UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Redefining the urban heat island Stewart, Iain Douglas 2011

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

Item Metadata

Download

Media
24-ubc_2011_fall_stewart_iain.pdf [ 16.95MB ]
Metadata
JSON: 24-1.0072360.json
JSON-LD: 24-1.0072360-ld.json
RDF/XML (Pretty): 24-1.0072360-rdf.xml
RDF/JSON: 24-1.0072360-rdf.json
Turtle: 24-1.0072360-turtle.txt
N-Triples: 24-1.0072360-rdf-ntriples.txt
Original Record: 24-1.0072360-source.json
Full Text
24-1.0072360-fulltext.txt
Citation
24-1.0072360.ris

Full Text

REDEFINING THE URBAN HEAT ISLAND  by IAIN DOUGLAS STEWART  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in The Faculty of Graduate Studies (Geography)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) October 2011  © Iain Douglas Stewart, 2011  Abstract The effect of urban development on local thermal climate is ostensibly well documented in scientific literature. Since the nineteenth century, observations of “urban-rural” air temperature differences (∆Tu-r), or urban heat island magnitudes, have been reported for hundreds of cities worldwide. The historical and geographical scope of the heat island literature is impressive. Over time, however, methodologists have raised concerns about the rigor and authenticity of that literature, especially regarding the definition, measurement, and reporting of heat island magnitudes. Indiscriminate use of “urban” and “rural” by heat island investigators to describe their field sites is a particular concern. Much confusion now surrounds the physical and cultural characteristics of so-called urban and rural sites in heat island literature. This thesis confronts these concerns through two approaches. The first approach synthesizes and evaluates a sample of 190 observational heat island studies from the period 1950 to 2007. The synthesis uses nine criteria of scientific method and communication to critically assess the experimental quality of each study. Results are discouraging: the mean quality score of the literature sample is just 50 percent, and nearly one-half of the reported heat island magnitudes are judged to be scientifically indefensible on account of incomplete or incompetent reporting. The second approach develops a landscape classification system to standardize reporting of heat island field sites and temperatures in all cities. The local climate zone (LCZ) system comprises 17 zones and is the first comprehensive climate-based classification of urban and rural landscapes for heat island investigators. Each zone represents an area that is local in scale and unique in land cover, building morphology, and screen-level thermal climate. Results show that the new classification leads to a  ii  more purposeful interpretation of heat island magnitude as ∆TLCZ , and thereby constrains the operational use of ∆Tu-r to climatologically defined and universally recognized urban and rural zones. The thesis concludes with a conceptual typology of urban heat island magnitudes, and a list of specific guidelines and recommendations to improve methodology and communication in heat island studies.  iii  Preface A version of Chapter 3 has been published: Stewart ID. 2007. Landscape representation and the urban-rural dichotomy in empirical urban heat island literature, 1950–2006. Acta Climatologica et Chorologica 40-41: 111–21. A version of Chapter 4 has been published: Stewart ID. 2011. A systematic review and scientific critique of methodology in modern urban heat island literature. International Journal of Climatology 31: 2000–17. A version of Chapter 5 has been submitted for publication: Stewart ID, Oke TR. 2011.  iv  Table of Contents Abstract................................................................................................................................ ii Preface................................................................................................................................. iv Table of Contents .................................................................................................................v List of Tables ...................................................................................................................... ix List of Figures..................................................................................................................... xi Acknowledgements .......................................................................................................... xiv Dedication ......................................................................................................................... xvi CHAPTER 1 INTRODUCTION ......................................................................................1 1.1 The problem ..............................................................................................................1 1.2 Research objectives...................................................................................................3 1.3 Thesis contents..........................................................................................................5 CHAPTER 2 OVERVIEW OF URBAN HEAT ISLANDS...........................................7 2.1 Defining the urban heat island ..................................................................................7 2.1.1 Heat island types ...............................................................................................7 2.2 Causes of the canopy layer urban heat island ...........................................................9 2.3 Heat island genesis..................................................................................................13 2.4 Observing the canopy layer urban heat island ........................................................15 2.4.1 Mobile surveys................................................................................................15 2.4.2 Stationary surveys...........................................................................................16 2.5 Magnitude of the canopy layer urban heat island ...................................................17 2.5.1 Definition ........................................................................................................17 2.5.2 Controls...........................................................................................................19 2.5.2.1 Weather ...................................................................................................19 2.5.2.2 Background climate ................................................................................20 2.5.2.3 Time ........................................................................................................21 2.5.2.4 Urban form and function.........................................................................23 2.5.2.5 Rural thermal admittance........................................................................25  v  CHAPTER 3 CRITICAL REVIEW OF URBAN HEAT ISLAND LITERATURE .........................................................................................27 3.1 Historical setting .....................................................................................................28 3.1.1 Pioneering the field: Early observations ........................................................28 3.1.2 A century of growth: 1850–1950...................................................................32 3.1.3 The modern period: 1950 to present ..............................................................37 3.1.4 Synopsis ..........................................................................................................48 3.2 Methodological and philosophical problems in the modern period........................49 3.2.1 Estimates of urban heat island magnitude ......................................................49 3.2.1.1 Incomplete reporting...............................................................................51 3.2.1.2 Flawed design .........................................................................................53 3.2.1.3 Inconsistent procedures...........................................................................55 3.2.2 Definitions of urban heat island magnitude ....................................................58 3.2.2.1 Scientific method and ∆Tu-r ....................................................................59 3.2.2.2 Urbanization theory and ∆Tu-r ................................................................64 3.2.3 Inter-city comparisons of urban heat island magnitude..................................72 3.3 Approach to the research problem ..........................................................................74 3.3.1 Synthesizing empirical urban heat island research .........................................74 3.3.2 Constructing a field site classification system for urban heat island observations ....................................................................................................76 CHAPTER 4 RESEARCH SYNTHESIS ......................................................................79 4.1 Methods...................................................................................................................79 4.1.1 Defining the “universe” of studies..................................................................80 4.1.2 Selecting the primary literature.......................................................................81 4.1.2.1 Eligibility criteria ....................................................................................81 (i) Characterisation of the heat island effect .................................................81 (ii) Principal aims...........................................................................................83 (iii) Date and source of print or publication....................................................84 4.1.3 Sourcing and retrieving the primary literature................................................87 4.1.3.1 Foreign-language literature.....................................................................90 4.1.3.2 Literature translation...............................................................................92 4.1.3.3 Sample size .............................................................................................93 4.1.4 Cataloguing the primary literature ..................................................................94 4.1.4.1 Urban heat island database......................................................................94 (i) Bibliographic inventory ...........................................................................94 (ii) Literature assessment ...............................................................................95 4.1.5 Evaluating the primary literature ....................................................................96 vi  4.1.5.1 Scientific criteria.....................................................................................97 4.1.5.2 Grading scheme ....................................................................................120 4.1.5.3 Rankings ...............................................................................................125 4.2 Results...................................................................................................................126 4.2.1 Describing the literature sample ...................................................................127 4.2.2 Analysing the primary literature ...................................................................131 4.2.2.1 Scientific criteria...................................................................................131 4.2.3 Grading the primary literature ......................................................................139 4.2.3.1 Tier placements and quality scores .......................................................139 4.2.3.2 Rankings ...............................................................................................143 CHAPTER 5 A NEW CLASSIFICATION SYSTEM FOR URBAN HEAT ISLAND OBSERVATIONS ..................................................................149 5.1 Methods.................................................................................................................149 5.1.1 Constructing the framework .........................................................................150 5.1.1.1 Taking stock of urban climate classification systems...........................151 5.1.1.2 Classification criteria ............................................................................159 5.1.1.3 Classification by logical division..........................................................161 5.1.1.4 Data sources ..........................................................................................163 5.1.1.5 Field site visits ......................................................................................165 5.1.2 Thermal differentiation of landscape classes................................................167 5.1.2.1 Observational approach ........................................................................168 (i) Uppsala, Sweden....................................................................................169 (ii) Nagano, Japan ........................................................................................175 (iii) Vancouver, Canada ................................................................................179 5.1.2.2 Numerical modelling approach.............................................................184 (i) Background ............................................................................................184 (ii) Oregon State University (OSU) atmospheric boundary layer model ....186 (iii) Coupled Atmosphere-Plant-Soil (CAPS) surface-layer model..............188 (iv) Town Energy Balance (TEB) model......................................................189 (v) Coupled model operations .....................................................................192 5.1.3 Revising the classification system and its landscape classes........................193 5.2 Results...................................................................................................................195 5.2.1 Local climate zones.......................................................................................195 5.2.1.1 Zone definitions ....................................................................................199 5.2.2 Thermal differentiation of local climate zones .............................................207 5.2.2.1 Observational approach ........................................................................207 (i) Uppsala, Sweden....................................................................................207 vii  (ii) Nagano, Japan ........................................................................................216 (iii) Vancouver, Canada ................................................................................222 5.2.2.2 Numerical modelling approach.............................................................230 CHAPTER 6 DISCUSSION OF RESEARCH RESULTS ........................................236 6.1 Synthesis of observational urban heat island literature.........................................236 6.1.1 Caveats..........................................................................................................236 6.1.2 Extracting generalizations.............................................................................238 6.1.3 Recommendations for improving method and communication in observational urban heat island literature .....................................................241 6.2 Local climate zone (LCZ) classification system...................................................245 6.2.1 Caveats..........................................................................................................245 6.2.2 Extracting generalizations.............................................................................248 6.2.3 Communication.............................................................................................252 6.2.4 Guidelines for use .........................................................................................253 6.3 Redefining urban heat island magnitude...............................................................262 6.3.1 Typology of urban heat island magnitudes...................................................265 6.4 Avenues for future research ..................................................................................268 CHAPTER 7 CONCLUSIONS.....................................................................................271 7.1 Summary of findings.............................................................................................271 7.2 Summary of contributions.....................................................................................273 References.........................................................................................................................276 Appendices........................................................................................................................294 Appendix A Abstraction forms for urban heat island literature translation................294 Appendix B List of studies evaluated in the research synthesis (by date of publication) ..........................................................................296 Appendix C Methodological data for the urban heat island literature sample ...........308 Appendix D Evidence to support pass/fail decisions in the heat island literature sample for scientific criteria 1 through 9 ...............................................312 Appendix E Pass/fail grades for scientific criteria 1 through 9 of the heat island literature sample .....................................................................................321 Appendix F Final rankings, tier placements, quality scores, and missing data index (MDI) values for the heat island literature sample .......................326 Appendix G Datasheets for local climate zones (LCZ) ..............................................335  viii  List of Tables Table 3.1  Arguments against ∆Tu-r as a trusted measure of heat island magnitude......................58  Table 3.2  Field sites representing “urban” and “rural” landscapes in observational heat island literature, 1950 to 2007...............................................................................60  Table 3.3  Illustrated examples of “urban” and “rural” field sites in modern observational heat island literature ......................................................................................................61  Table 4.1  Decision key for Criterion 1—Operational test and conceptual model are aligned.........................................................................................................................101  Table 4.2  Decision key for Criterion 2—Operational definitions are explicitly stated ..............102  Table 4.3  Decision key for Criterion 3—Instrumentation specifications are explicitly stated ...........................................................................................................................104  Table 4.4  Decision key for Criterion 4—Site metadata are appropriately detailed ....................106  Table 4.5  Decision key for Criterion 5—Field sites are representative of the local-scale surroundings................................................................................................................111  Table 4.6  Decision key for Criterion 6—Number of replicate observations is sufficiently large.............................................................................................................................112  Table 4.7  Decision key for Criterion 7—Weather effects are passively controlled....................115  Table 4.8  Decision key for Criterion 8—Surface effects are passively controlled .....................118  Table 4.9  Decision key for Criterion 9—Temperatures are measured synchronously ...............120  Table 4.10  Points-based grading scheme for assessing methodological quality in the heat island literature sample ...............................................................................................121  Table 4.11  Frequency distribution of the heat island literature sample by political state .............128  Table 4.12  Pass and fail ratios by scientific criterion and method of data collection ...................133  Table 4.13  Mean quality scores and missing data index (MDI) values by tier placement and method of data collection .....................................................................................141  Table 4.14  Studies in the heat island literature sample with sufficient site metadata to pass Criterion 4 ...........................................................................................................146  Table 4.15  Studies in the literature sample with best control of confounding factors on heat island magnitude .................................................................................................147  Table 4.16  Studies in the literature sample with weakest communication of definitions and procedures ............................................................................................................148  Table 5.1  Locations and dates of field site visits ........................................................................166  Table 5.2  Values of geometric and surface cover properties for local climate zones .................197  Table 5.3  Values of thermal, radiative, and metabolic properties for local climate zones .........198  Table 5.4  Metadata for classifying Sundborg`s observation sites into local climate zones ........209  Table 5.5  Metadata for classifying Taesler’s observation sites into local climate zones............213  ix  Table 5.6  Metadata for classifying Sakakibara and Matsui’s mobile observation sites into local climate zones...............................................................................................218  Table 5.7  Metadata for classifying Vancouver`s observation sites into local climate zones ......224  Table 5.8  Input parameter values for the soil-vegetation model (CAPS) ...................................231  Table 5.9  Input parameter values for the urban-surface model (TEB) .......................................232  Table 6.1  Comparison of modelled and observed diurnal temperature range in selected locations and local climate zones................................................................................251  x  List of Figures Figure 2.1  Two-layer classification of the urban atmosphere ........................................................8  Figure 2.2  Hemispheric sky views from Regina, Canada ............................................................12  Figure 2.3  Temporal variation in urban and rural air temperatures, heating/cooling rates, and heat island magnitude...........................................................................................14  Figure 2.4  Vehicle-mounted temperature sensors used in mobile urban heat island surveys of Utrecht, Netherlands, and Szeged, Hungary .............................................15  Figure 2.5  Instrument mounting used in stationary urban heat island surveys of Sendai, Japan, and Goteborg, Sweden.....................................................................................16  Figure 2.6  Relations between wind speed, cloud cover, and heat island magnitude in Orlando, Florida, 1999–2001......................................................................................20  Figure 2.7  Seasonal change in Mexico City’s monthly heat island magnitude, 1994–95............22  Figure 2.8  Relation between maximum heat island magnitude and sky view factor in European and North America cities............................................................................24  Figure 2.9  Relation between observed rural thermal admittance and potential heat island magnitude in Vancouver.............................................................................................26  Figure 3.1  Representativeness of observation points along Sundborg’s 1948–49 mobile survey route of Uppsala ..............................................................................................38  Figure 3.2  Chandler’s temperature traverse across London on October 11, 1961 .......................41  Figure 3.3  Distribution of climate stations and minimum temperatures in Mexico City, February 8, 1972 ................................................................................................44  Figure 3.4  Kumagaya “rural” site used by Yamashita to quantify heat island magnitude in Tokyo......................................................................................................................65  Figure 3.5  Eye-level and aerial views of “rural” sites defining heat island magnitude in region- and city-based urbanisms ...............................................................................69  Figure 4.1  Flow diagram illustrating the selection of literature for synthesis and evaluation....................................................................................................................82  Figure 4.2  Criterion-based scheme to determine tier placements in the heat island literature sample........................................................................................................123  Figure 4.3  Geographic distribution of heat island observations in the literature sample ...........128  Figure 4.4  Percentage frequency distribution of the literature sample by geographic realm, urban population, and number of heat island studies.....................................129  Figure 4.5  Frequency distribution of the heat island literature sample by decade and tier placement............................................................................................................130  Figure 4.6  Frequency distribution of the heat island literature sample by document source .....130  Figure 4.7  Frequency distribution of the heat island literature sample by scientific criterion and aggregate pass/fail ratios .....................................................................132  xi  Figure 4.8  Frequency distribution of the heat island literature sample by tier placement and method of data collection ...................................................................................139  Figure 4.9  Frequency distribution of the heat island literature sample by point-based quality scores and tier placement..............................................................................140  Figure 4.10  Frequency distribution of the heat island literature sample by missing data index (MDI) values and tier placement ....................................................................142  Figure 4.11  Distribution of the heat island literature sample by missing data index (MDI) values, points scores, and tier placement ..................................................................143  Figure 4.12  Distribution of the heat island literature sample by points scores, rank placement, and tier placement ..................................................................................144  Figure 4.13  Distribution of the heat island literature sample by missing data index (MDI) values, rank placement, and tier placement ..............................................................144  Figure 5.1  Chandler’s classification of local climate regions in the city of London..................152  Figure 5.2  Ellefsen’s urban terrain zone (UTZ) for “redeveloped core area”............................153  Figure 5.3  Oke’s Urban Climate Zone (UCZ) classification scheme ........................................154  Figure 5.4  Logical division of the universe class by roughness objects, object height, object density, and land cover...................................................................................163  Figure 5.5  View toward central Uppsala, circa 1948 .................................................................169  Figure 5.6  Vehicle and thermometer used in Sundborg’s 1948–49 traverses of Uppsala..........170  Figure 5.7  Standard weather screen at the Uppsala Meteorological Institute, one of nine sites used in Taesler’s temperature study..........................................................174  Figure 5.8  Vehicle and instrumentation used by Sakakibara and Matsui during mobile surveys of Nagano basin, 2001–02 ...........................................................................176  Figure 5.9  Elevated view of Nagano basin in winter .................................................................176  Figure 5.10  Street canyon, paddy field, and orchard sites in Nagano basin, 2006–07.................179  Figure 5.11  Vehicle and instrumentation used during mobile surveys of Vancouver, 1992–2006 ................................................................................................................180  Figure 5.12  Elevated view of central Vancouver and its coastal mountain setting......................182  Figure 5.13  Schematic representation of the coupled OSU-CAPS and OSU-TEB models for simulating diurnal temperature range in natural and built landscapes, ...............186  Figure 5.14  Local climate zones ..................................................................................................196  Figure 5.15  Observation sites of Sundborg’s mobile temperature survey of Uppsala, 1948–49 ....................................................................................................................208  Figure 5.16  Thermal differentiation of local climate zones in Uppsala, 1948–49 .......................210  Figure 5.17  Temperature cross-sections through local climate zones in Uppsala, 1948–49 ....................................................................................................................211  Figure 5.18  Observation sites of Taesler’s stationary temperature survey of Uppsala, 1975–76 ....................................................................................................................212  xii  Figure 5.19  Daily temperature extrema in local climate zones of Uppsala, 1976: (a) June 10; (b) September 22...................................................................................214  Figure 5.20  Time-temperature series for local climate zones in Uppsala, 20–23 September 1976 .............................................................................................215  Figure 5.21  Time-temperature series for local climate zones in Uppsala, 1976: (a) April 10–13; (b) June 9–11 .................................................................................216  Figure 5.22  Observation sites of Sakakibara and Matsui’s mobile temperature survey of Nagano basin, 2001–02.............................................................................................217  Figure 5.23  Thermal differentiation of local climate zones in Nagano basin, 2001–02 ..............219  Figure 5.24  Temperature cross-sections through local climate zones in Nagano basin, 2001–02 ....................................................................................................................220  Figure 5.25  Differences in the diurnal temperature range (DTR) of local climate zones in Nagano basin, 2006–07: (a) low plants vs. scattered trees; (b) compact lowrise vs. low plants; (c) compact lowrise vs. scattered trees ................................221  Figure 5.26  Observation sites from mobile temperature surveys of Vancouver, 1992–2010 ................................................................................................................223  Figure 5.27  Thermal differentiation of local climate zones using paired control sites in Vancouver, March 2008 and 2010............................................................................226  Figure 5.28  Thermal differentiation of local climate zones in Vancouver in March 2008 and 2010....................................................................................................................227  Figure 5.29  Thermal differentiation of local climate zones in Vancouver, 2005–06 ..................228  Figure 5.30  Temperature cross-sections through local climate zones in Vancouver, 1999 .........229  Figure 5.31  Thermal differentiation of local climate zones in Vancouver, 1992–96...................230  Figure 5.32  Simulated diurnal temperature range for local climate zones...................................235  Figure 6.1  Guidelines for classifying heat island field sites into local climate zones................254  Figure 6.2  Using wind roses to define “circles” of influence and local climate zones for a screen-height temperature sensor .....................................................................256  Figure 6.3  Standardized comparisons of measured inter-city heat island magnitudes and inter-zone temperature differences using local climate zones (∆TLCZ) ..............263  Figure 6.4  Typology of urban heat island magnitudes based on synchronous temperature differences between local climate zones (∆TLCZ) .................................266  xiii  Acknowledgements I am deeply indebted to many helpful and thoughtful people at UBC, but above all to my supervisor and mentor, Professor Tim Oke. Throughout my doctoral studies, Professor Oke has inspired me with his boundless wisdom of urban climates; he has motivated me with his passion and drive for groundbreaking research; he has engaged me with honest and sometimes humorous perspectives on scholarly life; and he has offered me endless opportunity, generosity, friendship, and counsel. For this, my horizons have opened wide and my gratitude extends beyond words.  I am indebted also to the members of my supervisory committee: Professors Anthony Brazel (Arizona State University), Jim Glassman (UBC Geography), and Mike Church (UBC Geography). Together their breadth of expertise enhanced all aspects of my research. I thank Professor Church in particular for his insightful teaching and invaluable feedback on early manuscripts of this thesis. My research has benefited enormously from the professional translations of many volunteers who committed long hours to this task. To those volunteers—Professors Andreas Christen, Mike Bovis, Mike Church, Robert North (all of UBC Geography), Etsuko Yasui (Brandon University), and Ms. Zheng Zheng (Michelle) Qu of Toronto—I am extremely grateful for your time and contributions. My gratitude is owed especially to Etsuko Yasui and Michelle Qu for their skilful interpretation of Japanese and Chinese materials used in this thesis.  Throughout my fieldwork, I was fortunate to visit many urban climatologists in many cities. To the hosts of these visits—Anthony Brazel, Phoenix (Arizona); Roger Taesler, Uppsala (Sweden); Janos Unger, Szeged (Hungary); Reinhard Bohm, Vienna (Austria); Mariusz Szymanowski, Wroclaw (Poland); Krzysztof Fortuniak, Lodz (Poland); Yasushi Sakakibara, Nagano (Japan); Shuji Yamashita, Tokyo; Kiyotaka Sakaida, Sendai (Japan); Yeon-Hee Kim, Seoul; and Janet Nichol, Hong Kong—I am sincerely thankful. My xiv  understanding of urban heat islands has grown immeasurably by your hospitality, and I will forever cherish our exchange of research, culture, and friendship.  I extend my gratitude to the urban climate research labs of Professors Tim Oke and Andreas Christen at UBC; Ake Sundborg and Roger Taesler at Uppsala University (Sweden); and Yasushi Sakakibara at Shinshu University (Japan). These labs provided valuable unpublished datasets that improved my research in countless ways. I am also thankful for the enthusiastic collaboration of fellow graduate student Scott Krayenhoff (UBC Geography), who took time away from his own research to develop and run the numerical models used in this thesis. The interest, dedication, and expertise that he brought to this work are truly inspiring. I warmly acknowledge the international graduate community of St. John’s College, UBC. My life as a doctoral student, and as a junior fellow of the College, has been greatly enriched by the beautiful setting and the intellectual spirit of this close-knit community.  Lastly, I give heartfelt thanks to the family and friends that surround me. It is through their love and support that I take such pleasure in research and writing. To Michelle Qu, I have the deepest appreciation for the sacrifices you have made: you have shown unimaginable patience, understanding, and selflessness throughout this long journey, and for that I am forever grateful.  This research was made possible through a Canada Graduate Scholarship and a Doctoral Fellowship (Discovery Grant to Professor Oke) from the Natural Sciences and Engineering Research Council of Canada (NSERC). Support was also provided through a PhD Tuition Award from the University of British Columbia, and Teaching and Research Assistantships from the Department of Geography and the UBC-Ritsumeikan (Japan) Academic Exchange Program.  xv  To Zheng Zheng (Michelle) Qu  xvi  CHAPTER 1 INTRODUCTION 1.1 The problem Urban development leads to radical land cover change. As cities expand into surrounding forests, grasslands, and deserts, the natural cover is replaced with roads, buildings, parks, and gardens. The environmental implications of this change are often subtle, as in primitive or sparsely populated settlements, but in most modern cities the implications are dramatic and long term. Land cover change in cities has significant effects on local climate: temperature, precipitation, humidity, wind, and, to a lesser extent, cloud and radiation are all noticeably different between a city and its countryside. These differences are sufficiently well documented in scientific literature that climatologists regard cities as having unique local climates, much like lakes, valleys, and coastlines. The one variable that best distinguishes city and country climates is air temperature. Substantive research points to the conclusion that cities are warmer, on average, than their natural surroundings. The region of warmth associated with cities is known as an urban heat island (UHI). In brief, the primary causes of the heat island are related to a city’s thermal, moisture, aerodynamic, and radiation properties, all of which are markedly different from those of the country. These differences arise from the replacement and screening of natural surfaces with perpendicular structures and heavy materials of low permeability and high heat capacity.  Pollutant emissions and  anthropogenic heat discharge into the urban atmosphere also contribute to “artificially” warm cities.  The combined effects of these urban-rural differences in surface  1  morphology, land cover, air quality, and human activity is that, as settlements become increasingly populated and built up, they emerge as “islands” of heat within the cool countryside. In contrast to this simplistic metaphor, the urban heat island is a complex phenomenon that operates at micro, local, and meso time and space scales: At the smallest scales, individual cars, trees, and houses create thermal microenvironments that can change in an instant; at the largest scales, metropolitan cities modify regional weather and climate patterns many kilometres to their lee. Implicit in this scale set are three heat island layers: (1) at and directly below the surface cover of the city; (2) immediately above the surface cover in the air enveloping the city and its buildings and people; and (3) in the urban boundary layer above and downwind of the city (Oke, 1976). This thesis focuses exclusively on urban heat islands occurring in layer (2): above ground but below building rooftops, or in what Oke (1976) defined as the “canopy layer.” Canopy layer heat islands are traditionally measured by synchronous screenheight (1–2 m above ground level) air temperature differences between so-called urban and rural field sites (∆Tu-r). This urban-rural temperature difference—known as the heat island magnitude—is the most widely recognized measure of city climate modification in the environmental sciences. Observations of heat island magnitude have a long and seemingly well-documented history in scientific literature. In two centuries since the first heat island “experiment” in London, England, heat islands have been studied in hundreds of cities and towns worldwide, including almost every major city in Europe, East Asia, and North America. Together these studies constitute a literature of great historical and geographical interest, and of unparalleled contribution to urban climatology.  2  However, despite the apparent worth of this literature, its universal expression of city climate influence—i.e., heat island magnitude—is arguably the most misrepresented in urban climatology. Heat island magnitude is conventionally defined as an “urbanrural” temperature difference (∆Tu-r), the meaning of which is intuitive but whose operational use in climate literature is subjective and overly simplistic. Compounding this problem is the geographic scope of the heat island literature, so broad as to encompass urban and rural landscapes of every description across the inhabitable world. “Urban-rural” classification thereby obscures the particular surface and near-surface temperature regimes that actually determine heat island magnitude. The urban-rural paradigm is well entrenched in heat island studies—investigators routinely design, interpret, and report their observations with reference only to “urban” and “rural” sites and temperatures. With no alternative, heat island investigators have had little choice but to follow this convention. Meanwhile, the urban-rural dichotomy has created methodological and philosophical problems in the literature that are now compromising the authenticity of reported findings.  