat island and oasis effects ofvegetative canopies: micro-meteorological field-measurements. Theoreticaland Applied Climatology, 44(2):123–138. → pages 10Tanaka, M., Nakazawa, T., and Aoki, S. 1985. Atmospheric carbon dioxidevariations in the suburbs of Sendai, Japan. Tellus B, 37(1):28–34. → pages 18Teske, M. E. and Thistle, H. W. 2004. A library of forest canopy structure for usein interception modeling. Forest Ecology and Management, 198(1):341–350.→ pages 34Tooke, T., Coops, N., Goodwin, N., and Voogt, J. 2009. Extracting urbanvegetation characteristics using spectral mixture analysis and decision treeclassifications. Remote Sensing of Environment, 113(2):398–407. → pages 32,75, 115Tyndall, J. 1859. Note on the transmission of radiant heat through gaseous bodies.Proceedings of the Royal Society of London, 10:37–39. → pages 1Tyndall, J. 1873. Heat a mode of motion. D. Appleton. → pages 2UN 2011. World Urbanization Prospects, the 2011 revision. , AccessedFebruary, 2014. → pages 3, 4, 5, 23Uno, I., Wakamatsu, S., Ueda, H., and Nakamura, A. 1988. An observationalstudy of the structure of the nocturnal urban boundary layer. Boundary-LayerMeteorology, 45(1-2):59–82. → pages 109, 130USCB 2014. U.S. Census Bureau World Population Counter., accessed Feb. 16, 2014. →pages 3USCOM 2009. U.S. Council of Mayors Climate Protection Center., Accessed February 16,2014. → pages 5Vaisala 2014. GMM220 User Manual.→ pages111van der Kamp, D. and McKendry, I. 2010. Diurnal and seasonal trends inconvective mixed-layer heights estimated from two years of continuousceilometer observations in Vancouver, BC. Boundary-layer meteorology,137(3):459–475. → pages 115, 118, 128166van der Laan, M. 2011. Scaling urban energy use and greenhouse gas emissionsthrough LiDAR. Master’s thesis, University of British Columbia. → pages 33,34, 73, 83, 102, 115VandeWeghe, J. R. and Kennedy, C. 2007. A spatial analysis of residentialgreenhouse gas emissions in the Toronto census metropolitan area. Journal ofIndustrial Ecology, 11(2):133–144. → pages 19Velasco, E., Pressley, S., Allwine, E., Westberg, H., and Lamb, B. 2005.Measurements of CO2 fluxes from the Mexico City urban landscape.Atmospheric Environment, 39(38):7433–7446. → pages 20, 25, 26, 72Velasco, E. and Roth, M. 2010. Cities as net sources of CO2: review ofatmospheric CO2 exchange in urban environments measured by eddycovariance technique. Geography Compass. → pages 20, 24, 36, 60, 67, 72,105Velasco, E., Roth, M., Tan, S., Quak, M., Nabarro, S., and Norford, L. 2013. Therole of vegetation in the CO2 flux from a tropical urban neighbourhood. Atmos.Chem. Phys. Discuss, 13:7267–7310. → pages 26Vickers, D. and Mahrt, L. 1997. Quality control and flux sampling problems fortower and aircraft data. Journal of Atmospheric and Oceanic Technology,14(3):512–526. → pages 174Villar, R., Held, A. A., and Merino, J. 1995. Dark leaf respiration in light anddarkness of an evergreen and a deciduous plant species. Plant Physiology,107(2):421–427. → pages 56Vogel, F. R., Tiruchittampalam, B., Theloke, J., Kretschmer, R., Gerbig, C.,Hammer, S., and Levin, I. 2013. Can we evaluate a fine-grained emissionmodel using high-resolution atmospheric transport modelling and regionalfossil fuel CO2 observations? Tellus B, 65. → pages 106Vogt, R., Christen, A., Rotach, M., Roth, M., and Satyanarayana, A. 2006.Temporal dynamics of CO2 fluxes and profiles over a Central European city.Theoretical and Applied Climatology, 84(1):117–126. → pages 20, 24, 36, 71,92, 98Voogt, J. 2002. Urban heat island. In Munn, T., Douglas, I., MacCracken, M. C.,Mooney, H. A., Timmerman, P., and Tolba, M. K., editors, Encyclopedia ofGlobal Environmental Change, volume 3, pages 660–666. → pages 5, 6167Wallace, A. R. 1904. Man’s Place in the Universe. Chapman and Hall. → pages 1Walsh, C. 2005. Fluxes of radiation, energy, and carbon dioxide over a suburbanarea of Vancouver, BC. Master’s thesis, Department of Geography, Universityof British Columbia. → pages 21, 27, 37, 75, 115Webb, E., Pearman, G., and Leuning, R. 1980. Correction of flux measurementsfor density effects due to heat and water vapour