"Arts, Faculty of"@en . "Geography, Department of"@en . "DSpace"@en . "UBCV"@en . "Runnalls, Kathryn Elizabeth"@en . "2009-11-12T04:40:39Z"@en . "2002"@en . "Doctor of Philosophy - PhD"@en . "University of British Columbia"@en . "The quality of historical climate data is a fundamental consideration in climate\r\nchange research. This thesis considers the problem of microclimatic biases in observed\r\nsurface-level air temperatures, and the effect of these biases on the homogeneity of historical\r\ntemperature records. An extensive literature review describes causes of potential biases in air\r\ntemperature records, and demonstrates that climate station siting biases, in the form of\r\nmicroclimatic change in the surrounding environment, are particularly difficult to detect by\r\nconventional homogeneity analysis techniques. Such biases are therefore potentially still\r\npresent in many of the records used to monitor global climate change. The microclimatic\r\nprocesses responsible for thermal biases in temperature records are reviewed to demonstrate\r\nthe relevant physical principles. A new technique to detect microclimatic inhomogeneities in\r\ntemperature records is presented. The technique is based on the construction of time series of\r\ncooling ratios derived from nocturnal cooling at neighbouring climate stations calculated\r\nfrom daily maximum and minimum temperatures. The 'cooling ratio' is shown to be a\r\nparticularly sensitive measure of the relative microclimatic differences between neighbouring\r\nclimate stations, because larger-scale climatic influences common to both stations are\r\nremoved by the use of a ratio and, because the ratio is invariant in the mean with weather\r\nvariables such as wind and cloud. Discontinuities in the time series of cooling ratios indicate\r\nmicroclimatic change in one of the temperature records. Hurst rescaling (Hurst, 1951) is\r\napplied to several time series of cooling ratios, and is shown to be helpful in identifying\r\ndiscontinuities in the time series. The technique is tested on several Canadian historical\r\ntemperature records, and proven to effectively identify subtle microclimatic changes such as\r\nminor station relocations, vegetation growth, and encroachment of buildings and parking lots.\r\nResults of this technique are compared to those from other homogeneity assessment\r\ntechniques, and the cooling ratio approach is shown to be better suited to the detection of\r\nmicroclimatic biases. The technique is also shown to be useful in the assessment of the\r\nhomogeneity of urban temperature records. Finally, since the research highlights the\r\nsignificance of having detailed station metadata, recommendations are made to improve the\r\nutility of such records, along with suggestions for future research."@en . "https://circle.library.ubc.ca/rest/handle/2429/14823?expand=metadata"@en . "17354649 bytes"@en . "application/pdf"@en . "DETECTING MICROCLIMATIC BIASES IN HISTORICAL TEMPERATURE RECORDS by K A T H R Y N ELIZABETH R U N N A L L S B.Sc., McMaster University, 1993 M . S c , University of British Columbia, 1995 A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF THE REQUIREMENTS FOR THE D E G R E E OF DOCTOR OF PHILOSOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES (Department of Geography) We accept this thesis as conforming To the required standard THE UNIVERSITY OF BRITISH C O L U M B I A December 2002 \u00C2\u00A9 Kathryn Elizabeth Runnalls, 2002 In presenting t h i s thesis i n p a r t i a l f u l f i l m e n t of the requirements for an advanced degree at the University of B r i t i s h Columbia, I agree that the L i b r a r y s h a l l make i t f r e e l y a v a i l a b l e f o r reference and study. I further agree that permission for extensive copying of t h i s thesis for s c h o l a r l y purposes may be granted by the head of my department or by h i s or her representatives. I t i s understood that copying or p u b l i c a t i o n of t h i s thesis for f i n a n c i a l gain s h a l l not be allowed without my written permission. The U n i v e r s i t y of B r i t i s h Columbia Vancouver, Canada Department of Jji 0~; J CL /, ABSTRACT The quality of historical climate data is a fundamental consideration in climate change research. This thesis considers the problem of microclimatic biases in observed surface-level air temperatures, and the effect of these biases on the homogeneity of historical temperature records. An extensive literature review describes causes of potential biases in air temperature records, and demonstrates that climate station siting biases, in the form of microclimatic change in the surrounding environment, are particularly difficult to detect by conventional homogeneity analysis techniques. Such biases are therefore potentially still present in many of the records used to monitor global climate change. The microclimatic processes responsible for thermal biases in temperature records are reviewed to demonstrate the relevant physical principles. A new technique to detect microclimatic inhomogeneities in temperature records is presented. The technique is based on the construction of time series of cooling ratios derived from nocturnal cooling at neighbouring climate stations calculated from daily maximum and minimum temperatures. The 'cooling ratio' is shown to be a particularly sensitive measure of the relative microclimatic differences between neighbouring climate stations, because larger-scale climatic influences common to both stations are removed by the use of a ratio and, because the ratio is invariant in the mean with weather variables such as wind and cloud. Discontinuities in the time series of cooling ratios indicate microclimatic change in one of the temperature records. Hurst rescaling (Hurst, 1951) is applied to several time series of cooling ratios, and is shown to be helpful in identifying discontinuities in the time series. The technique is tested on several Canadian historical temperature records, and proven to effectively identify subtle microclimatic changes such as minor station relocations, vegetation growth, and encroachment of buildings and parking lots. Results of this technique are compared to those from other homogeneity assessment techniques, and the cooling ratio approach is shown to be better suited to the detection of microclimatic biases. The technique is also shown to be useful in the assessment of the homogeneity of urban temperature records. Finally, since the research highlights the significance of having detailed station metadata, recommendations are made to improve the utility of such records, along with suggestions for future research. 11 TABLE OF CONTENTS Abstract i i Table of contents i i i List of tables vi List of figures vii Acknowledgments ix Chapter 1 Introduction 1 Chapter 2 Representativeness, homogeneity and the climatological use of screen-level air temperatures: background and motivation for the study 4 2.0 Introduction 4 2.1 Representativeness 4 2.1.1 Definition 4 2.1.2 Official guidelines for making representative measurements 7 2.2 Homogeneity 11 2.2.1 Definition 11 2.2.2 Sources of inhomogeneities in air temperature records 12 i) Methods of observing and data handling 12 ii) Instrument biases 14 iii) Thermometer screen errors 15 a) Type of screen 15 b) Placement of thermometer 20 iv) Siting biases 20 a) Station moves 21 b) Station surroundings 22 v) Urbanization 24 2.3 Techniques for the detection of inhomogeneities in temperature records 31 2.4 Implications for analysis of long-term temperature records 34 2.5 Summary of points providing motivation for the present thesis 37 Chapter 3 The microclimate of climate stations: the physical basis of siting biases in air temperature observations 40 3.1 Introduction 40 3.2 A surface energy balance approach 41 3.3 Radiative processes 43 3.3.1 Shortwave radiant energy 43 i) Shading 43 ii) Albedo 46 iii 3.3.2 Longwave radiant energy 48 i) Skyview factor 48 ii) Surface emissivity 49 3.4 Surface thermal properties 49 3.5 Convective processes 53 (i) Evaporation 53 (ii) Roughness and shelter effects 57 (iii) Advection 60 (iv) Cold air drainage 64 3.6 Summary of typical climate station siting biases 66 Chapter 4 Cooling ratios as a tool for homogeneity assessment: rationale and development of a technique 69 4.1 Introduction 69 4.2 Definitions 4.3 The nature of the cooling ratio: characteristics and behaviour 72 4.3.1 Premises underlying the usefulness of cooling ratios 72 4.3.2 Illustrative examples of the nature of cooling ratios 72 4.3.3 Effect of weather on cooling ratio variability 73 i) Temperature data 73 ii) The weather factor 73 4.3.4 Seasonal variations of cooling ratios 80 4.3.5 Role of soil moisture changes in cooling ratio variability 82 4.3.6 Relationship between absolute temperature and cooling ratios 83 4.4 Optimal determination of the cooling ratio 88 4.5 Summary of potential advantages of cooling ratios for homogeneity analysis 90 Chapter 5 Analysis of time series of cooling ratios: application of Hurst rescaling 92 5.1 Introduction 92 5.2 Hurst rescaling: background 92 5.3 Analysis and interpretation of rescaled cooling ratios 94 5.3.1 Example 1: LadnenSteveston 96 5.3.2 Example 2: Edmonton Municipal Airport: Edmonton International Airport 101 i) Effect of seasonal variations 101 ii) Identification of cooling regimes in the rescaled trace 103 5.3.3 Example 3: Woodbend: Ellerslie 115 5.3.4 Example 4: Regina A; Midale 122 5.4 The use of composite reference series 133 5.5 Comparison of Hurst rescaling results to other homogeneity analyses 137 iv 5.5.1 Double mass analysis and Parallel CUSUMS 137 5.5.2 Multiple linear regression 140 5.6 Summary of findings 144 Chapter 6 Application of cooling ratio analysis for homogeneity testing of urban temperature records 146 6.1 Introduction 146 6.2 Toronto (Bloor Street) 147 6.3 Montreal (McGill) 155 6.4 Vancouver 158 6.5 Edmonton 161 6.6 Summary 162 Chapter 7 Summary and conclusions 166 7.1 Summary of findings 166 7.2 Implications for the climatological use of historical temperature records 170 7.3 Recommendations for climate station meta-data requirements 172 7.4 Recommendations for future research 176 Bibliography 179 Appendix 1: Observations of microscale spatial variability of screen-level air temperatures: thermal influences of surface features 188 A 1.1 Introduction 188 A 1.2 Field site, instrumentation and methods 188 Al.2.1 Intercomparison of temperature sensors 192 A 1.3 Effect of a gravel road on surrounding air temperatures 193 A 1.4 Source Area Model (SAM) results 199 A l .5 Effect of a small building on surrounding air temperatures 202 A 1.5.1 Wall and ground surface temperatures 205 A 1.5.2 Air temperatures 206 Al.5.3 Discussion 209 A 1.6 Summary 212 Appendix 2: Algorithm for calculating cooling ratios and Hurst rescaling for homogeneity analysis of temperature records 214 A . Creation of a cooling ratio time series 214 B. Hurst rescaling of the cooing ratio time series 215 v LIST OF TABLES Table Page 3.1 Surface albedo values 46 3.2 Surface thermal admittance values 50 3.3 Summary of common climate station siting biases and their effect on daily temperatures 66 4.1 Comparison of statistics for subsets of YVR:Steveston cooling ratios 89 5.1 Station history for Ladner B C 96 5.2 Station history for Steveston, B C 97 5.3 Station history for Edmonton Municipal Airport 111 5.4 Station history for Edmonton International Airport 112 5.5 Station history for Woodbend, A B 121 5.6 Station history for Regina International Airport 131 5.7 Station history for Midale, SK 132 5.8 Statistical measures of cooling ratio time series analyzed 136 6.1 Station history for Toronto (Bloor Street), ON 148 6.2 Station history for Guelph, ON 148 6.3 Station history for Woodbridge, ON 149 6.4 Station history for Montreal (McGill), PQ 158 6.5 Station history for St. Hubert Airport, PQ 158 6.6 Statistical measures of urban records analyzed 162 vi LIST OF FIGURES Figure Page 4.1 Relation between cooling magnitude at Steveston and weather 75 4.2 (a) Relation between YVR:Steveston cooling ratio and weather 77 (b) Relation between Ladner:Steveston cooling ratio and weather 77 4.3 Relation between YVR:Steveston cooling ratio and cooling magnitude 79 4.4 Seasonal variation of LadnenSteveston and YVR:Steveston cooling ratios 81 4.5 Seasonal variation of Regina A.:Regina C D A cooling ratios 81 4.6 Variation of summer cooling ratio with precipitation (YVR and Ladner cf Steveston) 84 4.7 (a) Time series of daily cooling ratio for Y V R : Steveston 86 (b) Time series of daily cooling ratios for Ladner:Steveston 86 4.8 Relation between August cooling ration and mean maximum daily temperatures 87 5.1 Site photographs for Ladner, 1963 98 5.2 Site photographs for Steveston, 1959 99 5.3 Cooling ratios and Hurst rescaled trace for LadnenSteveston 100 5.4 Comparison of Hurst rescaled traces calculated from monthly and annual data for Edmonton Municipal:Edmonton International 102 5.5 Cooling ratios and Hurst rescaled trace for Edmonton Municipal: Edmonton International 104 5.6 (a) Site photographs for Edmonton Municipal Airport, 1956 105 (b) Site photographs for Edmonton Municipal Airport, 1967 106 (c) Site photographs for Edmonton Municipal Airport, 1973 107 5.7 (a) Site photographs for Edmonton International Airport, 1959 108 (b) Site photographs for Edmonton International Airport, 1979 109 (c) Site photographs for Edmonton International Airport, 1988 110 5.8 Site photographs for Ellerslie, 1981 116 5.9 (a) Site photographs for Woodbend, 1973 117 (b) Site photographs for Woodbend, 1997 118 5.10 Woodbend:Ellerslie cooling ratio time series and Hurst rescaled trace 119 5.11 (a) Site photographs for Regina Airport, 1964 124 (b) Site photographs for Regina Airport, 1983 125 (c) Site photographs for Regina Airport, 1993 126 5.12 (a) Site photographs for Midale, 1934 128 (b) Site photographs for Midale, 1967 129 (c) Site photographs for Midale, 1986 130 5.13 (a) Regina A: Midale cooling ratio and Hurst rescaled trace 1923-1991 134 (b) Regina A:Midale Hurst rescaled trace 1963-1991 134 5.14 Regina A: composite reference series cooling ratio and Hurst rescaled trace 135 5.15 (a) Double mass curve for Regina A and Midale cooling magnitudes 139 (b)Parallel C U S U M plots for Regina A and Midale 139 vii 6.1 (a) Cooling ratios and Hurst rescaled trace for Toronto (Bloor Street): Guelph 150 (b) Hurst rescaled trace for Toronto (Bloor Street):Guelph 150 6.2 Toronto:Guelph cooling ratio and Hurst rescaled trace, linear 152 trend removed 6.3 (a) Cooling ratio and Hurst rescaled trace for Toronto:Woodbridge 154 (b) Hurst rescaled trace for Toronto:Woodbridge 154 6.4 Cooling ratio time series and Hurst rescaled trace for Montreal (McGill):St. Hubert A . 