Especially problematic are  indiscriminate siting and classification of urban and rural field sites, poor communication of site metadata, and lack of scientific rigor in observational and analytical procedures.  1.2 Research objectives With this introduction, questions on which to build a research framework are numerous: •  To what extent are estimates of UHI magnitude in urban climate literature scientifically defensible?  •  By what criteria can such a scientific judgement be made?  •  What areas of methodology and communication in the literature are  3  particularly weak, or strong? •  How might the traditional method of heat island observation (i.e., ∆Tu-r) be improved so as to overcome these weaknesses?  At the core of heat island methodology are questions about the urban-rural dichotomy as a basis for definition and classification: •  Can the terms urban and rural be defined with universal understanding, and with greater relevance to climate science?  •  Can such a definition of urban and rural support a new classification of heat island field sites?  •  Can such a classification of field sites yield estimates of UHI magnitude that are more defensible than traditional estimates?  In this thesis, I address these questions through two separate methodologies. The first methodology uses “research synthesis” to integrate and evaluate a sample of observational heat island studies and their reported UHI magnitudes.  The specific  objectives of the research synthesis are (1) To statistically assess the quality of UHI magnitudes in observational heat island literature using criteria of experimental design and communication; (2) To establish guidelines and recommendations for improving methodological rigor in observational heat island literature. The findings of the heat island research synthesis are discouraging: a large proportion of the literature sample is compromised by poor scientific practice, with nearly one-half of all reported urban heat island magnitudes deemed scientifically indefensible. The second methodology uses the logic and theory of scientific classification to construct a new system for classifying heat island field sites worldwide. The system comprises 17 standardized local climate zones (LCZ), each characterised by a unique land cover, surface geometry, and thermal climate.  4  The specific objectives of the  classification system are (1) To provide a relevant, objective, and constrained interpretation of urban and rural landscapes for heat island investigators; (2) To provide a reliable and robust measure of UHI magnitude for climatologists and environmental scientists; (3) To support a typology of UHI magnitudes based on standardized temperature observations in local climate zones. Results show that the new classification leads to a more purposeful interpretation of ∆Tu-r by restricting its measurement to meaningfully defined local climate zones. The new classification also leads to a heat island typology representing UHI magnitude in any city, based on differences in the surface morphology and fabric of any two zones.  1.3 Thesis contents The thesis divides logically into three parts:  (1) chapters 2 and 3 cover  background material to the urban heat island effect and to the research problem; (2) chapters 4 and 5 constitute the main body of the thesis and account for its methods and results; and (3) chapters 6 and 7 generalize the thesis and summarize its results and contributions. Each chapter begins with a short introduction, followed by supporting arguments, evidence, and discussion.  In Chapter 2, I introduce the urban heat island as the  phenomenon of study. Its various types, causes, and methods of observation are defined, as are the specialized terms that appear throughout the thesis. In Chapter 3, I provide an historical review of observational heat island literature, with emphasis on the conceptual, technological, and regional advances of the nineteenth and twentieth centuries. I then take a critical look at the methodological and philosophical foundations of urban heat island studies.  Specific research questions arise from this critique and from the  5  highlighted weaknesses in heat island definition, observation, and reporting. The chapter closes with a formal statement of the research problem and the methodological approaches to solve that problem. In Chapter 4, I give a full account of the procedures used in research synthesis, and summarize the results in the form of merit-based grades and ranks. I develop a new field site classification system for heat island investigators in Chapter 5, first by taking inventory of existing urban climate classification systems, and then by extracting localscale climate-based classes from the urban-rural continuum. Thermal differences among local climate classes are evaluated with observed temperatures from Uppsala (Sweden), Nagano (Japan), and Vancouver (Canada), and with simulated temperatures from numerical surface-atmospheric models. In Chapter 6, I discuss the methods and results of the research synthesis and classification system for their limitations, broader meanings, and practical uses. I give recommendations to improve method and communication in heat island literature, and I stipulate guidelines for using local climate zones to define and quantify UHI magnitude in any city. I then develop a conceptual heat island typology that ties all strands of the thesis into a single, universal abstraction. The chapter closes with a short discussion of future research possibilities involving literature synthesis and the LCZ classification system.  Lastly, in Chapter 7, I summarize the main findings of the thesis and its  contributions to new knowledge.  6  CHAPTER 2 OVERVIEW OF URBAN HEAT ISLANDS This chapter begins with a simple definition of the heat island and its associated time and space scales. This division of scales leads to a classification of heat island types, the classic type being a canopy layer heat island, which is the subject of this chapter and the remainder of the thesis. The next section briefly examines the causes of the heat island through a systems perspective on energy and radiation exchange at and above typical urban and rural surfaces. Methodological approaches to observing canopy layer heat islands in the field are discussed, with a follow-up section on heat island magnitude, its accepted definition, and the factors that control it.  2.1 Defining the urban heat island A conceptual definition and classification of an urban heat island must stipulate the method (remote vs. ground based) and medium (air vs. surface) through which its time and space patterns are observed. Urban heat islands are generally considered localscale phenomena, meaning that horizontal and vertical distances of most relevance are in the range of 102 to 104 metres.  Micrometeorological heat island investigations use  sampling intervals of only tens of metres, while meso-scale investigations span tens or hundreds of kilometres between cities and countrysides. Time scales of most interest to heat island investigators range from hours to years.  2.1.1 Heat island types Urban heat islands can be identified by the scales and processes responsible for their formation, or by the methods and instruments used to observe them. While all urban heat islands are connected through exchanges of mass, energy, and momentum,  7  each heat island type involves different processes, measurement systems, and time and space scales.  Urban heat islands are divided first into surface, subsurface, and  atmospheric types depending on which of these media is sensed. The first type exists at the surface and is defined by temperature differences between urban roads and buildings, and rural soils, plants, and tree-tops. Surface heat islands are complicated by the many urban and rural materials that vary in temperature by several tens of degrees Celsius across a local area.  The second heat island type exists beneath the city surface.  Subsurface heat islands receive much less attention in climate literature. A warm city surface conducts heat down into the ground, and thus the subsurface temperature distribution generally corresponds with that above the surface. High temperatures are typically found below heavily built-up areas of the city, and low temperatures below natural areas (Taniguchi et al., 2008). The third and most traditional heat island type exists in the atmosphere (Figure 2.1). Oke (1976) distinguished two types of atmospheric heat islands, each associated with different, but still physically connected, layers of the urban atmosphere.  The  boundary layer urban heat island is governed by local and mesoscale processes and is  Wind Planetary Boundary Layer  Urban Canopy Layer Urban Boundary Layer (meso scale)  (local scale)  (macro scale)  Rural  Urban  Rural  Figure 2.1 Two-layer classification of the urban atmosphere (Oke, 1976; adapted by permission).  8  associated with that portion of the planetary layer whose climatic characteristics are affected by an underlying city. In a much shallower portion of atmosphere above ground but below the city roofline is the classical canopy layer urban heat island. Its climate, and in particular the exchange of energy and radiation, is governed by microscale processes operating among the streets, buildings, and other surface features of a city. Canopy layer heat islands are of great concern to geographers and climatologists because they affect the spaces of everyday human activity.  As an amalgamation of  microclimates, the urban canopy layer heat island is essentially a local-scale phenomenon.  2.2 Causes of the canopy layer urban heat island Urban and rural environments respond differently to processes of energy (solar heating), mass (precipitation), and momentum transfer (airflow). These differences are due primarily to unique combinations of radiative, thermal, moisture, and aerodynamic properties of particular sites, which influence how urban and rural surfaces regulate and partition available energy (Oke, 1987). Energy partitioning in turn influences the thermal climates of these surfaces and of the volumetric air layer to which they are coupled. The following treatment of surface energy budgets and air-volume temperature change is highly reductionist. The two sites chosen to represent urban and rural are a street ―canyon‖ and a bare field, respectively. A street canyon is a basic structural model commonly used in urban climatology to illustrate and understand city microclimates. It has been adopted universally to conceptualize typical building-street arrangements that comprise most urban environments. The canyon itself consists of building walls on either side of a paved street, with two open ends and an open top. For contrast with the canyon, the bare field site represents a typical agricultural (or undeveloped) tract of land. Its  9  surface is flat, dry, and isotropic, and thus not complicated by trees or non-uniform geometry. During the day, solar forcing is the main provider of surface energy to an urban canyon and a bare field site (Oke, 1987). On reaching the active surfaces of the canyon and bare field, the energy is partitioned differently. These differences are driven by surface geometry. City surfaces are structurally more complex, and therefore offer many horizontal and vertical surfaces for trapping and absorbing solar radiation. The blocklike geometry of the urban canyon also allows warm air to stagnate, especially if the canyons are deep and narrow. Lower wind speeds in the canyon inhibit evaporational cooling of its surfaces. Differences in urban and rural geometry therefore set differences in albedo, wind flow, and, ultimately, daytime heat storage in the surface substrate. Dissipation of energy in the urban canyon and bare field sites is further differentiated by surface materials and moisture availability. At the bare field site, the surplus is dissipated through evaporational cooling of the surface. Thus, any residual energy to warm the soil surface and generate upward sensible heat flux into the lower air volume is limited. Soil heat flux, which is normally directed into the ground by day, is reduced by the dominant role of evaporation in the surface energy balance. At the canyon site, the partitioning of energy into sensible and latent heat forms is entirely different. Paved roads and impervious walls and roofs quickly shed precipitation into catchment basins, leaving surfaces drier than in the countryside. The predominance of impervious cover in the city dramatically alters the partitioning of latent and sensible heat in urban energy budgets.  Dry, impermeable surfaces like canyon floors and walls  enhance sensible heat transfer, which often creates a relative evaporation deficit in the  10  city. Daytime surface energy and radiation balances at the canyon and bare field sites are directly coupled to temperature change in the adjacent air volume. At the canyon site, the floor and walls heat up when irradiated, and a surface temperature wave penetrates the air volume by conduction and convection (Oke, 1987). Heat flux convergence warms the canyon air volume and is an important driver in the formation of daytime thermal microclimates. At the bare field site, daytime air temperature change depends mainly on surface moisture availability and the potential for radiative and heat flux convergence or divergence at the surface and air layers. Assuming little or no advection, upward sensible heat flux from the surface into the lower air volume generally exceeds heat flux from the top of the air volume. Like the canyon, this imbalance leads to heat flux convergence and a subsequent rise in daytime air temperature. Inputs and outputs of radiation in the rural air volume are not significantly different, and so radiative flux convergence/divergence at the bare field site is roughly zero. Daytime rural air temperatures often exceed those in the urban canyon, depending on canyon geometry and rural soil moisture, but by sunset, heat uptake and storage in the urban surface will have almost certainly exceeded that of the rural surface. During the nighttime, the energy and radiation balances of the urban canyon and bare field take different directions and/or densities.  Solar radiation is no longer a  component of the energy balance, yet long-wave radiation becomes increasingly important in setting radiation budgets and temperature change in the urban and rural air volumes (Nunez and Oke, 1976). The long-wave energy balance favours the urban canyon because outgoing radiation from the floor is greatly reduced in comparison to the  11  bare field, due to the reduced ―sky view‖ (i.e., the proportion of sky hemisphere visible from ground level) from the canyon floor. Sky view factors (sky ) vary significantly in both the city and countryside and are closely related to the packing of trees and buildings (Figure 2.2). As a general rule, rural environments (except forests) have larger sky views than cities, and therefore experience greater escape of long-wave radiation to the open and radiatively cold sky. In contrast, buildings and trees reduce outgoing radiation flux, and thus there is less potential for heat escape from warm urban surfaces (Arnfield, 1990; Johnson et al., 1991). Sky view factor not only fixes the long-wave radiation budgets for the canyon and bare field surfaces, it also fixes nocturnal radiative and heat flux divergence/convergence in the air volumes (Nunez and Oke, 1976). At both urban and rural sites, the surface radiation balance becomes negative after sunset due to long-wave emissions. The surface temperature therefore drops and sensible heat is extracted from the lower air layer and the underlying ground. The net result is a post-sunset drop in screen-level air temperature. While the same radiative processes operate in the urban canyon, they are complicated by  Figure 2.2 Hemispheric sky views from Regina, Canada. Left: city outskirts, sky = 0.85. Middle: tree-lined residential street, sky = 0.55. Right: city centre, sky = 0.40.  12  vertical walls and obstructed sky views. Upward sensible heat flux from the canyon floor and inward flux from the walls partially compensate sensible heat flux through the canyon top. Heat flux divergence (i.e., cooling) in the canyon air volume is therefore less than in the rural air volume, and post-sunset air temperature response in the canyon more conservative than at the bare field. Counteracting the fall of air temperatures in the urban canyon is the release of stored heat from the fabric. Heat stored in the canyon fabric is greater than the bare field (dry) due to daytime solar uptake by the many vertical and horizontal surfaces of a building-street configuration. Urban structures of high thermal mass and inertia, such as asphalt streets and stone or concrete buildings, have especially large capacities to store heat by day, and to release that heat slowly through the night (Goward, 1981). The upward flux of stored heat toward the canyon floor greatly exceeds the counterpart flux at the bare field. This difference further reduces surface cooling of the canyon floor, and thus heat flux divergence in the air volume (Nunez and Oke, 1976). Reduced turbulent flux from the wind-shelter effects of a canyon also moderates surface and air cooling, as does anthropogenic heat and moisture flux (QF) from buildings, vehicles, industry, manufacturing, and human metabolism (Sailor, 2011).  2.3 Heat island genesis With this framework of urban and rural radiation and surface energy balances, the genesis of nocturnal urban heat islands in the canopy layer is easily anticipated. During the day, a dry, bare field takes less heat into storage than a structurally complex urban canyon, which has a larger surface area for heat uptake. At night, urban and rural temperatures diverge, and a heat island develops in the canopy layer.  13  The rural  environment cools quickly in the hours following sunset due to its open exposure and unabated radiation escape (Figure 2.3).  Cooling rates in the urban canyon are  comparatively sluggish due to the reduced sky view and the large diurnal heat store in the fabric (Arnfield, 1990; Johnson et al., 1991). The urban heat island normally persists through the night until shortly after sunrise when the daily solar cycle begins and the rural environment warms by insolation. For this reason, the canopy layer heat island is primarily a nocturnal phenomenon and its most important drivers are flux divergence and  Air Temperature  differential urban-rural cooling.  Urban  Heating/Cooling Rates  Rural  Urban Rural  Heat Island Magnitude  Tu-r  2K sunset  midnight  sunrise  Time (hr) (hr) Time  Figure 2.3 Temporal variation in urban and rural air temperatures, heating/cooling rates, and heat island magnitude (Oke, 1987; by permission).  14  2.4 Observing the canopy layer urban heat island Urban heat islands are traditionally measured via ground-based methods, which involve the mobile transport of temperature sensors across a city and its countryside, or a stationary network of sensors distributed in and around a city.  2.4.1 Mobile surveys Mobile surveys provide a simple, efficient, and inexpensive means of gathering temperature data across a city and its countryside. They are widely popular for assessing and quantifying canopy layer urban heat islands. Survey routes across an urban-rural area may be linear or circuitous, and are designed to obtain a dense sample of temperatures in a relatively short period of time. Automobiles are the favoured mode of transportation during a mobile survey, although trains, bicycles, motor scooters, and foot traverses have also been used. Temperature sensors are normally attached to roofs or bumpers of cars such that engine or exhaust heat is avoided while traversing the city (Figure 2.4). Mobile surveys are advantaged for sampling the microscale features of an urban heat island, such as a park or factory site, and for sampling a range of synoptic  Figure 2.4 Vehicle-mounted sensors used in mobile temperature surveys of Utrecht, Netherlands (Conrads and van der Hage, 1971), and Szeged, Hungary (Unger, 2006).  15  weather conditions and times of day. Single-vehicle surveys are especially advantaged because error from instrument calibration between two (or more) vehicles is eliminated. Despite the high spatial resolution of mobile surveys, they are limited in a temporal respect, with sampling periods rarely exceeding several hours. Upon completion of a mobile survey, the data are subjected to a time-correction scheme to account for regional temperature change during the time of survey. The rate of regional temperature change is usually determined from stationary urban and/or rural observatories in the study area, or from one or more crossover points along the traverse route. Simple regression techniques are then used to adjust the raw temperatures to a common base time.  2.4.2 Stationary surveys An alternative approach to a mobile traverse is a stationary survey involving fixed temperature sensors (Figure 2.5). A conventional arrangement of sensors consists of centrally located urban sites paired with nearby rural sites. The sensors may exist in situ at airports or meteorological observatories, or they may be installed at strategic points across a city and its countryside. If the network is sufficiently dense, it can highlight  Figure 2.5 Instrument mounting used in stationary temperature surveys of Sendai, Japan (Sakaida and Egoshi, 2006), and Goteborg, Sweden (Eliasson, 1994).  16  desired aspects of the urban-rural temperature field, such as the thermal effects of residential or agricultural areas. In contrast to mobile surveys, stationary surveys are favoured for monitoring the temporal rather than spatial aspects of urban heat islands. The challenge of finding secure sites at which to locate the stations is a drawback of this method. A stationary network may comprise temperature sensors and weather boxes that meet all WMO standards, or it may comprise a comparatively crude arrangement of radiation shields mounted to improvised support structures like utility poles, trees, fences, or buildings.  2.5 Magnitude of the canopy layer urban heat island 2.5.1 Definition The magnitude, or intensity, of the urban heat island is customarily defined as a synchronous screen-level (1–2 m agl) air temperature difference between mobile or stationary urban and rural thermal sensors (Tu-r). In most heat island studies, Tu-r is measured in one of three ways: (1) the temperature difference between the warmest and coolest points among a stationary network of sensors or along a mobile traverse route (e.g., Chandler, 1961); (2) the temperature difference between spatial averages of several urban and rural temperatures measured with stationary and/or mobile sensors (e.g., Sundborg, 1951); or (3) the temperature difference between an outlying rural sensor, often at an airport, and a sensor at or near the geographic centre of a city or its central business district (e.g., Runnalls and Oke, 2000). Measuring UHI magnitude via (3) is commonly regarded as the most accessible and convenient approach because airport weather stations are often located in or near most cities. All three measurements are equally simple to operationalize and require minimal time, labour, and equipment.  17  Alternatives to Tu-r as the universal standard for defining UHI magnitude were explored by Lowry (1977) in a classic study on experimental method in urban climatology. Lowry believed that the only means of obtaining accurate estimates of urban impact on thermal climate is through direct manipulation of a closed and controlled urban-atmosphere system. In reality, cities cannot be placed into or removed from a field experiment, and thus fully controlled heat island observations are impossible to carry out. Tu-r has therefore become a popular stand-in for heat island magnitude. Lowry argued that Tu-r is handicapped in two ways: (1) in most studies the surface climate of the chosen rural site is directly influenced by advected and entrained air properties from its neighbouring city; and (2) rural sites are often not representative of the natural, undisturbed environment that existed prior to urbanization. Implicit in this argument is a distinction between rural and preurban environments. Lowry allowed that empirical estimates of UHI magnitude can be accepted if rural temperatures are sampled instead of preurban temperatures, but warned that the observed urban-rural differences may not be a result solely of urbanization. Other non-urban influences on local temperature, like relief or water bodies, might confound the results. Estimates of UHI magnitude that compare urban and preurban temperatures from the same location but at different times, or from the same time but at different locations, are preferred. Lowry’s proposition that UHI magnitude be defined as an urban-preurban temperature difference is theoretically sound, but operationally problematic. Finding suitable preurban temperatures to compare with urban temperatures is a difficult challenge for most heat island investigators, given that the surroundings of most cities are no longer in native condition. Few studies have defined UHI magnitude with pre-urban temperatures (e.g., Hinkel et al., 2003; Yow and  18  Carbone, 2006).  2.5.2 Controls Urban heat island controls differ from causes in that they do not, in and of themselves, generate city heat. Instead they set the absolute temperature difference between urban and rural environments. Controls are important because they determine the perceived magnitude of the urban heat island. The following sections discuss five major controls on UHI magnitude. Although each is highly sensitive to site and setting, strong relations exist among all controls. Examining one control at the exclusion of others is therefore impossible. Regardless of the methods used to observe or measure a heat island, or one’s operational definition of that heat island, the meteorological, geographical, and structural controls on its magnitude remain the same.  2.5.2.1 Weather The most widely studied control on heat island magnitude is weather. Most heat island studies demonstrate that light winds and clear skies provide ideal conditions to maximize UHI magnitude (e.g., Sundborg, 1951). The prevailing weather conditions during heat island measurement, as well as the antecedent conditions leading up to a heat island event, have a strong influence on observed urban-rural temperature differences (Stewart, 2000). UHI magnitude shows close correlation with cloud type and amount (Yow and Carbone, 2006; Figure 2.6). If clouds are present, a cool, high, thin cover (e.g., cirrus) is associated with more intense heat islands than a warm, low, thick cover (e.g., stratocumulus). In Melbourne, Australia, for example, mean annual UHI magnitude from 1972 to 1991 was 2–3 K on calm, clear nights, but dropped to less than 1 K in windy, overcast conditions (Morris et al., 2001). The explanation for this drop is that daytime  19  Figure 2.6 Relations between wind speed, cloud cover, and heat island magnitude in Orlando, Florida, 1999–2001 (Yow and Carbone, 2006; by permission). Bolz cloud factor varies with cloud type and amount, and ranges from 0 (fog or low cloud) to 1 (clear sky or high cloud) (Bolz, 1949).  cloud cover limits heat stored in the urban fabric, and a nocturnal cover sets the strength of the sky heat sink. Open skies at night thus reduce incoming long-wave radiation to rural surfaces, and weak winds reduce the turbulent mixing and transport of urban and rural surface air.  Together these conditions promote sharp urban-rural temperature  gradients and maximum UHI magnitudes. Other weather variables such as humidity and atmospheric pressure have much less influence on UHI magnitudes (Duckworth and Sandberg, 1954).  2.5.2.2 Background climate Climates that promote large amplitudes in the diurnal temperature range produce large UHI magnitudes, in essence because magnitude is defined as a difference of urban and rural temperatures. A rise or fall in rural temperatures will therefore decrease or  20  increase measured urban-rural thermal differences even if urban temperatures remain constant. If rural temperatures fall dramatically after sunset while urban temperatures remain comparatively steady, nocturnal UHI magnitude will be large. Cities in desert or dry continental climates, like Phoenix, Arizona, report maximum UHI magnitudes exceeding 10 K (e.g., Hawkins et al., 2004). Coastal or humid cities are less likely to experience such intense heat islands. Maximum UHI magnitudes observed during 1 year of stationary measurement in the wet tropical city of Singapore, for example, averaged 3– 7 K (Chow and Roth, 2006). The relatively low magnitudes in Singapore are explained in part by the decreased amplitudes of the diurnal temperature range in wet climates, and especially in rural climates where atmospheric humidity is slightly higher than in the city. In addition to air properties, surface moisture also plays an important role in setting a reduced amplitude in daily temperature cycles, particularly in rural environments where natural surfaces predominate (see section 2.5.2.5).  2.5.2.3 Time Heat island magnitude changes across hourly, daily, and seasonal time scales. If boundary conditions are ideal, peak heat island values generally occur several hours after sunset when rural cooling rates are greatest and urban rates lag behind (see Figure 2.3). In New York City, seasonal magnitudes peaked at 3–4 K several hours after sunset, and dropped below 1 K by early morning (Gedzelman et al., 2003). At weekly time scales, heat island magnitudes vary with day-to-day cycles of human activity, that is, they diminish slightly on Sundays when traffic volume and energy consumption drop. In Seoul, weekend UHI magnitudes average 0.5 K less than weekday magnitudes (Kim and Baik, 2005). The contribution of anthropogenic heat and moisture to the urban energy  21  budget, and to measured UHI magnitudes in Seoul, is thought to be less significant during weekends due to lower human energy consumption, especially for transportation. Finally, at seasonal time scales, UHI magnitude fluctuates with warm and cold, or dry and wet, cycles. In high- and midlatitude cities, UHI magnitudes during the cold season may be greater than the warm season if anthropogenic heat flux from winter space heating increases significantly.  However, differences in the synoptic and surface moisture  conditions of warm and cold seasons may override any effects of anthropogenic heat flux on winter UHI magnitude. In tropical cities, UHI magnitude fluctuates with dry and wet cycles, particularly as the surface moisture status changes (Figure 2.7; see also section 2.5.2.5). Adebayo (1991) observed mean UHI magnitudes of 12 K in Ibadan, Nigeria, during its hottest and driest months, and of 6 K during its wettest month. Cyclical differences in surface and air properties at the urban and, especially, rural sites are  180  6 5.5 5 4.5 4 3.5 3 2.5 2 1.5 1  160 140 120 100 80 60 40 20 0 Jan  Feb  Mar  Apr  May June  Jul  Aug Sept  Oct  Nov Dec  Figure 2.7 Seasonal change in Mexico City’s monthly heat island magnitude, 1994–1995 (Jauregui, 1997). Bars = mean monthly precipitation; line = mean monthly heat island magnitude.  22  Heat island magnitude (K)  Mean monthly precipitation (mm)  responsible for these seasonal patterns.  2.5.2.4 Urban form and function Cities with tall, densely packed buildings have larger canyon height-to-width (H/W) ratios and smaller sky view values, and therefore experience less nocturnal cooling, than cities with open, low-density plans. Ceteris paribus, the narrow street canyons of a compact city will sustain higher nocturnal air temperatures (relative to the rural environment) than the exposed streets and open spaces of a low-density city. Oke (1981) hypothesized that urban geometry is partly responsible for the comparatively small UHI magnitudes reported in European cities, where the typical building profile of the core is lower and flatter than that of North American cities (Figure 2.8). Related to city form is city function. Urban geographers distinguish ―world cities‖ by their prominent role in the global economy (Friedmann, 1995).  World cites  represent local-scale nodes of international finance, business, and capital, and are associated with economic segregation at global, regional, and local spatial scales. An example of economic segregation at local scales is the ―citadel,‖ or cluster of gleaming towers and high-rise office buildings that occupy significant land fractions of world cities (Kleniewski, 2006). Urban economy is therefore a surrogate for urban form in much the same way that a city’s population is a surrogate for its morphology.  Economic  segregation in cities might indirectly couple with air temperature distribution so long as they are represented at similar spatial scales and in measures of urban structure like H/W ratio or impervious surface fraction. The influence of world city geometry (i.e., low sky view, high impervious fraction) on local thermal climate will likely produce strong heat islands in the  23  (K)  Figure 2.8 Relation between maximum heat island magnitude (Tu-r) and sky view factor (s) in European and North America cities (Oke, 1981; by permission).  citadels of these cities.  Industrial cities and factory towns might also have  characteristically strong heat islands if atmospheric pollution and anthropogenic heat and moisture flux density are sufficiently large to alter the urban radiation and energy balances. The potential for large UHI magnitudes also depends on the locations of urban and rural temperature measurement. In contrast to industrial cities, ―amenity‖ cities have low pollution and an abundance of open, natural space.  Amenity cities might be  associated with large sky view factors, low impervious fractions, and thus low UHI magnitudes, again depending on the locations of urban and rural temperature measurement. 24  2.4.2.5 Rural thermal admittance Rural surface properties exert a significant influence on measured UHI magnitudes because the rural environment determines the background temperature to quantify the heat island (Arnfield, 1990; Oke et al., 1991). Furthermore, the range of rural temperatures in a local or regional land area is often much larger than urban temperatures due to the spatial and temporal variability in land cover and soil properties. One such property is surface thermal admittance (), or the ease with which a surface takes up and releases heat. It is the product of the surface’s heat capacity and thermal diffusivity, and it is a useful indicator of the heat store potential and temperature wave dynamics of the surface (Oke, 1987). Surfaces of high admittance, like wet or saturated soils ( > 1,800 J m-2 s1/2 K-1), absorb large amounts of heat by day and subsequently release that heat slowly through the night. Air temperatures directly above that surface respond conservatively, and thus the potential for nocturnal cooling is reduced. Diurnal temperature range and nocturnal cooling potential are more extreme over surfaces of low admittance, such as snow or dry soils ( < 800 J m-2 s1/2 K-1). UHI magnitude quantified by urban-rural temperature differences increases if the rural reference temperature is obtained above dry or snow-covered surfaces, and decreases above wet or waterlogged surfaces (e.g., Hawkins et al., 2004). Although large UHI magnitudes are commonly perceived to result directly from more heat being produced in a city, they might, in this case, suggest only that rural admittance is low and daily temperatures range high relative to the city. Urban heat island magnitude—not urban heat—is strongly affected by rural admittance. In Vancouver, the antiphase relation between Tu-r and rural admittance is seen in observed seasonal heat island magnitudes (Figure 2.9). Maximum monthly heat  25  island magnitudes during the dry summer season average 4–6 K, but drop to 2–4 K during the wet winter season (Runnalls and Oke, 2000).  The changing moisture  conditions and thermal admittances of the rural environment are partly responsible for these seasonal differences in Vancouver’s heat island magnitude.  Figure 2.9 Relation between observed rural thermal admittance (µr) and potential heat island magnitude (K) in Vancouver. Potential heat island magnitude is the theoretically derived maximum expected value of ∆Tu-r under ideal wind and sky conditions. Smoothed lines are polynomial fits (Runnalls and Oke, 2000; by permission).  26  CHAPTER 3 CRITICAL REVIEW OF URBAN HEAT ISLAND LITERATURE 1 This chapter begins with an historical review of observational heat island literature and its development through two centuries. The review is restricted to urban heat island observations in the canopy layer, which formally began in the early 1800s. Scholarly papers that have made important contributions to the heat island literature are recognized, with special attention given to benchmark studies coinciding with major technological, methodological, and conceptual turning points in the field of urban climatology. The historical review consists of three major sections. The first examines the origins of formal heat island observation through the pioneering works of Luke Howard. The second section examines a long period of sporadic but steady growth in heat island literature from the mid-nineteenth to mid-twentieth century. The last section examines the modern period of observations from 1950 to the present day. The modern period experienced a steep rise in the number of heat island studies published worldwide, due partly to advances in instrumentation and to major theoretical and methodological achievements in urban climatology. The remainder of the chapter continues with a review of the modern period of heat island literature, but in a more philosophical light. The methods and results of this period are critically examined, and questions of epistemological concern are raised: To what extent does the literature serve the aims of science? By what judgment does it comprise ―sophisticated‖ observations? Can its measurements be trusted? One can  1  A version of Chapter 3 has been published: Stewart ID. 2007. Landscape representation and the urban-rural dichotomy in empirical urban heat island literature, 1950–2006. Acta Climatologica et Chorologica 40-41: 111–21.  27  quickly surmise from standard reviews of heat island literature that a response to these questions is not obvious, but that evidence on which to hypothesize is plentiful. Arguments developed from philosophy of science and urbanization theory uncover fundamental problems of using Tu-r to define UHI magnitude and to substantiate intercity comparisons of results. Specific examples from the published literature are given to support these positions and to expose recurring flaws in heat island methodology. These examples cast doubt over the consistency and authenticity with which heat island observations have been gathered and reported through history. The chapter closes with a methodological framework to address these problems and to answer the research questions.  3.1 Historical setting 3.1.1 Pioneering the field: Early observations References to warm cities and cool countrysides first appeared in historical annals of the seventeenth century, when urban-based thermometer networks detected ―artificial‖ warmth in European cities (Brooks, 1952; Meyer, 1991; Grimmond, 2006). At that time, the warm and polluted atmospheres of these cities were more of a health concern than a scientific one. Hot, dirty air in the city was thought to accelerate the spread of disease and endanger the health of urban inhabitants. However, these concerns were rarely supported by formal investigations into the presumed effects of urban heat and pollution, or by curiosity for their causes and characteristics. That changed in the nineteenth century when Luke Howard began a major study of the heat island effect in London, England. The beginnings of heat island research are therefore ascribed to Howard‘s London study, and for that study he is widely regarded as the founding father of urban  28  climatology (Mills, 2009). Luke  Howard  (1772–1864),  chemist  and  amateur  meteorologist,  first  distinguished himself in meteorology by his classification of cloud forms (Hamblyn, 2007). Since childhood, he had observed the sky and its cloud configurations with enough regularity to deduce that all clouds belong to one (or a combination) of three principal types: cumulus, stratus, or cirrus. Howard‘s cloud classification system is versatile yet remarkably simple—it was adopted by the WMO in 1896 and continues today as a universal system by which all clouds are named and described. Howard‘s appreciation for the universality of science and the complexities of nature brought him further distinction in urban climatology. He had the necessary interest and resources to understand urban effects on temperature, and he became the first person to conduct a lengthy, systematic investigation of the heat island phenomenon and to later publish his results (Mills, 2009). In 1806, Howard‘s observations in and around London—then a metropolis of more than a million people—began in earnest. Twenty-seven years later, his lifelong ―experiment‖ culminated in a monumental publication, The Climate of London (1833). In this book, the first of its kind in meteorology, Howard wrote of an ―imperfect‖ science with few long-term records or empirical facts on which to make discoveries and to advance the field. Howard‘s book was motivated by his belief that meteorology was ―less trodden‖ than other disciplines, and that it was lacking the ―regular and consistent form of a science.