transfer. Quarterly Journal ofthe Royal Meteorological Society, 106(447):85–100. → pages 29, 76, 180Weber, S. and Weber, K. 2008. Coupling of urban street canyon and backyardparticle concentrations. Meteorologische Zeitschrift, 17(3):251–261. → pages15, 71Wentz, E. A., Gober, P., Balling Jr, R. C., and Day, T. A. 2002. Spatial patternsand determinants of winter atmospheric carbon dioxide concentrations in anurban environment. Annals of the Association of American Geographers,92(1):15–28. → pages 18Williams, C. A., Scanlon, T. M., and Albertson, J. D. 2007. Influence of surfaceheterogeneity on scalar dissimilarity in the roughness sublayer. Boundary-layermeteorology, 122(1):149–165. → pages 149Wolman, A. 1965. The metabolism of cities. Scientific American,213(3):179–190. → pages 7Yang, P., Black, T. A., Neumann, H. H., Novak, M., and Blanken, P. 1999. Spatialand temporal variability of CO2 concentration and flux in a boreal aspen forest.Journal of Geophysical Research: Atmospheres (1984–2012),104(D22):27653–27661. → pages 67, 100168Appendix AEddy covariance data processingand quality control proceduresThis appendix describes data processing and quality control procedures used tocalculate local-scale eddy covariance turbulent fluxes of carbon dioxide in Chap-ters 2, 3, and 4. Two flux towers, ‘Vancouver-Sunset and ‘Vancouver-Oakridge,were operated in extensive residential areas composed of single-family homes. Arural reference flux station ‘Westham Island was located on flat, unmanaged andnon-irrigated grassland 16 km south of the two urban neighborhoods in an areadominated by intensive farming (Figure A.1).A.1 InstrumentationTurbulent fluxes (latent heat flux (QE), sensible heat flux (QH), and carbon diox-ide flux (FC) were measured at all three sites using Campbell Scientific CSAT-3dsonic anemometers (Table A.1) and open-path Li-COR Li-7500 infrared gas an-alyzers (IRGAs) (Table A.2) using the eddy-covariance method. Wind direction(u, v, and w components), acoustic air temperature (t), and CO2 (c) and H2O (q)concentrations were sampled continuously at 20 Hz.169Figure A.1: Photographs of the EC systems at a) Vancouver Sunset, b) Van-couver Oakridge, and c) Westham Island sites. Photos by: Rick Ketler(a) and Andreas Christen (b-c).170Table A.1: Set-up and settings of ultrasonic anemometer-thermometers used during this research. All sites used aCampbell Scientific CSAT 3D sonic anemometer. Azimuth refers to the direction the sensor head is facing towards,as seen from the mounting base relative to geographic North.Site Sensor height aboveground (m)Azimuth(◦) Serial Number Wicks SettingsVancouverSunset28.8 179.1 1394 (May 6, 2008 - June4, 2009); 1342 (June 4,2009 onward)Yes 60 Hz internal, 20 Hz out-put (SDM)VancouverOakridge29.0 305.0 (2008)321.6 (2009)1393 Yes 60 Hz internal, 20 Hz out-put (SDM)WesthamIsland1.8 (July 27, 2007 - June23, 2009); 2.2 (June 23,2009 - October 8, 2009)355.0 1342 (July 27, 2007 - June4, 2004); 1393 (June 4,2009 - June 11, 2009);1394 (June 11, 2009 - Oc-tober 8, 2009)Yes 60 Hz internal, 20 Hz out-put (SDM)171Table A.2: Set-up of infrared gas analyzers used during EPiCC. All sites used an open-path Licor, Inc. Li-7500 sensormodel.Site Sensor height aboveground (m)Azimuth rel-ative to sonic(◦)Distance tosonic (Vert,Hor (cm))Tilt fromvertical (◦)Serial NumberVancouverSunset28.8 300 0, 40 60 ◦ to South 1222 (May 6, 2008 - June24, 2009); 0561 (June 24,2009 onward)VancouverOakridge29.0 60 0, 18 30 ◦ to North 0561 (2008); 1222 (2009)WesthamIsland1.8 (July 27, 2007 - June23, 2009); 2.2 (June 23,2009 - October 8, 2009)350 0, 17 60 ◦ to North 0151(July 27, 2007 - Oc-tober 10, 2008); 0561(October 10, 2008 - June23, 2009); 0151 (June 23,2009 - October 8, 2009)172Figure A.2: Sonic anemometer intercomparison set-up at Westham Island(photo by Andreas Christen).A.2 Sensor calibrationsAs part of an NSERC DG research program, a field intercomparison of sonicanemometers was conducted from May 27 - June 15, 2009 at the Westham Is-land site including sonics used during EPiCC (Figure A.2). Full statistical resultsof the intercomparison are given in Crawford et