157 6.5 Vancouver PMO:Steveston cooling ratio and rescaled trace 161 A l . l (a) & (b) field site photographs 190 A1.2 Instrumentation photographs 191 A1.3 Meteorological conditions preceding measurement period, August 19, 1998 194 A1.4 Surface and air temperature differences for road and up- and down-wind fields. August 19-20, 1998 195 A1.5 Meteorological conditions during observation period 196 A1.6 Relationship between temperature differences and wind speed and direction 198 A1.7 Modeled source areas for temperature sensor 201 A1.8 Meteorological conditions during building observation period 203 A1.9 Plan view of air temperature variations in vicinity of building 207 A L I O Typical flow patterns around a solid barrier 211 viii ACKNOWLEDGEMENTS I am grateful to the following individuals and organizations for their contributions to this work. In particular, I thank my advisor and mentor, Professor T.R. Oke, for his support, guidance, and enthusiasm for my research, which have been a great motivation. The members of my research committee, Professors T.A. Black, M . Church and I.G. McKendry, have also been supportive, providing helpful advice at many stages of my programme. Thanks are due to the staff of the U B C Geography Department, particularly Vincent Kujala, who has rescued many hard-drives and files, and Sandy Lapsky, who has taken care of many details behind the scenes. Claybrook Holsteins of New Liskeard, Ontario generously provided the field site for this research, and Erica Ellis assisted with field work. Canadian climate data used in this thesis, and access to the climate station history archives, were provided by the Meteorological Service of Canada. Anna Deptuch-Stapf of Climate Services, and Lucie Vincent of the Climate Research Branch were particularly generous with their time and expertise, and their assistance is gratefully acknowledged. The Natural Sciences and Engineering Research Council of Canada provided funding in the form of a postgraduate scholarship and funding for T.R. Oke's research. Teaching and research assistantships from the Department of Geography are also acknowledged. Finally, I am grateful for all the encouragement I have received from friends and family, and especially to Brett and Brendan for their unwavering support. ix Chapter 1 INTRODUCTION \"Climate change is perhaps the most pressing and urgent environmental issue on the world's agenda.\" (NERC,1998). Over the last few decades, few scientific issues have attracted as much scientific and societal interest as global climate change. Scientific publications, conference proceedings, and the number of scientists contributing to all aspects of climate change research have grown exponentially since 1970, with current annual costs of this research estimated to be 3 billion U.S. dollars (Stanhill, 2001). With the diversity of interested parties ranging from insurance companies concerned about rising costs due to an increased frequency of extreme events such as floods, droughts and hurricanes, to ecologists concerned that species of flora and fauna may face extinction in a warmer world, it is not surprising that there is strong public support for funding climate change research encompassing the full spectrum of cause and effect. Yet, a question that persistently challenges the relevancy of this research is \"do we have proof that the climate is in fact changing?\" The most recent report by the Intergovernmental Panel on Climate Change (2001) claims there is now 'unequivocal' proof that Earth is getting warmer, citing evidence such as rising sea level, retreating glaciers, melting sea ice and changes to global circulation patterns (such as the increased frequency and intensity of warm-phase ENSO events). The IPCC also states that the instrumental temperature record shows \"the global average surface temperature has increased by 0.6 \u00C2\u00B1 0.2 \u00C2\u00B0C since the late 19 th Century.\" As a direct measure of a relevant climate variable, and with records extending back through the same period in which atmospheric concentrations of greenhouse gases have 1 increased, historical air temperature records are potentially the most useful of all forms of evidence to verify the nature and magnitude of climate change. Yet, this 'proof remains the subject of much debate. Indeed, when statements like that above \u00E2\u0080\u0094 implying that global temperature change can be determined to a certainty of two-tenths of a degree\u00E2\u0080\u0094 are paired with the IPCC (1996) statement that \"the balance of evidence suggests that there is a discernible human influence on the global climate,\" there is an inevitable backlash from those whose economic interests are threatened by the implications of these statements. While the process of questioning research findings is always necessary in order to advance understanding and improve research methodologies, curiously, very little of the opposing view regarding the accuracy of global temperature records makes its way into peer reviewed scientific media. Instead, much of the dissension appears in 'alternative' publications, such as Internet websites, where it can be easily dismissed as speculative or serving a particular agenda. This is an unfortunate situation, because many of the concerns of these self-described 'skeptics' may, in fact, be well founded. One need only consider the possible sources of error involved in measuring air temperatures, from observer errors and data handling, instrument errors, exposure and siting biases, to issues of spatial and temporal representativeness, in order to begin to question how it is possible to measure 'global' temperature change to a precision of 0.2\u00C2\u00B0C. This is not to say that errors are not recognized, and to a certain extent, 'corrected', before compiling global temperature data sets. There is an abundance of literature on this very subject (see Chapter 2). In fact, the compilers of large climate data bases such as the United States Historical Climatology Network (USHCN) do recognize that \"no station is completely free of changes that could possibly affect its instrumental record\" (Easterling et 2 al., 1999). Rather, the problem arises when dealing with thousands of individual records (the U S H C N contains over 1200 stations, the Global H C N : 7000), it is virtually impossible to identify every possible error, in every single record. Thus, it is only reasonable to be concerned that global temperature trends may have been calculated from flawed data. One potential source of bias in historical land-surface air temperature records that is particularly difficult to detect, often overlooked, but most importantly, extremely pervasive, is the microclimatic setting of the climate station itself. The purpose of this thesis is to examine these microclimatic biases in historical temperature records. Firstly, literature is reviewed to demonstrate that such biases are both significant, and underrecognized. Secondly, the physically relevant microclimatic processes responsible for creating thermal biases are discussed, with reference to studies that have described microscale temperature variability and associated physical controls, including original observations described in Appendix 1. Thirdly, a new technique is developed that is specifically designed to detect microclimatic, or siting biases in climatological temperature records. The technique is tested on a number of records to illustrate its utility for detecting these biases in a variety of settings, including urban environments. Finally, recommendations are made for improving climate station meta-data information that proves to be crucial to the identification of siting biases, and suggestions for future research are made. It is hoped that this work will begin the process of a more rational, scientifically based, but much needed critical discussion of the quality of global temperature records, so that necessary action can be taken to improve both our understanding of past climate change, and our ability to monitor it in the future. 3 Chapter 2 REPRESENTATIVENESS, HOMOGENEITY, A N D THE CLIMATOLOGICAL USE OF SCREEN-LEVEL AIR TEMPERATURES: BACKGROUND A N D MOTIVATION FOR THE STUDY 2.0 Introduction The concepts of 'representativeness' and 'homogeneity' are central to the interpretation of measured climate variables. This chapter seeks to define and discuss these concepts in the context of surface air temperature measurements. Sources of inhomogeneities in historical temperature records, methods used in their detection, and current views on the quality of 'homogeneity adjusted' climate records are reviewed and used to highlight points to be investigated in the thesis research. 2.1 Representativeness 2.1.1 Definition Although air temperatures are generally assumed to be less spatially variable than, for example, precipitation, obtaining a representative air temperature is not a simple task, particularly when the objective is to characterize global climate using data collected at a number of individual sites. Ideally, the representativeness of a potential measurement site should be assessed before observations are made in order to ensure that the observations adequately represent what they are intended to. Yet despite being a fundamental consideration in the proper measurement and interpretation of climate variables, representativeness is often an afterthought that occurs when the observations don't match with expectations. In some cases, this is unavoidable, such as when data collected for one 4 purpose are analyzed for an entirely different one, or when observations from the desired location do not exist and data from the next closest site must be used. However, the greatest problem arises from the fact that 'representativeness' is an extremely difficult concept to define. Ehinger (1993) notes that representativeness is ambiguous because it has both quantitative and qualitative aspects; and it applies to both sites and measurements. Wieringa (1997) states Representativity is not a universal quality like accuracy, not a fixed attribute of some (well executed) observation. Rather it depends both on the properties of the parameter... and on the application scales. Further he notes, Representativity of an observation means, that its exposure, instrumentation (if any) and sampling procedure are well matched to obtain a reliable average value of the observed parameter in a pre-determined space, which we want to represent. (Wieringa, 1997). In an attempt to promote more quantitative approaches to representativeness assessment, Nappo et al. (1982) define representativeness as The extent to which a set of measurements taken in a space-time domain reflects the actual conditions in the same or different space-time domain taken on a scale appropriate for a specific application. Statistical techniques are sometimes used in an attempt to quantify representativeness objectively by examining the spatial variability of observations in a given network (Zemel and Lomas 1976; Nappo et al, 1982). Such techniques are useful to identify the optimal spacing between stations for particular purposes (e.g. frost prediction in agriculture; Zemel 5 and Lomas 1976), however, they are not particularly useful to judge the representativeness of data being used for a purpose other than originally intended. Ehinger (1993) also suggests use of a more quantitative concept, 'area of representativeness', which is 'the land area, expressed in square metres, within which the values measured are identical to those given by the reference instrument'. He notes that this measure will vary from one parameter to another, with the kind of weather and with the seasons, and it must take account of topographical features near the site. Ehinger fails to suggest, however, how these multiple 'areas of representativeness' can be determined. As Nappo et al (1982) note, 'representativeness is and will remain a value judgment.' Pielke et al. (2000) and Pielke et al. (2002) examined the spatial representativeness of individual climate stations by comparing air temperature records from a number of sites in a relatively homogenous landscape. They found 'enormous' spatial and temporal variability, making it impossible to identify a regionally 'representative' climate station. They therefore caution against the use of single station data to infer regional climate, or the converse inference of site specific climate change predictions from coarse-scale general circulation models. Clearly, much of the current research investigating global climate change could benefit from a greater awareness of the importance of representativeness in interpreting climate data. An illustration of the importance of ensuring that measurements are \"well matched\" to what \"we want to represent\" (Wieringa, 1997) comes from the study of Bonan, (2000). This study is an attempt to determine the regional climate changes that have occured as a result of widespread replacement of natural forests with croplands over the last century in the United States. Climate trends in predominantly forested regions (Northeast US) were 6 compared to trends in predominantly agricultural regions (Midwest US). Stations were classified as either forest or cropland according to the predominant land cover in a 3 km x 3 km grid centred on the climate station. However, in an attempt to avoid microclimate biases, only stations that were classified as \"nonvegetated,\" \"open farmland, grasslands\", or \"small town less than 1000 population\" within 100 m of the station were used. That is, climate stations having \"forest\" land cover within 100 m were expressly excluded from the study. By selecting stations in this manner, Bonan (2000) contends that the stations represent 'regional' climate, yet it is unreasonable to assume that a climate station in a small town accurately portrays either forest or cropland climate. If the potential for microclimatic biases are not addressed in these types of analyses, conclusions drawn from them may be of limited value. 2.1.2 Official guidelines for making representative measurements Concerns about making representative temperature measurements are almost as old as the historical records themselves. The following quote appeared in the opening of Symons's Monthly Meteorological Magazine, October 1868: \"...I found that scarcely any observations had been made with thermometers placed under exactly similar circumstances, and without which unity no deductions can be drawn with any claim to accuracy. Some thermometers faced the north, some the north-east, some the north-west, &c,&c. Some were from three to five feet from the ground, some ten to twenty; some were embowered, some placed in a box, some sheltered by a high house or wall, some by a low wall, or by palings, some touching a wall, and others distant from it; some were in a angle of a high building (cool as a cellar), some exposed to the sun's rays either morning or evening (Lawson, 1843).