‖ In Howard‘s words, meteorology was ―in want of more data, of a greater store of facts, which…might guide [one] to more certain conclusions.‖ Howard was driven by the need not just for more facts, but for reliable facts. He  29  claimed that many of the existing meteorological records at the Royal Society in London were unworthy of publication, and that its instruments were unfit for daily use. In frustration, he forged a new approach to meteorological practice, one that advocated great care in observation and that was highly attentive to the placement of instruments and the precision of their measurements. He was meticulous in describing his thermometers and their physical surroundings, and, above all, he was sensitive to the nuances of scientific communication, promoting universal terms and measures that would allow all meteorologists to exchange data in a common tongue. This vision for a more truthful science, and this curiosity for the changing properties of the boundary-layer atmosphere, led Luke Howard along a remarkable path of observation, discovery, and publication. For more than two decades, Howard gathered daily air temperatures at city and country observatories in and around London. Temperatures that were representative of the city were gathered from the Royal Society in central London, and of the country from three observatories just a few kilometres from the urban outskirts. Country observatories at Plaistow, Tottenham, and Stratford were located in open and level surroundings that were representative of the local area. The microscale surroundings, however, were variable and non-conventional by modern standards. At Tottenham, Howard attached a thermometer to the door frame of a house; at Plaistow, he attached it to a post beneath a small tree.  Despite these seemingly peculiar sitings, which were common in the  nineteenth century because few standards existed for instrument mounting, Howard was forthright in acknowledging the potential inaccuracies that their exposures might generate.  In doing so, he carefully described the immediate environment of each  observatory, and the mounting and instrument specifications of each thermometer.  30  Howard‘s experimental set-up was simple and effective.  He extracted daily  maximum and minimum temperatures from a small network of ―common‖ (i.e., alcoholin-glass) thermometers, and with those data established consistent measures of an urbanrural temperature differential, known now as the urban heat island magnitude (Tu-r ). He was the first person to observe the heat island across an extended period of time, and to identify and characterise its changing diurnal and seasonal patterns (Mills, 2009). Through extensive and repeated observations, he concluded that city-country temperature differences in London were greatest at night, averaging 2 K. He further deduced that this temperature difference must be proportional to urban density and development, and thus regions of London with closely spaced buildings and sparse vegetation must be warmer, on average, than regions of widely set buildings and abundant vegetation. In these deductions, Howard correctly identified the temporal patterns that today characterise all urban heat islands, and the basis of spatial patterns as well. Howard‘s discovery of so much that now underpins modern-day urban climatology—the canopy layer heat island, the components of the surface energy balance, and the climatic significance of a street canyon—is remarkable for two reasons. First, he had little empirical support on which to base his discoveries, apart from daily maxima and minima from a sparse network of observatories. Second, he had no theoretical or conceptual model within which to gather, interpret, and classify his observations. In this regard, he was decades ahead of any future developments in urban climatology. Against these odds, Howard identified four of five factors that are inscribed in modern literature as universal causes of the heat island effect (Mills, 2008).  31  3.1.2 A century of growth: 1850–1950 Howard‘s research laid the groundwork for a long period of steady growth in urban climatology and heat island investigation. In 1862, Emilien Renou published the first heat island observations outside London, in the metropolis of Paris. Renou extracted city-country temperature differences from a 5-year climate record at the Observatory of Paris and a rural observatory 9.5 km south of the city (Brooks, 1952).  He found  nocturnal minima to be 1.2 K higher, on average, in the city than in the country, and he attributed this difference to air pollution, chimney fumes, human and animal respiration, and heat and moisture escape from houses.  In the half-century following Renou‘s  temperature study of Paris, other studies emerged from large cities across Europe. In each, evidence of the heat island effect was found in simple comparisons of city and country mean and minimum temperatures, usually among small networks of in situ observatories.  Simple two- and three-station configurations were employed in  temperature studies of Munich (Wittwer, 1860), Berlin (Perlewitz, 1890), Paris (Hann, 1895), Vienna (Topolansky, 1924), and Moscow (Bogolepow, 1928). In Mexico City, Moreno (1899) compared daily minima at the centrally located National Palace with minima from a rural observatory 6 km west of the city. Urban influences in temperature records of several years or longer were revealed in all of these studies, which together characterise the field of urban climatology in the early twentieth century. With the exception of city-country temperature comparisons reported in Mexico City (Moreno, 1899), St. Louis, Calcutta (von Hann, 1885), and New York City (Redway, 1919), few systematic heat island observations outside Western Europe had been documented or published prior to 1930.  32  In the coming decades—100 years after Howard‘s The Climate of London was published—heat island investigations emerged in growing numbers outside Europe. This growth was most dramatic in Japan, with investigators there developing new interest in heat island observations during the rapid urban-growth period of the 1930s (Shitara, 1978; Fujibe, 2011). Japanese heat island investigations flourished in mid-century, leading the way for a long tradition of urban climate research in many of its largest cities. Most of these early Japanese studies focused on the old core of Tokyo and its central meteorological observatory near the Imperial Castle. The earliest of these is Sasakura‘s (1931) winter investigation of the 1931 heat island. Sasakura isolated ―manmade factors‖ in Tokyo‘s temperature record by comparing daily minima at the central observatory with those in Shimokitazawa, a forested rural area 9 km west of the Castle. Temperature differences between Shimokitazawa and Tokyo were 1–2 K, on average, and were shown to fluctuate with prevailing weather conditions. Meanwhile, as Japanese investigators continued to investigate the Tokyo heat island, the course of urban climatology took significant turns.  Methodological  approaches to heat island observation were advanced on emerging data in cities of Central Europe. This important shift in region and method was motivated by growing interest in the subfield of micrometeorology, which transferred naturally into the city, with its many small-scale surface features and atmospheric processes (Kratzer, 1956). Urban-based micrometeorology spurred new and innovative techniques for acquiring city temperatures at small spatial scales. The most innovative of these involved temperature ―measuring trips‖ through city streets and countryside roads, with portable thermometers carried by hand or attached to moving automobile or bicycles. This technique was a  33  major breakthrough in urban climatology because it gave investigators access to temperatures that were previously difficult to acquire.  The data collected during  measuring trips could be used to produce detailed maps of urban and rural surface temperature distribution. Temperature profiles and isotherm maps revealed warm and cool spots in the city and provided new data for quantifying the urban-rural temperature difference. The first heat island measuring trip was conducted in 1927 by Wilhelm Schmidt (1930). On the night of May 12, Schmidt completed a 3-hour automobile journey across the city of Vienna and its communal limits. He attached a mercury thermometer to his car door and recorded air temperatures while passing through Vienna and its outlying villages, to a distance of 20 km.  Schmidt‘s innovative approach to heat island  observation was soon replicated by Albert Peppler (1929) in Karlsruhe, Germany. Peppler attached an Assmann aspirated thermometer to his automobile and gathered temperatures from a wide scatter of points across the small German city and its surrounding plain. Peppler advocated further testing of the measuring-trip approach, citing its superiority for retrieving temperatures at small spatial and temporal scales (Brooks, 1952). Through the 1930s and 40s, measurement journeys were made for the first time in the cities of Berlin, Hanover, Munich, and Freiburg (Brooks, 1952). Each of these journeys produced interesting discoveries of urban temperature distribution. In Munich, Budel and Wolf (1933) used a bicycle-mounted thermometer to show that the warmest sections of the city were invariably those of highest building density. In Hanover, Berg and Metzler‘s (1934) automobile observations revealed three thermal zones: the built-up  34  inner town with characteristically uniform temperatures; the city-country transition area with rapidly changing temperatures; and the open country, again with uniform temperatures. The results of these and other historical measuring trips confirmed the earlier deductions of Luke Howard, who, without access to automobile survey data, assumed with good reason that air temperatures around London would vary in sympathy with building height and density. The first measuring trip outside Europe was conducted in Toronto, Canada. Middleton and Millar‘s (1936) study mimicked earlier trips through European cities, but it differed slightly in that extraordinary temperature contrasts of up to 15 K were recorded along its relatively short route from the commercial district to the city‘s outskirts. The measurements started at Toronto‘s urbanized lakefront and progressed through several small and large valleys en route to the countryside. The authors explained that the dramatic temperature contrasts observed along this route were not caused by urban effects alone, but by significant relief and lake effects.  Thus the Toronto study is  important to heat island understanding because it provides an early and striking example of potentially confounding influences (urban and non-urban) on heat island measurement. Heat island investigations ceased during World War II, but resumed with heightened interest shortly after. The term urban heat island had then appeared for the first time in English-language literature. Its earliest known appearance is in Balchin and Pye‘s (1947) micrometeorological study of Bath, England, wherein reference is made to the ―characteristic heat island within a built-up area.‖  The term appears again in  Townshend‘s 1948 temperature-transect study of Aberdeen, Scotland, and in Parry‘s 1950 essay on town climates. The exact date or origin of the term is difficult to trace due  35  to the abundance of fugitive and foreign-language heat island literature predating World War II. More certain, however, is that first impressions of the city as an ―island‖ of heat originated in the popular measuring-trip studies in the early decades of the twentieth century. Unlike the observatory-based studies of the previous century, which relied entirely on pairs or small networks of instruments, measuring trips obtained a sufficiently dense sample of temperatures to discern the island-like morphology of a typical urban temperature field. In summary, the ―century of growth‖ following Howard‘s groundbreaking work has several defining characteristics. First, much of the heat island evidence of this period can be traced to a small number of Western European cities. Apart from a scatter of Japanese and North American studies, representation from other regions of the world was clearly lacking. Second, most heat island investigations used existing meteorological networks of city and country observatories—the measuring-trip approach was not conceived until the 1930s. Third, the quantity of published heat island literature was small, and its growth prior to mid-twentieth century slow and sporadic. Fourth, the literature was highly descriptive, offering no substantive or conclusive statements of heat island causation. The research paradigm that gave rise to this literature was one of observation and induction. Concern for the unique and individualistic nature of cities and regions was driving urban climate research, as were positivist influences from the broader scientific world. Empirical facts from an increasing number of cities were recorded, analysed, and classified, but these facts and the knowledge they created were not yet sufficient for scientific explanation of the heat island effect or for generalization of its behaviour.  36  Limitations in the methodological, geographical, and theoretical scope of the field resulted in slow but critical growth in the first century of heat island investigation. In the next half-century, these limitations gave way to a surge in heat island publication and understanding.  3.1.3 The modern period: 1950 to present The modern period of urban climatology began in the early 1950s with rapid and profound changes to heat island investigation, especially in methodology and conceptual understanding (Landsberg, 1981). These changes allowed heat islands to be examined and understood in greater detail than ever before, and for results to be published in unprecedented quantities. Modern heat island literature has been shaped by both intrinsic and extrinsic influences. Intrinsic influences include theoretical achievements in the field of urban climatology, and improved technology in computers and field instruments. Extrinsic influences include a gradual shift in the world‘s population toward an urban majority; rebuilding of war-torn cities in Europe and East Asia; heightened awareness of human influence on the environment; and expansion of metropolitan areas across the industrialized world (Oke, 1979; Landsberg, 1981; Yow, 2007). These external forces stimulated much interest in climatology and its potential to elucidate human influences on the atmosphere, and more specifically to inspire aesthetic and comfortable cities. Against a rising backdrop of ―publish or perish‖ attitudes in academia, these forces have generated a staggering number of publications in all branches of environmental science (Hunt, 1997; Valiela, 2001). The first heat island study of the modern period is one that is widely recognized as an urban climate classic. Sundborg‘s (1951) investigation of Uppsala, Sweden, is a  37  seminal study for its many ―firsts‖ in field methodology, quantitative analysis, and theory development (Oke, 1995). Highly impressive is that Sundborg rejected the use of climate observatories in the Uppsala area because their temperatures were not representative of the surrounding environment.  He instead designed a mobile survey route based  principally on objective (i.e., statistical) measures of site representativeness (Figure 3.1). Sundborg defined ―representativeness‖ as the extent to which air temperatures at each site deviated from those of the wider local area. This approach ensured that his temperatures were not unnecessarily biased by the effects of relief or forests. Sundborg was the first investigator to control his sites for unwanted non-urban influences, giving his study an experimental rather than purely observational basis.  Figure 3.1 Representativeness of observation points (black dots) along Sundborg‘s 1948–49 mobile survey route of Uppsala (hatched area). Representativeness varies inversely with dot size (Source: Sundborg, 1951).  38  Sundborg was also the first investigator to make statistical connections between heat island magnitude and prevailing weather conditions.  He developed a simple  regression model relating the city-country temperature differential to synoptic weather parameters (e.g., wind speed, cloud cover). With this model, Sundborg demonstrated the statistical connection of sky cover and advection with heat island genesis and control. He cautioned that these relations were not to be interpreted as cause-effect explanations, and that they could not expose the ―hidden physical realities‖ of the heat island effect. For that he developed a surface energy balance framework that included radiation exchange in street canyons, heat conduction at the earth‘s surface, and the release of ―artificial‖ combustion heat in cities. Sundborg consequently became the first urban climatologist to construct a theoretical framework that underpinned his empirical findings and that stimulated much thought into the causes of heat islands (Oke, 1995). This is a worthy achievement because Sundborg conceived this framework long before the earliest surface energy balance measurements were published. In the years following Sundborg‘s famous heat island study, several other benchmark studies were conducted. Duckworth and Sandberg‘s (1954) mobile heat island study of San Francisco, San Jose, and Palo Alto, California, was regarded as the best organized and most extensive measuring trip of its time (Kratzer, 1956). It exceeded all previous studies in area covered, number of sites sampled, and number of people and automobiles involved. The measured heat island magnitude of 11 K in San Francisco was the largest yet reported in the literature, and significantly larger than heat islands observed in European cities. The San Francisco study was also the first to explore relations between heat island magnitude and surrogates of urban structural density like  39  population and city area. The following year, Einarsson and Lowe (1955) produced a highly insightful study of the heat island in Winnipeg, Canada. Their study involved 16 measuring trips across the urban-rural area, which the investigators described as ―one of the flattest stretches of land in the world.‖ The effects of relief on temperature were therefore minimal, allowing Einarsson and Lowe greater control over non-urban factors such as local and microscale variability in the type, wetness, and cover of rural soils. Einarsson and Lowe discussed each of these surface conditions and their effects on air temperatures in and around Winnipeg. The novelty in their rural-based approach marks a profound shift in the perception of heat islands: for the first time, investigators had searched outside the city, in the soils and vegetation of the countryside, for answers to the fluctuating nature of heat island magnitudes.  During nights with strong radiational  cooling, Einarsson and Lowe observed intra-rural air temperature differences of 1–2 K among sites of wet, dry, clay, and peat soils, and among fallow, grassy, and wooded sites. They used a theoretical framework based on soil/air conductivity and heat capacity to explain the observed intra-rural temperature differences and their implications for the measured strength of Winnipeg‘s heat island. One of the earliest and most enquiring heat island studies conducted outside North America and Europe was Shitara‘s (1957) investigation of the winter heat island in Hiroshima, a large, coastal delta city of Japan. His automobile surveys in December 1955 gathered air temperatures at 103 sites in and around the city. Unlike Einarsson and Lowe‘s heat island study of Winnipeg, Shitara‘s investigation was challenged by complex interactions of river, ocean, and local topographic effects on temperature.  40  Shitara was especially interested in the thermal effects of building materials, noting that the urban centre, with its many concrete structures, was warmer than other districts with predominantly wooden structures.  Shitara recommended that future heat island  investigators quantify the thermal effects of housing materials more rigorously than he had done. Not long after Shitara‘s investigation, a second monographic study on the climate of London, England, appeared in the literature.  Using measuring trips and fixed  observatories at dozens of schools, colleges, and homes around London, Chandler (1965) depicted the urban temperature field as one that mirrors the distribution of buildings, from high- and low-density developments to open spaces (Figure 3.2). Chandler was intensely aware of the local and microscale climates that existed in the neighbourhoods, streets, and parks of London. He described the urban landscape as one that creates  Figure 3.2 Chandler‘s temperature traverse across London on October 11, 1961 (Source: Chandler, 1965).  41  intricate patterns of air temperatures ―pulsating‖ in time and space. He attributed these patterns to the ―kaleidoscope‖ of urban surfaces in London, to the ―honeycomb‖ geometry of its urban structure, and to changing synoptic weather conditions. Supported by many decades of urban observations in Europe and North America, and by extensive temperature records in his own city, Chandler concluded that the heat island effect and its microclimatic anomalies were now ―a matter of common experience.‖ To suggest that the heat island effect is ―a matter of common experience‖ evinces a belief in Chandler that the literature was sufficiently advanced for generalization and physically based reasoning. Descriptive heat island studies were rapidly accumulating from industrialised and midlatitude cities, and the time was opportune for constructing analytical models of the urban surface and its physical (especially thermal and aerodynamic) behaviour.  Chandler‘s plea in 1970 for greater emphasis on physical  theory and quantification of heat exchange mechanisms in cities ultimately demanded experimental rather than purely descriptive approaches to heat island studies. He insisted that properties of the surface (e.g., geometry, materials) and sky (cloud cover, wind speed) be parameterized, controlled, and isolated through more sophisticated observations to reach higher levels of inquiry.  New approaches would foster discovery of general  relations linking the heat island effect with physical properties of urban and rural surfaces, like roughness, wetness, reflectivity, and heating/cooling potential. Ironically, despite the views of many historical observers—including Howard, Renou, and Chandler himself— that heat islands were relatively simple phenomena to be anticipated in all large cities, physical understanding of urban influences on temperature remained largely superficial. Chandler‘s (1970b) prospective view on the field of urban climatology was soon  42  met with unprecedented growth in the heat island literature, not only in more structured and meaningful observations on which to base physical understanding, but through expansion into new cities and regions. Heat island studies in East Asia appeared in everincreasing numbers, beginning with several influential Japanese papers in the 1960s (e.g., Kayane, 1960; Kawamura, 1964), and growing to a substantial subliterature across the region three decades later. Major heat island investigations had then appeared in most metropolitan cities of East Asia, including Beijing, Shanghai, Taipei, Seoul, Tokyo, and Sapporo. In tropical regions, the first comprehensive heat island study was conducted in Singapore in the mid-1960s (Nieuwolt, 1966). Within twenty years, significantly more heat island studies had appeared in places like Malaysia, India, South Africa, and Mexico. The most inspiring paper to emerge from these regions was Jauregui‘s 1973 study of Mexico City. Although Jauregui‘s paper covers all aspects of Mexico City‘s urban climate, it devotes considerable attention to the heat island effect. Using temperatures from an extensive network of 30 climate observatories and a series of automobile measuring trips, Jauregui gathered new insight into the climatology of tropical heat islands, and in particular of wet and dry season heat islands (Figure 3.3). He drew special attention to the unique physical and cultural setting of Mexico City, noting that its basin topography and widespread factory and vehicle emissions were acute influences on the intense heat islands of the dry winter season. The 1970s and 80s are identified by many reviewers as an active period in tropical and subtropical heat island research (e.g., Jauregui, 1986; Oke, 1986; Roth, 2007). This activity is attributed to external pressures from urban population growth and  43  Figure 3.3 Distribution of climate observatories (left) and minimum temperatures (right, °C) in Mexico City, February 8, 1972 (Jauregui, 1973).  environmental degradation in the Global South. The expansion of tropical heat island literature during the 1970–80 period was part of a larger growth trend in urban climatology across the region, to Ecuador, Brazil, Kenya, Nigeria, Egypt, Israel, Hong Kong, and Taiwan. Studies in these and other tropical countries showed comparable results with the heat island investigations of midlatitude regions, such that UHI magnitude varied with urban morphology, surface thermal properties, and moisture availability. Generalizations to later emerge from this literature suggested that UHI magnitudes are lower in tropical and subtropical (excluding desert) cities than in midlatitude cities, primarily due to regional differences in urban geometry, urban albedo, and rural soil wetness (Oke et al., 1991). Despite the growth in heat island studies of the tropics and subtropics, these regions remained grossly underrepresented in urban climate literature until late in the  44  twentieth century. Jauregui (1986) estimated that prior to the 1980s, studies of tropical and subtropical cities comprised only 2 percent of the urban climate literature, with most of these studies concentrated in India, Mexico, Malaysia, and South Africa. By the turn of the century, the share of tropical and subtropical studies in urban climate literature had surpassed 15 percent, yet heat island papers comprised only a small fraction of that share (Roth, 2007). Advances in conceptual understanding of the heat island effect generated other important subliteratures within urban climatology.  The most significant of these  advances in the modern era has been Oke‘s (1976) classification of urban canopy and boundary layer heat islands. His classification reveals fundamental differences in scale, process, and measurement, and it marks a clear theoretical divide across empirical heat island literature. Major developments linking canopy layer UHI magnitude to thermal properties of buildings and soils, to urban and rural surface energy balances, and to canyon geometry were put forward by Myrup (1969), Goward (1981), Oke (1982), and Arnfield (1990), among others (see Chapter 2).  These groundbreaking studies  established the theoretical framework to improve understanding of the physical causes behind heat island formation—a framework that would encourage further testing through city-based observations. Many heat island studies were also conducted on the relation between UHI magnitude and synoptic weather conditions, in cities of South America (e.g., Hannell, 1976), Europe (e.g., Bohm and Gable, 1978), Australia (e.g., Nunez, 1979), Northeast Asia (e.g., Yamazoe and Ichinose, 1994), North America (e.g., Runnalls and Oke, 2000), and Southeast Asia (e.g., Chow and Roth, 2006). One relation in particular, that of UHI magnitude versus city population,  45  generated an especially large subliterature, initiated by Oke‘s (1973) temperature surveys through variously sized settlements in the St. Lawrence Lowland region of Montreal. Oke quantified a simple and intuitive relation between the size of a settlement, given by its population, and the magnitude of its measured heat island. In a subsequent article, Oke (1981) expounded this relation, stating that population itself is not a causal factor of heat island formation. Oke offered population only as a surrogate for city form, which he expressed through geometric measures like sky view factor and canyon aspect ratio. This development triggered a wave of studies worldwide that tested the universality of Oke‘s relation between urban geometry and urban heat island magnitude. As the descriptive heat island literature proliferated, especially in midlatitude regions, new studies were establishing stronger links among heat island causes, processes, and effects. A more analytical approach to heat island observation was taking shape, as investigators sought statistical correlations between heat island magnitude and physical parameters like rural lapse rate (e.g., Ludwig, 1970), urban and rural cooling rates (Oke and Maxwell, 1975), atmospheric stability (Bowling and Benson, 1978), building height and spacing (Eliasson, 1994), rural land use (Hawkins et al., 2004), vegetative cover (Jonsson, 2004), and sky view factor (e.g., Unger, 2004). Understanding of heat island causation advanced on these and other studies that had tested the earlier and more important discoveries of Sundborg (1951), Chandler (1965), and Oke (1981), all of which clarified the relations between heat island magnitude and urban form, surface cover, and weather conditions. By the turn of the twenty-first century, correlation studies had been repeated and published in such large volumes that any ―new‖ results were more confirmatory than exploratory.  46  Technological advances in computers and data acquisition systems is the last of many intrinsic influences on the dramatic rise in heat island publications during the modern period. Temperature observations were now made with a variety of reliable, accurate, and highly precise measuring devices. The introduction of digital thermometers, dataloggers, portable computers, and automated weather stations greatly enhanced data acquisition, storage, and analysis (Grimmond, 2006). The compactness of this equipment facilitated easy installation on cars and other forms of transportation for conducting mobile surveys. Stationary surveys also benefited, as street posts, building facades, and previously inaccessible or insecure fixtures in the city were now able to support compact measurement and recording systems. The spatial and temporal resolution of heat island investigations increased considerably, yielding unprecedented detail in diurnal, seasonal, and annual heat island patterns. Automated dataloggers gave investigators freedom to design their studies with greater purpose, and to configure appropriate sampling networks for answering increasingly sharp and provocative questions. Dataloggers also reduced the need for time-temperature data correction because the measurements could easily be taken synchronously. In contrast, heat island studies of the previous century had access only to small networks of weather observatories with standard maximum/minimum thermometers, or to short measuring trips into the countryside. Their datasets were often much smaller, fitting neatly into simple tables of two-point temperature differences between city and country. Modern investigators now have the means to configure dense networks of automated sensors across a city, and to design mobile survey routes with small time and space intervals between sample points. In summary, the modern period of heat island investigation can be characterised  47  as an era of growth and maturation from its traditional roots in midlatitude Europe. The field expanded rapidly not only in its geographic reach but in its methods, instruments, and conceptual sightlines. By the 1980s, urban climatology had found its niche in environmental science, with urban heat island literature holding the identity and status of that niche. Observations in tropical, subtropical, arid, polar, and subpolar regions were now part of a worldwide stock, one study after another testing the heat island controls and characteristics discovered decades earlier in midlatitude cities.  The breadth of  observations accumulated during the modern period was sufficient to differentiate regional heat island characteristics and to support a continuing trend toward more predictive and process-oriented urban temperature studies.  3.1.4 Synopsis After more than two hundred years of heat island observations worldwide, Luke Howard‘s wish for a ―greater store of facts‖ has certainly been fulfilled. Heat island studies similar to Howard‘s in London have appeared in hundreds of cities and towns around the world. Achievements in theory and technology have carried the field from its modest beginnings among casual observers, to its current position as a research domain of much scholarly attention.  A remarkably diverse—yet still relatively small—  community of scientists has come to this domain, drawing expertise from geography, meteorology, architecture, town planning, landscape ecology, urban engineering, physiology, and public health. Equally diverse is the range of scales, resources, and measurement systems that these experts use to observe and report heat islands. At one extreme, low-budget studies have responded to simple research questions with relatively crude instruments and shortened time scales of days or weeks. At the other extreme,  48  large, cooperative projects of international scope have responded to multifaceted questions with expensive and elaborate measurement systems and lengthened time scales of many years. The contributions of these studies should be deduced not from the quantity of their data or the novelty of their techniques, but from successful advances of knowledge and critical tests of theory. Heat island research in the twenty-first century has reached a scientific milestone in surpassing the critical stages of recognition, observation, and quantification (Lowry and Lowry II, 2001; Arnfield, 2003; Mills, 2009). However, despite the impressive achievements of this research, its descriptive outlook on the many controls of heat islands has generated an excessive number of individual case studies in the published literature. Many of these studies are inconsequential to the current status of the field. Given the broad base of heat island observations and generalizations now in place, additional evidence of spatial and temporal heat island behaviour serves local interests well ahead of regional or global concerns.  Efforts to conduct and publish repeated heat island  observations in towns and cities worldwide (not including less developed regions) are consequently much less profitable to urban climatology than they were a half-century ago. Efforts should instead fulfill a more philosophical purpose by reflecting on the existing literature and the extent to which it provides ―reliable facts‖ for scientific progress.  3.2 Methodological and philosophical problems in the modern period 3.2.1 Estimates of urban heat island magnitude The following sections examine three concerns of mine regarding the methodological rigor with which estimates of canopy layer UHI magnitude are measured and reported in modern heat island literature. The first concern is that heat island  49  investigators often overlook, or deliberately withhold, pertinent information that is needed to understand and trust the reported findings, and to make valid generalizations and inter-city comparisons. The second concern is that heat island studies are often flawed or poorly designed in their approach to estimating UHI magnitude. Use of nonrepresentative sites and poor control of non-urban effects on heat island magnitude are common problems in the literature. The third concern is that heat island investigators inconsistently adhere to appropriate time and space scales, measurement systems, operational definitions, and reporting protocols.  I examine each of these concerns  separately, citing evidence in the literature to support my arguments and to underscore the need for this thesis. Before proceeding to the evidence, I first give standard dictionary meanings of ―urban‖ and ―rural‖ so as not to confuse the reader with my understanding of these terms or my interpretation and critique of the literature. Dictionary meanings are ideal for this purpose because most researchers of urban and rural matters can generally agree on their nature, and thus common ground is easily established. Oxford English Dictionary defines urban as ―constituting, forming, or including a city, town…or part of such,‖ with town being a ―densely populated area…opposed to the country or suburbs,‖ and characterised physically as a ―cluster of dwellings or buildings.‖ Suburbs are defined as ―residential parts belonging to a town or city that lie immediately outside and adjacent to its wall or boundaries.‖ Rural, in contrast to urban and suburban, is defined as ―agricultural or pastoral…characteristic of the country or country life,‖ with country being ―the parts of a region distant from cities.‖ From these definitions, I interpret ―rural‖ landscapes to be less populated than  50  ―urban‖ landscapes, and to have fewer built structures and more abundant natural space, particularly for agricultural production. I interpret urban landscapes to have significantly more built structures and larger populations for predominantly non-agrarian activities. ―Suburban‖ landscapes are transitional and are understood to have a mixture of built structures and natural/agricultural spaces with population densities lower than cities but higher than the country. While these traditional definitions may now be superseded by modern or more regionalized interpretations of the landscape—which allow rural areas to be more diverse and densely populated than urban areas, depending on political and historical circumstances—they are adequate for developing the research problem and a method of approach to that problem.  3.2.1.1 Incomplete reporting When reporting the results of an observational heat island study, investigators must fully account for the methods and assumptions behind their UHI estimates. Giridharan et al.‘s (2007) study of the Hong Kong heat island is one such investigation with an incomplete report. Conducted with good intent, the investigators measured canopy layer heat island magnitude in a dense Hong Kong housing estate. The motive behind their study is to promote climate-sensitive city planning in Hong Kong. Screenheight temperatures were gathered from stationary sensors at 17 ―urban‖ sites and a single ―rural‖ reference site. Urban sites were interspersed among compact residential towers. The report provides extensive metadata—including photographs, instrument specifications, and measures of surface cover and geometry—for each site.  The  investigators stratified all heat island estimates by weather type so as to control for the effects of cloud on UHI magnitude. After five months of observation, the investigators  51  concluded that UHI magnitudes in high-density highrise districts of Hong Kong average 2–3 K. The study recommends that this result be incorporated into design plans for improving human comfort and energy efficiency in Hong Kong. By most criteria of scientific reporting, the Giridharan et al. study is well documented. The instruments and observations are openly described and the findings are succinctly reported. However, first impressions of heat island studies can be misleading. Closer examination of the Hong Kong study reveals two areas of weakness. First, it lacks the most fundamental aspect of experimental field science: an operational definition of the study phenomenon. Despite its stated aim of quantifying UHI magnitude in a Hong Kong housing estate, the study fails to explain how ―magnitude‖ is defined or measured in the field. Neither the temperature variables used to measure UHI magnitude nor the times of those measurements are known to the reader. Second, the study lacks critical information regarding the location and setting of the ―rural‖ reference site. The exposure and physical nature of the site and its surroundings are therefore completely unknown to the reader, yet the reported UHI magnitudes are said to represent ―urban-rural‖ temperature differences. Problems of incomplete reporting, particularly of the kind revealed in the Giridharan et al. study, are common in observational heat island literature. Many studies fail to translate ―UHI magnitude‖ into operational procedures.  In some cases, the  locations of the urban and rural field sites defining UHI magnitude are undisclosed, or poorly documented, as in Garnett and Bach (1967), Eagleman (1974), Padmanabhamurty and Bahl (1982), and Pinho and Orgaz (2000).  