al. (2010).The IRGAs were calibrated regularly in the UBC Biometeorology lab (TableA.3). First, H2O and CO2 readings are zeroed using dry N2 gas with 0 ppm CO2.Next, the H2O readings are spanned with air at a dew point temperature of 8.5◦ Cusing a dew point generator. Finally, CO2 is spanned using reference gas calibratedby Environment Canada.A.3 High frequency quality controlHigh frequency (20 Hz) data pass through several quality control filters beforecovariances and fluxes are calculated:173Table A.3: Calibrations and software settings of IRGAs used during EPiCC.Model Serial Number CalibrationDatesSensor OSLi-7500 0151 5/12/2007,23/10/2008,10/6/2009Windows Interfacev3.0.2, Internal v3.0.1Li-7500 0561 26/6/2008/,24/6/2009Windows Interfacev3.0.2, Internal v3.0.1Li-7500 1222 4/3/2008,15/6/2009Windows Interfacev3.0.2, Internal v3.0.1A.3.1 Sonic anemometer diagnostic valueThe CSAT sonic anemometer reports if the sonic anemometer path is obstructed,the path length has been altered, or for up to 10 seconds after the sonic has justbeen powered on (CSI, 2011). Individual high frequency u, v, w, and t data pointsare withheld from further processing when diagnostic values are triggered.A.3.2 IRGA diagnostic valuePrecipitation, condensation, fog, insects, etc. in the optical path of the IRGA mayinterfere with measurements of c and q. The automatic gain control (AGC) outputof the IRGA registers a change in value if the optical path of the IRGA is blocked(Li-COR). The AGC value was recorded at 20 Hz at all sites and ranges from 0100 (the optical path is completely obscured at 100). If there is any change in AGCvalue, or if the AGC value is greater than 90, H2O and CO2 high frequency datafor ±5 s around that point are withheld from further processing.A.3.3 High frequency spike detectionRandom electronic noise and short-term data spikes are filtered out of high fre-quency data sets using a dynamic iterative standard deviation filter (e.g. Vickersand Mahrt 1997). First, individual 20 Hz data points are flagged if they fall outsidea physically justified, realistic data range for each variable (Table A.4). Individual17420 Hz data points are then flagged as spikes and withheld from further processingif they are above or below a variable-specific standard deviation threshold froma 30-minute mean (Table A.4). Consecutive passes are then performed with thestandard deviation threshold raised by 0.3 each time until no spikes are detected.Spikes must also be less than 0.3 s in duration, otherwise they are considered real.175Table A.4: Summary of quality control limits and thresholds used for high-frequency eddy covariance data.u (m s−1) v (m s−1) w (m s−1) t (◦C) q (mmol m−3) c (mmol m−3)Physically-basedmin/max thresholds(variable units)-30/30 -30/30 -5/5 -20/40 100/1500 12/40Spike threshold(standard deviations)6 6 8 8 10 10Standard deviation(min/max)0.5/4.0 0.5/4.0 0.02/1.5 0.01/2.0 0.01/150.0 0.001/2.0Skewness (min/max) -3.0/3.0 -3.0/3.0 -2.0/2.0 -2.5/2.5 -5.0/5.0 -5.0/5.0Kurtosis (min/max) -2.0/5.0 -2.0/5.0 -2.0/15.0 -2.0/15.0 -2.0/15.0 -2.0/15.0176Figure A.3: Difference of wind speed measurements (vector mean) betweenwind tunnel and CSAT 3d sonic anemometer at 4 m s−1 dependent onazimuth and tilt (%) (reproduced with permission from Christen et al.2001).A.3.4 Flow distortion by the sensor headWind tunnel measurements of CSAT-3d anemometers show that flow is stronglydistorted when ± 7◦ from directly behind the sonic mounting block (Figure A.3).30-minute flux averaging periods are flagged as questionable if more than 25% of20 Hz wind directions fall within ± 7◦ of 180◦ from the sonics azimuth (TableA.5).A.3.5 High frequency statistics checkStatistics of 30-minute standard deviation, skewness, and kurtosis are calculatedfor u, v, w, t, q, and c. Empirical limits were determined for each variable and datafrom periods with values outside these limits are flagged as questionable (TableA.4).177Table A.5: Wind directions influenced by sonic anemometer mounting block.The Oakridge Tower sonic anemometer was mounted with different ori-entations during 2008 and 2009.Site Wind directions withheld