\" 7 Symons's concern for this 'perpetuation of anarchy' was the motivation for the Strathfield Turgiss experiments, an elaborate comparison of 13 different thermometer exposures, the results of which led to the adoption of the Stevenson Screen as a standard by the Royal Meteorological Society in 1875. Other recommendations made at the same time included placement of the thermometer screen a minimum of 10 feet from any wall, and the preparations of ground plans at all stations showing the positions of the instruments and all surrounding objects (Meteorological Society, 1875). A standard measurement height of 4 feet above ground was apparently common practice by that time as well. Shortly thereafter, the Stevenson screen became a standard in many countries around the world, and remains so to this day. (See Sparks, 1972, for a review of different exposures in use). Despite this apparent 'uniformity' of exposure, in many respects the 'perpetuation of anarchy' which concerned Symons, still persists today, and continues to pose problems in the analysis of historical temperature records. Even the modern day W M O Guidelines for Siting of Meteorological Instruments (which are also the basis for most national meteorological service guidelines) place too much reliance on the experience and good sense of the station installers and operators in the subjective interpretation of representativeness. For example: For general meteorological work, the observed temperature should be representative of the free air conditions surrounding the station over as large an area as possible, at a height of between 1.25 and 2 m above ground level....The best site for the measurements is, therefore, over level ground, freely exposed to sunshine and wind and not shielded by, or close to, trees, buildings and other obstructions (WMO, 1996). Take [temperature] measurement under a standard shading apparatus 2 m aboveground. Ground to be covered with regularly mowed grass. Thermometer to be installed far from heat sources or water 8 bodies, unless these are \"representative\" of the region. Distance to obstacles: 4 to 10 times their height (Ehinger, 1993). For automatic weather stations, W M O guidelines suggest \"Location on ground typical of surrounding terrain. No obstruction should be closer than 4 times the height of that obstruction\" (WMO, 1997). Canadian guidelines suggest that temperature measurements at a distance of 2 times the height of an obstruction are acceptable (AES, 1992). Some of the most specific and quantitative guidelines come from the French Meteorological Service. For example, recommendations include a distance of more than 100 m from any artificial heating source or body of water, and that such surfaces should not occupy more than 10% of the area subtended by a 100 m radius, or 5% of 10-30 m radius, or any area within 10 m (Leroy, 1998; Meteo France, 1997). While such moves towards more exact specifications suggest that significant improvements in instrument siting can be expected for recently installed and future climate stations, concerns about the representativeness of historical records remain. Despite the fact that many historical temperature records come from stations which are assumed to comply with guidelines for representative measurements, as previously discussed, representativeness can only be judged with respect to a specific purpose, or application. As Wieringa (1997) points out, W M O guidelines for representativeness were originally based on the ability to make accurate weather forecasts, and therefore, the stations are not necessarily ideally sited to give a representative record of the surrounding climate. Thus, it is essential to re-examine the representativeness of such records before attempting to interpret the data for a different purpose. 9 In fact, the existence of guidelines for representative measurements is a double-edged sword from the perspective of climatological analysis of historical temperature records. On the one hand, many 'official' stations do not comply absolutely with the guidelines. The guidelines describe an 'ideal' site, but often the ideals compete with practical considerations, such as costs, accessibility, and security. In other cases, initial compliance may have occurred, but changes around the station (e.g. urbanization) may have changed the representativity over time. Unfortunately, ancillary information to document aspects of the station's surroundings is often non-existent, making assessment of representativeness after the fact, an impossibility. On the other hand, if only the records from stations proven to conform to the guidelines for representativeness were used in climatological analyses, we would then have a historical record of climate over 'flat, regularly mown grass,' but as the earth is composed of many surface types, other than 'flat, regularly mown grass', such records cannot provide a representative 'global average.' Clearly, it is desirable to obtain temperature records from all kinds of different environments, including urban areas, forests, and mountainous regions, despite the inherent difficulties in making representative measurements there. The central focus of this thesis is the problem that arises when the representativeness of individual climate stations changes, abruptly or gradually, and then temperature records from these stations are used to describe regional or global scale climate change. Thus, a change in representativeness, is an important consideration in the assessment of climate data homogeneity, which is the subject of the following section. 10 2.2 Homogeneity 2.2.1 Definition Attempts to identify temporal changes in climatological time series are limited by problems of homogeneity of the records. Conrad and Pollack (1962) define a homogeneous time series as one in which variations are caused only by weather and climate. Since it is almost impossible to assess the homogeneity of an individual time series, the concept of 'relative homogeneity' is also important. A climatological series is relatively homogeneous with respect to a synchronous series at another place if the differences (or ratios) of pairs of homologous averages constitute a series of random numbers that satisfies the law of errors (Conrad and Pollack, 1962). In other words, true regional- to global-scale climatic changes should be reflected in neighbouring station records in the same way, whereas differences between stations must be due to peculiarities at one of the individual sites. In the absence of good quality information (meta-data) about a climate station's operations or changes in its surroundings, relative homogeneity assessment is often the only means of sorting out 'real' and 'artificial' components of a climate record. The longer a station has been in operation, the more likely the possibility that it has been subject to some discontinuity, or non-climatic influence, in its observed record. For example, gradual changes in instrument calibration over time, replacement of old, or broken instruments, 'weathering' of the thermometer screen followed by repainting, and changes around a climate station site all contribute to apparent changes in the record, which could potentially augment, offset, or completely obscure 'real' changes in the climate itself. Thus the ability to detect inhomogeneities is crucial to climate studies attempting to infer climate change from instrumental records. 11 Mitchell (1953) presented a comprehensive framework for the classification of potential causes of secular changes in surface-temperature records. He classified potential causes of temperature change into two groups: real and apparent. The 'real' category is further classified into local effects, direct climatic effects, and other 'indirect' climatic effects. Although Mitchell's (1953) work pre-dates present-day concerns about the detection of 'Global Warming', the issues raised are still relevant to modern analyses. Many 'non-climatic' factors causing discontinuities, or inhomogeneities, have been addressed in recent attempts to compile high quality global databases. The following discussion is not an attempt to provide a comprehensive review or critique of all of the work which has been done to identify inhomogeneities in climate records, but rather to update Mitchell's (1953) overview with more recent work on this topic. It also identifies crucial omissions in past considerations of homogeneity analyses, and these will form the basis of the present thesis. The discussion is structured to consider potential biases in temperature records at increasing scales from the measurements and instruments themselves, up to the influence of the surrounding environment. Thus, the focus of the following discussion is to clarify the nature and cause of biases in temperature observations, rather than the purely statistical manifestation of these biases in the resulting temperature records. 2.2.2 Sources of inhomogeneities in air temperature records i) Methods of observing and data handling Errors in observing, recording, and processing data are inevitable. Many of these errors are often caught early by quality control checks (e.g. misplaced decimals, transposed numbers). Others, such as improper rounding of decimals are unrecoverable, and can only be 12 assigned a potential error margin. Other problems have been noted due to the conversion between Fahrenheit and Celsius scales (Guttman, 1998). A significant methodological error to affect a majority of records, at least in the United States, was a change in the method of computing mean daily temperatures. While mean daily temperature is generally calculated as (max+min)/2, originally, it was derived from spot readings (Mitchell, 1953). Mitchell (1953) estimates this may contribute possible errors of up to 2\u00C2\u00B0F in a 100 year record. Karl et al, (1986) describe the 'time of observation' bias resulting from a change after 1940, when observers began recording the 24-hour max and mins in the morning, rather than at sunset. Karl et al. (1986) estimate this bias to be on the order of 2\u00C2\u00B0C, and developed a model to estimate the possible bias at any location. At principal stations across Canada, a similar observation schedule change occurred in 1961, when the definition of the reporting day changed from the 00Z-00Z to 06Z-06Z. This change is estimated to negatively bias annually averaged minimum daily temperatures by 0.6 to 0.8\u00C2\u00B0C in Eastern Canada, and by 0.2\u00C2\u00B0C in Western Canada (Vincent and Gullett, 1999). As DeGaetano (2000) reports, at volunteer climate stations, the observations schedule is determined for the convenience of the observer, and therefore biases are more difficult to identify. DeGaetano found up to 96 per cent of stations included in the monthly H C N data set have incomplete records regarding observation times. Bootsma (1976) also reported daily minimum temperature differences of 5\u00C2\u00B0C or more, depending on the time of day when observations were made, which translated to differences of 1.7\u00C2\u00B0C for the monthly mean minimum temperature. Janis (2002) examined the effect of the time of observation bias on daily temperature records. Although large temperature differences did occur in a few circumstances, Janis (2002) contends that correcting daily temperature series should be avoided, since the magnitude of the bias depends on the specific observation schedule, as well as the particular weather conditions that lead to the error (e.g., a morning observation of minimum temperature could experience a positive or negative bias depending on whether a cooler day preceded a warmer day or vice versa). Janis (2002) presents a technique for identifying those situations in which the time of observation bias is most likely to occur. ii) Instrument Biases Mitchell (1953) estimated changes in thermometer calibration over time to contribute to an increase of slightly less than 1\u00C2\u00B0F per 100 years, and thermometer replacement to be on the order of \u00C2\u00B10.3\u00C2\u00B0F. More recently, many national weather services have replaced mercury-in glass thermometers with electronically recording temperature sensors. Quayle et al. (1991) estimated positive biases in large-scale averages of 0.4\u00C2\u00B0C for maximum temperatures, 0.3\u00C2\u00B0C for minimum temperatures and 0.7\u00C2\u00B0C for mean diurnal temperature range, but only 0.06\u00C2\u00B0C for mean daily temperatures, caused by nearly one-half of the US National Weather Service stations replacing liquid-in-glass thermometers with thermistor-based sensors in different housing. Gall et al, (1992) and Kessler et al. (1993) subsequently reported errors at individual stations of 1 to 2\u00C2\u00B0C in maximum temperatures as a result of a design error in the new housing which prevented proper aspiration rates of the sensor. Biases caused by instrument housing will also be discussed in a following section. 14 Guttman and Baker (1996) examined biases introduced by replacing 'conventional' measurements (HO-83 hygrothermometer) at most stations reporting hourly observations in the US with automated observing systems using model-1088 hygrothermometers. A comparison of 10 stations running both systems in parallel suggested a potential negative bias of a few tenths of a degree Fahrenheit (larger in the daytime) caused by the introduction of the automated systems. However, results also showed that this bias was minimal compared to potential biases of 'a couple of degrees' caused by siting differences, which could be either positive or negative. Siting problems are discussed in more detail in a following section. iii) Thermometer Screen Errors (a) Type of screen The use of screens or shelters to house temperature sensors is necessary to provide protection from the elements, and to prevent direct solar radiation from falling on the sensors causing the sensor to register a temperature much higher (up to 25 K; W M O , 1996) than that of the surrounding air. However, the presence of the shield itself alters the micro-climate around the sensor. If the shield is warmed by solar radiation, it acts as an additional source of radiation to the sensor, and can also warm the air inside the screen, relative to the surrounding outside air. Such errors can be minimized if the ventilation of the screen is adequate. However, under most conditions, adequate ventilation can only be achieved by the use of artificial aspiration, which is not available at many climate observing stations. Therefore, most thermometer screens in use are capable of introducing errors into the observed air temperature. 