In other studies, the measurement  variables used to define and quantify UHI magnitude are withheld, as in Woollum and  52  Canfield (1968), Lee (1975), and Capelli de Steffens et al. (2001). In each case, the reader has no choice but to speculate on the nature of the temperature data—whether it be hourly, daily, monthly, or seasonal—behind the reported UHI magnitudes. Incomplete reporting affects other areas of methodology as well. Overlooked or downplayed in many heat island studies are the weather conditions in which heat islands are observed.  Cloud, wind, and other meteorological data are crucial to one‘s  interpretation of a heat island report. Sani (1987), Tsuji and Fukuoka (2004), and Wong and Yu (2005) each failed to provide these data in their reports.  Failure to give  instrument specifications in a heat island report is equally crucial to a reader‘s interpretation of its finding. Instrument type or precision, for example, is not discussed in heat island studies by Townshend (1948), Sanderson et al. (1973), Lyall (1977), and Colacino (1980). In studies by Lazar and Podesser (1999) and Hogan and Ferrick (1998), the number and frequency of heat island observations is not reported. In studies by Sekiguti (1963), Suckling (1981), and Chiang et al. (1995), the structural characteristics of the field sites used to define UHI magnitude are not sufficiently discussed, while Eagleman (1974), Park (1986), and Robaa (2003) overlook the effects of relief on the reported urban-rural temperature differences. The extent to which these studies are representative of a wider trend in the literature is not fully known, but one might expect the list of published heat island papers lacking critical information to be a long one.  3.2.1.2 Flawed design The second methodological concern surrounding heat island observations in published literature relates to flawed design. I cite the heat island study of Syrakova and Zaharieva (1998) to demonstrate the nature of this problem.  53  These investigators  collected 10 years of hourly temperature data from a pair of stationary thermographs at ―typical urban and rural stations‖ in Sofia, Bulgaria. The recorded temperatures were stratified by wind and cloud conditions so as to remove unwanted weather influences from computed UHI magnitudes.  The investigators operationalized UHI magnitude  through hourly and mean monthly temperature differences between the urban and rural field sites. Their results indicate that mean annual UHI magnitudes in Sofia average 1.5 K, while daily maximum magnitudes exceed 4 K in clear, calm weather. The Syrakova and Zaharieva study appears to be well conducted, but in probing its methods and results more carefully, flaws emerge. The choice of a non-representative urban field site to define UHI magnitude is the first such flaw. Whereas the rural site was appropriately situated in a cultivated field far from city climate influences, the urban site was situated in a botanical garden described as ―a small park with tall trees…in the central part of the city.‖ The city site is not representative of the local urban area, which the investigators portray as a ―square [with] pavement and streets…as well as densely [packed] tall buildings.‖  The investigators have confused the microscale climatic  influences of a vegetated garden with the local-scale influences of the surrounding streets and buildings. The ―urban‖ surroundings that they intended to represent are therefore not well served by the essentially rural characteristics of a botanical garden. The use of non-representative ―urban‖ and ―rural‖ field sites to quantify UHI magnitude is widespread in the literature. Placing thermal sensors in urban parks, for example, with the intent of characterising the wider built-up area is a flawed approach to UHI estimation. Heat island studies by Heino (1978), Leduc et al. (1980), Figuerola and Mazzeo (1998), and Gedzelman et al. (2003) each use non-representative urban or rural  54  field sites to compute UHI magnitude. Equally flawed is the use of field sites at airports to represent urban and/or rural settings. The microscale surroundings at most airport weather stations are open, grassy, and flat, as per WMO guidelines. The local-scale surroundings, however, are often characterised by buildings, airstrips, roadways, and parking lots. Empirical evidence shows that airport temperatures are potentially nonrepresentative of either urban or rural settings (Runnalls and Oke, 2006), yet these settings are consistently used in heat island investigations to represent the city (e.g., Adebayo, 1991), the countryside (e.g., Hage, 1972), and both city and countryside in the same investigation (e.g., Nakamura, 1966). The effects of weather and relief on observed UHI magnitudes is another challenge faced by all heat island investigators. If non-urban effects are not in some way controlled, they will obscure the desired urban effects in reported UHI magnitudes. Woollum (1964), Nkemdirim and Truch (1978), Unwin (1980), and Kim and Baik (2005) report ―urban‖ heat islands that are not sufficiently urban-induced because the influences of weather and topography have not been regulated. Estimates of UHI magnitude based on undersized datasets, such as one or two evenings of observation, are also misleading because they are not controlled for the effects of time. Heat island studies by Louw and Meyer (1965), Deosthali (2000), and Saaroni et al. (2000) each lack the statistical power to generate valid inferences of a heat island effect because sample size is too small.  3.2.1.3 Inconsistent procedures The third concern with published estimates of heat island magnitude relates to inconsistent procedures. ―UHI magnitude‖ is so loosely defined in the literature that a variety of approaches (or operational definitions) can be used to quantify the same  55  phenomenon. If each of these approaches were tested at the same time in a single city, estimates of UHI magnitude would differ dramatically (e.g., Hedquist and Brazel, 2006). One can only imagine the scatter of results ensuing from a test of all approaches across all urban and rural settings reported in the heat island literature. This prospect alone highlights the problem of inconsistent methodology and inter-city comparison of UHI magnitudes. Demonstrating these methodological inconsistencies are two published heat island studies that take different approaches to identical questions. The first study examines the heat island of Taipei, Taiwan, and was published by Lin et al. (1999) in the Journal of Architecture (China). The study characterises the nocturnal canopy-layer heat island in the basin city of Taipei, based on two spatial surveys in the summer of 1998. Temperatures were recorded with thermistor resistors mounted to a fleet of seven motorscooters. The precision of the thermistors is reported to be +/- 0.5°C. The scooters traversed through the streets of Taipei, starting in the city centre and progressing outward for 2 hours toward the countryside. Air temperatures were sampled at approximately 500-m intervals across the 350-km2 study area, yielding 640 data points.  The  investigators defined UHI magnitude as the synchronous temperature difference between the single warmest and coolest points among the 640 measurements.  The highest  temperatures were observed near a shopping mall in the city centre, and the lowest in the heavily vegetated hills southeast of the city. The average UHI magnitude of the two spatial surveys, conducted under partly covered skies and light winds, is reported to be 4.5 K. With an identical research motive, but an entirely different approach, Robaa  56  (2003) quantified Cairo‘s nocturnal heat island and reported his findings in Atmosfera. He employed a simple approach to quantify UHI magnitude with only three field stations in the relatively flat study area.  Each station was serviced by the Egyptian  Meteorological Authority and consisted of a standard weather screen with dry- and wetbulb mercury thermometers at 2 m agl. The ―rural‖ station was located among cultivated fields of irrigated grass 13 km northwest of central Cairo. The ―urban‖ station was situated on the banks of the Nile River near central Cairo, among factories, compact buildings, and heavy traffic. A third station, called ―suburban,‖ was located at the Cairo international airport 29 km northeast of the city centre. That station was surrounded by desert on all sides except southwest. Temperature observations at each station were made twice daily at 0600 and 2200 hrs for a period of 5 years (1995–2000). Robaa defined UHI magnitude as a synchronous temperature difference between urban-rural and urbansuburban station pairs, with monthly averages of 3–5 K and 1–2 K, respectively. Methodological differences of the kind demonstrated in the Taipei and Cairo studies are common in published heat island literature.  Instrumentation (precision,  mounting), time and space scales (e.g., sampling density, period of observation, study area), definitions of terms (urban, rural, UHI magnitude), and control of extraneous influences (weather, relief, time) are so inconsistent, if not incommensurable, that the reported magnitudes can rarely be compared or generalized in a meaningful way. With each new published heat island study, these methodological differences become more apparent. Given the individuality of method and setting in urban climatology, and the lack of standardized procedures, placing confidence in reported UHI magnitudes is best done case by case.  57  3.2.2 Definitions of urban heat island magnitude Review of observational heat island literature exposes an alarming diversity of urban and rural field sites defining heat island magnitude around the world. The common practice of classifying field sites as indiscriminately ―urban‖ or ―rural‖ obscures the array of surface and near-surface climates that actually determine UHI magnitude.  The  heuristic value of the urban-rural dichotomy as a basis for describing and classifying field sites in heat island investigations therefore needs to be brought into question. With recourse to constructs of the scientific method and urbanization theory, the arguments that I posit (Table 3.1) are meant to (1) elicit experimental flaws generated by Tu-r , (2) challenge Tu-r as a trusted definition of UHI magnitude, and (3) underpin a redefined expression of UHI magnitude in more meaningful, defensible, and operable terms. In the following sections, I expand on each of these arguments.  Table 3.1 Arguments against Tu-r as a trusted definition of heat island magnitude. Scientific method  Urbanization theory  Tu-r is neither unequivocal nor objective as a testing operation of heat island magnitude. It is susceptible to variable interpretation by individuals, groups, and cultures.  The urban-rural dichotomy is an outdated and overly simplistic construct for periurban landscape classification, particularly in less developed regions of the world.  Tu-r leads to stipulative definitions of its antecedent terms—urban and rural—and by this account gives investigators unwarranted freedom to manipulate field-site configurations.  The urban-rural dichotomy is deeply embedded in traditional city-based urbanism and does not easily transfer to places of region-based urbanism.  The classification of field sites as vaguely ―urban‖ and ―rural‖ allows investigators to design heat island experiments that protect their central thesis, Tu > Tr .  Urban and rural are diluted concepts due to evolving and inconsistent jurisdictional definitions and usage.  Tu-r gives way to circular definitions of its constituent terms on the faulty assumption that they are antecedently and universally understood.  Tu-r is irrelevant in cities of polynucleated, dispersed, and decentralized forms.  58  3.2.2.1 Scientific method and Tu-r The first and most damaging attack on the use of Tu-r in heat island studies is that the selection of urban and rural field sites violates one of science‘s most fundamental tenets: objectivity. The selection of urban and rural sites is bound by human perception and prejudices that are time and space dependent and that vary with individuals, groups, and cultures. As a testing operation (or ―field definition‖), Tu-r fails to give unequivocal and objective estimates of UHI magnitude because its constituent terms—urban and rural—evoke different images for different people. In Philosophy of Natural Science, Hempel (1966) stated that regardless of how intuitively clear or familiar a term may be, it must invoke an operational procedure that can be ―unequivocally carried out by any competent observer,‖ and the result of which can be ―objectively ascertained.‖ It is doubtful that Tu-r meets these criteria as the operational procedure of UHI magnitude. Due to historical, geographical, and cultural factors, the selection of urban and rural field sites to represent Tu-r has allowed a confusing portrayal of landscapes in heat island literature. All landscapes and field sites described or expressly classified in the literature as ―urban‖ or ―rural,‖ or sites otherwise representing these environments in the computation of UHI magnitude, are listed verbatim in Table 3.2. Specific examples from this list are highlighted with photographs in Table 3.3. Each of the studies in Table 3.3 seeks an empirical estimate of canopy layer UHI magnitude based on urban-rural temperature differences from stationary or mobile surveys. The problem in this portrayal is not the diversity of sites defining Tu-r, but the representation of these sites by a vague and subjective dichotomy. Conventional heat island methodology defines these sites as universally urban or rural despite their obvious seasonal and spatial variations in surface  59  Table 3.2 Field sites representing ―urban‖ and ―rural‖ landscapes in observational heat island literature, 1950 to 2007. ―Urban‖  ―Rural‖  ―Urban‖and ―rural‖  botanical gardens pollution monitoring sites golf courses fire halls hospitals railyards shipyards industrial storage yards shopping centres water treatment plants football stadiums street markets waterworks reservoirs highrise housing estates high-density housing sites low-density housing sites church grounds temple grounds arboretums parking lots amusement parks health laboratories conference centres police stations residential yards/gardens street canyons wharfs/piers/bridges parks/greenbelts/green spaces traffic intersections cemeteries cathedral squares old commercial quarters town squares open squares/courtyards/plazas central business districts commercial highrise districts sports fields/sports grounds old city cores new city cores housing blocks harbour fronts terrace housing sites factory sites industrial ports skyscraper districts hamlets coastal villages heavy industrial sites highrise hotel districts lowrise housing projects mixed residential/agricultural areas government administrative regions  astronomical observatories agricultural academies agricultural field stations agricultural villages fruit farms paddy fields tree nurseries ecological preserves archaeological ruins rubber plantations dairy farms wildlife refuges experimental farms ranchlands cocoa plantations dryland grain farms tropical rainforests deserts tundra pine flatwoods moorland open fields rural towns wetlands bare fields fallow fields greenfields snow-covered fields rice fields orchards mixed farmland open farmland cultivated plains irrigated farmland irrigated vegetable farms alluvial plains marshland swampland lagoons woodlands rough grass fields open grassy areas open hinterland grassland meadowland muskeg pine forests proving grounds military aerodromes hydroelectric dams bushlands  weather observatories meteorological institutes university campuses college campuses airports airstrips air force bases elementary-school grounds high-school grounds middle-school yards city outskirts coastal settlements hamlets villages towns suburban housing sites satellite cities  60  Table 3.3 Illustrated examples of ―urban‖ and ―rural‖ field sites in modern observational heat island literature. ―Urban‖ sites  ―Rural‖ sites  Street canyon  Farmyard Goteborg, Sweden  Vienna, Austria  (Eliasson, 1994)  (Bohm, 1998)  Housing estate  Floodplain Hong Kong  Tainan, Taiwan  (Giridharan et al., 2005)  (Lin et al., 1999)  Desert airstrip  Airport grounds Phoenix, USA  Lodz, Poland  (Hedquist & Brazel, 2006)  (Klysik & Fortuniak, 1999)  Building rooftop  Apple orchard Seoul, Korea  Nagano, Japan  (Kim and Baik, 2005)  (Sakakibara & Matsui, 2005)  Public park  Bushland Vienna, Austria  Akure, Nigeria  (Bohm, 1998)  (Okpara, 2002)  Highrise district  Snow fields Vancouver, Canada  Obuse, Japan  (Runnalls and Oke, 2000)  (Sakakibara, 1999)  61  geometry, cover, and wetness. The urban-rural paradigm that governs site classification gives heat island investigators little room for individuality. Compounding the first argument in Table 3.1 that Tu-r is highly subjective, and therefore unscientific, is a second argument that it leads to stipulative definitions of its constituent terms.  Stipulative definitions are advantageous in science for attaching  special meaning to concepts (such as UHI magnitude) that are used in experimentation or theoretical argument. The disadvantage is that they cannot be qualified as true or false, unlike standardized dictionary meanings that adhere to customary usage (Black, 1952). The testing of freely proposed concepts puts sampling design at risk of bias insofar as it changes with the peculiarities of place, region, and people. In this case, the definiendum, or UHI magnitude, has been defined in the literature through variable interpretations of its definiens, Tu-r . The different field sites shown in tables 3.2 and 3.3 each characterise the same antecedent terms of the definiens, and together evince a disorderly approach to scientific testing. Tu-r is sufficiently vague and inconsistent in meaning that ―clean results, sharp inferences, and rapid retesting‖—each necessary for accelerating scientific progress—are difficult to achieve (Platt, 1964). The third argument against Tu-r is that classifying field sites as vaguely urban or rural enables investigators to deliberately design their investigations to meet desired outcomes. Observations designed to protect the central thesis that Tu > Tr are borne by the broad and inclusive nature of the terms urban and rural.  These terms seduce  investigators to select and classify field sites on the basis of known temperature differences at the sites prior to empirical testing. Airports and other locations having both city and country characteristics can be classified loosely as urban or rural—  62  regardless of their local and microscale surroundings—to guard the Tu > Tr expectation. Selecting field sites that are assured of confirming an (ostensible) urban heat island effect is an objectionable approach to scientific testing. Finally, Tu-r promotes fallacies of definition. Investigators who qualify their urban and rural sites with a description of their surroundings often succumb to circular, or tautological, definition. Throughout the literature are clumsy definitions using redundant words to describe urban and rural sites. Moreno-Garcia (1994), for example, defined her rural site as ―typical of the urban limits,‖ and gave no further indication of what those ―urban limits‖ look like. Other investigators are equally remiss. Louw and Meyer (1965) described their rural sites only as belonging to the ―open country‖ and to the ―peri-urban area.‖ Morris et al. (2001) gave a repetitious account of their rural site locations as ―open environments with vegetated surfaces similar to those found in natural landscapes.‖ None of these definitions has merit: they are fallacious because they assume prior understanding of what characterises ―urban,‖ ―rural,‖ or ―natural‖ in a particular setting. The investigators give no useful information about the sites or their surface cover and exposure characteristics that influence thermal climate. This problem arises in part because urban climatology has no stated or understood definitions of urban and rural for heat island investigators. Urban geography, on the contrary, distinguishes urban from rural through standard referents like population (size, density, heterogeneity), territorial limits, division of labour, and various forms of material culture (Dewey, 1960; Gugler, 1997). While urban-rural thresholds for these referents may not be defined or accepted universally, they are at least quantifiable and unequivocal.  63  3.2.2.2 Urbanization theory and Tu-r I now posit arguments from urbanization theory that further weaken Tu-r as a reliable measure of UHI magnitude (see Table 3.1). The most persuasive of these arguments is that in view of today‘s rapidly evolving urban systems, the overly simplistic urban-rural dichotomy is an outmoded construct for landscape classification, particularly in developing and densely populated regions of the world. Decades ago, urban theorists abandoned the urban-rural dichotomy as a policy paradigm in the developing world, arguing that the space economy in peri-urban regions could no longer be distinguished by a clear city-country divide (McGee and Robinson, 1995; Gugler, 1996). On the urban periphery of developing cities, in situ population densities are extremely high, traditional and non-traditional land uses co-exist, and people, capital, commodities, and information flow continuously between city and country. Urban theorists therefore contend that the spatial demarcation between urban and rural is artificial, and that the relation between these two constructs is more accurately described as a continuum, or a dynamic, rather than a dichotomy. By rejecting these peripheral spaces as universally urban or rural, urban geographers instead use descriptors like ―development corridors,‖ ―growth triangles,‖ and ―extended metropolitan regions‖ (Chu-Sheng Lin, 1994). Urban climatologists have long encountered, but not yet resolved, the problem of classifying complex landscapes with a simplistic dichotomy.  Yamashita (1990)  confronted this problem in his study of the Tokyo heat island. Admitting to the difficulty of locating ―rural‖ stations on the Kanto Plain, Yamashita paired his ―urban‖ site in central Tokyo with a meteorological station 60 km to the north, in the local ―city‖ of Kumagaya (Figure 3.4).  Despite the physical distance between these two sites,  64  Yamashita‘s rural station was located well within the mixed urban-rural surroundings of the Kanto Plain: Heat island intensity [in Tokyo] is very difficult to define because of the uncertainty in defining rural areas. In general, the rural-urban boundaries are often not distinct at all. The Tokyo Metropolitan Area is particularly obscure in its boundaries because Tokyo is continually expanding [due to] the sprawl phenomenon, and it is no exaggeration to say that the whole area of the Kanto Plain, where Tokyo is located in its southern part, is more or less urbanized. (p. 100)  N  100 m  500 m  Figure 3.4 Kumagaya ―rural‖ site used by Yamashita (1990) to quantify heat island magnitude in Tokyo. Urban influences are evident at micro (top) and local (bottom) scales. Red circle indicates location of site. Kumagaya map (bottom-right) shows settlement density on the Kanto Plain.  Yamashita‘s use of a local ―city‖ to represent ―rural‖ in Tu-r is further confused by an earlier Japanese study published by Kawamura in 1964, titled ―Analysis of the Temperature Distribution in Kumagaya: A Typical Example of the Urban Climate of a Small City.‖ In that study, the Kumagaya field site is classified as ―urban,‖ while sites 65  further out on the Kanto Plain are classified as ―rural.‖ The overlap between Kawamura and Yamashita‘s urban and rural areas is instructive for two reasons. First, it originates from different time periods and from different levels of the urban hierarchy—it thus exposes the subjectivity of human perception, and the cultural, geographical, and historical nuances of landscape classification. Second, it demonstrates that the urban hierarchy of the Kanto Plain cannot possibly be reduced to such a simplistic and dichotomous interpretation of the landscape. Urban and rural therefore do not transfer easily into the Tokyo vernacular (S. Yamashita, pers. comm., 2007). Like Japan‘s Kanto Plain, rice-bowl regions of Southeast Asia are characterised by dense settlement patterns that are distinctive in theory but blurry in reality. Ginsburg et al. (1991) described these regional patterns as desakotas, an Indonesian term for village-city. Desakotas accommodate intensive mixtures of small-holder agricultural (usually wet rice) and non-agricultural activity, and extend along transportation corridors for tens of kilometres into a densely populated countryside. Desakota regions now surround nearly every major city in the Asia Pacific. Although peculiar to wet-rice regions of Asia, desakotas exist more generally as ―extended metropolitan regions‖ in other parts of the world where peri-urban environments coincide with areas of welldeveloped infrastructure and concentrated mixtures of urban and rural land use. In the desakota regions of Singapore, Wong and Yu (2005) attempted to define UHI magnitude in a way that preserves its universal interpretation as an urban-rural temperature difference. They alluded to the challenge of classifying field sites beyond the built-up city centre as categorically ―rural.‖ The opening of their study acknowledges  66  that the settlement patterns surrounding central Singapore do not classify easily as urban or rural: Singapore is a garden city without a distinct borderline between urban and rural areas. [Only] the two existing predominant green areas, the primary forest of 75 ha in the middle of the island and the recreation area in the northeast…can be considered as ―rural‖ [for measuring heat island magnitude]. (p. 549)  The term rural is used sparingly throughout the remainder of the Wong and Yu study, which eventually concludes with an appropriately nuanced statement that the maximum temperature difference in Singapore was observed between the urban core and ―wellplanted areas.‖  Kim and Baik (1995) also struggled with the term rural in their  investigation of the Seoul heat island. Their baseline field sites were classified not as rural but as ―suburban,‖ only because there is no undeveloped countryside adjacent to the city (Y. Kim, pers. comm., 2007). ―Suburban‖ classification offers no improvement over either urban or rural—it merely highlights the awkwardness with which heat island investigators in Seoul and other cities of the world use Tu-r to define heat island magnitude. These characterisations of Tokyo, Singapore, and Seoul lead to a second theoretical argument against Tu-r.  Chu-Sheng Lin (1994) and Gilbert (1996)  characterised urbanization in core and peripheral regions of the world through ―regionbased‖ and ―city-based‖ urbanisms. In the Asia Pacific, region-based urbanism describes a peri-urban landscape that is fundamentally integrated with its densely populated peasant periphery, such that the urban-rural divide is virtually indistinguishable.  This  phenomenon is attributed to transportation technology, economic decentralization, rural industrialization, and massive rural-urban migration. The ―rural‖ surroundings of Asian cities differ profoundly from those of Europe, whose compact and localized towns and  67  cities are seemingly detached from a sparsely settled countryside. The divide between city and country is comparatively clear and abrupt. Urbanism of this kind—described as city-based—lends itself to the simplistic testing of Tu-r . City- and region-based urbanisms have influenced the representation of urban and rural field sites in heat island literature throughout its history. Opposing views of social and economic life have also influenced the way in which reported UHI magnitudes are interpreted and exchanged in the literature. The Tu-r test was born in localized European cities like London, Paris, Berlin, and Vienna, where the distinction between urban and rural was seen not only in culture and language, but in the physical plan of the towns and cities. In Howard‘s The Climate of London, for example, ―city‖ and ―country‖ are diametrically opposed throughout the entire study. Differences in region- and city-based urbanisms are also evident in modern heat island studies of East Asian and European cities (Figure 3.5). In Hong Kong and Tokyo, urban activity extends well into the ―rural‖ surroundings, but in smaller and more localized European centres like Wroclaw (Poland) and Szeged (Hungary), it is well contained within the city limits. Finding rural sites to define UHI magnitude in cities of region-based urbanism is therefore difficult because the undeveloped countryside lies many tens of kilometers from the city centre. In these cases, the physical distance between city and country is sufficiently large to undermine the validity of UHI magnitudes defined by standard urban-rural temperature comparisons. The third argument against Tu-r is that the terms urban and rural are diluted concepts due to evolving and inconsistent jurisdictional definitions and usage. Gugler (1996) highlighted China, in particular, for its evolving patterns of urbanization and for  68  5 00 m  Figure 3.5 Eye level and aerial views of ―rural‖ sites defining heat island magnitude in regionand city-based urbanisms. From top: Hong Kong (Giridharan et al., 2007); Tokyo (Sakakibara and Owa, 2005); Szeged, Hungary (Unger, 1996); and Wroclaw, Poland (Szymanowski, 2005). Red circles indicate site locations.  69  inconsistent criteria defining urban and rural territories. Chinese cities typically overspill their administrative boundaries into surrounding villages and counties, and thus peripheral settlements defined as ―rural‖ by jurisdiction might appear more urban in form. In India, the criterion for official ―urban‖ designation varies widely among states, but is generally based on population density and labour division (Ginsburg et al., 1991). Settlements that are urban-like in physical structure may have no such status because in function they fail to meet the necessary criteria. McGee (1995) noted that Vietnam is especially sophisticated in its urban hierarchy, having five official designations of ―urban‖ ranging in populations from 4,000 to 1 million. In the Vietnamese countryside, population densities can exceed those of North American cities. These definitions of urban and rural in China, India, Vietnam, and elsewhere lead to unpredictable relations between landscape form and function. Regional inconsistencies in urban and rural definition undoubtedly confuse one‘s perception of representative field sites for quantifying Tu-r . Establishing scientific standards for landscape definition only by means of a simple, dichotomous construct is complicated by the inherent political and social character of cities and their countrysides. This complication is at the forefront of Hogan and Ferrick‘s (1998) article, ―Observations in Nonurban Heat Islands.‖ These investigators attempted to quantify UHI magnitude in the hamlet of Lyme, New Hampshire, using ―nonurban‖ field sites in the hamlet itself and ―rural‖ sites in ―open, quasi-flat fields‖ a few kilometres away.  Lyme might be  considered ―nonurban‖ in regional or economic function, but it is arguably urban in physical form. Hogan and Ferrick describe Lyme as a cluster of houses, businesses, roads, and other structures. By this description, the field sites in Lyme are likely no  70  different, at local scales, than those of low-density towns or cities. Although ―rural‖ and ―nonurban‖ accurately describe the regional setting of the Lyme investigation, they poorly describe its local sites of observation. Here the investigators have confused not only the distinction between urban and rural form and function, but also the scales of influence in their so-called nonurban study area. Lastly, substantive literature points to increasingly complex metropolitan forms in both the developed and developing worlds. Polynucleated, decentralized, and dispersed cities have become definitive features of global urbanization (Lo and Yeung, 1998). These complex urban forms present many challenges to heat island investigators. Sani (1973), for example, described his study area of Kuala Lumpur as ―a sprawling commercial area consisting of a multiplicity of landuses with no definite city centre in the classical sense.‖ Kumar et al. (2001) described their study area of Mumbai as an ―urban octopus whose tentacles encompass towns outside the municipal limits.‖ These and many other investigators were justifiably cautious in their use of ―urban‖ and ―rural‖ to quantify and characterise the heat island. Urban and rural, and thus Tu-r , are hardly relevant to landscape classification and heat island assessments in cities like Kuala Lumpur, Mumbai, Tokyo, Seoul, and Singapore. Although regional models of settlement transition, such as the desakotas of Southeast Asia, are of limited use at scales of interest to urban climatologists, a global understanding of settlement patterns and processes is essential for a universal solution to a problematic urban-rural dichotomy.  71  3.2.3 Inter-city comparisons of urban heat island magnitude Due to the methodological and philosophical weaknesses of Tu-r, and to the uncertainty that these weaknesses create in published UHI estimates, few tenable intercity comparisons of heat island magnitude can be made in urban climate literature. Despite these weaknesses, comparison and generalization is a standard feature of nearly every published heat island report. Investigators are quick to cite inter-city differences in UHI magnitude without considering the conceptual and operational design of the studies being compared. Hartley (1977), for example, concluded his study of the Glasgow heat island with the broad generalization that measured magnitudes in that city compare favourably to those of other European and North American cities. This statement is baseless unless it is qualified by the methodological features of his and other studies that might support (or refute) the validity of that comparison. The study by Louw and Meyer (1965) also gives unqualified comparisons of heat island magnitudes in Pretoria, London, San Francisco, Hiroshima, Tokyo, and elsewhere. Padmanabhamurty (1990/91) closes his study of heat islands in India with the general conclusion that no discernable difference can be found among heat island magnitudes reported in coastal and inland cities. He based that conclusion on the classification of the Indian studies only as ―inland‖ or ―coastal,‖ ignoring factors other than location—such as sample size, operational definition, weather control, and site exposure—that might otherwise explain the observed geographic differences in UHI magnitude.  Inter-city comparisons not  supported by these considerations are difficult to accept. Many more studies in the literature are remiss for drawing unsupported heat island comparisons (e.g., Lyall, 1977; Travis et al., 1987; Steinecke, 1999).  72  A commonly used criterion thought to support inter-study comparisons of UHI magnitude is city population. Sweeney (1987), for example, confirms his observed maximum UHI magnitude of 6.5 K in Dublin as a reasonable estimate largely because it relates well with magnitudes observed in cities of similar population in North America and Europe. Countless other investigations have relied on population statistics to validate their observed heat island magnitudes, or to substantiate inter-city comparisons (e.g., Norwine, 1976; Rastorguyeva, 1979; Chiang et al., 1995; Sakakibara, 1999; Torok et el., 2001).  The underlying assumption in these comparisons is that cities with larger  populations produce more intense heat islands. Comparisons on this assumption alone are doubtful unless the investigators carefully screen the studies for differences in scale, definition, site character, and sampling procedures. Population is a surrogate for urban form and does not directly implicate important methodological or climatological aspects of heat island investigation. Settings that are sparsely populated, such as industrial factory districts, might otherwise produce significant heat islands for reasons unrelated to population density. Likewise, densely settled rural areas are unlikely to produce intense heat islands despite having large populations. Estimates of UHI magnitude in observational heat island literature become increasingly incommensurate as hundreds more published studies across many decades and geographic regions are examined. Differences in methodology have affected the outcomes of these studies to the extent that comparisons of UHI magnitude cannot be made without careful consideration of scale, definition, and sampling procedures. Scientific progress relies heavily on inter-study comparisons to corroborate or refute results, and thus drawing comparisons of UHI magnitude with previous studies is a  73  necessary and worthwhile pursuit; doing so without methodological support, however, is counterproductive.  3.3 Approach to the research problem If urban heat island observation and reporting continue in the undisciplined manner shown in this chapter, the literature will become increasingly fragmented. A structured and comparative view of heat island observations from cities worldwide, and from studies past and future, is now both timely and necessary. A research framework that addresses this need must therefore synthesize a geographically vast and methodologically diverse literature for critical analysis. Within this framework, two approaches to the research problem are taken. First, a summary assessment, or ―research synthesis,‖ of heat island literature is conducted by statistically evaluating a representative sample of studies on methodological and experimental criteria. Second, a new and more purposeful site classification is developed to succeed the overly simplistic urban-rural dichotomy and its operational test, Tu-r. The new system will embody a standardized and universally applicable test of UHI magnitude.  3.3.1 Synthesizing empirical urban heat island research Synthesis of research from a large collection of individual studies is formally known as meta-analysis (Cooper and Hedges, 1994a). Meta-analysis was developed in the mid-1970s in response to the increasingly disparate methods and results of a growing number of studies in the behavioural, medical, and social sciences.  Its earliest  practitioners declared it ―a new method of discovering knowledge…that lies untapped in completed research studies‖ (Glass, 1976). Through meta-analysis, questions can be asked that are not normally testable in single primary studies. Finding consistencies  74  among inconsistencies, and placing greater certainty in combined outcomes than in individual studies, meta-analysis answers to ―why‖ studies of the same intent produce different, and often conflicting, outcomes.  In tackling this provocative question,  methodologists in many disciplines use meta-analysis to extract overarching patterns or problems from a body of disorderly and fragmented literature, with the broader aim of bringing scientific results closer to the ―truth.‖ Meta-analysis is therefore best conducted on research topics supporting a substantial volume of accumulated studies. The distinction between meta-analysis and research synthesis is hereafter one of statistical rigor and output. While the former is a statistical reconciliation of outcomes, and the latter an aggregate summary of primary descriptive data, both are deemed ―higher‖ analyses of literature and follow similar review protocols (Cooper and Hedges, 1994b). Practitioners assert that only experimental studies with control/treatment groups and randomized trials are suitable for meta-analysis, as are studies of high, or at least uniform, quality (Fink, 2005). On this basis, synthesis of heat island literature is not properly a meta-analysis because the sample studies do not involve control/treatment groups and are not uniform in quality. Furthermore, the general findings of empirical heat island literature are not in dispute—it is the quality and consistency of methods used to measure and report UHI estimates that need scrutiny. The intent of the UHI research synthesis is to provide a quantitative assessment of methodological quality in modern heat island literature. The synthesis develops outcome measures to explain discrepancies in methods, results, and overall merit, and gives a broader and more coherent perspective on heat island investigation than any single primary study can deliver. Its final output identifies those studies that have produced  75  sound and reliable estimates of UHI magnitude. Despite the rapid growth of heat island literature during the modern period, a formal assessment of its primary studies has never been attempted.  