from flux processingVancouver Sunset 352◦ - 6◦Vancouver Oakridge 118◦ - 132◦ (2008), 134.6◦-148.6◦ (2009)Westham Island 168◦ - 182◦A.3.6 Sonic-IRGA time lagDue to the separation of the sonic anemometer and IRGA, measurements of q andc are not exactly correlated with measurements of wind velocity because of thetravel time of an air parcel between the sonic and IRGA. The magnitude of thislag was calculated by shifting q and c time series relative to w and determining themaximum covariance (Table A.6). Lag times vary by wind direction and were cal-culated for eight wind sectors at each site, but only average lag times are reportedhere.For all sites, lag times were less than the 20 Hz measurement resolution (i.e.less than 1 record length), so it is doubtful that the lag can be accounted for byshifting the sonic and IRGA time-series relative to each other with out introduc-ing additional uncertainties. To test this, covariances (w￿c￿ and w￿q￿) were calcu-lated with IRGA c and q time series shifted forward by 1 record length relativeto the sonic wind vector data. These covariances were then compared with co-variances calculated from un-shifted data. Seven half-hour periods (20 July 2009,0900-1200) measured at Sunset Tower were used as a test period representative ofsummer, clear sky conditions with flow from the NW. Covariances calculated fromshifted time series were 0.4% different for w￿c￿ and 0.3% different for w￿q￿ thanun-shifted covariances. This difference is deemed negligible, so record-shifting isnot implemented during flux data processing.178Table A.6: Average sonic-IRGA lag in milliseconds and number of records by site and IRGA for CO2 and H2O.Site and Date IRGA SN CO2 (ms) CO2 (records) H2O (ms) H2O (records)Sunset6/5/2008 - 23/6/2009 1222 -42.1 -0.84 -41.1 -0.8225/6/2009 - 20/4/2010 0561 -41.8 -0.84 -40.0 -0.80Oakridge9/7/2008 - 31/8/2008 0561 -30.7 -0.61 -31.2 -0.6227/6/2009 - 31/8/2009 1222 -32.8 -0.66 -25.4 -0.50Westham Island27/7/2007 - 31/12/2008 0151 7.2 0.14 -5.29 -0.1111/10/2009 - 23/6/2009 0561 20.76 0.42 12.99 0.2524/7/2009 - 8/10/2009 0151 -38.1 -0.76 44.3 -0.89179A.4 Block average calculation and coordinate rotationIn a second step after initial high frequency data quality control filters, mean valuesand higher-order moments (including covariances) are calculated if a 30-minuteperiod has greater than 75% of possible u, v, w, TA, c, and q 20 Hz data points.Wind components are rotated two times so that the x-axis of the new soniccoordinate system is aligned with the mean 30-minute wind direction, and themean vertical wind w is zero (e.g. McMillen 1988, Finnigan et al. 2003). Follow-ing Reynolds decomposition, 30-minute statistics are calculated based on a simpleblock-average.A.5 Post-processing corrections applied to turbulentcarbon dioxide flux1. 30-minute FC is calculated as:FC = w￿c￿ρ (A.1)2. FC values are then corrected for density effects according to Webb et al.(1980):FC = w￿c￿+MAMq·ρcρA·w￿q￿+￿1+MAMq·ρqρA￿·ρcT·w￿T ￿a (A.2)where:ρq =e ·MqT ·R(A.3)and:ρA =(P− e) ·MAT ·R(A.4)and based on the Clausius-Clapeyron equation:e = RH ·￿e0 · exp￿LVRV·￿1273−1T￿￿￿(A.5)3. FC values are also corrected to account for high frequency flux losses based180on IRGA and sonic anemometer path length and horizontal separation be-tween the IRGA and sonic anemometer sensors (Moore, 1986). Inputs tothis correction include:• Instrument specific path lengths of the sonic (11.6 cm) and IRGA (12.5cm).• Site specific sonic and IRGA horizontal separation (Table A.2).• Measurement height (Table A.1).• Measured horizontal wind speed from sonic anemometer.• Zero-plane displacement (zd), estimated as 2/3 the average height ofthe canopy (effective canopy height = 10.6 m for Sunset, 8.0 m forOakridge, variable canopy height (0.1-1.8 m) at Westham because ofgrowing grass).