15 The problem of thermometer screen design has received considerable attention dating back to the Strathfield Turgiss experiments which compared 13 different types of thermometer screens (Gaster, 1879). The motivation for modern research in this area has been partly due to the recognition that a wide variety of instrument exposures were in use before the Stevenson screen became widely adopted in the early 1900's, and therefore the continuity of the historical record is in question as a result of these changes. Furthermore, recent changes to automated systems and the high cost of Stevenson screens has led to comparisons of a variety of new screens in an attempt to avoid the introduction of further discontinuities to the record when these new systems are adopted throughout networks. Sparks (1972) surveyed the variety of thermometer screens commonly used around the world, and summarized aspects of screen design and exposure that affect the measurement of air temperature. The following features were identified as important considerations: colour and material of the screen, exposure (e.g. location, including height above ground), non-uniformity of temperature within the screen, and thermal inertia, or 'lag' of response. The Stevenson screen was shown to have a number of defects including overheating in bright sunshine, with the resulting size of errors depending on both wind speed and direction, artificial cooling of the air when the louvres get wet, and problems with blockage of louvres and reflection from snow in colder, environments. (Also see Sparks, 1970 for a summary of studies reporting errors in Stevenson Screen temperature measurements). Kiel (1996 in Strangeways, 1999) reported that simply painting an old screen could lower temperature measurements by 1\u00C2\u00B0C. Andersson and Mattisson (1991) investigated 7 screens (2 types of Stevenson screens, new and old, and 3 commercially available designs with different mountings) Comparisons 16 with an aspirated Teledyne screen (reference) for a period of 1 year were made. In general, they found errors due to thermal inertia of the screens and radiation errors. Although extreme differences of 3\u00C2\u00B0C were occasionally observed, average differences over months or longer periods were of the same order as the measurement accuracy. The larger, wooden screens deviated the most from the reference, with the largest errors observed in the old screen in poor condition. Above wind speeds of 1 m/s, and during cloudy periods, differences between screens tended to zero. However, during calm winds and clear skies, all screens averaged 0.5\u00C2\u00B0C too high during the day, and 0.5\u00C2\u00B0C too low at night compared to the reference, with extreme values of 3.0\u00C2\u00B0C and -2.0\u00C2\u00B0C occurring at sunset and sunrise. Andersson and Mattisson (1991) also summarize their results for characteristic weather types occurring at the location of their study in Sweden. It is important to note that such errors are not generalizable to other locations. Chenoweth (1992) attempted to estimate biases in the US historical record caused by early unscreened 'north wall' exposures of thermometers which were common before the introduction of the 'Cotton Region Shelter' (CRS), a screen similar to the Stevenson type. Temperatures were observed several times daily for nearly 3 years. His results indicated that the CRS screens measured average annual temperatures 0.5\u00C2\u00B0C warmer than the north wall exposure, mainly due to differing radiation regimes. The differences had an annual cycle, with the CRS being warmer from October to April, and cooler from May to September. Minimum temperatures averaged over the year were no different for either exposure, while maximum temperatures averaged 1.0\u00C2\u00B0C higher in the CRS than the north wall. For a shorter period, a second identical CRS was placed in the back yard near the north wall exposure. This comparison showed that proximity to the building was a more 17 important cause of temperature differences than the exposure of the thermometers. A 'heat island' of 0.4\u00C2\u00B0C was responsible for warmer temperatures observed in the summer, particularly at night, in the vicinity of a brick house compared to the open CRS. The difference was negligible in the winter. Parker (1994) documented historical changes in thermometer screens around the world, and summarized typical seasonal and annual effects caused by these changes. Results vary with the location, and specific type of screen. The largest warm bias of 0.4\u00C2\u00B0C in the mean annual temperature was identified for tropical locations due to thermometer exposure inside thatched sheds in the early days. In some locations, seasonal errors tended to cancel in the annual mean. Parker (1994) also noted significant effects on the diurnal temperature range caused by the introduction of new screens, with a tendency for the diurnal range to be reduced by the introduction of Stevenson screens in most locations, except where the change was from north wall exposure, as indicated by Chenoweth (1992). Nicholls, et al. (1996) documented regional changes in Australian thermometer exposure, and indicated that the introduction of Stevenson screens in Australia probably created an artificial drop in observed temperature. However, Hughes (1995) cautions that the records used by Nicholls et al. (1996) also contained other discontinuities in the form of instrument deterioration and site moves, which casts doubt on their conclusions. Nordli et al. (1997) reviewed the problem of thermometer screen changes in Nordic countries. They noted the following biases in monthly mean temperatures: 0.2 to 0.4\u00C2\u00B0C summer warm bias caused by single louvred screens compared to double louvred; up to 0.3 \u00C2\u00B0C warm bias in wall-screens compared to open locations in summer (no effect in winter), 18 and no significant difference due to new automatic sensors compared to free-standing screens. Effects on maximum and minimum temperatures were not considered. Lefebvre (1998) and van der Meulen (1998) describe intercomparison studies of several modern sensor shields (aspirated and unaspirated) in order to determine what effect a change to these screens would have on the continuity of climate records. The results vary from study to study, but in general, the 'cup-like' screens, such as the Vaisalla, tend to behave most like the Stevenson screen. Not surprisingly, aspirated screens perform the best, and the higher the rate of aspiration, the better, in terms of avoiding radiation errors. However, these studies indicate that discontinuities in the climate record are caused when these newer systems replace existing Stevenson screens, with biases varying from site-to-site. In a W M O report Barnett, et al, (1998) conclude that modern screens represent air temperature in a more realistic way than traditional wooden Stevenson screens, but that they undoubtedly introduce discontinuities into the record. Barnett et al. (1998) also noted the significance of solar elevation and azimuth angle in explaining radiation errors. The largest errors were seen when the solar radiation is aligned in the same direction as the wind (i.e. the heated surface is upwind of the thermometer), and when high insolation occurs with low wind speeds. Clearly, the effect of changes in screen design on the continuity of climate records varies with the climatic regime, as well as with the peculiarities of the individual site and exposure. Thus any changes in instrumentation and shelter must be done with side-by-side comparisons of the new and the old system if the errors introduced by such changes are to be known. Barnett et al. (1998) also noted that despite a large number of studies attempting to quantify thermometer-screen errors, the lack of uniformity in methodology, and 19 inadequate consideration of the effects of local climate, ground reflectivity and wind conditions prevents generalization of the results. In general however, the foregoing results indicate that changes in screen design do introduce discontinuities into the climate records but, often, the positioning of the instrument within the screen and the relative location of the screen itself are at least as important as the screen design. (b) Placement of thermometer within the screen This problem has received considerably less attention than issues of screen design. Parry (1962) showed that even with adequate ventilation inside a wooden screen (Cotton Region Shelter), there was as much variation horizontally within the screen itself as between the inside and outside of the screen (differences of up to 0.7\u00C2\u00B0C inside the screen were observed, comparable to results reported in Sparks 1972). Other studies have noted vertical differences within the screen of 0.2\u00C2\u00B0C (Sparks 1972). Lin and Hubbard (1996) have also investigated temperature variations within thermometer screens. iv) Siting biases Although the question of biases in temperature records due to screen design has been recognized for some time, it is a problem that is somewhat manageable because changes between screen designs tend to have occurred around the same time within individual countries, and such changes are often documented. Unfortunately, the same cannot be said for biases introduced by the siting of the screens themselves. While there are, and have always been, guidelines suggesting proper siting and exposure of climate stations, the potential sitings of screens are as numerous as the number of climate stations themselves, and 20 often, although conformity with the official guidelines is assumed, deviations from the guidelines, and exact details of the true siting characteristics are not readily known. Furthermore, this particular problem has received less research attention and is often therefore overlooked in the assessment of homogeneity of records. It often emerges as a footnote to research into other problems where it is revealed that microclimatic differences between sites are often as important as the problem under question (e.g. Guttman and Baker, 1996; Chenoweth, 1992). This problem manifests itself in two ways in long-term climate records. The first is biases introduced when a station is physically moved (e.g. from roof-top to ground, from city to airport), and the second is the introduction of biases caused by physical characteristics of the site itself, particularly if these characteristics change over time. (a) Station Moves The longer a station has been in operation, the more likely is the possibility that the instruments have been relocated at some point in the station's history. Ideally, records of any relocations and parallel observations at the old and new sites would allow identification of the exact timing and magnitude of correction which might be needed to maintain continuity of the record. Unfortunately, the need for such information becomes evident only years later, when obvious jumps in the station record are detected. Often, station moves will affect each station individually, depending on the relative differences between the old and new sites. However, some network wide changes have had more far reaching implications. For example, it was common practice in the United States to place temperature sensors on roof-tops rather than at ground level, but there was a gradual 21 relocation of such roof-top stations down to the ground surface. Also in the US, a large number of urban stations were relocated to airports in the 1940s and 50s, which likely produced an apparent 'cooling' as a result. Station moves introduce abrupt jumps into the record and modern homogeneity analyses are usually able to identify this type of discontinuity. (b) Station Surroundings Although it is widely recognized that a station's surroundings can bias the observations made there, such influences are extremely difficult to quantify. W M O , and most national weather service guidelines for siting new stations are intended to ensure that siting biases are avoided, that is, that observations are representative of the climate of the surrounding area, not just the site itself. Problems with the guidelines with respect to representativeness for climatological purposes were noted in the preceding discussion of representativeness. A confounding problem is that changes over time in the surrounding environment introduce biases to long-term records at stations which may have been ideally sited when first installed, but which have grown to become unrepresentative. Such biases are particularly problematic because they tend to introduce gradual changes, or trends in the record which are indistinguishable from true climatic trends, and which are difficult to detect with even modern statistical tests for homogeneity. Problems of this nature range in scale from the growth of a single tree next to the thermometer screen, to large scale urbanization around a climate station. Although 'urban biases' in temperature records are generally considered to be more significant than 'minor' 22 changes around a station by most who have considered the homogeneity of long term records, it is clear that seemingly insignificant changes at individual sites are capable of introducing significant biases. Mitchell (1953) gave possible error estimates of up to 1\u00C2\u00B0F/100 years in the mean annual temperature caused by 'local growth of foliage', and errors of 0.5 to 1.3\u00C2\u00B0F caused by building and pavement construction. While there have been few specific studies to estimate the potential biases in long-term temperature records due to site changes other than large-scale urbanization, there is evidence to suggest that seemingly minor site changes close to the station itself are potentially just as important as larger-scale changes farther from the site. For example, Hogg (1949) reported an average increase of 0.4\u00C2\u00B0F in winter minimum temperatures after two glass houses were erected 66 feet away from the Stevenson screen. Norwine (1973) showed air temperatures around an isolated shopping complex (a large building and parking lot) were as much as 4 degrees Fahrenheit greater than the surrounding rural areas. Manley (1974) suggested that the relative warming of the long-running Radcliffe Observatory temperature record compared to other more rural sites (about 0.4\u00C2\u00B0F in mean minimum temperatures), was probably due to the gradual encroachment of buildings into the area surrounding the observation site, causing a notable impediment to outgoing radiation. Manley (1974) also mentioned evidence suggesting that changes in the proportion of ploughed to grassed land at the Rothamsted agricultural site had perceptible effects on climate observations made there. As previously noted, while investigating differences between north-wall exposures and free standing screens, Chenoweth (1992) found that microclimate differences caused by 23 the presence of a brick house were more important than the differences between thermometer exposures. A warming of 0.4\u00C2\u00B0C in minimum temperatures was observed in the screen near the house, compared to an identical screen in a nearby open field. Guttman and Baker (1996) also noted microclimatic differences over short distances between a low-lying swampy location and another next to a runway at an airport. The differences were largest during calm, clear nights, indicating the importance of cooling differences on minimum observed temperatures. They indicated 'siting' biases on the order of several degrees, which was more significant than biases due to changes from conventional to automatic observations. In their guidelines for classifying stations, Meteo France (1997) indicates that errors of at least 1\u00C2\u00B0C can be expected when measurements are within 10-30 m of an artificial heating source, (building, parking lot) or water; at least 2\u00C2\u00B0C errors within 10 m of a heat source; and 5\u00C2\u00B0C or more when instruments are situated directly over the source. Concerns about the location of many temperature observation sites at airports have also been raised. Extensive tarmac, buildings and waste heat from the jet engines are all potential warming influences. Kochar and Schmidlin (1990) found temperature differences of up to 6.3\u00C2\u00B0F between Akron-Canton Regional Airport and its rural surroundings, with the largest differences occurring on calm, clear summer nights. Mays (1997) also found night winter temperatures observed at Willow Run Airport, Michigan, were slightly higher than those of the rural surroundings. Although not directly related to problems of climatological observations of air temperature, further evidence of the difficulty to obtain representative air temperatures comes 24 from the extensive theoretical and observational studies of advection. Although not reviewed here, evidence suggests that air temperatures can be affected at significant distances downwind of different surfaces: e.g. tarmac (Rider et al. 1963); irrigated vegetation (de Vries, 1959); crops/roughness changes (Munro and Oke , 1975); lakes (Antonioletti et al, 1980; Goulter, 1990); and isolated vegetation canopies (Taha et al, 1991). The problem of siting biases, and the microclimatic processes relevant to screen level temperature observations are considered in more detail in Chapter 3. v) Urbanization The extensive literature on the 'urban heat island' phenomenon has demonstrated that air temperatures within cities are often substantially warmer than their rural surrounds. The effect is due to a number of factors, including the replacement of natural surfaces and vegetation with artificial, impermeable, and