3.3.2 Constructing a field site classification system for urban heat island observations In recent years, urban climatologists have advocated greater consistency in the use of scales, symbols, techniques, and concepts (e.g., Oke, 2004, 2006; Grimmond, 2006). Urban heat island magnitude is one such concept that needs standardization. Communication among heat island investigators has continued for decades without a standardized measure of heat island magnitude, but not without harmful consequences. Urban climatology‘s long-standing paradigm for space classification has not only blurred the physical and cultural peculiarities of the field sites used to define UHI magnitude, it has diverted attention away from their more important structural and physical characteristics that directly influence local thermal climate. The urban-rural paradigm has created an impression among heat island investigators that all estimates of UHI magnitude originate from field sites of similar character, and that those estimates are necessarily amenable to comparison. Heat island investigators were forewarned decades ago of the dangers in urbanrural classification and unqualified inter-city comparison. In 1967, Parry issued this statement about proper scale and communication in heat island studies: [Since] the early years of last century…many comparisons of urban and rural temperatures have been based on records of orthodox climatological station sites in open exposures on the lawns of large gardens or parks. Quantitative conclusions drawn from these [records]…suffered from the confusion of two different scales of climatic consideration, in that stations intended to be representative of a region were being used to provide comparisons of a local nature. The use of records from roof-top stations in some American cities further complicated the issue. (p. 617)  76  Parry continued with a foretelling remark that ―the diverse ways in which information has been gathered regarding urban ‗heat islands‘ compels one to doubt if the same feature has been measured in all cases.‖ His doubts were founded in ―unrepresentative siting‖ and ―conditions of exposure‖ at fixed recording stations, and in the widespread misunderstanding of the ―micro-climatic character‖ of urban climates. Similar doubts were raised by other methodologists of the time, including Ludwig (1970) : The measurement site becomes very important in the determination of urban-rural temperature differences. If the temperature is measured at street level among densely packed, tall buildings, the result may be different from a case in which the measurement is made just outside the central area….Furthermore, rooftop temperatures might be quite different from street level temperatures. If all of these factors are not taken into consideration, misinterpretation of urban-rural temperature differences is quite possible. (p. 89)  At the 1968 WMO Symposium on Urban Climates and Building Climatology, Chandler (1970b) closed the proceedings with a definitive statement regarding standardized communication: Observation sites need defining much more systematically and rigorously to ensure comparability between different investigations…We must define standard…sites for our urban observations. (p. 377)  Several years later, Bohm and Gabl (1978) delivered the same message, but with a greater sense of urgency: Given the wide variety of sampling and analysis methods [in heat island literature], the possibility of comparing the urban temperature effect between different cities must now be seen as critical…To allow a worldwide comparison of the influence of cities on temperature fields, that might eventually lead to confirmed physical relationships…we must first develop an international standard to measure and analyse the urban effect on temperature distribution. (p. 236; translation by Andreas Christen)  Implicit in these early and repeated warnings are fundamental differences among the regional scale urban-rural dichotomy, the local scale heat island effect, and the microscale signal of a screen-height temperature sensor. The variability in scale, method,  77  and setting alluded to by Parry (1967) and his contemporaries suggests that a standardized approach to site classification and heat island definition is long overdue. Just as urban theorists have argued for decades that the urban-rural dichotomy has lost its heuristic value as a research paradigm, I have contested on parallel grounds that this same dichotomy has also lost its value as an operational test of heat island magnitude. Landscape representation is fundamental to the definition, interpretation, and physical explanation of urban heat islands, yet no attempts have been made to develop a classification that reconciles differences of scale or that facilitates reliable and reasoned estimates of heat island magnitude. In this thesis, I address this methodological gap by developing a new site classification system for heat island investigators. The general aim of the new system is not to purge the urban-rural dichotomy from heat island discourse, but to encourage a more appropriate and constrained role for the dichotomy while curtailing baseless comparisons of heat island magnitude. In developing the new system, I divide the terms urban and rural into appropriately scaled and defined classes for quantifying urban heat islands. The system of classes to evolve from this division provides the conceptual framework for a new system of local-scale climate zones, each differentiated on a generalized set of surface properties that directly influence thermal climate.  The  divisional structure of the zone system is given physical support from observed temperatures in three midlatitude cities, and from the simulated temperatures of numerical surface-atmospheric models. Estimates of UHI magnitude from these cities are then related to standardized local climate zones, rather than to arbitrarily defined urban and rural field sites.  78  CHAPTER 4 RESEARCH SYNTHESIS 1 4.1 Methods The purpose of standard literature review is to describe current knowledge on a topic and to explain recent research findings. Urban climatologists have long invested in this tradition, with many reviews examining urban heat island literature and its rapid growth through the twentieth century (e.g., Kratzer, 1937; Brooks, 1952; Peterson, 1973; Oke, 1979; Landsberg, 1981; Nakagawa, 1996; Arnfield, 2003; Roth, 2007). Standard reviews, however, are seldom critical, they rarely engage the quality of the original studies, and generate little if any new knowledge. Despite these limitations, literature reviews are the most widely cited articles in science (Cooper and Hedges, 1994a). Research synthesis differs from standard review in that it integrates a body of literature by methodically extracting data from a representative sample of studies (Hunt, 1997). The extracted data are then combined into a single “super study” with quantitative and decisive findings.  Unlike more pedestrian reviews, research synthesis probes  fundamental differences in method to explain variances in outcome. In serving that aim, research synthesis must be (a) systematic, to meet the specific needs of the review; (b) reproducible, to give researchers opportunity for retesting its methods and re-evaluating its findings; (c) explicit, to ensure that the research questions and screening criteria are clearly defined; and (d) comprehensive, to examine the full literature and give justification for why certain studies are included while others are rejected (Fink, 2005). 1  A version of Chapter 4 has been published: Stewart ID. 2011. A systematic review and scientific critique of methodology in modern urban heat island literature. International Journal of Climatology 31: 2000–17.  79  A research synthesis follows five crucial steps that conform to the review protocols of a meta-analysis (Hunt, 1997). First, hypotheses are formulated in response to the specific goals of the meta-analysis. Second, the population, or “universe,” of studies about which the synthesis aims to generalize is defined by strict eligibility criteria. Third, a representative sample of that universe is retrieved from the literature through a logical search strategy. Fourth, essential information from each eligible item is extracted, coded, and combined into statistical outcome measures. Fifth, the methods, results, and theoretical implications of the analysis are reported and discussed. Research synthesis by this design is as much a scientific enterprise as the primary research it evaluates.  4.1.1 Defining the “universe” of studies The “universe” of studies is the complete body of literature about which a reviewer or synthesist wishes to generalize (Hunt, 1997). An entire universe of studies is normally impossible to assemble because many studies are not listed in standard indexes or databases and are not easily retrieved. A more practical approach is to select a representative sample of studies through fixed and unequivocal search conditions. A representative sample improves the likelihood that inferences of the study universe are valid. Furthermore, samples that are well defined are more likely coterminous with the universe of reality they are said to represent. This synthesis aims to generalize the methodological strengths and weaknesses of ground-based observational heat island studies and their estimates of canopy layer UHI magnitude. A universe of this description includes many hundreds of studies and extends well beyond the capacity of any single structured review. Strict eligibility criteria are therefore necessary to reduce the study universe to a workable size for evaluation. The  80  search for a representative and homogeneous sample of studies is a crucial first step to research synthesis, and it demands as much consistency and impartiality as later stages of the synthesis. Its importance to the external validity of the review cannot be overstated.  4.1.2 Selecting the primary literature “Primary” studies contain original research and are targeted works for literature synthesis (Hedges, 1994). Selecting eligible studies for a heat island synthesis is a multistep process that begins with screening the literature by three criteria: (i) characterisation of the urban heat island effect, (ii) principal aims, and (iii) date and source of print or publication (Figure 4.1).  Studies successfully meeting these criteria were deemed  eligible for synthesis and retained for inventory and appraisal; studies failing one or more of the criteria were declared ineligible for synthesis and rejected from candidacy. Eligibility of the candidate studies by criteria (i), (ii), and (iii) was decided by specific inclusion and exclusion terms. Inclusion terms alone are broad in scope and require a complementary set of exclusion conditions to restrict the sample to a manageable size. In principle, the eligibility criteria are practical rather than methodological, and descriptive rather than substantive.  4.1.2.1 Eligibility criteria (i) Characterisation of the heat island effect The first eligibility criterion targets observational canopy layer heat island studies of local-meso time and space scales (Figure 4.1). All studies using stationary or mobile temperature surveys that span one or several neighbouring urban settlements for the purpose of observation, description, or explanation of the nocturnal heat island effect were successful in meeting the first eligibility criterion. Local-meso scales were confined  81  Literature universe - Observational canopy layer urban heat island studies -  English-language literature  Foreign-language literature  First filter: Characterisation of the heat island effect Time and space scales: local to meso Data collection: ground based, mobile or stationary Measurement variable: nocturnal surface-air temperature  FAIL  Reject study  FAIL  Reject study  FAIL  Reject study  Eligibility criteria  PASS  Second filter: Principal aims Empirical measurement of an urban-rural or city-country air temperature difference, or of heat island "magnitude" or "intensity" PASS  Third filter: Date, source, and format Date: printed or published between 1950 and 2007 Source: original works of scholars and researchers Format: journal articles, book chapters, research reports, technical documents (conference papers and theses excluded)  PASS  Literature translation  Literature sample  Figure 4.1 Flow diagram illustrating the selection of literature for synthesis and evaluation.  to horizontal distances of 102 to 104 metres, and to time periods of days, months, or years. The first eligibility criterion disqualifies all investigations defining heat islands by larger or smaller scale sets, or by alternative sampling methods or sensing media. Immediately rejected from the sample were studies of boundary layer heat islands, remotely sensed heat islands, surface or subsurface heat islands, daytime heat islands, and non-urban heat 82  islands. Also excluded were studies consisting of regional, continental, or global datacollection networks across tens or hundreds of urban settlements, and studies of historical (i.e., decadal) temperature-time series in urban areas. Time-scale considerations were included in the first filter because a shortened timeframe sidesteps problems of data homogeneity (station moves, changes in instrumentation, etc.) and regional or global climate change. Although these issues are important to temperature analyses at any time scale, they were considered secondary to the evaluative criteria of the research synthesis.  All studies meeting the  “characterisation” filter were assumed to have reasonably homogeneous temperature datasets. Studies that include long- and short-term data series qualified for synthesis on their latter series, not their former. Studies incorporating time scales of several years or more also passed the first filter, provided that trend analysis of UHI magnitude was not a principal aim. (ii) Principal aims The second eligibility criterion targets all studies aiming to quantify the spatiotemporal features of the nocturnal UHI magnitude, or intensity, in a specified city, town, village, or other local-scale settlement (see Figure 4.1). This aim invokes empirical measurement of a nocturnal air-temperature differential across city and country, urban and rural, or otherwise built and non-built landscapes.  “Nocturnal” is restricted to  observations between sunset and sunrise, or in the early morning hours immediately following sunrise. If the candidate study had no intent of quantifying UHI magnitude, or if this intent was not its principal aim, it was withdrawn from the sample. The “principal aims” filter also removes studies of intraurban thermal patterns or  83  profiles. Studies that do not explicitly or implicitly refer to the “urban heat island effect,” its “magnitude” or “intensity,” or to the heat island conceptual model were similarly excluded.  Likewise, studies failing to pair “city” or “urban” temperatures with  “background,” “baseline,” “rural,” or otherwise ex-urban observations were disqualified from the sample. Theoretical and numerical modeling studies of the heat island effect are also ineligible by the “principal aims” filter, as are microclimate studies of small-scale urban or non-urban heat island features such as shopping malls, airports, lakes, and parks. Studies of relief effects rather than urban effects on city temperatures were excluded, although studies with the explicit aim of investigating both relief and urban effects on city temperature patterns qualified for evaluation if the researchers attempted to separate these effects. Investigations that give no particular attention to temperature as separate from wind, humidity, or precipitation were rejected from the study sample, as were urban climate surveys using secondary data sources for a particular city or region. (iii) Date and source of print or publication Luke Howard (1833) conducted the first scientific investigation of the heat island effect nearly two hundred yeas ago. The period between his investigation and those of the modern era covers significant conceptual progress and technological advances in urban climatology. However, research synthesis is best carried out with a literature sample that spans shortened periods during which no major theoretical or methodological shifts or revolutions have changed the field or its experimental ideals. The chosen evaluation period for this review is the modern period, from 1950 to 2007. Although many historical studies pre-dating 1950 made valuable contributions to heat island methodology, they lack the same conceptual basis that now underlies modern-  84  day investigations. The modern period also captures first usage of the term urban heat island in English-language literature, as well as the majority of published heat island studies worldwide, including those of tropical and developing regions.  All studies  printed or published between 1950 and 2007 passed the time filter of the third eligibility criterion. The third criterion further restricts the literature sample to studies of a desired source, format, and availability. Only the original works of scholars and researchers were included in the research synthesis, meaning that editorials, testimonials, surveys, and reviews of heat island literature were excluded. Restricting the synthesis to original research promotes accurate representation and generalization of the study universe, whereas the inclusion of secondary reviews, interpretations, and assessments introduces unwanted subjectivity from external reviewers into the evaluation. Multiple papers by the same author and of the same study area were eligible only if the heat island magnitudes in those papers were derived from different time periods or data-collection methods. Duplicate papers appearing in different sources or languages were immediately disqualified from the sample. Studies that are public and frequently cited in the literature were favoured for synthesis, partly to allow readers the option of retrieving those studies for their own use. Given the nature and intent of a research synthesis, its methods and results must be open to public scrutiny; the sample items should therefore be accessible to the public. Large quantities of primary research can be found in “fugitive” literature, which by definition is not widely distributed or indexed and for that reason difficult to locate (Rosenthal, 1994). Fugitive literature is not necessarily synonymous with unpublished  85  literature. An unpublished study that is easily retrieved may not be considered fugitive; likewise, a published study that is not widely distributed may be considered fugitive if it is difficult to procure. Theses, dissertations, manuscripts, conference papers, interim reports, and newsletters are examples of fugitive literature that exist in large quantities and that were excluded from the heat island sample. Other fugitive works such as government reports or institute papers were included only if they were original and met all remaining eligibility criteria. In addition to practical screening criteria (e.g., date, publication type), methodological criteria are often used to identify an appropriate sample for evaluation. Methodological screening criteria remove studies of seemingly inferior quality from evaluations, thereby increasing homogeneity in the literature sample (Fink, 2005). The “source” filter of the heat island synthesis is not intended to restrict the literature sample to works of a distinguished quality, despite the prevalence of traditional published media (e.g., journal articles, technical reports) in heat island literature. “Quality” criteria were excluded from the synthesis because the motive behind the evaluation is itself the disparate worth of the heat island literature. Screening for quality would defeat the purpose of the synthesis. The level of scrutiny across peer-reviewed work, whether fugitive or otherwise, is by no means standardized. Many investigations have never been peer reviewed but are of excellent quality and are frequently cited in the literature. In spite of the apparent quality of a reported investigation—as judged by the stature of its source—choices were inevitably made for or against inclusion of particular studies in the research synthesis. Studies that are cited widely in the literature and that passed the previous two eligibility  86  criteria were preferred, as were studies expanding the geographic scope of the literature sample into cultural and physiographic regions not well represented. This gives the research synthesis wider scope for generalization. When using peer-reviewed literature in a research synthesis, publication bias is a concern. Publication bias arises with the tendency of editors to publish significant results ahead of seemingly insignificant or “null” findings (Hunt, 1997). The resulting bias in published literature skews a reader’s understanding of the phenomenon because the results, as reported in refereed journals, are frequently exaggerated. Hunt suggests that prestigious journals are subject to greater publication bias than lesser-known sources. Reviewers of literature must therefore resist the assumption that studies published through lesser-known or non-conventional sources, or studies not published at all, are inferior to those in more prestigious publications. Research reviewers can lessen the effects of publication bias by including unpublished studies in the review sample. To ensure that a variety of document formats were represented in the heat island literature sample, non-reviewed, reviewed, unpublished, and published items were equally qualified for the research synthesis.  4.1.3 Sourcing and retrieving the primary literature The study sample was sourced primarily through online and print-accessible abstracts, article indexes, and bibliographic databases, both public and private. The literature search followed a logical strategy to ensure that little time was wasted on irrelevant or missed citations. Using subject key words, author names, and study titles, a preliminary list of citations was compiled from multiple indexes and databases, including General Science Index, Web of Science, GEOBASE, GeoRef, Meteorological and  87  Geoastrophysical Abstracts, SpringerLink, Wiley InterScience, and WorldCat. Hundreds of bibliographic references and article summaries relevant to the subject area were screened for eligibility. Based on title and summary content alone, a large proportion of these were disqualified from the sample for failing one or more of the eligibility criteria. Some studies that met all eligibility criteria were also disqualified, but for logistical problems with document recall. Additional references were obtained through “ancestry searching,” or manually retrieving citations from bibliographies and reference lists of books, conference proceedings, literature reviews, articles, serials, and so on.  Ancestry searching is  meticulous and time consuming, but it is no less rewarding than online or “paperless” searches in its yield of eligible items. As the ancestry search progressed, study titles deemed eligible for evaluation were indexed until the complete work could be retrieved. The decision to accept or reject a particular work was based on its title and summary (if available). On this basis alone, each study was judged to be either (a) outside the universe of study and of no value to the research synthesis, or (b) inside the study universe and of potential use to the research synthesis. Studies belonging to (b) were conditionally accepted until screened in full by the eligibility criteria. References from early decades of the evaluation period not listed in electronic databases and indexes were discovered primarily through ancestry searches. After a list of preliminary citations had been assembled, digital or paper copies of the qualifying documents were sought. Open-access journals were retrieved online via UBC Library’s full-text databases.  Bound copies of earlier journal editions were  manually recovered from library shelves, or retrieved through the Library’s automated  88  storage system. Facsimiles were then made of available works meeting all eligibility criteria. Book chapters, government reports, and technical notes were also acquired from the Library collections. Publications from more obscure sources were requested from the Canadian Institute for Scientific and Technical Information (CISTI) in Ottawa. If CISTI failed to meet a search request, UBC inter-library loans (ILL) was, in almost all cases, successful in procuring the requested document. As paper and electronic reprints amassed, ancestry searches of reference lists were sustained on their own “success.” Once a study was acquired, it was assessed for its eligibility in the literature sample. Although many of the citation titles first appeared promising, after careful examination of their content they proved less helpful and were removed from the sample. Of the hundreds of citations selected for retrieval, only a small number were unrecoverable through ILL or CISTI. Unrecovered documents were then requested from the original authors, who were contacted for reprints of their studies. Some of these requests were successful, while others met with no response. Contact information for some authors, particularly those of the early years in the evaluation period, could not be found. In addition to formal methods of literature retrieval, informal searches led to the discovery of eligible items. Expert consultation, for example, was an important link to undiscovered literature. Colleagues and urban climate investigators were informed of the heat island literature search and were consulted for citations that could usefully expand the study sample. Personal communication with conference and workshop participants uncovered many non-circulating works and historical pieces not indexed in public or private databases. As these works were acquired, ancestry searching of reference lists  89  uncovered additional citations to pursue.  4.1.3.1 Foreign-language literature Language barriers pose a significant obstacle to information transfer. Given the international reach of scientific activity, foreign-language communication is essential for disseminating useful data for good purpose.  However, foreign-language material is  rarely cited or consulted in most subject areas of science (Large, 1983). Citation analysis of major bibliographic indexes reveals that English is overwhelmingly the language of communication among contemporary heat island investigators. A keyword search for “urban heat island” in either of two comprehensive environmental science indexes— GEOBASE and Meteorological and Geoastrophysical Abstracts—returns more than 700 journal citations for the period 1970 to 2007. Restricting the search to languages other than English, only 9 percent of the original 700 citations remain. Inclusion of foreignlanguage material in a research synthesis ensures that its results are valid beyond an English-only universe of studies. However, linking these studies to their original sources, obtaining paper or digital copies, and translating their content can be a long and costly process. To avert lengthy and fruitless searches, and to avoid excessive translation costs, only a small number of foreign-language studies were included in the heat island research synthesis. Scores of heat island citations in many languages were amassed from foreignlanguage journals, abstract journals, article reference lists, and on-line and printaccessible indexes and databases. Most foreign-language studies had English titles and abstracts, and thus quick decisions could be made for or against inclusion of the work in the study sample. Bibliographic information not offered in English was translated with  90  online text-translation software. Computer-assisted translations were adequate only for deciding if foreign-language citations were worthy of fuller inspection. Throughout the screening process, foreign-language citations having English duplicates by the same author(s) but under different titles were excluded from the sample.  Citations not  recoverable from ILL, local library collections, or personal communication with colleagues or the authors themselves were also purged from the candidate list. Foreign-language citations were selected and screened by the same eligibility criteria as the English-language literature (see Figure 4.1). Preference was given to major languages of international scientific communication, which historically are Russian, German, and French, and more recently Japanese, Chinese, and Spanish (Large, 1983). Although heat island studies exist in many other languages, including Hungarian, Polish, Italian, Finnish, Slovenian, Portuguese, and Korean, these items were not retrieved unless, for reasons previously cited, they were of special significance to the synthesis. The list of foreign-language studies was culled to several dozen prior to acquiring reprints. Most of the candidate studies were retrieved from UBC library shelves or through ILL, CISTI, or the original authors themselves. Citations not available through these sources were discarded from the sample. Reprints with English translation of the study title, summary, and figure/table captions often contained enough detail of their methods and results to be declared (in)eligible for evaluation and further translation. In other cases, reprints were withdrawn from the sample prior to translation if evidence was sufficient that, once translated, the papers would fail the eligibility criteria.  91  4.1.3.2 Literature translation Scientists infrequently translate foreign-language studies because of the considerable time, effort, and costs involved.  A document must therefore be rated  sufficiently valuable to a research synthesis before its acquisition and translation can be justified. Foreign-language studies with English abstracts and figure captions required only partial translation, whereas studies with no English content demanded fuller translation. All studies were translated in ad hoc fashion, meaning that only specific information pertaining to its hypotheses, definitions, procedures, sampling conditions, and results was selected for translation depending on the English-language content of the paper. Professional translation services were avoided for practical and financial reasons. The costs required to commission a freelance translator or professional agency were difficult to justify given the intended private and pragmatic use of the translated material. Instead, students and faculty members who were sensitive to the scientific and geographic expression of the papers were sought as lay translators. Translators were expected to have the necessary reading and comprehension skills in the source language, as well as competency in the target language.  The search for translators started  internally, with UBC faculty members, post-doctoral fellows, and graduate students in scientific or environmental disciplines. Translations were done as objectively as possible by assigning one translator to a single source language and its corresponding studies. This avoided inconsistencies arising from multiple translators working in a single source language. Translators were generally non-experts in urban climatology. Prior to translation they were given written explanation of (a) the research synthesis and its purpose, (b) the  92  heat island effect and conventional methods for its observation, and (c) specialized language commonly encountered in heat island reporting. The translators were also given standardized abstraction forms (Appendix A) for retrieving important details from each heat island study. The forms served two main purposes: (1) to ensure that all translators followed an identical abstraction protocol, and (2) to keep translators focused on targeted information and to discourage them from devoting excessive time to superfluous detail or to full translation. This protocol was the same as that used for the English-language literature. Translators were advised first to survey the general content of the studies, then to target and translate verbatim those excerpts responding directly to the critical questions contained in the standardized abstraction forms. Translators were warned not to make subjective inferences from vague or missing information in a study, but instead to convey its factual content as accurately as possible. If, due to insufficient evidence, translators were unable to compile a response to one or more of the critical questions, they were instructed to reply in negative, rather than inferred, terms. A negative response is critical to the assessment of a study because it demonstrates lack of information or pertinent detail.  4.1.3.3 Sample size The point at which a sufficiently large and representative sample of studies has been gathered for synthesis is difficult to know. The longer a search continues, the larger the sample grows. The trade-off between continuous searching and critical sample size is one that burdens all research reviewers (Hunt, 1997). While a search typically continues well after the evaluation of material has begun, the returns gained—in terms of support  93  for or against a particular outcome—gradually diminish after a threshold of studies has been surpassed.  A critical number of studies beyond which returns diminish is  impossible to determine, but a sample of 200 or 300 items is unlikely to contribute significantly more to the aims of a research synthesis than one of 100 items, so long as the sample accurately reflects the terms of eligibility (i.e., it has internal validity) and the universe from which it was retrieved (i.e., external validity) (Valiela, 2001).  4.1.4 Cataloguing the primary literature 4.1.4.1 Urban heat island database An urban heat island database was constructed in two stages: (i) bibliographic inventory of the literature sample, and (ii) methodological assessment of the literature sample. (i) Bibliographic inventory Bibliographical descriptors from each primary study were extracted, condensed, coded, and assembled into standardized database forms.  These forms allowed the  information to be manipulated quickly and systematically as new data were added and existing data modified. Each study’s title, date, source, language, document type, and author name(s) were extracted from its bibliographic citation.  Document type was  classified into one of six categories: scholarly journal articles (peer reviewed vs. nonreviewed), professional/trade journal articles, research reports (monographs, occasional papers, technical notes, etc.), magazine articles, and book chapters. Scholarly journals were sometimes difficult to classify as either peer reviewed or non-reviewed due to the many levels of scrutiny available to a research paper. To standardize this process, the review status of all serials in the literature sample was classified according to Ulrich’s  94  Periodical Directory, which is an authoritative source of bibliographic and publisher information for thousands of serials worldwide. The local study area of each primary report was classified by the towns and cities under investigation, and the regional study area by de Blij and Muller’s (2003) division of world geographic realms, which includes North  America,  Middle  America,  South  America,  Europe,  Russia,  North  Africa/Southwest Asia, Subsaharan Africa, the Austral-Pacific, South Asia, Southeast Asia, and East Asia. Lastly, the field methods employed in each investigation were classified as either mobile or stationary. (ii) Literature assessment Unlike bibliographic data, which are plainly displayed on a study’s cover page, methodological data require a more discriminating eye to locate and interpret. The sample studies were carefully scrutinized for their experimental design features, or for evidence leading to reasonable inferences about those features. Guesswork was generally avoided, although in some cases judgements were made from minimal information. If the required information could neither be found nor inferred from a study’s content, a “null” entry was conceded to the database. Standardized abstraction forms were used for all studies in the sample, regardless of differences in document source, setting, language, or method. A standardized form facilitates data entry, manipulation, and reporting, and reduces error and subjectivity in the research synthesis. The following methodological details were abstracted from each sample study: operational definition(s) of UHI magnitude; physical character and spatial configuration of field sites; instrument specifications; provision of metadata; sampling period and procedures; control of extraneous influences on UHI magnitude; methods of  95  data correction; observed UHI magnitudes; inter-city comparisons; and other results or design features useful to the literature synthesis. Separate fields in the database were then assigned to each of the nine scientific criteria developed in section 4.1.5.  4.1.5 Evaluating the primary literature The following criteria were developed specifically to assess methodological quality in the heat island literature sample. The sample studies were assigned “pass,” “fail,” or “unknown” grades for each criterion: • Operational test and conceptual model are aligned; • Operational definitions are explicitly stated; • Instrument specifications are explicitly stated; • Site metadata are appropriately detailed; • Field sites are representative of the local-scale surroundings; • Number of replicate observations is sufficiently large; • Weather effects are passively controlled; • Surface effects are passively controlled; • Temperatures are measured synchronously.  Unlike the eligibility criteria in section 4.1.2.1, which screened the literature for practical concerns such as study aims, measurement scales, and date/format of publication, the scientific criteria screen the literature for more substantive concerns. These criteria were conceived from (1) well known methodological and conceptual frameworks in urban climatology (e.g., Landsberg, 1970; Oke, 1976, 1982, 1988; Lowry, 1977; Goldreich, 1984; Wanner and Filliger, 1989; Lowry and Lowry II, 2001; Szymanowski, 2005); (2) World Meteorological Organization (WMO) guidelines for weather and climate observations (e.g., WMO, 1983, 2006; Oke, 2004); and (3) classical interpretations of scientific method (e.g., Hempel, 1966; Valiela, 2001). Included in (3) are the hallmark  96  features of science: a problem statement, comprising a conceptual model, operational definitions, and research hypotheses; and systematic measurement, comprising a defined study area and controlled and repeated observations. Primary literature generally differs widely in the completeness of its reporting (Pigott, 1994). Any information that is not available in a report, but that is needed to respond positively or negatively to a particular criterion, is termed “missing data.” Missing data include all features of experimental design relating to a study’s definitions, assumptions, procedures, instruments, field sites, and outcomes. The grading of each primary study by the scientific criteria was based only on evidence contained in its original source document. No supplementary information or “missing data” to favour the decision process were retrieved from external sources, such as other publications or the authors themselves. Likewise, no studies were dropped from the research synthesis on account of their missing data. Instead, if the sample studies were lacking the information necessary for evaluation, “failing” or “unknown” grades were assigned to the relevant criteria. In this way, poor communication is synonymous with methodological weakness: writing accurate and detailed reports is as important to the scientific process as observation itself. Each study’s success with the scientific criteria is therefore balanced on sound methodology and effective communication.  4.1.5.1 Scientific criteria This section discusses the constituent terms of each scientific criterion. The terms leading to “pass,” “fail,” or “unknown” judgments of these criteria are displayed in decision keys, which show multiple outcome scenarios (tables 4.1 to 4.9). Boolean operators and, or, and not broaden or narrow or the range of conditions associated with  97  the pass, fail, and unknown grades of each criterion. Placement of the operators is crucial to the evaluation process and to matching experimental conditions with appropriate grades. Criterion 1: Operational test and conceptual model are aligned. The operational test of the investigation is aligned with the conceptual model of a canopy layer urban heat island. The test for this model invokes air temperature measurement below roof level in urban environments, and in the turbulent surface layer of rural environments. Criterion 1 builds on the first and second eligibility criteria of section 4.1.2.1, which limit the research synthesis to empirical studies of the canopy layer urban heat island. As a latent first condition of the scientific method, the conceptual model enables investigators to formulate hypotheses and develop suitable tests. The canopy layer heat island model provides a theoretical framework for observed urban-rural air temperatures differences in the surface layer.  Although this model is often poorly described or  understood in the literature, it underpins each sample study and gives operational basis to its measurements. Criterion 1 stipulates that the operational basis of each study must properly align with its conceptual model. Having the stated or understood aim of measuring UHI magnitude in the canopy layer, each study must invoke a suitable test of these concepts. If a study’s conceptual model and test are not well aligned, investigators cannot properly answer the questions they wish to ask. The operational test required of the canopy layer heat island model is surface air temperature measurement in urban and rural, city and country, or otherwise built and non-built environments.  