• Monin-Obukhov Length, calculated as:L =−￿￿u￿w￿2+ v￿w￿2￿0.25￿3k · gT ·w￿T ￿a(A.6)4. Comparisons of potential temperature measured from both the sonic anemome-ter and T/RH sensor (HMP, Vaisala Inc., Finland) at each site were used as acheck on sonic performance. During and after precipitation events when thesonic path length may be obstructed by water, measurements of wind veloc-ities (and therefore FC) may be unreliable. During these periods, the HMPsability to measure temperature is assumed to be unaffected.For each site and sonic, potential temperature was calculated from the sonicand HMP temperature measurements and compared. If the difference be-tween measurements fell outside of empirically determined limits for eachsite/sensor configuration, 30-minute values of FC were flagged as question-able (Table A.7).181Table A.7: Sonic-HMP quality control temperature departure limits.Site and Date Sonic SN Low/High limit (K)Sunset06/05/2008 - 04/06/2009 1394 -1.5/1.504/06/2009 - 31/04/2012 1342 -1.5/1.5Oakridge30/06/2008 - 31/08/2009 1393 -1.5/1.5Westham Island01/07/2007 - 11/06/2009 1394 -1.5/1.511/06/2009 - 08/10/2009 1394 -2.0/2.0A.6 Comparison of eddy covariance processing softwareThis section compares output from the EPiCC eddy-covariance data processingcode developed and used at UBC Geography (ubc, Version 3.04, EPiCCversion, IDL based processing) against Eddy Pro (3.0) which is distributed byLicor, Inc.Five weeks of high-frequency data (20 Hz) measured on Sunset Tower werecompared (Feb 7 to Mar 23, 2012) were processed using the settings listed in TableA.1 and the output of turbulent fluxes of sensible heat, latent heat, and carbon-dioxide.Eddy covariance data was measured by a separate system consisting of a CSICSAT-3 ultrasonic anemometer, a Li-7500A (CO2/H2O open path analyzer), andLi-7700 (CH4 open path analyzer, not used here), all operated at 28.7 m above localground (tower base). The CSAT was pointing with its undisturbed sector towards206◦ from geographic North. The sensor separation between the CSAT-3 measure-ment volume and the Li-7700 was 45 cm horizontal and 7 cm vertical (Li-7700 ishigher than CSAT-3 measurement volume) and the Li-7500 was installed to the NE(53◦) of the CSAT-3. The sensor separation between the CSAT-3 measurement vol-ume and the Li-7500A was 35 cm horizontal and 0 cm vertical and the Li-7500Awas installed to the North (356◦) of the CSAT-3 volume. This system was set apartfrom the towers long-term system (at 90 cm horizontal and 10 cm vertical distanceto the measurement volume of the long-term system).182Table A.8: Summary of software settings during the EC software compari-son.Setting Eddy Pro ubc mmd.proTilt correction (axisrotation)Double rotation (v=0,w=0)Double rotation (v=0,w=0)Detrending Block averaging Block averagingTime lag compensa-tionNone NoneDespiking Yes YesStatistical tests Absolute limits, skew-ness and kurtosisAbsolute limits, skew-ness and kurtosisTests for AnalyzerQualityYes AGC test for Li-7500A. Only datawhen signal strength> 20% for Li-7500Compensations fordensity fluctuationsWebb-Pearman-Leuning (open path)Webb-Pearman-Leuning (open path)Sonic temperature cor-rection for humidityvan Dijk et al. (2004) Yes (based onSchotanus et al.,1983)Filtering of flow dis-tortionNone Remove data fromdisturbed sector fromblock averageAngle of attack correc-tionsYes NoSpectral corrections Yes (standard) Sensor separation only(Moore, 1986)The period of the comparison experienced extensive periods of rain. First, theentire period was compared. Secondly, the analysis was restricted to three dry days,March 6, 00:30 to March 8, 24:00 without rain (Table A.9). From both time framesall valid data was used (Eddy Pro QC = 0, 1 or 2).183Table A.9: Comparison between fluxes calculated using and Eddy Pro during February 7 to March 11,2012 in terms of a linear regression with slope and r2, and expressed as median absolute difference (MedAE).The entire comparison period experienced extensive intervals of rain, but the March 6 - 8 period was dry. Valuesis brackets are the error estimates relative to the average flux in the given period. Data includes all data that wasoutputted with various quality flags.Date Slope r2 MedAE(µmol m−2s−1)MedAE relative toaverage fluxFeb. 7 - Mar. 11, 2012 (all) EddyPro = 1.003 · UBC 0.989 0.30 1.4%Mar. 6 - Mar. 8, 2012 (dry) EddyPro = 1.016 · UBC 0.996 0.34 1.7%184


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