heat storing materials such as concrete, as well as changing geometry of the surface by the addition of tall buildings, and the production of waste heat from human activities. In general, the more intense the urbanization or modification of the natural environment, the greater the warmth of the city. It stands to reason, therefore, that observations of air temperatures at a climate station which is surrounded by urban growth, will reflect a gradual warming over time. Many climate stations, particularly those with the longest running records, were initially sited in cities, or have had cities grow up around them during the course of their record. In the 1940s and 1950s, particularly in North America, many urban climate stations were relocated to airports outside the city to meet the growing demand for weather observations for aviation purposes. While these moves were initially to 'cooler', non-urban surroundings, many such airports have since been surrounded by built features. The possibility of warm biases in such 25 temperature records is now widely recognized, although there is still much uncertainty regarding the magnitude of contamination of large-scale (hemispheric or global scale) average temperatures. At the scale of individual climate stations, potential urban biases are undeniable, and readily demonstrated by comparing synchronous temperature records from nearby non-urban stations. Depending on the desired use of the record, its representativeness can then be judged. However, the difficulty of urban biases arises in the possibility that when thousands of individual temperature records are averaged together in hemispheric or global scale studies of climate change without a priori knowledge of each station's exact location, the entire average could be contaminated by urban warming biases. The detection of such a bias is particularly problematic in climate change studies, because urbanization causes a gradual warming over time, which is exactly the signal sought after as evidence that 'global warming' is occurring. Secondly, because urban effects on temperature are typically largest at night, and often insignificant by day, urbanization is likely to have a much greater effect on minimum temperatures than on mean or maximum. Many recent studies have demonstrated that historical temperature records also display a similar pattern: that is, minimum temperatures are increasing at a much faster rate than mean and maximum (e.g. Karl et al, 1984; Karl et al, 1991; Karl et al 1993a; Easterling and Peterson, 1995a). (Karl (1984) also found that the most significant trends in minimum temperature occurred in the summer months, which is also typically the time of year when urban heat islands are largest in midlatitude cities). While the authors of these studies believe that any significant urban bias has been removed from their compiled data sets, the previous discussion included evidence that even individual buildings can create 'heat islands' 26 which influence temperatures (Chenoweth, 1992). It is clear that the threshold that defines 'urbanization' should be lowered significantly from the standards currently used to define it. A discussion of some recent methods to remove urban biases follows. Dronia (1967) was one of the first to suggest that urban effects distorted the record of Northern Hemisphere temperatures. However, apparent methodological problems have led to concerns about the reliability of Dronia's (1967) conclusions (Jones et al. 1986a). Jones et al. (1986a & b) claimed to have compiled the first 'homogeneous' global data base of air temperatures. Because of the lack of available station history data, and the large number of records included, they assessed relative homogeneity of mean monthly temperature records by comparing neighbouring stations. They identified urban warming in 2% of the 2666 stations, mostly from North America, which they calculated to contribute to a 0.0087 \u00C2\u00B0C per year bias. However, the fact that the neighbouring stations used to identify urban trends in this study could also contain urban biases probably led to an underestimation of the potential bias. In comparing urban-rural station pairs, Kukla et al. (1986) concluded that North American cities warmed at a rate of 0.12\u00C2\u00B0C per decade faster than their rural neighbours. In this study, the distinction between urban and rural sites was based entirely on population. Most of the 'urban' stations were located at municipal airports, with surrounding populations of more than 100,000, whereas 'rural' stations had populations, typically less than 7,000, but in some cases, up to 40,000. Their estimate, is again an underestimate of potential urban bias because of the possibility that some of the rural stations were also influenced by urbanization. 27 Karl et al. (1988) attempted to estimate the potential urban bias in historical temperature records by developing regression equations which related temperature change to population growth. Although they found detectable warming biases in small towns (populations less than 10,000), they concluded that the net effect of urbanization in the US Historical Climate Network was small (0.06\u00C2\u00B0 C in the 20 t h century for mean annual temperatures, larger for minima) in comparison to the total observed temperature change. Their conclusion was based on the fact that 85% of the U S H C N stations came from settlements with populations of less than 25,000. Since then, these regression equations based on population have often been used a method of correcting large-scale temperature averages for urban-induced inhomogeneities (Karl et al, 1991; Karl et al, 1993a; Easterling etal, 1996). Population has also been used to differentiate between 'urban' and 'rural' records in attempts to demonstrate the magnitude of urban biases in large data bases (Jones et al, 1990; Peterson et al, 1999). In these studies, it is assumed that a data base composed of 'rural' stations is free of urban biases, and therefore represents the true 'climate' record. However, when rural stations with local populations of up to 100,000 (Jones et al, 1990) are considered 'unaffected' and therefore a suitable reference with which to compare urban stations, it is not surprising that the urban-affected, and 'rural' data bases are so similar. Peterson et al (1999) compiled a separate global data base of 'rural' stations to compare with the full Global Historical Climatology Network (Peterson and Vose, 1997) data set which presumably includes some urban bias. In this case, rural and urban stations were distinguished on the basis of population ('rural' is a population less than 10 000), land use classification on Operational Navigation Charts (1:1 000 000 scale) created by the US 28 Department of Defense, and the intensity of surface 'night lights' observed from satellites. As both 'rural' and 'urban affected' data sets showed similar climate trends, the authors concluded the G H C N time series \"is not significantly impacted by urban warming,\" which supported similar statements of urban biases by Jones et al, (1990) and Easterling et al, (1997). However, Hansen et al. (2001) further classified stations into three categories based on nighttime brightness: urban, periurban and unlit. They found evidence of 'urban' effects on temperature records from \"small town\" stations, previously thought to be unaffected by urban warming. The use of population as a reliable indicator of the degree of urbanization around a climate station is questionable (Oke, 1997; Changnon, 1999). Gallo et al. (1996) found that the land use within 100 m of a station was largely responsible for the observed temperature trends, although some influence of the effect of land use up to 10,000 m away was also detected. Furthermore, Bohm (1998) showed for the city of Vienna that, although the population remained constant, re-development in areas within the city led to increased energy consumption, changes in building density and form, and reduced greenspace. Significant temperature trends of 0.6 K in 45 years were observed in Vienna, but the value varied depending on the local surroundings of the site within the city. Changnon (1999) compared temperatures observed at a rural agricultural research site to those from the nearby University of Illinois, which experienced rapid growth between 1890 and 1950. Over the 64 year period examined, temperatures increased 0.6\u00C2\u00B0C at the university site, but only 0.4\u00C2\u00B0C at the rural site. Changnon (1999) points out that the heat island effect of the university is much larger than would be predicted by Karl et al.'s population based method. Changnon (1999) also cautions that a residual urban bias may be 29 present in H C N data set because many of the records were observed in small towns generally assumed to be free of urban biases. Pielke et al. (2002) compared temperature trends at several stations in Eastern Colorado, using population as an indicator of surrounding urbanization influences. However, they found significant increasing trends in minimum temperatures at one site that, despite its 'rural' population, had a very 'urban-type' exposure. These analyses confirm that population is an inappropriate measure of the potential 'urban' bias in temperature records, and imply that the immediate surroundings of the station exposure are more relevant. Knappenberger et al. (1996) examined temperature records from 15 'non-urban' airports across the United States. Because they observed consistent trends of a decreasing diurnal temperature range at all stations examined, the authors contend the trends cannot be the result of urbanization, since they selected their stations to avoid 'urbanization'. However, the authors make no note of the immediate surroundings of each station, and therefore their results may be more indicative of the effects of airport growth on temperatures, rather than strictly \"urbanization\". Clearly, it is not reasonable to draw conclusions about the causes of climate trends unless the exact nature of the climate stations surroundings are considered. Airport climate stations are just as subject to microscale warming biases as those surrounded by large-scale urbanization. The compilers of recent 'homogenized' data sets contend that their methods of correcting urban biases are adequate for large scale averages, but that urban biases would still be present in some local and regional scale analyses. This paradoxical position is difficult to understand. If individual records are themselves contaminated by urban biases because 30 current methods for detecting them are inadequate, how is it possible to claim that large averages made up of such records are not biased? The need to consider urban biases in temperature records from the perspective of the physical processes causing the biases is well illustrated by Camilloni and Barros (1997). These authors define a 'second urban bias,' in addition to the warming bias caused by urban growth, which apparently results from the negative correlation of urban-rural temperature differences and the rural temperature itself. That is, Camilloni and Barros (1997) contend that, during periods in which regional temperatures experience a warming trend, urban heat islands tend to be smaller, and therefore the urban bias in the record is smaller than during cooling periods when urban heat islands are larger. They claim that this 'second urban bias' therefore offsets the urban bias caused by urban growth during climate warming periods and augments it during cooling trends, making it more difficult to identify true climate change. Camilloni and Barros' (1997) assertion that urban biases will somehow offset themselves as the climate warms, supports the notion that 'urban warming' is not responsible for the observed warming trends in historical temperature records. This conclusion is, however, incompatible with the basic fact that urban, or any other siting biases, are the result of the physical nature of the immediate surroundings of the measurement site itself. An unrepresentative observation is an unrepresentative observation: if a thermometer is located close to a heated building, the bias in the observed temperature will not 'disappear' if the regional climate warms. Camilloni and Barros' (1997) study, and others which attempt to describe urban biases in the absence of information about the measurement sites themselves, indicate that a demonstration of the importance of the physical processes responsible for the biases in temperature records is warranted. 31 2.3 Techniques for the detection of inhomogeneities in temperature records The preceding discussion clearly shows that there are numerous potential sources of bias in long term temperature records, and that they must be carefully considered before a reliable interpretation of the records can be undertaken. Recognition of this fact has led to the development of numerous techniques for both the detection and correction of inhomogeneities in temperature records. Peterson et al. (1998) present a thorough review of many of these techniques being used around the world. Because the approach of the present thesis is the identification of inhomogeneities from the perspective of the physical processes causing them, rather than a purely statistical approach, the various statistical techniques of homogeneity adjustments will not be discussed in detail. Rather, homogeneity analysis will be discussed in general, with an emphasis on the strengths and inadequacies of modern approaches, in the context of the motivation for this research. Undoubtedly the most valuable aid to the identification of inhomogeneities in climate records is the meta-data about the observations, if they exist. Station history files, inspectors reports, site maps, photos and detailed records of station moves, instrument, observer, or procedural changes can provide valuable insight into the exact timing of potential discontinuities in a station record. In the ideal case, the potential discontinuity may have been recognized ahead of time and appropriate measures may have been taken to ensure the continuity of the record. For example, if a station is moved, parallel measurements at both the old and new sites would give an indication of any differences introduced by the change in location. Unfortunately, this foresight is rare, and meta-data files are often lacking in appropriate detail, inaccurate, or non-existent, and fail to provide the necessary information 32 to identify all potential problems in a record. Therefore, it is necessary to resort to other methods of identification of such discontinuities by examining the climate data directly. Peterson et al. (1998) note that subjective judgment by experienced climatologists plays an important role at various stages of homogeneity analysis. However, it is impossible to judge the success of a subjective method which undoubtedly varies with the amount and type of experience of the individual. The majority of modern approaches to homogeneity assessment are based on the principle of 'relative homogeneity' of Conrad and Pollack (1962), in which a candidate station's record is compared to neighboring station records, with the assumption that differences between the records are due to inhomogeneities at one of the sites. In the simplest case, such a comparison could take the form of graphing the candidate and the reference time series and visually comparing them for divergence. However, differences