This model is implicit in  Howard’s (1833) historical analysis of London’s heat island, but is developed and systematized more formally by Oke (1976, 1982, 1988) in modern literature. Studies that  98  met this fundamental condition passed Criterion 1. Studies that failed to measure air temperature at approximately screen height (1–2 m agl), or at least below roof level, and at sites broadly defined as urban and rural, are poorly aligned with their conceptual model. These studies failed Criterion 1. If sufficient detail of instrument height was not found in a report, or could not be inferred from its text, tables, or figures, Criterion 1 was graded “unknown.” WMO (2006) specifies that “screen height” for air temperature measurement is 1.25 to 2 m agl. This height-standard is expected of measurements in the rural areas of all heat island investigations in the literature sample. In the city, height restrictions can be relaxed because instrument security is a concern to all investigators, who therefore tend to position their sensors high above street level, discreetly removed from danger but still within the canopy layer (Oke, 2004). This arrangement is acceptable because at night the air volume in street canyons and other interstitial spaces is sufficiently mixed to prevent strong microscale temperature variations (Nakamura and Oke, 1988). Temperatures high above the ground but below roof level are generally representative of those at screen height.  However, siting of instruments on or over rooftops is  unacceptable. Rooftop sites by definition are above the urban canopy layer and are exposed to a different airflow, energy balance, and temperature regime than at ground level (Nakamura and Oke, 1988; Offerle et al., 2007). Comparisons of above– and below–roof level temperatures should not be used to quantify UHI magnitude because the cause of those temperature differences cannot be attributed to urban effects alone. If temperatures above roof level are used, the conceptual model and its operationalization are not aligned and Criterion 1 fails.  99  In judging each study’s success with Criterion 1, the height-placement of instruments was deduced from descriptors like “orthodox meteorological stations,” “standard weather screens,” “official climate stations,” or “first-order observatories.” One assumes from these descriptors that instruments are compliant with WMO guidelines for height placement, and are suitably positioned for surface air temperature measurement. However, in using these descriptors, investigators often neglect to report instrument heights. A “standard” or “conventional” weather screen, for example, may be compliant with most WMO guidelines, but if placed on the rooftop of a multi-storey building, it is not appropriately sited for measuring UHI magnitude in the canopy layer. Despite the possibility of non-standard siting, it was assumed that investigators using these descriptors measured surface air temperatures suitably below roof level or at approximately screen height. During mobile surveys, instrument heights vary from one to several metres above ground level.  If instrument height is not explicitly stated in a mobile study, this  information can be inferred from its described mounting (see Criterion 3). For example, a thermistor sensor mounted horizontally from a car roof or window is presumably 1–2 m agl. If sufficient detail of instrument height during a mobile survey is not provided, or cannot be inferred from the discussion or presentation of data in a report, Criterion 1 fails.  100  Table 4.1 Decision key for Criterion 1—Operational test and conceptual model are aligned. Experimental conditions  Grade  The operational test is aligned with the conceptual model of a canopy layer urban heat island. The test consists of air temperature measurement in the urban and rural surface layers.  Pass  The operational test is not aligned with the conceptual model of a canopy layer urban heat island. The test does not consist of air temperature measurement in the urban and rural surface layers.  Fail  The extent to which the operational test is aligned with the conceptual model of a canopy layer urban heat island is not reported. The test consists of measurements at unknown heights or in unknown environments.  Unknown  Criterion 2: Operational definitions are explicitly stated. Operational definitions of urban heat island magnitude or intensity are explicitly stated in the report, or made implicit through discussion or presentation of data. Operational definitions reveal the measurement variables and field sites used to compute urban heat island magnitude. Operational definitions translate concepts into procedures (Cooper and Hedges, 1994a). However, a universal procedure for measuring canopy layer UHI magnitude does not exist.  Investigators must therefore contrive and communicate appropriate  ad hoc procedures of their own to quantify the magnitude of a canopy layer heat island. Their chosen procedures, or “operations,” must be explicitly stated in the report, or made implicit through discussion or presentation of data. Operational definitions are essential for proper interpretation of a study’s findings, because without definitions the findings are effectively meaningless. Criterion 2 makes no judgment on the appropriateness of the operational definition, only on the provision of that definition. “Unknown” grades are therefore not possible with Criterion 2. 101  Criterion 2 requires two conditions of an operational definition: (1) the location and number of field sites used to quantify UHI magnitude, and (2) the measurement variables (e.g., annual mean temperature, daily minimum temperature) obtained at those sites. In passing Criterion 2, a study must satisfy both conditions. Many heat island investigators operationalize UHI magnitude with the familiar expression ∆Tu-r, largely because it appears widely in the literature and is universally recognized. In doing so, they neglect further definition and explanation of its antecedent terms (i.e., urban and rural). Investigators must give sufficient operational basis to these terms for all heat island settings and definitions used in their studies. ∆Tu-r is a vague representation of UHI magnitude, and explicit definition is usually required to contextualize its meaning to particular settings. “Insufficient definition” gives rise to failing grades in Criterion 2.  Table 4.2 Decision key for Criterion 2—Operational definitions are explicitly stated. Experimental conditions  Grade  Operational definitions of UHI magnitude are explicitly reported or made implicit through discussion or presentation of data; and  Pass  Operational definitions reveal the measurement variables and field sites used to quantify UHI magnitude. Operational definitions of UHI magnitude are not explicitly reported or not made implicit through discussion or presentation of data; or  Fail  Operational definitions do not reveal the measurement variables or field sites used to quantify UHI magnitude.  Criterion 3: Instrument specifications are explicitly stated. Instrument specifications are explicitly stated in the report, or made implicit through discussion or presentation of data. Specifications include type, mounting, and measurement precision of the thermal instruments used to quantify urban heat island magnitude.  102  Investigators are expected to give the technical specifications of their thermal instruments used to quantify UHI magnitude. These include thermometer type (e.g., liquid-in-glass, electrical), mounting (i.e., the object/apparatus to which the thermometer is attached), shielding (i.e., enclosure), and measurement precision (i.e., uncertainty). Investigators siting instruments in the urban canopy layer for the purpose of surface air temperature measurement should avoid WMO (2006) guidelines stating that thermometers be mounted “over level ground, freely exposed to sunshine and wind and not shielded by, or close to, trees, buildings and other obstructions.” These guidelines are inappropriate for city-based instruments because the thermal influences of trees, buildings, cars, and paved surfaces are the very same influences that the instruments are intended to capture (Oke, 2004). Open, grassy sites are often not representative of urban settings. If possible, instruments that are intended to represent the city environment should be mounted in a microscale setting that is typical of the local area, such as a street canyon or paved lot (more about site representativeness in Criterion 5). WMO guidelines also stipulate that temperature sensors be placed in standardized weather boxes, or thermometer screens. Again, this guideline can be relaxed for citybased instruments whose security in public spaces demands discreet placement above ground level, often extending from sign posts or building walls. Regardless of position, the instruments should be housed in mechanically aspirated, or otherwise well-ventilated, radiation shields or screens that can be mounted to a utility pole, sign post, or fence. Instruments that are mounted to cars, trains, or bicycles should be done so in a way that ensures adequate ventilation of the sensor and that shelters it from vehicle or body heat. Descriptions of instrument mounting are often good indicators of microscale  103  environment, and are useful in judging the representativeness of temperatures. Instrument mounting must be clearly described or illustrated in every heat island report. The measurement precision of the instruments, whether stationary or mobile, must also be stated in a heat island report.  A statement of measurement precision, or  uncertainty, gives readers greater confidence in a study’s estimates of UHI magnitude. WMO (1983) is unequivocal in its stance on measurement precision: “No statement of the results of a measurement is complete unless it includes an estimate of the probable magnitude of the uncertainty,” which is normally expressed as the interval of values “within which the true value of a quantity can be expected to lie.”  Heat island  investigators must be explicit in disclosing the measurement precision of their instruments. If measurement precision was stated in a report, as was instrument type, Criterion 3 passed. If instrument type was stated but with no reference to its precision, Criterion 3 failed.  Finally, if sufficient detail of instrument mounting (including  shielding) was not found in a report, or could not be inferred from its text, tables, or figures, Criterion 3 failed.  Table 4.3 Decision key for Criterion 3—Instrumentation specifications are explicitly stated. Experimental conditions  Grade  Instrument specifications (type, mounting, and precision) are explicitly reported, or made implicit through discussion or presentation of data in the report. Instrument specifications (type, mounting, and precision) are not explicitly reported, or not made implicit through discussion or presentation of data in the report.  104  Pass  Fail  Criterion 4: Site metadata are appropriately detailed. Site metadata are appropriately detailed in the report. Metadata include a local- or regional-scale map, sketch, or photograph of the study area, and one or more quantitative indicators of local or microscale surface exposure, roughness, or cover at the field sites used to quantify urban heat island magnitude. Broadly defined, metadata are information about the observed data, and site metadata are circumstantial (i.e., non-technical) information about the place where data are observed. The provision of site metadata in a heat island report is essential to a complete and accurate portrayal of (1) the study area and its main geographic features, (2) the field sites and their locations within the study area, and (3) the surface and exposure characteristics of the field sites. Without site metadata, generalizations about the reported heat islands are difficult to make, as are inferences about their causation. According to WMO guidelines on climate metadata, all meteorological measurements should include specification of station identity, geographical location, local environment, instrumentation, observing practices, data processing, and station history (Aguilar et al., 2003).  Supplementary WMO guidelines for meteorological  measurements in urban areas stress that local environment and historical events are especially important due to the complex and dynamic nature of cities (Oke, 2004). The former includes maps, sketches, and aerial and terrestrial photographs of the field sites and their instruments, and the latter the long-term physical and cultural changes to the landscape that influence surface climate. The conditions of Criterion 4 are relaxed from WMO guidelines, which are too detailed for a single heat island report. For the purpose of the heat island research synthesis, “historical” metadata are not expected of a report because the time scales already imposed by the eligibility criteria generally preclude significant landscape  105  change. The majority of studies in the literature sample have observation periods of several months or years, and thus “historical” metadata are less relevant than “local environment” data to the reported UHI magnitudes. The first condition of Criterion 4 stipulates that site metadata include a local- or regional-scale illustration (e.g., photograph, plan map, site sketch) of the study area. The illustration should portray major physical and cultural features of the region, such as mountains, valleys, water bodies, transportation routes, built-up areas, and other terrain characteristics that influence local and regional surface climate.  Also expected of this, or another,  illustration are the relative locations of the field sites used to quantify UHI magnitude. The second condition of Criterion 4 stipulates that site metadata include one or more measurable and climatologically relevant indicators of local or microscale surface exposure, roughness, or cover of the field sites used to quantify UHI magnitude. Possible indicators include sky view factor, height or aspect ratio of buildings or trees, built or natural surface fractions, and thermal admittance of built or natural surfaces. If either of these two conditions was not met in a heat island report, Criterion 4 failed.  Table 4.4 Decision key for Criterion 4—Site metadata are appropriately detailed. Experimental conditions  Grade  Site metadata include a map of the study area depicting its major physical and cultural features and the locations of its field sites; and Site metadata include one or more quantitative indicators of surface exposure, roughness, or cover at the field sites used to quantify UHI magnitude. Site metadata do not include a map of the study area depicting its major physical and cultural features or its locations of field sites; or Site metadata do not include quantitative indicators of surface exposure, roughness, or cover at the field sites used to quantify UHI magnitude.  106  Pass  Fail  Criterion 5: Field sites are representative of the local-scale surroundings. The microscale settings of the field sites used to quantify urban heat island magnitude are approximately representative, in surface morphology, land cover, and human activity, of the local-scale surroundings. The role of scale in Criterion 5 is paramount. UHI investigators are expected to place screen-height instruments in locations where the local-scale “circle,” or area, of influence (also known as fetch, footprint, or source area) is relatively uniform in surface morphology, land cover, and human activity. The shape and radius of the area of influence are difficult to estimate because they change with building density and atmospheric stability (Kljun et al., 2002). Empirical evidence suggests that the source area for a screen-height thermal sensor in a stable atmosphere is no more than a few hundred metres in radius (Tanner, 1963; Parry, 1967; Chandler, 1970a; Mizuno et al., 1990/91; Gallo et al., 1996; Runnalls and Oke, 2006). If the microscale (< 104 m2) setting of the sensor is reasonably uniform, but the local-scale (104 to 108 m2) surroundings are conspicuously varied or more heterogeneous, then the measured temperatures are not spatially representative, or accurate, beyond the microscale area. Investigators who extrapolate temperatures beyond regions of uniformity into wider, more diverse, and more complex surroundings are confusing the scales of influence behind their measurements. This, in turn, confuses one’s understanding of heat island causation and behaviour.  “Confusion of scales” is a common flaw in heat island  investigation and it amounts to failure of Criterion 5. Assessing Criterion 5 in most heat island studies is problematic because site metadata are incomplete and building densities vary—as do the areas of influence—at the field sites. Criterion 5 was therefore judged not on rigorous statistical measures but on  107  qualitative evidence from site maps, photographs, sketches, station names and locations, and descriptions of the study area and its field sites. As a general rule, sites were judged to be representative if the surface morphology, land cover, and human activity they intended to characterise are reasonably well distributed in all directions from the sensor, and they represent more than half of the surrounding local area. Representativeness at high-density urban sites (specifically compact street canyons) was judged at micro to local scales because local thermal influences are often blocked by buildings. An urban site said to be representative of a high-density, lowrise residential neighbourhood should have most of its local surroundings dominated by roads and small, compact buildings. If that site is situated on the edge of a large park, with a local fetch of trees and vegetated surfaces to one side, the site and its temperatures are unrepresentative of the dense residential area (depending on wind direction). Representative urban field sites include street canyons, shopping centres, commercial areas, parking lots, public squares, residential gardens or terraces—the possibilities are many, so long as the natural and built surroundings of the instruments are reasonably homogenous at local horizontal scales. City airports and urban parks are not ideal for the placement of instruments because surface character typically varies across micro and local scales. Among potentially representative rural sites are agricultural fields, natural parks and preserves, and otherwise undisturbed land. Instruments sited in or near rural villages, industrial or recreational parks, or other facilities, institutions, or settlements in the urban-rural transition zone can be problematic due to lack of surface homogeneity. Instrument siting near small-scale cultural features is equally problematic: temperatures measured next to a factory or on a bridge, for example, may not be  108  representative of the local fetch if those features are themselves anomalies of the surrounding urban or rural area. Paved roads can have microscale thermal effects that are not representative of the local (especially rural) surroundings. However, roads cover only a small fraction of land in non-urban areas, and their effects on screen-height thermal climate are usually minimal due to microscale advection from the immediate environs. Roads are nevertheless an unavoidable feature of mobile surveys. Similarly, stationary networks of meteorological observatories are not easily reconfigured to avoid the disturbing features of a local landscape, in which case the investigators should eliminate all unrepresentative sites from computations of UHI magnitude. Sampling design provides helpful clues to assess the representativeness of a field site. Interval, or systematic, sampling is least conducive to obtaining representative temperatures. Mobile surveys that gather data at regular spatial or temporal intervals (especially microscale intervals of metres or seconds) typically produce dense arrays of temperatures.  But with no allowance for the underlying surface character, interval  sampling does not guarantee that the recorded temperatures will spatially coincide with homogenous source areas, or that the temperatures will represent thermal conditions beyond the microscale surroundings. Random sampling designs are also disadvantaged because they neglect the natural and built configuration of the study area. “Convenience” sampling, which obtains samples only from those sites for which data are easily available (e.g., in situ climate stations), is disadvantaged for the same reason. Stratified sampling is the best approach to establish representative field sites. A stratified sample is one based on homogeneous “strata” that are known to influence the outcome measure (Fink, 2005). If the outcome measure is UHI magnitude, strata include  109  local areas differentiated by building morphology, land cover, or human activity. Neglecting to stratify a sample may produce results that are confounded by supposed causal factors. Studies based on stratified sampling designs are more likely to pass Criterion 4 than those based on interval, random, or convenience designs. In judging the representativeness of each study’s field sites, special attention was given to a controversy known among research reviewers as the “expectancy effect.” The expectancy effect arises in primary research when investigators induce, through contrived means, a desired or exaggerated response from an experimental test (Hunt, 1997; Chalmers, 1999).  Experiments so designed are sometimes advantageous:  passive  control, for example, of synoptic weather effects (Criterion 7) allows urban effects to be accentuated and separated from non-urban effects, and the heat island phenomenon to be more “accessible” to observation. Other times, exaggerating an observed heat island effect can be injurious to an experiment, for example, through the common practice of using field sites that are known a priori to exhibit maximum urban-rural temperature differences, regardless of their location or exposure characteristics. In this case, the expectancy effect is adequate warning that field sites may not be representative. Insufficient metadata to allay this warning constitutes an “unknown” grade for Criterion 5.  110  Table 4.5 Decision key for Criterion 5—Field sites are representative of the local-scale surroundings. Experimental conditions  Grade  Metadata are sufficient to depict the surface morphology, land cover, and human activity of the field sites used to quantify UHI magnitude as spatially representative of micro and local horizontal scales; or The measurement sites used to quantify UHI magnitude are sampled by stratified design using surface morphology, land cover, and human activity; or  Pass  The field sites used to quantify UHI magnitude are reported to be spatially representative of micro and local horizontal scales. Evidence is given to support this claim. Metadata are sufficient to depict the surface morphology, land cover, and human activity of the field sites used to quantify UHI magnitude as spatially unrepresentative of micro and local horizontal scales; or  Fail  The field sites used to quantify UHI magnitude are sampled at fixed microscale time or space intervals; or The field sites used to quantify UHI magnitude are reported to be spatially unrepresentative of micro and local horizontal scales. Metadata are not sufficient to depict the field sites used to quantify UHI magnitude as spatially representative of micro and local horizontal scales.  Unknown  Criterion 6: Number of replicate observations is sufficiently large. The number of replicate heat island observations in a report is sufficiently large to meet the stated aims of the study and to yield representative and reliable estimates of urban heat island magnitude. Regular and repeated measurement provides control over random variation and increases the probability of obtaining representative values of a desired effect at a chosen time and place (Valiela, 2001).  Regular measurement also gives reliable basis to  inferences. While no critical limit to the duration of a heat island investigation or to its number of replicate observations exists, the sample size must be sufficiently large to stabilize variance and to meet the stated objectives of the study.  Investigators  traditionally follow a reductionist approach to heat island assessment, meaning that observations are made under simplified conditions. 111  This approach diminishes the  quantity of data needed to meet a study’s objectives. Investigators aiming to characterise UHI magnitude during clear, calm summer weather, for example, may require only a small dataset—e.g., a dozen or so mobile transects—to meet this aim, assuming all observations are appropriately controlled (more about control in criteria 7, 8, and 9). Investigators who instead aim to characterise seasonal patterns of UHI magnitude require a larger dataset—e.g., one year or more of stationary measurement—to meet their aims. Studies boasting large sample sizes were not automatically judged superior to ones with small sample sizes. However, studies with extremely small samples, such as one or a few nights of observation, failed Criterion 6 regardless of their stated aims. Small samples lack statistical power, meaning that inferences based on that sample are weak and unreliable. Small samples also allow little or no control over measurement error, especially random errors from inaccurate, deformed, or improperly configured instruments, or from “blundered” observations (WMO, 1983; Church, 2011).  Table 4.6 Decision key for Criterion 6—Number of replicate observations is sufficiently large. Experimental conditions  Grade  The number of replicate heat island observations is sufficiently large to meet the stated aims of the study and to obtain representative values of UHI magnitude.  Pass  The number of replicate heat island observations is not sufficiently large to meet the stated aims of the study or to obtain representative values of UHI magnitude.  Fail  The number of replicate heat island observations is not disclosed or made implicit through discussion or presentation of data in the report.  Unknown  112  Criterion 7: Weather effects are passively controlled. The extraneous effects of weather on urban heat island magnitude are passively controlled. Computations of urban heat island magnitude use temperatures measured in relatively steady-state weather: no passing fronts, strong advection, or precipitation. Unlike laboratory tests or numerical models, field experiments cannot normally achieve active control over weather and other environmental conditions. Passive control, however, can reduce the effects of these extraneous influences on observed phenomena. Criterion 7 is the first of two criteria stipulating control of environmental conditions. Heat island investigators must passively control weather to reduce the risk of confounding “real” urban heat islands caused by urban effects with “fictitious” ones caused by precipitation or air mass advection (Lowry, 1977; Lowry and Lowry II, 2001). Passive control of weather can be gained through preconceived sampling designs or through post hoc data selection.  Preconceived sampling avoids frontal passage or  precipitation events during data retrieval. Post hoc selection excludes from computations of UHI magnitude those data retrieved during frontal passage or precipitation events, or at least acknowledges their effects on reported UHI magnitudes. Szymanowski (2005) discussed the difficulty of identifying and removing frontal events from temperature datasets. He argued that city-based weather-station networks rarely have the spatial resolution needed to detect small-scale weather changes associated with passing fronts. Precipitation and strong-advection events must therefore be filtered from heat island datasets, leaving the residual data for computing true urban effects on temperature.  Correcting datasets in this way does not imply that urban effects on  temperature cease to exist during unsettled weather conditions. Urban effects do persist through cloudy, windy, and/or rainy conditions, and there are many situations in which  113  heat island investigators might be interested to compute a mean UHI magnitude (i.e., averaged over all weather conditions). However, the presumed urban effects on the heat island during frontal or precipitation events may actually be weather effects if the observations are not partially controlled. A reductionist approach is ideal for isolating and observing the desired urban effect. Although this approach yields a reduced subset of heat island observations, these observations ultimately furnish more reliable and representative estimates of UHI magnitude. Each of the sample studies was inspected for evidence of precipitation or frontal passage in its heat island dataset. Larger study areas and longer-term datasets of several years or more are generally at greater risk of weather “contamination” than smaller study areas and shorter-term datasets. A popular strategy for separating weather and urban effects on UHI magnitude is to stratify observed temperature data by wind, cloud, and other atmospheric variables, and then to correlate these variables with computed UHI magnitudes. This strategy does not guarantee that the effects of precipitation and frontal passage have been removed from the dataset or from computations of UHI magnitude.  114  Table 4.7 Decision key for Criterion 7—Weather effects are passively controlled. Experimental conditions  Grade  Data collection avoids frontal passage and precipitation events; or  Pass  Temperatures that are affected by frontal passage or precipitation events are acknowledged as such or excluded from computations of UHI magnitude. Temperatures that are affected by frontal passage or precipitation events are neither acknowledged as such nor excluded from computations of UHI magnitude. Weather conditions during data collection are not reported in sufficient detail to suggest that temperatures are affected by frontal passage or precipitation events; or  Fail  Unknown  The extent to which computations of UHI magnitude include temperatures affected by frontal passage or precipitation events is not reported.  Criterion 8: Surface effects are passively controlled. The extraneous effects of surface relief, elevation, and water bodies on urban heat island magnitude are made sufficiently small through planned sampling design, or made sufficiently known through discussion and recognition of their influences on observed heat island magnitudes. Cities are often situated on the complex terrain of river valleys, coastlines, basins, and lakeshores. Lowry (1977) and Goldreich (1984) suggested that the terrain effects of these locations are a “nuisance” that too few heat island investigators address when designing and reporting their observations. In general, urban and terrain effects on surface air temperatures are nearly impossible to separate. Heat island investigators must nevertheless adopt an appropriate design strategy to counteract unwanted terrain, or surface, influences, otherwise the computed city-country temperature differences are not sufficiently urban-induced to warrant classification as urban heat islands. “Surface” here refers to all local and mesoscale landforms and water bodies that generate variable effects on the temperature regime of a study area.  115  Extraneous surface effects on heat island magnitude can be eliminated or avoided through experimental design (Wanner and Filliger, 1989; Lowry and Lowry II, 2001). Placing urban and rural field sites at similar elevations and within relatively uniform local to mesoscale settings effectively eliminates surface effects, and thereby isolates the urban contribution to observed heat islands. Instruments should be sited away from slopes, gullies, cliffs, or ridges, and configured parallel—not perpendicular—to elongated surface features such as valleys and shorelines so as to eliminate their variable effects on the sampled temperatures. Microscale surface features like depressions, hollows, and hills are less of a “nuisance” than those exerting local or regional effects (e.g., mountains, valleys, coastlines), provided that the instruments are not sited in the immediate area of the surface feature.  Temperature sampling should also avoid excessive horizontal  (> 102 km) or vertical (> 102 m) distances between individual measurement sites. Finally, the mitigating effects of local sea-breeze circulations on UHI magnitude can be avoided by sampling during months and seasons when sea-land temperature gradients are lowest (Runnalls and Oke, 2000).  Each of these design strategies greatly reduces  variable surface effects across a sampled area. Most urban and rural locations have unwanted surface effects that cannot be avoided, in which case corrective measures can be performed on the temperature data after they have been collected. Two post hoc techniques can improve isolation of the urban effect in the temperature data of complex terrain (Goldreich, 1984). The first technique regresses temperature against height to determine a representative lapse rate for a particular study area. The observed temperatures can then be normalized to a standard level using the measured lapse rates. The weakness of this approach is that lapse rates  116  are not always linear, especially when the dynamic effects of surface cooling and air drainage have complicated the vertical temperature profile.  Moreover, profile  measurements are not easily attainable. The second technique regresses temperature against distance inland to determine a representative sea-land profile for a particular study area. Variable sea effects on urban and rural temperatures can then be reduced by normalizing the observed temperatures to a standard distance from the shoreline. The obvious weakness in this approach is that the regression coefficients are highly unstable outside the particular environment in which the relation was established. Given the serious drawbacks of these post hoc techniques, they should be used cautiously, if at all, to correct estimates of UHI magnitude Each study was assessed for its success with Criterion 8 on evidence gathered from its discussion and illustration of the study area and on the individual field sites used to quantify UHI magnitude.  If, through planned sampling design, heat island  investigators were unable to avoid the disturbing surface features of a particular study area, they should account for the surface factor in other ways. At minimum, they should qualify their estimates of UHI magnitude by appropriately recognizing unwanted surface effects on measured heat island magnitudes. Recognition of these effects may include one or more of the post hoc regression techniques previously discussed.  Post hoc  correction by itself, however, does not constitute a passing grade—it must be part of a broader treatment of the surface factor that qualifies the purported “urban” heat island estimates as over- or underestimates by way of unavoidable land-surface features. Given the difficulty and uncertainty of establishing control over the effects of surface relief, elevation, and water bodies on UHI magnitude, qualitative treatment of the topoclimatic  117  effect constitutes a passing grade for Criterion 8.  Table 4.8 Decision key for Criterion 8—Surface effects are passively controlled. Experimental conditions  Grade  The extraneous effects of surface relief, elevation, and water bodies on UHI magnitude are made sufficiently small through planned sampling design; or  Pass  The extraneous effects of surface relief, elevation, and water bodies are made sufficiently known through discussion and recognition of their influences on reported UHI magnitudes. The extraneous effects of surface relief, elevation, and water bodies on UHI magnitude are not made sufficiently small through planned sampling design; or The extraneous effects of surface relief, elevation, and water bodies are not made sufficiently known through discussion and recognition of their influences on reported UHI magnitudes.  Fail  Surface relief, elevation, and water bodies in the study area are not reported in sufficient detail to make their extraneous effects on UHI magnitude known; or The extent to which computations of UHI magnitude include temperatures affected by the extraneous influences of surface relief, elevation, or water bodies is not disclosed; or  Unknown  The locations of the measurement sites defining UHI magnitude are not disclosed.  Criterion 9: Temperatures are measured synchronously. Temperatures used to quantify urban heat island magnitude are measured synchronously. Inhomogeneities resulting from non-synchronous measurement are acknowledged as such and standardized to a common base time. Criterion 9 highlights the importance of time control during heat island measurement. If the temperatures used to quantify UHI magnitude are not synchronous, or adjusted so as to be synchronous, urban induced heat islands may be confounded with time (and therefore weather) induced heat islands. During a mobile temperature survey, the regional temperature may fall, rise, fluctuate, or remain constant. The surveyed  118  temperatures must therefore be subjected to a time-correction scheme to account for these changes.  Likewise, investigations that use temperature minima to quantify UHI  magnitude must apply time-temperature corrections to their data. Minimum temperatures are convenient for quantifying heat islands because they are observed at most official climate stations and are easily extracted from historical datasets. However, temperature minima are not normally synchronized across a spatial network of instruments, especially over complex urban and rural topography or in non-steady weather (Oke and Maxwell, 1975; Szymanowski, 2005). Investigators must therefore adjust their data to equalize differences in the time of daily minima, otherwise the data are subject to the influences of changing weather (e.g., cloud cover, wind speed, wind direction). If the precise times of occurrence for daily minima are known at each measurement site, as is regional temperature change across the study area and throughout the observation period, the temperatures can be adjusted to a standard base time through simple regression. Simultaneity is also expected of all long-term temperature series used to quantify UHI magnitude. If the temperature series do not coincide, conditions of simultaneity are breached and Criterion 9 fails.  119  Table 4.9 Decision key for Criterion 9—Temperatures are measured synchronously. Experimental Conditions  Grade  Temperatures used to quantify UHI magnitude are measured synchronously; time-temperature corrections are unnecessary and are not performed on the data; or Temperatures used to quantify UHI magnitude are not measured synchronously; time-temperature corrections are necessary and are performed on the data; or  Pass  Temperatures used to quantify UHI magnitude are not measured synchronously; time-temperature corrections are shown to be unnecessary and are not performed on the data. Temperatures used to quantify UHI magnitude are not measured synchronously; time-temperature corrections are necessary but are not performed on the data.  Fail  Synchroneity of temperatures used to quantify UHI magnitude is not disclosed; or Temperatures used to quantify UHI magnitude are not measured synchronously; time-temperature corrections are necessary but are not disclosed; or  Unknown  Temperatures used to quantify UHI magnitude are not measured synchronously; neither the necessity for timetemperature corrections nor the corrections themselves are disclosed.  4.1.5.2 Grading scheme A points-based grading scheme was designed to quantify methodological quality in the heat island literature sample (Table 4.10). Underpinning the grading scheme are the nine scientific criteria discussed in the previous section. Each sample report was graded and ranked by a conventional “vote count” procedure in which points were awarded for passing a criterion and no points for failing a criterion (Glass, 1976). A study earned a maximum of 18 points for passing all nine scientific criteria. Few, if any, studies were expected to earn full points for all nine criteria. Challenges with data acquisition, site access, time and labour constraints, and variable weather and terrain put this total beyond the reach of most investigators. Despite these unavoidable challenges,  120  Table 4.10 Points-based grading scheme for assessing methodological quality in the heat island literature sample. Maximum points allotted  Fail  Unknown  Pass  3  0  0  3  No  3  0  …  3  No  1  0  …  1  ¼ point each for mounting and shielding; ½ point for precision  Least important  1  0  …  1  5. Site representativeness Most important 6. Number of replicates Least important  3  0  0  3  ½ point each for site map and quantitative indicator No  1  0  0  1  No  Criterion  Weight class  1. Conceptual model  Most important 2. Operational definitions Most important 3. Instrument Least specifications important  4. Site metadata  Points allotted by grade  Partial points  7. Weather control  Somewhat important  2  0  0  1 or 2  1 point for post hoc treatment; 2 points for planned sampling  8. Surface control  Somewhat important  2  0  0  1 or 2  9. Synchroneity  Somewhat important  2  0  0  1 or 2  …  18  0  0  15–18  1 point for post hoc treatment; 2 points for planned sampling 1 point for nearsynchronous measurement or time-temperature correction; 2 points for synchronous measurement ...  Total  18 points is the “gold standard” by which all reports of UHI magnitude were judged. The number of points assigned to each scientific criterion is based on its importance in generating reliable and reasoned estimates of UHI magnitude. The nine criteria were separated into three weight classes: criteria 1 (conceptual model), 2 (operational definitions), and 5 (site representativeness) are deemed “most important”  121  to a UHI estimate and consequently weigh heavily in the grading scheme; criteria 3 (instrument specifications), 4 (site metadata), and 6 (number of replicates) are deemed “least important” and weigh less; and criteria 7 (weather control), 8 (surface control), and 9 (synchroneity) are deemed “somewhat important” and carry intermediate weight (Table 4.10). Success with the “most important” criteria co-varies with the success of the other six criteria: failure of criteria 1, 2, or 5 increases the likelihood of failing criteria 3, 4, and 6–9. Thus the ceiling on a study’s potential point score drops significantly with failure of one or more of criteria 1, 2, or 5. The sample studies were sorted into three tiers—top, middle, and bottom—based on their overall success with the nine scientific criteria (Figure 4.2). Standards for tier placement are most demanding in the top tier and least demanding in the bottom. The studies were assigned to these tiers based on their success with only the “most important” and “somewhat important” criteria; the “least important” criteria had no bearing on tier placement and were used only for assigning points within tiers.  The tiers are  differentiated not by the sum of criteria met or by point totals, but by the combinations of criteria met. Studies with similar point totals but different tier placements are therefore alike only in the quantity of criteria passed, not in the combination of criteria passed. Top-tier studies, having passed five (or more) of the six “most important” and “somewhat important” criteria, are of highest methodological quality in the literature sample. They boast minimum scores of 11 and maximum scores of 18, depending on their success with the “least important” criteria. Notwithstanding point scores, heat island investigations in this tier yield reliable estimates of urban effects—as separate from non-urban effects—on surface-layer air temperatures. These studies follow the  122  UHI literature sample  NO  Does the study pass three "most important" criteria and two or more "somewhat important" criteria?  Does the study pass two or more "most important" criteria and one or more "somewhat important" criteria?  NO  YES  Top tier YES  UHI estimates are acceptable  Middle tier UHI estimates are conditionally acceptable  Bottom tier UHI estimates are unacceptable  "Most important" criteria 1. Conceptual model 2. Operational definitions 5. Site representativeness  "Somewhat important" criteria 7. Weather control 8. Surface control 9. Synchroneity  "Least important" criteria 3. Instrument specifications 4. Site metadata 6. Number of replicates  Figure 4.2 Criterion-based scheme to determine tier placements in the heat island literature sample.  scientific method to the extent that the conceptual model and operational test are aligned, operational definitions are clearly communicated, field sites are approximately representative of the local environment, and extraneous influences such as precipitation, frontal advection, relief, elevation, water bodies, and time are carefully controlled. Only those studies near the top of the points range give full account of instrument specifications and site metadata, and gather a sufficiently large sample to control random variation. Middle-tier studies, having passed three (or more) of the “most important” and “somewhat important” criteria, have minimum scores of 7 and maximum scores of 15, depending on their pass/fail record with the “least important” criteria.  123  Middle-tier  estimates of UHI magnitude are less refined than the top tier, yet more refined than the bottom tier. They are acceptable only on the condition that weaknesses or uncertainties in method are acknowledged. Methodologically, middle-tier studies lack rigor in some of the following areas: aligning the conceptual heat island model with the operational base, stating operational definitions, using field sites that are representative of the local environment, and controlling confounding effects on UHI magnitude. Only those studies at the high end of the points range give full account of instrument specifications and site metadata, and gather sufficiently large samples to control random variation. Bottom-tier studies, having passed few, if any, of the “most important” and “somewhat important” criteria, bear minimum scores of 1 and maximum scores of 12. Their methods for observing UHI magnitude are the least sophisticated of the literature sample, and thus their empirical estimates are unacceptable regardless of any success with the “least important” criteria.  These studies are crudely designed and yield  indefensible heat island estimates.  Without control over confounding effects like  weather, relief, and time, the reported UHI magnitudes are induced as much through nonurban effects as through urban effects. Bottom-tier studies are consequently at high risk of attributing false cause to observed heat island magnitudes. Furthermore, field sites are not likely representative of their local surroundings, and operational definitions are often not known to the reader.  Studies with high point scores in the bottom tier earn  recognition for successfully reporting instrument specifications and site metadata, and for gathering a sample large enough to control random variation. During the grading process, each study was tagged with a Missing Data Index (MDI) to measure its completeness and efficiency of reporting. MDIs were determined  124  by tallying the number of points lost to “unknown” grades, which was then converted to a percentage of the total number of “unknown” points possible (13). For example, a study receiving an “unknown” grade for Criterion 5 (site representativeness) accumulated 3 points toward its MDI total because “3” is the number of points forfeited to incomplete or incompetent reporting (see Table 4.10).  The accumulation of MDI points with  “unknown” grades differs with each criterion and its assigned weight. MDI values were normalized from 0 to 1.  Values approaching unity indicate a detrimental lack of  information in a report, and raise the possibility of unconventional or unrepresentative instrument siting and/or lack of experimental control. Values approaching zero indicate full and competent reporting. One might argue that studies with excessive reporting gaps (i.e., many “unknown” grades) should be removed from a research synthesis because they cannot be rated fairly against those with more complete reporting. Given that reporting itself is a measure of research competence, the argument to remove studies that are weak in communication is immaterial to this review’s aims and desired output.  4.1.5.3 Rankings In the final stage of evaluation, the quality scores of each primary study were converted to rank equivalents. By convention of standard competition ranking, rank equivalents were determined first by tier placements and second by quality scores. Accordingly, studies in the top tier were ranked above those in the middle and bottom tiers, and studies with high scores above those with low scores. If two or more studies had identical tier placements and scores, the study earning more points from the three “most important” criteria was ranked higher. If the studies earned an equal number of points from the “most important” criteria, the study earning more points from the  125  “somewhat important” criteria was ranked higher. Studies still equal in point earnings from the “somewhat important” criteria were assigned shared ranks. This convention leaves gaps in the sequence following shared rank assignments. The number of gaps is always one less than the number of “competitors” sharing the same ranking.  The  positions of all competitors ranked below the shared ranking are unaffected.  4.2 Results More than 500 candidate papers and online articles and abstracts were screened for inclusion in the research synthesis. Of this total, 177 studies were declared eligible for assessment. Eighty-eight stationary and 102 mobile subsamples were extracted from these eligible studies, giving an aggregate sample size of 190 heat island studies. The number of eligible studies and the total sample size are different because studies classified by method of data collection as both “mobile” and “stationary” were graded twice. A complete listing of bibliographic information for the 177 eligible studies is given in Appendix B. The studies are listed chronologically, as indicated by their ID codes. In Appendix C, the codes are cross-listed with the language, survey method, and survey area of the 177 studies. The internal composition of the literature sample is presented first, followed by the aggregate results from the grading and ranking of the literature.  A variety of  graphical and non-parametric statistical measures are used to characterise the literature sample.  This opening section is important because the validity behind any  generalizations later extracted from the research synthesis (see Chapter 6) hinges on the representativeness of the 177 sample studies. Frequency distributions of the sample are presented by geographic realm, political state, language, method of data collection, and  126  year and source of print or publication.  4.2.1 Describing the literature sample The heat island observations reported in the literature sample are distributed across 11 continental realms, 41 political states, and 221 cities and towns (Appendix C). Geographic breakdown by political region puts the United States in the frequency modal class, with 29 studies, followed by the United Kingdom (20), Japan (17), Canada (15), and India (12) (Table 4.11). The reported populations of the cities and towns in the literature sample range from 250 in Realitos, Texas, to 23 million in Tokyo. In more than half of the 177 sample studies, the observations originate from European and North American cities, and in one-quarter of the studies they originate from East and South Asian cities (figures 4.3 and 4.4).  The remaining 7 geographic realms are each  represented by 10 or fewer studies. The sample’s modal class is Europe, with 55 studies, followed by North America and East Asia, with 44 and 30 studies, respectively. Russia and Middle America are represented by only 2 studies each. Continental realms having a larger percentage of the sample’s total urban population are not necessarily represented by greater frequencies of heat island studies (Figure 4.4).  Europe and North America, for example, are over-represented in the  sample, while North Africa, Southwest Asia, South America, and Middle America are all underrepresented. This distribution is partly a function of retrieval bias, which arises from a reviewer’s inability to access and retrieve all research equally. It is also a function of an underlying geographic bias in the literature universe, which is a result of European and North American research dominating the field since its beginnings in the nineteenth century. Studies from equatorial and subtropical regions are historically  127  Table 4.11 Frequency distribution of the heat island literature sample (n = 177) by political state. State USA Japan England Canada India China Spain Malaysia Hungary Germany Singapore Australia Sweden Austria Taiwan South Korea Poland Israel Russia Kenya Nigeria  Frequency 29 17 17 15 12 7 6 6 4 4 4 4 4 3 3 3 3 3 2 2 2  State South Africa Romania Bulgaria Finland Mexico Ireland Chile Netherlands Greece Scotland Italy Iceland Portugal Egypt Argentina Ecuador New Zealand Botswana Sri Lanka Brazil  Frequency 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1  Figure 4.3 Geographic distribution of heat island observations in the literature sample.  128  30 25 20 15 10 5  er ic a Su SE bSa As ha ia ra n M Af id ric dl a e Am er ic a R us Au si st a ra lP ac ifi c  ic a  Am  N  Am er  As ia  S  SW  As ia  N  Af  ric a/  S  ro p  Eu  E  e  0 As ia  PERCENTAGE FREQUENCY  35  Figure 4.4 Percentage frequency distribution of the literature sample (n = 177) by geographic realm, urban population (black bars), and number of heat island studies (grey bars).  underrepresented in urban climate literature and comprise only 23% of the total heat island sample. Seven foreign languages are included in the literature sample: English is the modal class with 152 studies, or 86% of the sample, followed by Japanese (8), German (5), Chinese (5), Spanish (3), French (2), Russian (1), and Korean (1). Twenty of the 25 foreign-language studies have English titles, abstracts, and figure captions, and required only partial translation of their content. The remaining 5 studies have no English content (except titles) and required fuller translation. Frequency distribution by year of print or publication is positively skewed across the 58-year sample period (Figure 4.5). The first and last decades are represented by the lowest and highest frequencies, respectively, in the study sample. Frequency distribution by document source (Figure 4.6) is biased toward academic journals because the eligibility criteria screened all candidate papers for “the original works of scholars and researchers.” The most accessible outlet for work of this kind is the scholarly journal.  129  50  FREQUENCY  40 30 20 10  20 00 -0 7  19 90 -9 9  19 80 -8 9  19 70 -7 9  19 60 -6 9  19 50 -5 9  0  Figure 4.5 Frequency distribution of the heat island literature sample (N = 190) by decade and tier placement. Black bars = bottom tier; grey bars = middle tier; white bars = top tier.  120  FREQUENCY  100 80 60 40 20 0 A  B  C  D  E  F  Figure 4.6 Frequency distribution of the heat island literature sample (n = 177) by document source. A = peer-reviewed scholarly journals; B = non-reviewed scholarly journals; C = professional/trade journals; D = research reports; E = book chapters; F = magazines.  130  By method of data collection, the literature sample is comprised of 89 “mobile” and 75 “stationary” studies, or 50 and 42%, respectively, of the total sample. The mobile studies primarily used automobiles to transport their temperature sensors across an urban-rural area, although trains, motor-scooters, and bicycles were also used. Stationary studies used in situ or purpose-built networks of urban and rural temperature sensors.  4.2.2 Analysing the primary literature An aggregate summary of the UHI research synthesis and its descriptive data is provided in Figure 4.7 and Table 4.12. Pass/fail frequencies are illustrated and tabulated for each of the nine scientific criteria and for both the mobile and stationary subsamples. Appendix E contains the pass/fail grades for each of the nine scientific criteria for all 190 sample studies. The studies are listed by ID codes only, which are cross-listed with bibliographic information in Appendix B. The nine scientific criteria each comprise antecedent parts. In some cases, failure of one or more antecedent parts constitutes automatic failure of the criterion; in other cases, success of one or more antecedent parts constitutes partial success of the criterion. In following section, pass/fail frequencies are examined to identify the sources of success or failure behind each criterion. Examples of evidence to support the pass/fail decisions in the literature sample for scientific criteria 1 through 9 are given in Appendix D.  4.2.2.1 Scientific criteria Criteria 1 (conceptual model) and 2 (operational definitions) have the highest aggregate pass ratios, at 75 and 78%, respectively, of all nine criteria. Twenty-two percent of the sample studies failed Criterion 2, meaning that nearly one-quarter of all studies in the literature sample provide no definition or explanation of UHI “magnitude”  131  Figure 4.7 Frequency distribution of the heat island literature sample (N = 190) by scientific criterion and aggregate pass/fail ratios.  132  Table 4.12 Pass and fail ratios by scientific criterion and method of data collection.  Criterion 1. Conceptual model Mobile Stationary Aggregate 2. Operational definitions Mobile Stationary Aggregate 3. Instrument specifications Mobile Stationary Aggregate 4. Site metadata Mobile Stationary Aggregate 5. Site representativeness Mobile Stationary Aggregate 6. Number of replicates Mobile Stationary Aggregate 7. Weather control Mobile Stationary Aggregate 8. Surface control Mobile Stationary Aggregate 9. Synchroneity Mobile Stationary Aggregate  n  No. of “passing” grades  No. of “failing” grades  No. of “unknown” grades  “Passing” ratio (%)  “Failing” ratio (%)  “Unknown” ratio (%)  102 88 190  99 43 142  1 3 4  2 42 44  97 49 75  1 3 2  2 48 23  102 88 190  72 77 149  30 11 41  … … …  71 88 78  29 12 22  … … …  102 88 190  33 10 43  69 78 147  … … …  32 11 23  68 89 77  … … …  102 88 190  12 9 21  90 79 169  … … …  12 10 11  88 90 89  … … …  102 88 190  9 15 24  13 27 40  80 46 126  9 17 13  13 31 21  78 52 66  102 88 190  40 76 116  60 10 70  2 2 4  39 87 61  59 11 37  2 2 2  102 88 190  72 32 105  2 53 54  28 3 31  71 36 55  2 60 29  27 4 16  102 88 190  57 46 103  19 18 37  26 24 50  56 52 54  19 21 20  25 27 26  102 88 190  82 54 136  7 33 40  13 1 14  80 61 72  7 38 21  13 1 7  or “intensity,” nor any evidence on which to base a reasonable inference of that definition. Despite an aggregate passing ratio of 75% for Criterion 1, the discrepancy between ratios for the mobile and stationary subsamples is large (97 vs. 49%, respectively). The success rate for mobile studies is high because simple description of their modes of transport (e.g., automobiles or motor-scooters) makes implicit the fact that all air temperatures 133  were made below roof level. Clear statements of instrument height are therefore less critical to the success of Criterion 1. In contrast, the use of stationary surveys involving fixed weather boxes or towers requires more explicit description from investigators as to instrument height. Too often this height is not specified in a heat island report and thus the suitability of its operational test cannot be determined. Despite giving passing grades to stationary studies that use descriptors like “official climate stations” and “standard weather shelters” to convey the origin of the temperature measurements, the frequency of “unknown” grades in the literature sample is disproportionately higher in the stationary subsample, at 48%, than in the mobile subsample, at 2%. For Criterion 2, individual passing ratios for the mobile (71%) and stationary (88%) subsamples are not greatly different. The stationary ratio is slightly higher, suggesting a greater tendency for these investigations to make their operational definitions of UHI magnitude known. With fewer measurement sites, on average, stationary studies tend to have simpler and more communicable definitions of UHI magnitude than mobile surveys. Aggregate pass and fail ratios for criteria 3 (instrument specifications) and 4 (site metadata) are inversely proportional to those of 1 and 2. At 77 and 89%, respectively, criteria 3 and 4 have the highest failure rates of all nine criteria. Only 43 of a total 190 studies, or 23% of the sample, provide full details of their instruments. Neglecting to report the precision of instruments used to quantify UHI magnitude is by far the most frequent cause of failing grades. Of the 147 studies that failed Criterion 3, 97% gives no indication of instrument precision, 39% no indication of instrument type, and 40% no description of instrument mounting or shielding.  The use of stationary surveys to  quantify UHI magnitude has a greater tendency for incomplete reporting of instrument  134  specifications than mobile surveys. Investigators using mobile surveys to observe UHI magnitude often employ instruments that are designed or configured purposely for vehicular transport, and whose specifications are easily known and described in a heat island study. The aggregate pass/fail ratios for Criterion 4 are nearly identical to the component pass/fail ratios for its mobile and stationary subsamples. Of the 169 studies that failed Criterion 4, only 8% failed on account of incomplete or incompetent cartographic or photographic representation of the study area. Ten percent failed because the major physical and cultural features influencing local surface climate in the study area were not depicted in regional maps or illustrations, while 17% give no depiction whatsoever of the field site locations quantifying UHI magnitude. Accounting for a much larger fraction of the failing grades in Criterion 4 is the deficiency of quantitative descriptors of local or microscale site character (e.g., surface exposure, roughness, and cover). Of the 169 studies failing Criterion 4, 168 provide no quantitative description of the field sites defining UHI magnitude. Most of these studies instead use qualitative expressions like “green fields” or “downtown” to describe their sites and local settings. Thirty-three percent of the literature sample, or 62 of 190 studies, gives neither qualitative nor quantitative descriptions of their field sites and settings. These studies use only “urban” and “rural,” or other similar terms, to describe their sites. In other studies, quantitative descriptors are provided for one or several of the sites defining UHI magnitude, but not for all sites. The most frequently cited quantitative descriptor of site character in the primary literature is fractional coverage of built and natural surfaces, followed by the height of roughness elements (e.g., buildings, trees), and, lastly, sky view factor. These descriptors are each cited in less than 10% of the 190  135  sample studies. The high failure rate for Criterion 4 (site metadata) has negatively influenced the outcome of Criterion 5 (site representativeness).  Sixty-six percent of the literature  sample, or 126 of a total 190 studies, has field sites of unknown representativeness. Only 13% of the literature sample provides enough detail of its sites to earn passing grades for Criterion 5. Conversely, one-fifth of the sample was judged to have unrepresentative field sites quantifying UHI magnitude. Investigators in six of the 40 studies failing Criterion 5 openly confess that their field sites are not representative of the local surroundings and give evidence to support that claim. More than half of the 190 sample studies, or 66%, were graded “unknown” for inadequate reporting of field site characteristics. Barring any evidence for or against site representativeness, studies employing fixed-interval or grid sampling techniques for site selection were graded “unknown.” The ratios of “unknown” grades for Criterion 5 are significantly different between the stationary and mobile subsamples (52 vs. 78%, respectively). This discrepancy is due in part to the much lower spatial resolution of temperature sampling associated with stationary data collection. The likelihood of a stationary study describing the character of its sites in sufficient detail to pass Criterion 5 is therefore greater than that of a mobile study. Aggregate pass ratios for criteria 6 (number of replicates), 7 (weather control), and 8 (surface control) range from 54 to 61%. Thirty-seven percent of the sample, or 70 of a total 190 studies, was judged unsuccessful in carrying out observations of sufficient duration or frequency so as to meet the stated aims of their investigation (Criterion 6). Fifty-nine percent of these failing studies involve mobile surveys, and only 11%  136  stationary surveys. Mobile surveys are comparatively labour and resource intensive and are therefore disadvantaged by data of poor temporal resolution. The number of mobile surveys conducted in each study of the literature sample ranges from 1 to 140. Stationary surveys, in comparison, are operationally simple and thus favoured for replicate, frequent, and long-term observations of UHI magnitude. The duration of fixed sampling among the remaining 88 stationary studies in the literature sample ranges from 1 day to 20 years. Studies graded “unknown” give no indication of sampling frequency or duration and comprise just 2% of the total sample, or 4 out of 190 studies. Criteria 7 (weather control) and 8 (surface control) met similar passing ratios of about 55%. Of the 105 studies that passed Criterion 7, 78% use planned sampling designs to deliberately manipulate the conditions of observation, that is, to avoid frontal passage or precipitation events during the data collection period, and 28% use post hoc treatments to remove data that are compromised by these unwanted weather influences. Forty-five percent of the literature sample failed to sufficiently control the disturbing effects of weather on UHI magnitude, or to communicate the extent to which control was taken. Many of these studies have committed the fallacy of false cause, meaning they have incorrectly assumed that the spatial (or temporal) correlation of high temperatures with urban areas necessarily implies a causal relation. The significantly lower failure rate in the mobile subsample is explained by the freedom that investigators have to control the time of a mobile survey. If weather effects are thought to potentially distort the measured heat island signal, investigators can abandon or delay the survey until more desirable conditions develop. In this way, investigators can “control” weather effects on the recorded heat island magnitudes. Stationary surveys require investigators to manually  137  remove the distorting effects of weather from their datasets because the heat island signal is recorded continually through all weather types. This passive approach to weather control is often overlooked in stationary heat island studies. Over half of the 190 sample studies successfully met the conditions of Criterion 8 (surface control). Of the 103 studies that passed, 54% succeeded on account of planned sampling design and 46% on post hoc data correction.  The 37 studies that failed  Criterion 8 were judged inadequate in their attempts to recognize and separate surface from urban influences in their estimates of UHI magnitude.  The potential for  confounded urban and non-urban influences in the UHI estimates of these studies is unacceptably high, as is the risk of attributing false cause to the magnitudes of these estimates. Criterion 9 (synchroneity) has pass and fail ratios of 72 and 21%, respectively. Mobile studies that passed Criterion 9 standardized their observations to a common base time, or conducted their surveys when change in background temperature was negligible. All studies that failed Criterion 9 derived UHI estimates from non-synchronous temperature measurements, which raises the possibility of confounding spatial change in temperature with temporal change.  The remaining 7% of the sample was graded  “unknown” for incomplete reporting of time/temperature data, or of attempted corrections to those data. Failing rates for the stationary and mobile subsamples in Criterion 9 are significantly different, at 38 and 7%, respectively. The stationary studies that failed Criterion 9 use minimum temperatures to quantify UHI magnitude, whereas the mobile studies use temperature data that were left unadjusted for temporal inhomogeneities despite survey times lasting several hours.  138  4.2.3 Grading the primary literature Final rankings, tier placements, quality scores, and MDI (missing data index) values for individual studies are tabulated in Appendix F. ID codes can be crossreferenced with those in Appendix B to link author names with final rankings.  4.2.3.1 Tier placements and quality scores Frequency distribution by tier placement favours the bottom and middle tiers of the grading scheme (Figure 4.8). The distribution of studies across the three tiers changes slightly between the mobile and stationary subsamples. Given that the original stationary subsample is slightly smaller than the mobile subsample, the weighted distribution of stationary studies favours the bottom and top tiers.  Figure 4.8 Frequency distribution of the heat island literature sample (N = 190) by tier placement and method of data collection.  139  Quality scores for the study sample range from 1 in the bottom tier to 18 in the top tier (Figure 4.9). Only 2 of 190 studies earned maximum scores of 18. The mean quality score for the entire sample is 9.3 (Table 4.13), and the modal class is the 10 to < 11 point range, with 25 of 190 studies. The distribution of scores around the mean is symmetrical, with progressively fewer studies toward higher and lower point ranges. In the bottom tier, the distribution of studies is positively skewed: 35 of a total 85 studies (or 41%) earned totals in the range of 6 to < 8 points, and an additional 37 studies (44%) earned totals in the ranges 4 to < 6 and 8 to < 10. Only three studies earned less than 2 points. More than one-quarter of the bottom-tier studies were printed or published between 1990 and 1999 (see Figure 4.5). In the middle tier, the distribution of quality scores is more symmetrical around the mean. The majority (53%) of studies earned points in the range 10 to < 12 (Figure 4.9). Five of 86 studies earned quality scores of less than 9, and only 1 study a score of 14 or greater. In the top tier, with only 19 studies,  Figure 4.9 Frequency distribution of the heat island literature sample (N = 190) by point-based quality scores and tier placement. Black bars = bottom tier; grey bars = middle tier; open bars = top tier. Mean quality score = 9.3  140  Table 4.13 Mean quality scores and missing data index (MDI) values by tier placement and method of data collection.  Tier Top Mobile Stationary Aggregate Middle Mobile Stationary Aggregate Bottom Mobile Stationary Aggregate All Mobile Stationary Aggregate a  n  Mean quality scorea  9 10 19  14.6 16.4 15.5  .09 .02 .05  58 28 86  10.6 11.2 10.8  .28 .19 .25  35 50 85  6.9 5.9 6.3  .36 .39 .38  102 88 190  9.6 8.8 9.3  .29 .28 .29  Mean MDI value  Out of 18 points  the distribution frequencies are low in all point-range classes. Frequencies in the top tier generally rise as point scores increase: only 3 studies earned fewer than 14 points, while 16 studies earned 14 or more points (Figure 4.9). Distribution of tier placements by decade shows that almost half of the top-tier studies were printed or published between 2000 and 2007 (see Figure 4.5). MDI values in the heat island literature sample range from 0 to 0.85 (Figure 4.10). The frequency modal class, with 35% of the sample studies, is the .2 to < .3 range of values. Sixteen percent of the sample, or 30 of 190 studies, has MDI values of 0. Communication in these studies—which belong almost entirely to the top and middle tiers—is complete, and therefore grading of their content was not impaired by missing information. Twelve percent of the sample has MDI values of 0.5 or greater. In these  141  70 FREQUENCY  60 50 40 30 20 10  -1 .9  -<  .9  .8 .8  -< .7  .6  -<  -<  .7  .6  .5 .5  -<  .4 .4  -< .3  -<  .3  .2 .2  -<  .1  0  -<  .1  0  MDI VALUE  Figure 4.10 Frequency distribution of the heat island literature sample (N = 190) by missing data index (MDI) values and tier placement. Black bars = bottom tier; grey bars = middle tier; open bars = top tier. Mean MDI value = 0.29  studies—which belong mainly to the bottom tier—most of the information needed to pass criteria 1 and 5–9 was missing. Grading of these studies was severely impaired by reporting deficiencies. No studies in the sample were assigned MDI values of 1. The overall distribution of studies across the three tiers and the 18-point score range is related to the calculated MDI values (Figure 4.11; Table 4.13). In general, studies in the top tier have lower MDI values than those in the middle and bottom tiers. The relation between MDI values and quality scores for the 190 sample studies is negative (r = -.66), suggesting that higher MDI values are associated with lower scores.  142  18  POINT SCORE  15 r = -.66 12 9 6 3 0 0.0  0.2  0.4  0.6  0.8  1.0  MDI VALUE  Figure 4.11 Distribution of the heat island literature sample (N = 190) by missing data index (MDI) values, points scores, and tier placement. Black circles = bottom tier; grey circles = middle tier; open circles = top tier.  4.2.3.2 Rankings A strong association (r = .97) exists between rank number and point scores for the entire literature sample, suggesting that the latter are good predictors of the former (Figure 4.12). The stability of this relation weakens in the point-range 7 to 11, as shown by the spread in data points corresponding with the overlapping scores of the bottom and middle tiers. A study with a score of 10.5 points out of 18, for example, could be placed into either the bottom or middle tier (and therefore assigned a different rank number) depending on the combinations of criteria passed. The correlation between rank numbers and MDI values is negative (r = -.64), meaning that high rank/tier placements are associated with efficient reporting, and low rank/tier placements with incomplete or incompetent reporting (figures 4.10 and 4.13). As MDI ratings increase, the range in rank placements greatly diminishes, such that incomplete reporting has a stronger influence on rank placement than complete reporting. The explanation for this pattern is 143  1  RANK  r = .97  100  199 190 0  3  6  9  12  15  18  POINT SCORE  Figure 4.12 Distribution of the heat island literature sample (N = 190) by points scores, rank placement, and tier placement. Black circles = bottom tier; grey circles = middle tier; open circles = top tier.  1  RANK  r = -.64  190 .0  .2  .4  .6  .8  1.0  MDI VALUE  Figure 4.13 Distribution of the heat island literature sample (N = 190) by missing data index (MDI) values, rank placement, and tier placement. Black circles = bottom tier; grey circles = middle tier; open circles = top tier.  144  that complete reporting does not guarantee a study’s success with the scientific criteria, whereas incomplete reporting necessarily guarantees a study’s poor performance with the criteria. The rankings (Appendix F) are crude indicators of each study’s success in observing and communicating reliable estimates of UHI magnitude. A high-ranking study has few significant threats to its validity, whereas a low-ranking study faces serious threats. Studies receiving similar or identical rankings are potentially quite different in their methodological strengths and weaknesses. This curiosity is a result of the grading scheme and its aim to provide a cumulative index of research quality. Studies in the literature sample having commendable design features—though not necessarily high rankings—are listed in tables 4.14 and 4.15. Table 4.14 provides a complete list of all studies providing sufficient site metadata to pass Criterion 4. Studies recognized for exceptional control of confounding factors on UHI magnitude are listed in Table 4.15. Only 22 studies in the heat island sample passed all three “control” criteria. Lastly, studies noted for serious deficiencies in reporting are listed in Table 4.16. These studies lack crucial information about field procedures and operational definitions, and immediately stand out for their high MDI values. Foreign-language studies are slightly disadvantaged in the assessment of reporting deficiencies. The high MDI values of these studies may in part be attributed to information lost or misrepresented during translation. Three of the 10 weakest studies in primary reporting are foreign-language documents (Table 4.16).  145  Table 4.14 Studies in the heat island literature sample with sufficient site metadata to pass Criterion 4. ID code  Ranka  Tier  Study title  Source  Year  Survey method  058  1  Top  Study of the subarctic heat island at Fairbanks, Alaska  Report, Environmental Protection Agency  1978  Stationary  076  1  Top  Studies of the development and thermal structure of the urban boundary layer in Uppsala  Report, Meteorological Institute (Uppsala)  1980  Stationary  167  3  Top  Urban-rural contrasts of meteorological parameters in Lodz  Theoretical and Applied Climatology  2006  Stationary  129  5  Top  The relationship between heat island intensity and rural land coverage in Obuse, Nagano  Tenki  1999  Mobile  175  6  Top  The urban heat island and local temperature variations in Orlando, Florida  Southeastern Geographer  2006  Stationary  126  8  Top  Temporal and spatial characteristics of the urban heat island of Lodz, Poland  Atmospheric Environment  1999  Stationary  166  9  Top  Influence of urban morphology and sea breeze on hot humid microclimate: The case of Colombo, Sri Lanka  Climate Research  2006  Stationary  165  10  Top  Temporal dynamics of the urban heat island of Singapore  International Journal of Climatology  2006  Stationary  047  12  Top  Urban heat island dynamics in Montreal and Vancouver  Atmospheric Environment  1975  Mobile  137  15  Top  Dynamics and controls of the near-surface heat island of Vancouver, British Columbia  Physical Geography  2000  Stationary  144  23  Middle  Temperature cross-section features in an urban area  Atmospheric Research  2001  Mobile  155  23  Middle  Intra-urban relationship between surface geometry and urban heat island: Review and new approach  Climate Research  2004  Mobile  156  23  Middle  The relationship between built-up areas and the spatial development of the mean maximum urban heat island in Debrecen, Hungary  International Journal of Climatology  2005  Mobile  014  39  Middle  Analysis of the temperature distribution in Kumagaya city  Geographical Review of Japan  1964  Mobile  088  61  Middle  On relationships between heat island and sky view factor in the cities of Tama River basin, Japan  Atmospheric Environment  1986  Mobile  146  70  Middle  Characteristics of the urban heat island in the city of Salamanca, Spain  Atmosfera  2003  Mobile  123  84  Middle  Observing an urban heat island by bicycle  Weather  1998  Mobile  168  93  Middle  Urban, residential, and rural climate comparisons from mobile transects and fixed stations: Phoenix, Arizona  Journal of the ArizonaNevada Academy of Science  2006  Mobile  024  99  Middle  The urban "heat island"  Biometeorology  1967  Stationary  154  128  Bottom  Heat-island in Helsinki: A higher latitudes city  Japanese Journal of Biometeorology  2004  Mobile  104  160  Bottom  The heat island and land uses in Guadalajara  Serie Geografica  1994  Mobile  a  Out of 190 studies  146  Table 4.15 Studies in the literature sample with best control of confounding factors on heat island magnitude. ID code  Ranka  Tier  Study title  Source  Year  Survey method  076  1  Top  Studies of the development and thermal structure of the urban boundary layer in Uppsala  Report, Meteorological Institute (Uppsala)  1980  Stationary  058  1  Top  Study of the subarctic heat island at Fairbanks, Alaska  Report, Environmental Protection Agency  1978  Stationary  157  3  Top  Pseudovertical temperature profiles and the urban heat island measured by a temperature datalogger network in Phoenix, Arizona  Journal of Applied Meteorology  2005  Stationary  167  3  Top  Urban-rural contrasts of meteorological parameters in Lodz  Theoretical and Applied Climatology  2006  Stationary  129  5  Top  The relationship between heat island intensity and rural land coverage in Obuse, Nagano  Tenki  1999  Mobile  161  6  Top  Relation between heat island intensity and city size indices/urban canopy characteristics in settlements of Nagano basin, Japan  Geographical Review of Japan  2005  Stationary  126  8  Top  Temporal and spatial characteristics of the urban heat island of Lodz, Poland  Atmospheric Environment  1999  Stationary  001  10  Top  Climatological studies in Uppsala  Geographica  1951  Mobile  165  10  Top  Temporal dynamics of the urban heat island of Singapore  International Journal of Climatology  2006  Stationary  057  21  Middle  The urban heat island in dependence of different meteorological parameters  Arch. for Meteorology, Geophysics, and Bioclimatology  1978  Stationary  155  23  Middle  Intra-urban relationship between surface geometry and urban heat island: Review and new approach  Climate Research  2004  Mobile  156  23  Middle  The relationship between built-up areas and the spatial development of the mean maximum urban heat island in Debrecen, Hungary  International Journal of Climatology  2005  Mobile  144  23  Middle  Temperature cross-section features in an urban area  Atmospheric Research  2001  Mobile  163  28  Middle  Interactions between thermal advection in frontal zones and the urban heat island of Wroclaw, Poland  Theoretical and Applied Climatology  2005  Stationary  028  29  Middle  Urban temperature fields  WMO Technical Note  1970  Mobile  119  41  Middle  Some aspects of the three-dimensional heat island in Moscow  International Journal of Climatology  1997  Stationary  112  82  Middle  Influences of the city of Bucharest on weather and climate parameters  Atmospheric Environment  1999  Stationary  084  106  Bottom  Heat island of a city in snow-covered area  Tenki  1985  Stationary  038  107  Bottom  City size and the urban heat island  Atmospheric Environment  1973  Mobile  118  110  Bottom  Heat island development in Mexico City  Atmospheric Environment  1997  Stationary  147  113  Bottom  Mesoscale aspects of the urban heat island around New York City  Theoretical and Applied Climatology  2003  Stationary  030  113  Bottom  The form of the urban heat island in Hamilton, Canada  WMO Technical Note  1970  Mobile  a  Out of 190 studies  147  Table 4.16 Studies in the literature sample with weakest communication of definitions and procedures.  