between stations are often more easily detected when cumulative sums of the record are plotted, or, when the differences of paired observations are plotted directly. More common now, is the use of more sophisticated tests which can provide a statistically robust indication of significant changes in climatological time series. These methods also rely on the use of a reference time series which serves as a comparison for the candidate station. Thus, the success of such methods is entirely dependent on the choice of a reliable reference series. While some homogeneity analyses (Jones et al. 1986) relied on the intercomparison of several neighboring records to reveal major inhomogeneities, it has been suggested that the reliance on individual records as reference series places too much faith on the quality of individual records, and that a single composite series derived from several 33 individual series would make a more homogeneous reference (Potter, 1981; Alexandersson, 1986; Peterson and Easterling, 1994). Keiser and Griffiths (1997) contend however, that it is uncommon to find enough suitable neighboring stations to create a composite series, and that the use of one station with high quality data as a reference is more realistic. Another approach to the problem of inhomogeneities in the reference series is to create a first difference series, where the original time series is converted to a series of year-to-year differences (dT/dt) in an attempt to isolate the effects of potential discontinuities in only one year (Peterson and Easterling, 1994; Easterling and Peterson, 1995). Clearly this method is helpful for minimizing effects of discontinuities in the form of jumps (e.g. when a station moves, or an instrument is changed), but it cannot eliminate problems caused by trend inhomogeneities such as those resulting from gradual changes around a station. Peterson et al. (1998) discuss other methods aimed at minimizing or eliminating inhomogeneities in reference series. The relative merits of individual techniques for the detection and correction of inhomogeneities in temperature records will not be discussed here. In fact, the merits of a particular approach can only be considered on a case by case basis. Peterson et al. (1998) outline approaches used in a number of countries, and it is clear that some techniques are better suited to particular problems depending on the amount and type of data available, as well as the resources available to perform the analysis. Peterson et al. (1998) also note that different techniques applied to the same data can lead to different results. Thus, there is no universally 'correct' method for dealing with inhomogeneities in climate records. 34 2.4 Implications for analysis of long-term temperature records Those who use these techniques to detect and correct climate data in the compilation of large climate databases contend that the modern techniques are adequate to detect all major inhomogeneities which affect large scale climate averages. The fact that in large scale (hemispheric to global) averages, adjusted data sets are very similar to non-adjusted sets is used to support this contention (Easterling and Peterson (1995a); Peterson et al. 1999). In other words, any inhomogeneities undetected are insignificant when enough records are averaged together. Nevertheless, the same researchers acknowledge that inhomogeneities at individual stations, or even in regional averages, can be significant and that, without correction, they can significantly alter the perceived climate trends. In general, the techniques used in homogeneity analysis of large data bases are inappropriate for smaller scale studies. While it might be argued that such statistical techniques are the best available solution to homogeneity problems in the absence of appropriate meta-data, or the inability to analyze detailed meta-data when thousands of stations are being considered, there are still a number of limitations inherent in previous analyses which raise concerns about the reliability of historical climate records. While there is general consensus that the best (and presumably homogeneous) data indicate that the global mean temperature increased 0.4\u00C2\u00B0-0.8\u00C2\u00B0C in the twentieth century, there has been a recent shift in research focus towards examining trends in the frequency and intensity of extreme climate phenomena. The need for caution regarding the homogeneity of temperature records is perhaps even greater when 'extreme' values are interpreted. 35 Easterling et al. (1999a) note some of the problems associated with 'extreme' data including instrument changes and station moves (e.g. moving from a shaded to sunlit site could significantly increase the probability of extreme maximum daily temperatures being observed). Biases such as these may not be detected in monthly or annual mean records, but could certainly affect the frequency of days with temperatures exceeding a given threshold value. As many analyses of extreme temperatures are based on daily temperature values, or indices derived from them (Folland et al, 1999; Jones et al, 1999), the issue of the homogeneity of daily temperature records is paramount to the correct interpretation of such records. Allen and DeGaetano (2000) corrected annual extreme threshold exceedences for inhomogeneities, but it is well recognized that detecting and correcting for inhomogeneities in daily records is a more difficult task. Vincent et al. (2002) homogenized daily maximum and minimum temperatures for Canadian stations, but inhomogeneities were first identified in annual data, and daily adjustments were interpolated from previously applied monthly adjustments. Many other studies of temperature extremes also rely on homogeneity tests of monthly data, (i.e. the methods described in Peterson et al., 1998), assuming that if the monthly data pass the homogeneity test, then daily data can also be considered homogeneous. However, such assumptions may not be valid and, in fact, the opposite may be true: siting biases are probably strongest when extreme temperatures occur (e.g. calm clear nights allowing maximum cooling, or sunny days with strong radiative input). Regardless of the underlying homogeneity (or lack thereof) of data analyzed, results of the numerous studies examining temperature extremes (e.g. Bonsai et al., 2001 (Canada); Easterling et al, 1999a (North and South America); Durre and Wallace, 2001, 36 Knappenberger et al, 2001, (USA); Serra et al, 2001 (Spain); Horton et al, 2001 (Central England and Globe); Tuomenvirta et al, 2000 (Fenno-Scandia and Greenland); Salinger and Griffiths, 2001 (New Zealand)), arrive at a common result: daily minimum temperatures are increasing at a faster rate than daily maximum temperatures, and extreme 'cold' occurrences are becoming less common, while extreme hot events are either not changing, or increasing much less noticeably. (Regional and seasonal variations to this finding do occur, however.) Although most researchers dismiss inhomogeneities as the sole explanation for these trends, it is difficult to ignore the possibility that inhomogeneities, particularly those due to siting biases, may at least be contributing factors. Bonsai et al. (2001) summarize their findings as follows: \"Canada is not getting hotter, but rather \"less cold.\"\" Discussions in Chapters 3 and 4 of this thesis examine the microclimatic processes relevant to siting biases, particularly their role in decreasing nocturnal cooling, demonstrating that siting biases are indeed capable of contributing to these \"less cold\" conditions. The following section summarizes the considerations which form sufficient motivation for the present thesis. 2.5 Summary of points providing motivation for the present thesis \u00E2\u0080\u00A2 A compilation of many records can be no more homogeneous than the individual records themselves. The fact that 'adjusted' and 'unadjusted' large databases agree (Peterson et al 1999, Easterling and Peterson, 1995a) is no more an indication of the robustness of the large scale database than of the possibility that the 'adjusted' database still contains many undetected inhomogeneities. 37 Although methods for detecting inhomogeneous trends in temperature series are improving (e.g. Alexandersson and Moberg 1997; Vincent 1998), non-climatic trends remain more difficult to identify than abrupt discontinuities. Gradual changes over time in a station's surroundings are likely to introduce trends in the climate record, and are therefore likely to escape detection. If a non-climatic trend is superimposed on a climatic trend of similar characteristics (such as when urban warming affects a record) it is difficult to separate the observed trend into its real climatic and 'artificial' components. Essentially all modern techniques are based on comparisons of test and reference stations. Although the problem of inhomogeneities in the reference series is acknowledged, it is possible that a candidate and its neighbouring stations could all be influenced by the same type of inhomogeneity (e.g. urbanization), which would not be detected by examining 'differences' between stations. Existing techniques are most successful at detecting inhomogeneities in smoothed time series, such as monthly, or annual averages, and any homogeneity adjustments are usually applied on the monthly, or annual series. However, few discontinuities are likely to manifest themselves uniformly throughout the year, or even over the whole day. Hence following the initial analysis of global temperatures by Jones et al. (1986), other studies examined maximum and minimum temperature trends (e.g. Karl et al. 1984, 1991, 1993a, Easterling and Peterson, 1995a). Keiser and Griffiths (1997) showed that necessary adjustments were significantly different when series of maximum and minimum monthly temperatures were analyzed compared to mean monthly temperatures. They suggested that substantial errors are introduced to a time series if these variations are not considered when homogeneity adjustments are performed. 38 \u00E2\u0080\u00A2 Potential biases introduced into a temperature record as a result of changes in the station's surroundings are generally considered insignificant in large-scale averages. However, as indicated in the preceding discussion, there is sufficient evidence to suggest that such assumptions of 'negligible' influence should be reconsidered. \u00E2\u0080\u00A2 Climate station siting biases are likely to have a significant effect on extreme temperatures, and therefore studies which examine daily exceedences of thresholds or temperature indices are particularly susceptible to these kinds of influences. \u00E2\u0080\u00A2 Previous considerations of biases resulting from 'station surroundings', especially urbanization, have used statistical approaches based on surrogate measures of potential biases (e.g. population, light intensity measured from satellite-borne sensors; Peterson et al, 1999), rather than more physically relevant measures of the characteristics of the sites themselves. Thus, a more thorough examination of station siting biases, giving consideration to the physical processes responsible for the biases, is warranted. Therefore, the research which follows was undertaken to elucidate the nature of biases in historical air temperature records caused by the nature of the environment surrounding a climate station. This thesis seeks to demonstrate the magnitude of such biases in historical temperature records, and to indicate the potential impact of neglecting such biases in the analysis of historical temperature records. 39 Chapter 3 THE MICROCLIMATE OF CLIMATE STATIONS: THE PHYSICAL BASIS OF SITING BIASES IN AIR TEMPERATURE OBSERVATIONS 3.1 Introduction In his classic text, The Climate Near the Ground, Rudolph Geiger, (1965) provides numerous examples of the incredible variety of microclimatic conditions that can occur over short distances as a result of variations in the near-surface physical environment. Geiger (1965) presents observations of significant air temperature differences arising from influences such as topography, soil type and treatment, proximity to water bodies, gardens, crops, orchards, vineyards, woodlands, and meadows. Yet, when it comes to assessing the homogeneity of historical air temperature records, conventional analyses all but ignore the microclimatic conditions around climate stations themselves, unless the station is situated in a major urban centre. The conventional thinking seems to be that as long as enough temperature records are averaged together, non-urban microclimatic biases will offset each other and allow the 'representative' global temperature signal to emerge. The purpose of the current chapter is to challenge this notion, by considering potential microclimatic biases in climatological records from a physical, rather than a purely statistical perspective. In particular, the microclimatic processes capable of influencing daily maximum and minimum screen-level air temperatures are considered. The discussion which follows is intended to illustrate the kinds of thermal biases inherent to climatological temperature observations, rather than a comprehensive review. Illustrative examples of observed temperature differences or anomalies reported in published studies, along with the field observations described in Appendix 1 of this thesis are included as empirical evidence of the role of microscale influences on air temperatures. The majority of 40 observational studies discussed here are based on simultaneous temperature differences between sites that have different site characteristics. It is assumed that these results are generally transferable to the case of temporal temperature changes at a single site whose physical characteristics change over time. Because of the difficulty to obtain experimental control in field observations, observed between-site temperature differences are usually the result of many confounding influences. Mathematical modeling studies are useful in isolating individual influences, and therefore modeling results are included in this discussion to provide further insight into the role of surface controls. The discussion of the physical processes is followed by a summary organized around the most common siting biases, and their probable effect on daily maximum and minimum temperatures. 3.2 A surface energy balance approach Any given air temperature observation is the result of a variety of controls acting at many different scales. Thus, the task of separating microscale, or site-specific temperature influences from larger scale influences such as latitude, altitude, continentality, topography, and synoptic variability (air mass characteristics, cloudiness), to name a few, seems an onerous one. However, the role of site specific features can be demonstrated by considering their influences on the individual components of the surface energy balance. Thus, the discussion which follows is organized around the microclimatic processes relevant to the surface energy balance: radiation, convection, advection, evaporation, and the role of surface thermal properties. 