a b  Study title  Source  Year  Survey method  Bottom  An urban climate analysis of Graz and its significance for urban planning in the tributary valleys east of Graz (Austria)  Atmospheric Environment  1999  Mobile  128  Bottom  Heat-island in Helsinki: A higher latitudes city  Japanese Journal of Biometeorology  2004  Mobile  0.62  189  Bottom  Effects of urbanization on climate with special reference to temperature in the Kuala Lumpur– Petaling Jaya area, Malaysia  Tropical Ecology and Development  1980  Stationary  149  0.62  188  Bottom  Urban climate and air quality in Trier Germany  2003  Stationary  077  0.62  184  Bottom  Some physical features of heat and humidity islands at Delhi  International Journal of Biometeorology Mausam  1982  Mobile  111b  0.62  182  Bottom  A study of the urban heat island effect in Western Taiwan  Journal of Geographical Science  1995  Stationary  081  0.62  181  Bottom  Isothermal and isohyetal patterns at Delhi as a sequel of urbanization  Mausam  1984  Stationary  009  0.62  160  Bottom  The changing form of London's heat island  Geography  1961  Stationary  176  0.62  155  Bottom  Urban design factors influencing heat island intensity in high-rise highdensity environments of Hong Kong  Building and Environment  2007  Stationary  110b  0.62  113  Bottom  An analysis of urban heat islands in Tokyo and its environs: The comparison of cloudless nights with cloudy nights in autumn  Geographical Review of Japan  1994  Stationary  ID code  MDI  Ranka  Tier  127  0.85  182  154b  0.69  070  Out of 190 studies Foreign-language document  148  CHAPTER 5 A NEW CLASSIFICATION SYSTEM FOR URBAN HEAT ISLAND OBSERVATIONS 1 5.1 Methods Heat island investigators traditionally classify their field sites as dichotomously “urban” and “rural.”  This paradigm for site classification is at the core of the  methodological and philosophical framework behind urban climatology. The framework is so seductively simple, so fundamental that it aims to separate urban from non-urban effects on local climate (e.g., Lowry 1977), a distinction that has been instrumental in the overall development and understanding of urban climatology. However, results from the research synthesis show that method and communication in UHI literature have suffered critically by the popular use of “urban-rural” classification. This finding underscores the need for a more refined approach to site description and metadata exchange. Lowry (1977) began the process of dissecting the urban-rural dichotomy into descriptive parts for urban climatologists. Through that process came the notion of “preurban” temperatures and landscapes, which Lowry defined as representative not of an urban or rural space, but of a time before human disturbance to the landscape. Lowry allowed that temperatures from modern preurban space (i.e., land adjacent to the city yet still undisturbed) are reasonably analogous to those of preurban time. This was deemed acceptable because temperature records of preurban time in present-day cities are difficult to find. Lowry’s framework provides only the beginning of a more formal  1  A version of Chapter 5 has been submitted for publication: Stewart ID, Oke TR. 2011.  149  redefinition of urban and rural landscapes that is now needed in the study of urban climates and urban heat islands.  The remainder of this chapter builds on Lowry’s  framework and its role in defining UHI magnitude. The new classification is constructed in three steps: (1) taking stock of existing landscape classifications used by heat island investigators; (2) dividing the terms urban and rural into appropriately defined and scaled classes for landscape description and local temperature observation; and (3) differentiating these classes with observational temperature data and numerical modelling data.  5.1.1 Constructing the framework The purpose of classification is to establish order among phenomena. Order, in turn, is established by grouping phenomena with common properties into classes, which then reveal larger patterns and relations. In Geography, spatial uniformity of physical and/or cultural properties and patterns defines the concept of region. An example of a regional system in physical geography is the Koppen classification of world climates, which defines regions as macroscale surface areas that are relatively uniform in temperature, precipitation, and vegetation. The field site classification system for heat island observations will define region at much smaller scales, and will find uniformity in the surface processes and features that influence screen-height temperature. The greatest challenge in this process is to organize the surface features—buildings, trees, roads, and soils—into meaningful patterns, and to represent these patterns in a most effective way. Local-scale urban climate classification has several precedents in climate literature. These precedents have determined class membership by one or more surface climate criteria, as have regional systems like Koppen’s. Among the precedents are  150  general land-use and land-cover classifications, as well as urban climate and urban morphological systems, all of which are commonly used by heat island investigators to characterise their field sites. The first step toward a new climate classification is to weigh these systems and identify the relevant, restrictive, and advantageous features of each. None of the systems was designed specifically for classifying heat island field sites, and none claims to serve that purpose.  Therefore, what is taken as advantageous, or  restrictive, with these systems relates only to the aims of the new classification system.  5.1.1.1 Taking stock of urban climate classification systems Chandler (1965) was the first urban heat island investigator to develop a climatebased classification of the city. He classified the Greater London area into four local regions, each distinguished for its climate, physiography, and built morphology (Figure 5.1). The most distinct region was the city itself, having sharp boundaries in temperature, wind, and humidity with the surrounding country. Within the city, differences were less marked and boundaries more gradual, but evidence was sufficient to see four ―broad climatic regions‖:  (I) ―central‖ for high-density commercial areas, which were  consistently warmer than (II) ―inner suburban‖ and (III) ―outer suburban‖ areas of highand low-density residential and commercial development; and (IV) ―northern heights‖ for open spaces and low-density residential areas, which were consistently cooler than the rest of the city. All four regions were warmer than undeveloped rural areas. In 1978, following Chandler’s lead, Auer proposed a similar classification system for the city of St. Louis, USA. He distinguished ―meteorologically significant‖ land-use classes on the basis of vegetation and building characteristics in the St. Louis region. Altogether, 12 land-use types were associated with observed thermodynamic, kinematic,  151  Figure 5.1 Chandler’s classification of local climate regions in the city of London (I = Central; II = Inner suburban; III = Outer suburban; IV = Northern heights). (Source: Chandler, 1965).  and radiative ―anomalies‖ across St. Louis.  Meteorologically speaking, the most  anomalous classes (relative to the background rural climate) are ―heavy industrial,‖ ―commercial,‖ and ―compact residential,‖ each with an abundance of built land cover and human-made structures; the least anomalous classes are ―estate residential,‖ ―metropolitan natural,‖ and ―agricultural rural,‖ each with extensive grass, tree, and/or crop cover. Unlike Chandler’s classification, Auer’s is highly general and does not use location as a distinguishing class property. Ellefsen (1990/91) devised a system of neighbourhood-scale ―urban terrain zones‖ (UTZ) based on building morphology, street configuration, and construction materials of U.S. cities. His system characterises ten such cities, including Boston, 152  Denver, Seattle, and Pittsburgh.  Urban terrain zones were derived from empirical  measures of building structure (e.g., height, roof/wall area) and materials (brick, steel, glass).  An important differentiating property of Ellefsen’s classification system is  building ―type,‖ which subdivides into ―attached‖ and ―detached‖ forms. He identified 17 zones on this distinction alone, including ―redeveloped core area,‖ ―shopping centers,‖ and ―outer city‖ (Figure 5.2). Incorporating features of both Auer’s and Ellefsen’s systems, Oke (2004) designed a simple city-based climate classification scheme to improve siting of meteorological instruments in urban areas (Figure 5.3). His system divides urbanized terrain into discrete, homogenous regions called ―urban climate zones‖ (UCZ). The zones are distinguished by their surface structure, cover, metabolism, fabric, and overall  Figure 5.2 Ellefsen’s urban terrain zone (UTZ) for ―redeveloped core area‖ (Ellefsen, 1990/91).  153  Figure 5.3 Oke’s Urban Climate Zone (UCZ) classification scheme (Oke, 2004; by permission).  potential to ―disturb‖ the natural surface climate. Zones with greatest potential have large values for building-aspect ratios, impervious surface fractions, and aerodynamic roughness. Zones with least potential have small values for aspect ratios, impervious fractions, and roughness. Oke ranked and numbered his zones from UCZ1 (―intensely developed urban‖) to UCZ7 (―semi-rural‖). In addition to the schemes of Chandler, Auer, Ellefsen, and Oke, comprehensive  154  land-cover and land-use classification systems have also been used to classify heat island field sites. The U.S. National Land Cover Dataset (NLCD), for example, divides the coterminous United States into 16 land-cover classes (Homer et al., 2007). Twelve of the 16 classes are designated ―undeveloped‖ (i.e., rural), and four ―developed‖ (i.e., urban). Developed classes include ―low intensity,‖ ―medium intensity,‖ and ―high intensity‖; undeveloped classes include ―open water,‖ ―evergreen forest,‖ ―grassland,‖ ―woody wetlands,‖ and ―cultivated crops.‖ The system specifies impervious surface fraction and tree canopy coverage at 30-metre intervals across the U.S. The NLCD is a simplified and updated version of an earlier land-use and land-cover classification system that was developed by Anderson et al. (1976) for the U.S. Geological Survey.  That system  consists of 37 classes, 7 relating to urban land uses and 30 to rural land covers. In China, a similar land-use and land-cover coding system exists and has been used to classify field sites in that country (He et al., 2007). The National Land Cover Database (NLCD) of China consists of 10 satellite-based land-use classes, including ―urban,‖ ―rural inhabitable,‖ ―waste land,‖ ―dry land,‖ ―seashore,‖ ―forest,‖ ―paddy field,‖ and ―water field.‖ The spatial resolution of China’s NLCD is 1 km. Census population criteria are also used to support ―urban‖ and ―rural‖ designations.  Field sites in areas with populations exceeding an arbitrarily defined  threshold (e.g., 100,000) are deemed ―urban,‖ while those below a lower threshold (e.g., 10,000) are deemed ―rural‖ (e.g., Kulka et al., 1986). Other precedents for ―urban‖ and ―rural‖ classification include satellite-derived indices, such as the Normalized Difference Vegetation Index (NDVI) and night-light frequency metadata. The NDVI measures the vigor and density of green vegetation on Earth’s surface. Urban sites are assumed to be  155  less vegetated than rural sites and to have lower NDVI values (Henry et al., 1989). Night-light metadata are said to yield more accurate classifications of field sites than population criteria and the NDVI index, largely because they can differentiate the intensity of residential, industrial, or commercial activity surrounding a site (Owen et al., 1998). Night-light, vegetation, population, and land-cover and land-use criteria, as well as the more formal classifications of Chandler, Auer, Ellefsen, and Oke, together provide the structural framework for a new classification system that aims to redefine urban and rural landscapes. The relevant features amenable to this aim must be extracted from these previous classifications and transferred to the new system.  Chandler’s  classification of London, for example, expresses the urban landscape through heights, densities, and materials of buildings, as well as distributions of trees, parks, and open spaces. Anderson et al.’s land-cover classification of the USA is equally descriptive, with thoughtful and unambiguous class definitions. Auer’s classification of St. Louis transmits information clearly and concisely using oblique aerial photographs of urban and agricultural landscapes and a descriptive summary table of the entire system. Ellefsen’s urban terrain zones provide a complete inventory of building characteristics related to materials and height, as well as composite figures depicting aerial views and plan/profile sketches of each zone. His scheme outranks all others for its graphic zone templates, and for the wealth of information conveyed in a compact space. Oke’s Urban Climate Zone system accurately represents local-scale climates in the urban environment. Each climate zone is supported by empirical measures of building geometry, land cover, and aerodynamic roughness. His system portrays a generic regionalization of urban space,  156  and therefore functions far beyond any single city. National land-use and land-cover classification systems, like Anderson et al.’s for the USA, also have advantageous features. First, they are accessible and comprehensible to a broad range of users. These systems provide wide coverage of urban and rural landscapes across large regions, and can be applied at local, regional, or national scales. The classes are easily distinguishable and include a diversity of land covers, especially in rural environments. Population and satellite-derived metadata (e.g., NDVI, night-light intensity) are favoured for similar reasons: they are easily interpreted and fairly accessible at local and regional scales. Despite the many advantages of these classification systems for heat island observation, each is limited in key areas that make it unsuitable for that purpose (albeit a purpose for which the systems were never intended). These areas are (1) climatic relevance, (2) urban and rural representation, (3) nomenclature, and (4) origin and scope. First, only one of the systems incorporates into its class criteria the full set of surface properties that influences thermal climate. A full set should include properties of surface structure, cover, fabric, and metabolism. The other systems instead use subsets of surface properties, such that the climate regime or surface characterisation of each class is greatly simplified. The U.S. NLCD distinguishes its urban classes primarily by impervious surface fraction. Auer (1978) defines his land-use types primarily by vegetative fraction, with little consideration for surface exposure and soil wetness.  Satellite-derived  vegetation indices, night-light indices, and population metadata are all disadvantaged by class criteria that are surrogates for urbanization and therefore not directly related to the local and microscale surface properties that control thermal climate.  157  Cultural and  regional discrepancies in urban-rural threshold values for population and night-light frequency further undermine the appropriateness of these criteria for classifying field sites. As a surrogate for urbanization, land use similarly conveys little in the way of surface geometry or cover. Anderson et al.’s classification of the urban landscape is based entirely on land use, not surface form. ―Residential‖ classes, for instance, include contrasting geometries like high-density towers and low-density houses; ―commercial‖ includes highrise office buildings and lowrise shopping centers. Second, the reviewed classification systems are unsuitably balanced in their urban and rural representation of the landscape for heat island investigators. A system without proportional representation of city and country is not well suited to field site classification. Oke’s scheme, for example, has just 1 ―semi-rural‖ class among 7 climate zones. Auer’s system is similarly biased, with just 4 rural classes among 12 land-use types.  Chandler’s classification has no rural representation at all.  National land-  cover/land-use classification systems are biased in opposite favour to the rural environment: just 7 of 37 categories in Anderson et al.’s classification, and 1 of 10 in China’s NLCD, are urban-oriented. The urban detail in these national systems is not sufficient to delimit and characterise local and microscale surface climates in the city. Third, the reviewed systems are limited in nomenclature.  Class names and  definitions are open to misinterpretation by heat island investigators because they couch terms that vary with culture, time, and/or location.  References to ―suburban,‖  ―metropolitan,‖ ―semi-rural,‖ ―inner/outer,‖ and ―downtown‖ appear throughout the schemes of Chandler, Auer, and Oke, yet these references are culture and place specific and thus largely irrelevant to surface climate. Moreover, land uses like ―agricultural‖ and  158  ―institutional‖ can take many forms—compact, sparse; treed, open—that vary with time and region. A complete and objective system of class names and definitions does not yet exist in urban climate literature. Lastly, the classifications reviewed here are limited in origin and scope. Chandler’s system, for example, functions only in the original setting in which it was designed, namely London. Auer’s and Ellefsen’s systems are slightly more general but still function poorly outside the USA. The national land-cover classifications of China and the USA also function poorly beyond their borders. Oke’s UCZ scheme is the most universal of the reviewed classifications, yet its outlook inevitably favours the form and function of modern, developed cities. The use of Oke’s scheme in diverse cultural and economic regions, particularly of the developing world, is uncertain. Shantytowns and block-tower housing developments have no definitive place in the UCZ system, or in any of the reviewed system, yet these settlement types are characteristic of many of the world’s largest cities.  5.1.1.2 Classification criteria In a classic paper on the method, theory, and logic of classification, Grigg (1965) listed several criteria that all classification systems should meet. First, classifications must invoke a simple and logical nomenclature by which objects/areas can be named and described. Nomenclature is critical to a classification’s validity and acceptance. Second, the classification must facilitate information transfer among its users by associating individual objects/areas in the real world with an organized system of classes. Users of the system can then make comparative statements about the members belonging to each class. This criterion led Grigg to his third and most important feature of classification:  159  inductive generalization. If properly constructed, classification systems should simplify the objects/areas being studied, and thereafter promote theoretical statements about their properties and relations. In meeting these criteria, classification fulfills a pre-eminent role in science by generating hypotheses and guiding investigations (Haines-Young and Petch, 1986). To Grigg’s criteria, several others are added that relate specifically to a field site classification system for urban heat island observation. Many of these criteria were addressed earlier in the thesis. Given their theoretical nature, the criteria inherently overlap. 1. Accessible: The system must be functional and accessible to a broad spectrum of users, especially those who have limited resources to classify field sites or conduct sophisticated observations. 2. Comprehensible: The system must be manageable and easily understood. Too many classes makes the system awkward and difficult to decipher, while too few makes it overly simplistic and poorly defined. 3.  Objective:  The system must incorporate measurable and observable class  properties. These properties ensure that group members are classified and named consistently and accurately. 4. Inclusive: The system must be universal in scope so as to avoid regional or cultural biases in nomenclature, definition, and illustration. 5. Exhaustive: The system must ensure that all landscapes are assignable to one class. 6. Standardized: The system must be structured and coherent in its depiction and  160  quantification of the landscape so as to encourage universal recognition, acceptance, and adoption. Classes must be distinguished by standardized names, scales, symbols, definitions, and illustrations. 7. Relevant: The system must include class properties that are germane to surface thermal climate at micro and local scales. All class properties should relate directly to the system’s purpose of classifying field sites for heat island observation. 8. Reproducible: The system should yield the same classifications of the same (or similar) areas, regardless of who is using the system. Reliable and reproducible outcomes are assured if the classification system meets all preceding criteria.  5.1.1.3 Classification by logical division Scientific classification is essentially a process of definition. It begins with the division of a ―universe‖ class, which is then divided into subclasses (Black, 1952). The basis for division at each class level is a differentiating principle or property of theoretical interest, called a differentia. The differentia separate one class from another and require all items in a class, or subclass, to possess the same defining characteristics. The end product of logical division is a system of names for groups (or classes) showing regular properties. The universe class for heat island observation is ―landscape,‖ which Oxford English Dictionary defines as ―a tract of land with its distinguishing characteristics and features, especially considered as a product of modifying or shaping processes and agents.‖ The Dictionary further states that although the ―processes and agents‖ that shape the land are usually natural, the ―works of man‖ can be expressed through a  161  ―cultural‖ landscape.  For purposes of logical division, the dictionary meaning of  ―landscape‖ is modified to that of a local-scale tract of land (i.e., 100 m to 10 km in length) with physical and/or cultural characteristics that have been shaped by physical and/or cultural agents. The landscape universe is then divided according to properties that influence screen-height thermal climate, namely, surface morphology (object height and density) and land cover (pervious or impervious).  Surface morphology affects local climate  through its modification of airflow and the transport of heat in the air, while land cover modifies the albedo, moisture availability, and heating/cooling potential of the ground. Division of the landscape with respect to these properties generates 26 hypothetical groupings that gave the final classification its structural basis (Figure 5.4). Several groupings were considered highly improbable in the real world (e.g., close buildings on pervious cover), or logically unacceptable (e.g., close trees on impervious cover). These groupings were removed from the system, while new groupings were added to represent landscapes defined not by their morphological or land cover characteristics, but by building materials and anthropogenic heat emissions.  All eventual groupings were  quantified by their surface properties and assigned simple, concise names. References to location (e.g., suburban, core), function (agricultural, institutional), time (old, modern), and cultural norms (common, regular) of the landscape were avoided in most cases.  162  Figure 5.4 Logical division of the universe class by roughness objects (buildings, plants), object height (high, mid, low, nil), object density (O = open; C = close), and land cover (I = impervious; P = pervious). Sample derivative classes are shown at bottom. The close highrise class (bottom left) derives from landscape division into buildings of maximum height and density, and surrounding surfaces of impervious cover.  5.1.1.4 Data sources Qualitative data to adapt the class groupings to the real world were chosen from the urban design literature, which describes traditional, modern, colonial, socialist, African, Chinese, and Latin American cities, among many others (Brunn and Williams, 1983; O’Connor, 1983; Vance, 1990; Kostof, 1991; Potter and Lloyd-Evans, 1998). This literature gives qualitative attributes to urban form through expressions of ―fabric,‖ ―texture,‖ and ―morphology‖—the same expressions to which urban climatologists give  163  quantitative attributes. The common vernacular between urban climate and urban design literatures was especially helpful to assimilate regional urban form into the classification system, and to balance its temporal (old vs. modern) and spatial (core vs. periphery) representation. These data also supported culturally non-biased definitions for the class groupings. Additional sources of qualitative data were found in the classifications of Anderson et al. (1976), Auer (1978), Ellefsen (1990/91), and Oke (2004). Many features of these schemes were replicated in the new system. Personal visits to field sites in the heat island literature sample were equally valuable as sources of qualitative data (see section 5.1.1.5). Quantitative data to characterise the class groupings by their surface properties were selected from urban observational and numerical modelling literature. Measured and estimated values of geometric, thermal, radiative, metabolic, and land cover properties for urban and rural field sites worldwide were gathered. These sites include old city cores (e.g., Christen and Vogt, 2004), high-density suburbs (e.g., Kanda et al., 2005), low-density periurban regions (e.g., Offerle et al., 2005), traditional desert settlements (e.g., Ali-Toudert et al., 2005), lightweight housing settlements (e.g., Pearlmutter and Meir, 1995), medium-density residential neighbourhoods (e.g., Grimmond and Oke, 1995), light industrial sites (e.g., Nunez and Oke, 1976), industrial/refinery sites (e.g., Petersen, 1997), modern highrise districts (e.g., Voogt and Krayenhoff, 2006), shantytowns (e.g., Silva and Ribeiro, 2006), paddy fields (e.g., Susaki et al., 2007), orchards (e.g., Valancogne et al., 2000), croplands (e.g., Oguntoyinbo, 1970), and forests (Grimmond et al., 2000). Additional quantitative data were retrieved from the observational heat island  164  literature, which was of limited use because its field site descriptions are often brief. Among studies with complete reporting of site metadata are those of Uppsala, Sweden (Taesler, 1980); Lodz, Poland (Klysik and Fortuniak, 1999); Nagano, Japan (Sakakibara, 1999); Vancouver (Runnalls and Oke, 2000); Szeged, Hungary (Unger, 2004); Singapore (Chow and Roth, 2006); Colombo, Sri Lanka (Emmanual and Johansson, 2006); Phoenix, Arizona (Hedquist and Brazel, 2006); and Orlando, Florida (Yow and Carbone, 2006). Supplementary metadata were extracted from the classification systems of Anderson et al. (1976), Auer (1978), BMBau (1980), Ellefsen (1990/91), Theurer (1999), and Oke (2004). Lastly, general texts of physical climatology (e.g., Oke, 1987; Hartmann, 1994), remote sensing (e.g., Barrett and Curtis, 1992), and soil science (e.g., Sumner, 2000), as well as detailed reviews of empirical urban climate literature (e.g., Wieringa, 1993; Grimmond and Oke, 1999), provided useful data for estimating surface property values for each class grouping.  5.1.1.5 Field site visits From 2005 to 2010, site visits were made to cities and towns of Europe, North America, and East Asia (Table 5.1). The purpose of these visits was to observe the diversity of urban and rural field sites used in heat island literature, and to document their physical characteristics. A secondary aim was to assess investigators’ prevailing attitudes to the experimental design of heat island studies and to the selection and classification of field sites. The visits informed all aspects of the thesis, especially the synthesis and evaluation of heat island literature (Chapter 4), the construction of a field site classification system (this chapter), and the recommendations for improving heat island methodology (Chapter 6).  165  Table 5.1 Locations and dates of field site visits.  a  Corresponding studies in the heat island literature samplea [089] [151] [157] [168]  Location Phoenix, USA  Date September 2005  Uppsala, Sweden  June 2006  [001] [076]  Stockholm, Sweden  June 2006  …  Goteborg, Sweden  June 2006  [106]  Szeged, Hungary  June 2006  [115] [144] [155]  Wroclaw, Poland  June 2006  [163]  Lodz, Poland  June 2006  [126] [167]  Vienna, Austria  June 2006  [079]  Toronto, Canada  August 2006  [045]  Regina, Canada  August 2006  [139]  Sendai, Japan  November 2007  …  Tokyo, Japan  November 2007  [097] [162] [006]  Seoul, South Korea  November 2007  [086] [133] [159]  Nagano, Japan  November 2007  [161]  Obuse, Japan  November 2007  [129]  Hong Kong  May 2008  [176]  Vancouver, Canada  May 2010  [047] [053] [137]  See Appendix B for bibliographic information  Regions having an established tradition of heat island research were favoured for visits, as were cities having one or more reputable heat island studies in the published literature. The visits were initiated by contacting the lead investigator of a selected study, who was then requested to host such a visit. Visits usually began with a reconnaissance survey of the urban-rural study area, followed by collection of metadata (e.g., maps, sketches, photographs) at field sites of particular interest, and then discussion of the methodological challenges facing heat island investigation in that city or town. The metadata were later used to estimate surface property values of corresponding classes in the new field site classification system.  Investigators were questioned about their  rationale for site choice and their criteria for site classification. These questions and the  166  ensuing discussions often focused on sites near the outskirts of large cities—e.g., Tokyo, Seoul, Vienna—that are frequently used in heat island investigations but are problematic to classify. Finally, the field visits uncovered additional heat island studies (published and unpublished) that were not previously found through conventional literature searches, but that contained valuable metadata for the new classification system. Several days were spent in each city to complete these activities.  5.1.2 Thermal differentiation of landscape classes Each class grouping that was derived through logical division has a unique building morphology, tree geometry, and/or land cover. The extent to which these classes have unique thermal climates was assessed with observational and numerical modelling data.  Mobile observations are ideal for differentiating simultaneous  temperatures across many local-scale landscapes, while continuous stationary observations are ideal for differentiating the thermal responsiveness (as measured by diurnal temperature range—DTR) of two or more landscapes. Observational data were acquired from canopy layer heat island investigations in three representative cities of Europe, East Asia, and North America. Numerical models are advantageous because they offer control over surface and boundary layer conditions (e.g., wind speed, solar radiation, surface geometry), and they can generate temperatures for any weather and landscape type. Modelled data were generated from a simple mesoscale atmospheric boundary layer scheme, which was coupled to surface-atmosphere transfer schemes for built (urban) and natural (rural) surfaces.  When combined, the observational and  numerical modelling data provide a powerful test of the classification system and its landscape classes.  167  5.1.2.1 Observational approach Observational temperature datasets were obtained from heat island investigations in Uppsala, Sweden (59°N, 17°E); Nagano, Japan (36°N, 138°E); and Vancouver, Canada (49°N, 123°W). These locations were selected because they offer a broad range of city, country, and climatic settings that are characteristic of the heat island literature. Uppsala (pop. 100,000) is an old, compact European city with a dense core, flat building profile, and clear urban-rural divide. It has a humid continental climate with cold winters and warm summers. The mean annual temperature in Uppsala is 9°C, with monthly means of -4°C in February and 17°C in July. Nagano (pop. 300,000) is a medium-sized East Asian city with intensely mixed urban and agricultural land uses, both within and outside the city. Like Uppsala, Nagano has a humid continental climate with four distinct seasons. Its mean annual temperature is 12°C, with monthly means of 0°C in January and 25°C in August. Vancouver (pop. 2 million) is a large North American city with a modern highrise core, low-density residential neighbourhoods, and an extensive urban forest. Vancouver’s climate is maritime temperate, and is therefore less extreme than the climate of Uppsala and Nagano. Vancouver has a mean annual temperature of 10°C and monthly means of 4°C in February and 16°C in August. Mean annual anthropogenic heat flux density at city scale in all three locations is < 25 W m-2. At local and seasonal scales, winter anthropogenic heat flux densities in the city cores range from 50 to 150 W m-2.  The midlatitude climates of these locations are ideal for studying the  seasonal effects of snow, soil wetness, tree coverage, and anthropogenic heat flux density on local climate formation. Each of the three test locations has a well established tradition of urban climate  168  research, and is the site of many reputable heat island investigations. In addition to basic requirements concerning the precision of measurement systems, these investigations ensured that (1) the confounding effects of time, weather, relief, and surface wetness on UHI magnitude are controlled; (2) surface cover, structure, and human activity across the local-scale fetch of the observation sites are reasonably uniform; (3) site metadata are fully and properly documented; and (4) the number of observations is sufficiently large to support statistical inferences. (i) Uppsala, Sweden Observational temperature datasets for Uppsala were extracted from two distinguished heat island studies of that city. The first is Sundborg’s classic urban temperature study of 1951 (Figure 5.5). Sundborg conducted 207 automobile traverses across Uppsala between May 1948 and May 1949.  The traverse route covered 15  observation sites that were carefully selected by Sundborg to yield locally and  Figure 5.5 View toward central Uppsala, circa 1948 (Source: Lindberg, 1998).  169  statistically representative temperatures (see Figure 3.1). Covering a distance of less than 10 km, the traverse route began in the countryside 1.5 km southeast of Uppsala’s built-up area and progressed northwest toward the city centre, finishing at the Meteorological Institute southwest of the old core. Sundborg conducted his traverses under ―certain interesting or especially typical weather situations,‖ such that all seasons, synoptic weather types, and hours of day were represented. Sundborg used a platinum-wire resistance thermometer and double radiation shield attached to the edge of an automobile roof at 150 cm above ground (Figure 5.6). The precision of the thermometer, which was aspirated by the steady movement of the automobile, was ± 0.2 K. Sundborg strategically designed his traverse route to cross the southeast sector of the city because relief was especially slight, and topoclimatic influences on temperature were minimal.  Elevation differences between the highest and lowest points of his  traverse route were less than 40 m. Sundborg was confident that his route through the  Figure 5.6 Vehicle and thermometer used in Sundborg’s 1948–49 traverses of Uppsala (Source: Sundborg, 1951).  170  city was the best possible design for the purpose of assessing urban influences on local thermal climate. He deemed his chosen observation sites to be more representative at local scale than any other sites in and around the city. Raw temperatures were extracted from Sundborg`s original unpublished thesis, along with dates, times, locations, and wind and cloud conditions, for each of his 207 surveys. Sundborg had corrected these data for changes in regional temperature during the time of survey. The 207 traverses were filtered for those conducted (1) after sunset but before sunrise, (2) in relatively calm and clear weather, and (3) during 24-hour periods of no antecedent precipitation. This ensured that maximum thermal contrasts would be found along the traverse route and among its landscape classes, and that advection of thermal and moisture properties across landscape boundaries would be minimal. In meeting criterion (2), Oke’s (1998) weather factor (W) was used to identify nights with ideal conditions for local climate formation. Oke’s weather factor is ideally suited to filtering weather effects from heat island datasets because it accounts for the cumulative effects of cloud amount, cloud type, and wind speed on UHI magnitude. W expresses the degree to which wind and sky conditions reduce the heat island magnitude below its maximum value on any given night. It ranges from 0 (no heat island effect) to 1 (maximum heat island effect) and assumes wind speeds > 1 m s-1. W is derived from non-linear relations between UHI magnitude and cloud and wind conditions: W = u-1/2 (1-kn2) where u-1/2 is an empirical relation between heat island magnitude and regional wind speed (m s-1) (Oke, 1973), and 1-kn2 is the Bolz correction for net long-wave radiation under a given cloud type (k) and cloud amount (n) (Bolz, 1949). High winds combined  171  with low, thick (i.e., warm) cloud are associated with small W values and thus low UHI magnitudes (see Figure 2.6). Conversely, calm winds and thin, high-level (i.e., cold) cloud or clear skies are associated with large W values and maximum UHI magnitudes. A computed weather factor of ≥0.7 at the approximate time of Sundborg’s 207 traverses defined ―ideal‖ nights for thermal differentiation of landscapes classes. After a subset of ideal nights had been identified, the approximate areas of influence for the 15 observation sites along the traverse were quantified for their general surface properties. In dense urban surroundings, like the city centre, the area of influence was set to 100 m (radius); at sites with an open fetch, like the countryside, it was set to 200 m. Landscape classes that best matched the measured or estimated surface properties were then selected. The best, not the exact, representation of the Uppsala field sites was sought using all available metadata. Reconstructing Sundborg’s original traverse route and its observation sites was not possible with only the metadata contained in his original monograph. Archival research methods were therefore used alongside other strategies and tools for retrieving supplementary metadata. These metadata included historical photographs and land-use / land-cover maps (circa 1948), images from Google Earth / Maps, site photographs from personal visits to the city in 1998 and 2006, and site descriptions from modern investigations of Uppsala’s urban climate (e.g., Taesler, 1980).  In some areas of  Uppsala, the urban form has changed significantly from Sundborg’s time, but the old core and its close surroundings, as well as the countryside to the southeast, have changed little. Recent sources of metadata, like Google Earth, were used to reconstruct these segments of the traverse despite the temporal gap between Sundborg’s temperature measurements  172  and the Google Earth images. Other segments of the route that changed significantly in structure or cover were difficult or impossible to reconstruct. Mobile surveys were Sundborg’s only means to assess urban effects on temperature because a network of fixed meteorological stations did not exist in Uppsala in 1948. In 1976, however, Taesler (1980) designed such a network for a study with similar purpose to Sundborg’s. His network consisted of nine fixed stations across Uppsala, each located in areas of different building density and land cover. The city’s general urban structure in 1976 was similar to 1950, except that the built-up area had expanded considerably to the north and east (residential), and to the southeast (light industrial). Building heights and densities in the city centre increased marginally, giving the core a slightly more compact form than in 1950. At each of Taesler’s fixed stations, thermometric instruments were enclosed in standard meteorological screens 1.5 m above ground. At the residential and city centre sites, thermographs obtained semi-continuous 24-hour temperature readings. The white, louvered enclosures were not mechanically aspirated but were sufficiently exposed to natural airflow. Ventilation and radiation errors in the data are therefore considered minimal, i.e., 1 K or less, for calm nights (Taesler, 1980; Tanner, 1963). At the rural site and the Meteorological Institute southwest of the city centre, standard thermometers recorded hourly air temperatures in force-ventilated screens (Figure 5.7). Taesler regularly calibrated and corrected al