41 The surface radiant energy balance can be expressed as Q* = K[-Kt+ Ll-Lt (3.1) where Q* is the net radiation, K is shortwave radiation, L is longwave radiation, and down-and up- arrows indicate incoming and outgoing components, respectively. An alternative expression for the surface radiation balance which clearly indicates the role of surface properties to the balance is Q* = (l-a)Kl + e(> L | - e\u00E2\u0080\u009EoT,/ (3.2) where a is the surface albedo, e0 is the surface emissivity, a is the Stefan-Boltzmann constant (5.67 x 10 \" 8 W m\"2 K\" 4), and T() is the surface temperature (K). The simplest form of the surface energy balance is Q* = QH+ QE+ QG (3.3) where QH is the sensible heat flux, QE is the evaporative heat flux, and QG is the ground heat flux. In certain situations, additional terms must be included in the balance to account for additional energy sources and sinks such as artificial heat sources, positive or negative storage (in the ground or within air volumes), energy used for photosynthesis, or energy used or released in melting or freezing. The role of site specific features is not obvious in this expression of the surface energy balance, but characteristics such as surface roughness and its influence on wind speed and turbulent mixing, near surface temperature and moisture gradients, surface moisture conditions, and surface thermal properties, determine the partitioning of Q* into the different energy forms, thereby influencing the resulting microclimate and air temperature. 42 3.3 Radiative Processes 3.3.1 Shortwave radiant energy The amount of shortwave radiation received at a site depends on latitude and time of year, atmospheric characteristics and weather (especially clouds), as well as the angle of the receiving surface with respect to the direction of radiation. Site specific controls on the net solar radiation relevant to climate station siting include shading effects and surface albedo characteristics. While the input of solar radiation is clearly linked to daytime temperatures, the amount of storage during the day also affects nocturnal cooling, so the receipt of shortwave radiation has an effect on both maximum and minimum daily temperatures. (i) Shading The receipt of shortwave radiation can be reduced by obstructions that fully, or partially block incoming solar radiation (e.g. walls, buildings, trees). A reduction in incoming radiation is capable of reducing both daytime, and nighttime temperatures. Myrup et al. (1993) measured solar radiation and air temperature differences between four suburban sites and an unobstructed rural reference site during the summer at Davis, California. They found a strong correlation between the suburban-rural differences (both temperature and radiation) and the canopy height at the suburban site, with all sites showing maximum temperature differences compared to the control site in the morning and evening, (i.e. when shading is more significant due to the solar angle). Myrup et al. (1993) also concluded that the increase in turbulent mixing due to canopy height played as important a role in temperature differences as the shade effect (See section 3.5 (i)), sometimes producing positive temperature anomalies. Because the shaded sites were 'suburban' other site 43 characteristics such as the proportion of paved surfaces, and the local urban influence would also contribute to the observed temperature differences. Bogren et al. (2000) compared daytime road surface temperatures between sun-exposed and 'screened' sites in Jonkoping county, Sweden. The shading was caused by vertical walls of a rock-cut that the road passed through. They found surface temperature differences of up to 13\u00C2\u00B0C, depending on meteorological conditions and the solar altitude (i.e. temperature differences increased with the solar elevation). They did not report air temperature differences for the same sites. Bogren et al. (2000) also studied the screening effect under different cloud conditions. The increased proportion of diffuse radiation under cloudy skies might be expected to reduce the shading potential and the resulting temperature differences. They found that for up to 3-octas of cloud cover, sun-lit and shaded surface temperature differences were approximately the same as for clear skies, and that surface temperature differences were still significant when 4-6 octas of cloud cover were present. Surface temperature differences were still detectable (approximately 2\u00C2\u00B0C) under complete cloud coverage . The effect of clouds on the shading potential was most significant with greater sun elevations. It is significant to note that shading influences on surface temperatures are nearly always present, that is they are not exclusively a clear sky problem. Bogren et al. (2000) also found evidence suggesting that screening of solar radiation in the daytime reduced nighttime surface temperatures because less heat was stored during the day to offset the nocturnal cooling. However, they note that variations in surface thermal properties may also have contributed to the nocturnal temperature differences. Their results indicate that because of the complex topography of their particular study area, cold air 44 ponding is more significant to nighttime temperature differences than the influence of daytime shading. While solid objects such as walls and buildings completely obstruct the receipt of direct shortwave. radiation for certain periods of the day (depending on the geometry in relation to the sun angle), shading from vegetation is slightly more complicated because some solar radiation is transmitted, and the shading potential varies seasonally with the amount of foliage present, and between coniferous and deciduous species. Heisler (1986) measured the reduction in solar radiation received on a vertical wall due to shading by individual trees. It was noted that reductions on horizontal surfaces are expected to be slightly higher. Results varied depending on the species of tree, and with the orientation of the sun with respect to the receiving surface. Radiation reductions of up to 80 percent when the trees were in leaf, and up to 40 percent for leafless trees were reported. These results demonstrate that shading by trees is not only a summertime phenomenon, but that significant reductions in solar radiation can also occur for some species, even when the trees are leafless. The above discussion implies that shading of a climate station site would reduce the daytime maximum air temperature, and may also slightly lower nocturnal temperatures. It should also be noted that shading of the Stevenson Screen, and the instruments within, will reduce the radiation errors of the measurement by preventing radiation absorption and subsequent warming of the instrument shelter. This effect of shading would augment the cooling (or warming) of the site itself when shade increases (or decreases). It is possible in some cases, that an object providing shade at certain times of the day may actually reflect solar radiation towards the surface at other times, thereby partially offsetting the shading effect, (e.g. an east facing wall to the west of the site would reflect 45 solar radiation in the morning, but block the afternoon receipt of KJ,). However, the net effect of shading will be a reduction in solar radiation compared to an unobstructed site, leading to lower maximum and minimum temperatures, (ii) Albedo The role of the surface albedo (defined as the proportion of incident solar radiation that is reflected by the surface) in determining the microclimatic characteristics of a site is fairly straightforward: the higher the albedo, the less solar radiation absorbed, and therefore the lower the energy status of the environment. Hence, if the albedo of a site increases over time, temperatures are likely to decrease, and vice versa. The recommended surface cover for climate station siting is short grass, having an albedo value of approximately 0.26. Albedos of other surfaces likely to be found in the vicinity of climate stations are listed in Table 3.1. Table 3.1 Surface albedo values (Sources: Oke, 1987, Campbell and Norman, 1998) Surface Albedo Short grass surface (0.02 m): 0.26 Grass, 1 m: 0.16 Agricultural crops : 0.18-0.25 Deciduous forest 0.10-0.20 Coniferous 0.05-0.15 Soil wet, dark: 0.05 Soil dry, dark: 0.13 Soil, wet light 0.1 Soil dry, light: 0.18 Asphalt: 0.05-0.20 Concrete: 0.10-0.35 Brick: 0.2-0.4 Stone: 0.2-0.35 Tar and Gravel roofs: 0.08-0.18 Window glass: 0.08-0.52 (depending on zenith angle) Snow fresh: 0.95 Snow, old: 0.4 46 Many of the site conditions that commonly occur around climate stations appear to reduce the overall albedo, creating a potential warming bias in the temperature record. Examples of typical surface conditions around instrument sites include muddy conditions when the site is first established, construction of asphalt, concrete, gravel or wooden walkways to the instruments; unchecked growth of grass surface; growth of shrubs and trees; encroachment by roads, parking lots, and runways and taxi-ways, and buildings. The presence of snow can significantly increase the surface albedo, but as observers repeatedly walk to the instruments, trampling and dirtying of the snow will reduce the albedo compared to a pristine 'fresh' snow surface, and likely lead to earlier melting (or a shorter snow-covered period). Also, surrounding roads and parking lots are likely to be cleared of snow, decreasing the albedo of the surrounding area. Because of the effect of surface albedo on the overall energy status of the surface, changes to this variable are likely to affect daily maximum and minimum temperatures in the same way: that is, reduced albedo values will lead to warming of both daytime and nighttime temperatures, although the magnitude of the effect will be largest during the daytime. Hopkins (1977) examined minimum temperature differences between similarly exposed stations in East Anglia, U K . He found that temperature differences increased linearly with distance between stations (at an average rate of 2\u00C2\u00B0C per 80 km), however additional variances of 1\u00C2\u00B0C resulted from albedo and soil property differences between the sites. 47 3.3.2 Longwave radiant energy (i) Skyview factor The loss of longwave radiation by the surface at night causes surface cooling. The minimum temperature that occurs at the end of the night, is therefore a function of the nocturnal radiation deficit, which in turn is strongly affected by the geometry of the site characteristics, or the skyview factor. Objects in the surface's field of view, such as buildings and trees, will reduce the net-loss of longwave radiation and therefore limit the amount of cooling compared to an unobstructed site with a skyview factor closer to 1. A reduction in skyview factor as a result of urban structures is recognized as one of the fundamental causes of the urban heat island (the increased warmth of urban areas compared to their rural surroundings), but even small reductions in the skyview factor from a single tree or building can have detectable effects on nocturnal temperatures. Oke et al. (1991) used a simple 1-D energy balance model to determine the role of skyview factors in generating nocturnal surface temperature differences for calm cloudless conditions. Surface temperature differences relative to an unobstructed surface ranged from 1.1 \u00C2\u00B0C to 5.2\u00C2\u00B0C for skyview factors of 0.8 and 0.2, respectively. It is expected that air temperature differences would follow the same pattern, but be smaller in magnitude. Taha et al. (1991) reported nighttime screen-level air temperatures within an orchard in Davis, California were 1 to 2\u00C2\u00B0C higher than in nearby open fields. They attributed the relative warmth to the reduced skyview factor amongst the 5 m high trees. Any encroachment on the instrument site by buildings, walls, fences or other obstructions, as well as vegetation growth is likely to introduce a warming influence to the minimum daily temperatures observed at that site as a result of reduced nocturnal surface 48 cooling. The daytime impact is less significant relative to solar shading, but should reduce net longwave radiation, (ii) Surface Emissivity The expression of the radiation balance in equation 3.2 indicates that the surface emissivity, e0, is also important to the surface net radiation. Because the range of surface emissivity values for most natural, and many artificial surfaces is relatively small (typically 0.90 - 0.98), emissivity changes over time would not be expected to contribute to large temperature changes. Oke et al. (1991) modeled the effect of emissivity variations from 0.85 to 1 on nocturnal surface temperatures, and found the effects were on the order of less than 0.5\u00C2\u00B0C. Again, the impact on the air temperature would be less. 3.4 Surface thermal properties Soil microclimates are influenced by a number of thermal properties including heat capacity, thermal conductivity, and thermal diffusivity. However, thermal admittance is the most appropriate thermal property to consider with respect to surface temperature variations. Defined as the ability of a surface to accept or release heat in response to a change in heat flux, and calculated as the square root of the product of heat capacity and thermal conductivity, thermal admittance plays a key role in the diurnal variation of temperature, and in setting the maximum and minimum values. That is, surfaces with high thermal admittances are less thermally responsive, because heat input during the day is readily accepted without a significant increase in surface temperature. At night, when the surface loses longwave radiation, a large thermal admittance allows stored heat to be released, limiting the amount of surface cooling. 49 Urban-rural thermal admittance differences have been shown to contribute to the nocturnal urban heat island, by affecting the surface cooling rates in both environments (e.g. Goward, 1981; Oke, 1981; Arnfield, 1990; Oke et al, 1991; Runnalls and Oke, 2000). Most construction materials (asphalt, concrete, etc.) have large thermal admittances, while admittances of 'natural' surfaces vary depending on soil composition, soil moisture content, pore space, vegetation cover, and snow cover. Typical values are listed in Table 3.2. Table 3.2 Thermal admittance values for different surfaces (J m\"2 s-1/2 K\" 1). Sources: Oke, 1981; Oke 1987. Surface Material Thermal admittance Typical values/ ranges (J i n 2 s-1/2 K\"1) Sandy soil, 40% pore space Dry 620 Saturated 2550 Clay soil, 40% pore space Dry 600 Saturated 2210 Peat soil, 80% pore space Dry 190 Saturated 1420 Rocks 1700-3400 Snow Fresh 130 Old 595 Water (20\u00C2\u00B0C) 1580 Air, 10\u00C2\u00B0C Still 5 Turbulent 390 Building Materials Concrete 150 -2370 Asphalt 1230-1680 Brick 1070 Stone 2220 Glass 1110 Wood 200-540 50 If the surface thermal admittance around a climate station increases over time, a smaller diurnal variation in temperatures would result: that is lower daily maximum and higher daily minimum temperatures would occur. Increases in thermal admittance would occur with the addition of construction materials, compaction of the soil due to trampling, increased soil moisture, decrease in vegetation height, or vegetation removal, and trampling or compaction of snow surfaces. Decreases in thermal admittance, and subsequent increase in daily maximum/decrease in daily minimum temperatures would be expected with a decrease in soil moisture or growth of vegetation. In the study mentioned above (section 3.3.2 (i)), Oke et al. (1991) modeled the effect of thermal admittance differences on surface cooling for calm clear nights. A range of thermal admittances were modeled to represent typical urban and rural surface materials. Their results showed that, following a twelve hour cooling period, the between-surface temperature differences were greater when the absolute value of thermal admittance was low. (e.g. surfaces temperatures with thermal admittances of 600 J m 2 s-1/2 K\" 1 and 1000 J m\"2 s-1/2 K \" \ differed by 2.6 \u00C2\u00B0C at the end of the cooling period, compared to a 0.8\u00C2\u00B0C difference for thermal admittances of 1800 J m\"2 s-1/2 fC'and 2200 J m\"2 s-1/2 K\" 1). In the context of climate station site influences, these results imply that small changes to the thermal admittance at a site where the thermal admittance is low (e.g. dry soils, snow), would have a greater impact on the temperatures than the same relative change in thermal admittance where the original value is higher (e.g. moist soils, bare rock). Thus, the addition of a concrete sidewalk leading up the instruments would have a larger effect on the observed temperatures at a dry site than at a moist site for two reasons: the absolute thermal 51 admittance change would be larger at the dry site, and the low thermal admittance of the dry site is more sensitive to the change. As mentioned, thermal admittance is a strong control on surface temperatures, and it is assumed, that under the calm, stable nocturnal conditions, the surface cooling has a strong influence on the temperature of the near surface air as well. Zdunkowski and Trask (1971) modeled surface and screen-level nocturnal cooling for four different soil types, for calm conditions. Their results showed that following 8 hours of cooling, screen-level air above a quartz sand had cooled 9\u00C2\u00B0C more than air above a rocky soil, compared to a 12\u00C2\u00B0C difference in surface temperatures. These results indicate that surface thermal properties at climate stations are a significant control on nocturnal cooling and therefore on minimum daily temperatures, particularly under calm conditions. Oke et al. (1991) examined the combined effects of thermal admittance and surface geometry (skyview factors), and found the resulting surface temperature differences were equally sensitive to both controls. Although the surface thermal admittance affects both warming and cooling rates of the surface, its influence is most pronounced on the nocturnal cooling phase, particularly on calm clear nights, because inversion stability concentrates effects in a shallow layer of air. Thus, it is anticipated that changes to the thermal admittance around climate stations would have the greatest effect on minimum temperatures occurring at the end of the nocturnal cooling period. Increases in thermal admittance are likely to reduce the amount of cooling, leading to increasing minimum temperatures. 52 3.5 Convective processes The foregoing discussions have identified significant site controls on surface temperatures and it is assumed that screen-level air temperatures would be well correlated to the surface temperatures. While air temperature variations are largely the result of surface forcings, convective processes essentially determine how well the surface and air temperatures are linked. The relative proportioning of net radiation into sensible and latent heat forms (QH and QE), the degree of convective mixing in the air, horizontal advection of relatively warm or cool air, down-slope drainage and artificial sources of heat are all significant to screen-level air temperatures, and are also significantly influenced by site characteristics, (i) Evaporation If surface net radiation, Q*, is preferentially channeled into latent (QE), rather than sensible heat (QH), less energy is available to heat the air. This partitioning is determined by the availability of moisture at the surface, but also by the air-surface moisture gradients, and the degree of convective mixing. For example, Myrup et al. (1993) found that increased turbulent mixing due to surface roughness elements brought relatively dry air down to the surface, and enhanced evaporative cooling of the air, compared to sites with less surface roughness, and therefore less turbulence. The presence of vegetation also influences the evaporative flux at a site. Souch and Souch (1993) found temperature reductions of 0.7 to 1.3\u00C2\u00B0C close to individual trees compared to an open site in the summer time at Bloomington, Indiana. Temperature differences were greatest in mid-afternoon, and least in early morning. While part of the observed cooling is attributable to shading effects* humidity measurements made at the same 53 time (indicating increased vapour pressure near the trees) imply that increased transpiration by the tree was partly responsible for the lower air temperatures. Heilman and Gesch (1991) studied potential cooling effects of turf grass evaporation on external building temperatures in Texas, to determine if microscale evaporation exerted a noticeable cooling effect on surrounding air temperatures. They concluded that only mesoscale evaporative cooling affected screen-level air temperatures. However, McPherson et al. (1989) had previously demonstrated that microscale evaporative cooling was relatively more influential on air temperatures in arid environments. Again, these findings illustrate an important point: some environments are more sensitive to small microclimatic changes than others, and therefore the magnitude of a potential temperature bias caused by a particular site feature depends on the initial characteristics of the environment. Taha et al. (1991) studied microclimatic effects of a small orchard (150 m across in the prevailing wind direction) in summer at Davis, California. They found significant screen-level air temperature reductions of 2\u00C2\u00B0C on average (with maximum differences of 6\u00C2\u00B0C) within the canopy, with cooling effects extending over 110 m downwind of the orchard (approximately 5 times the tree height). Taha et al. (1991) attributed the cooling effects to both shading and evaporation. There is considerable interest in local and regional climatic effects of vegetation changes, such as deforestation, reforestation and agricultural practices. Lewis (1998) found-that Western Canadian ground surface temperatures increased by 1-2 K immediately following deforestation, either by logging or forest fires. This temperature increase was attributed mainly to a reduction in evaporation and transpiration following deforestation. Air temperature differences are likely to be smaller in magnitude than surface temperature 54 differences, as shown by Morecroft et al. (1998), who compared both surface and air temperatures between two types of deciduous woodlands and an open grassland location for a three year period at Oxford, U K . They reported maximum air temperature reductions of 3\u00C2\u00B0C under the canopy on sunny, summer days. However, average summer temperature reductions were 0.6 to 0.9\u00C2\u00B0C, and winter differences were not significant at all; similarly, nighttime temperature differences at all times of the year were not significantly different. The summer air temperature differences that did occur were strongly correlated to solar radiation, the presence and state of development of the canopy, and wind speed, indicating that the effect of vegetation on air temperatures cannot be attributed to a single factor. Schwartz and Karl (1990) examined the regional effect of spring green-up, or leaf emergence on maximum air temperatures in the North-Eastern United States. They estimated that a cooling of 3.5\u00C2\u00B0C can occur following the first appearance of leaves in the spring, and the subsequent increase in evapotranspiration. Fitzjarrald et al. (2000) subsequently examined turbulent flux measurements made above a deciduous forest in the North-Eastern United States. They found that sensible heat flux drops, and latent heat flux increases at the time of spring leaf emergence, leading to measurable decreases in air temperature and increases in humidity. The implications for climate station environments are that changes in vegetation (growth, removal, or changes in species composition) have the potential to alter the relative partitioning of energy into sensible and latent forms, thereby affecting the thermal and moisture characteristics of the site. The availability of soil moisture can also influence the amount of evaporation, and therefore the near-surface air temperature. Morecroft et al. (1998) found that soil moisture variations were not strongly correlated to the temperature differences between a deciduous 55 woodland and open grassland, being secondary in importance to solar irradiation, wind and leaf presence, as noted above. However, Durre et al. (2000) showed that antecedent soil moisture conditions at climate stations throughout the United States could explain the occurrence of many 'extreme' temperature variations, i.e. record breaking high temperatures tended to occur with extremely low soil moisture. The relationship between soil moisture and air temperatures was strongest in the southeastern United States, where local evapotranspiration is thought to be more relevant to the water balance. In addition to natural soil moisture variations, irrigation practices are a potential source of increased evapotranspiration that may be particularly significant to climate stations located in agricultural regions where irrigation is common. The effect of evaporation on air temperature is the same, regardless of whether the soil moisture is natural in origin, or a result of irrigation, but a few studies have looked specifically at the effect of irrigation on air temperature. De Vries (1959) compared screen-level air temperatures in a dry pasture to a nearby irrigated pasture for summertime conditions in Australia. He found that the average air temperature in a two week period was 2.4\u00C2\u00B0C lower over the irrigated field. De Vries (1959) also developed a mathematical model that showed the effect of irrigation on air temperatures extends a distance downwind similar to the along-wind length of the irrigated surface. Stohlgren et al. (1998) used a regional climate model to explore climatic effects of changes to irrigation practices in Colorado by comparing summertime conditions for natural grasslands, dry agricultural crops, and highly irrigated crops. Model results indicated that irrigation led to temperature decreases of 0.5 to 0.9\u00C2\u00B0C. Summertime air temperature trends at several climate stations in the most highly irrigated areas were also examined, and found to 56 have experienced significant long-, and short-term cooling trends. However, it is not apparent that the individual station records used were assessed for homogeneity prior to the analysis. Geerts (2002) demonstrated that large scale irrigation in an arid region of Australia decreased the annual range of monthly-mean temperatures by 1-2 K. The main effect was a reduction of summertime temperatures. An observed increase of summertime dewpoint temperatures indicates that increased evaporation as a result of the irrigation was likely responsible for the lower summer temperatures. In summary, typical site specific changes around climate stations capable of affecting evaporation include changes to soil moisture status as a result of irrigation changes, or changes to ground water table, as well as changes to vegetation and replacement of natural surfaces with impermeable artificial surfaces. In general, increased moisture and/or increased vegetation should increase the amount of evaporation and lead to lower daytime maximum temperatures. Few studies have shown any significant evaporative effects on nighttime minimum temperatures, (ii) Roughness and shelter effects The convective transport of sensible heat towards or away from the surface affects near surface air temperatures. In general, increased roughness enhances convective transport. For typical daytime situations enhanced convection results in the removal of warm surface air and downward mixing of cooler air from aloft, thereby reducing daytime near surface air temperatures. Increased turbulent mixing can also enhance surface evaporation, leading to further cooling. As Myrup et al. (1993) demonstrated, surface roughness led to vigorous downward mixing of very dry ambient air, significantly increasing surface evaporation. 57 Under calm, clear nighttime conditions, the surface and near-surface air cool by longwave radiation emission, creating a surface based temperature inversion. Turbulent mixing transports warmer air from above down towards the surface, offsetting the radiative cooling and leading to higher minimum temperatures. The exact positioning of the temperature measurement site with respect to the roughness elements will determine whether the site experiences enhanced convective transport as a result of increased turbulence, or reduced convective transport as a result of sheltering. For example, Gustavsson (1995) and Gustavsson et al. (1998) found that air temperature differences between wind-sheltered (forested) and exposed sites developed quickly after sunset, and were maintained throughout the night, with between-site temperature differences being greatest at intermediate winds speeds. Gustavsson (1995) and Gustavsson et al. (1998) concluded that wind shelter in the forested sites prevented the downward mixing of warmer air, and therefore lower nighttime temperatures occurred compared to the exposed sites where the development of strong surface inversions was disrupted by convective mixing. Greater cooling at the sheltered site is contrary to the expectation that cooling would be reduced because of the lower skyview factor. However, Gustavsson (1995) notes that the skyview factor is a significant control on nocturnal surface temperatures, but that air temperatures are not nearly as responsive to that control, except under very calm conditions. The microclimate effects of windbreaks, or shelterbelts, have long been exploited for agricultural purposes. McNaughton (1988) reviewed research on the aerodynamic effects of windbreaks on the adjacent microclimate. It has been shown that a triangular 'quiet zone' exists downwind of the windbreak to a distance approximately eight times the height of the 58 barrier. In this zone of reduced turbulence, evaporation decreases and air temperatures increase. McNaughton (1988) reports that some observations have demonstrated temperature increases of up to 2.8\u00C2\u00B0C or more in the quiet zone. Downwind of the quiet zone is a zone of increased turbulence, the 'wake zone', where turbulent fluxes are enhanced, and temperatures reduced approximately the same magnitude as the quiet zone perturbations. The wake zone is more horizontally extensive, with some studies finding temperature reductions of greater than 1\u00C2\u00B0C, at a downwind distance 23 times the barrier height. The dimensions of the quiet and wake zones vary with stability (i.e. the values given here are for neutral conditions, and are slightly less for unstable conditions), as well as with the porosity of the windbreak. If the windbreak is extremely porous, the quiet zone does not develop, and the most recent findings indicate that denser barriers lead to larger sheltered regions due to larger reductions in wind speed (McNaughton, 1988). Wang and Klaassen (1995) studied the effects of an extensive rectangular network of windbreaks in Anhui Province, China. They observed a slight warming of the near-surface air during the day time as a result of decreased turbulence between the wind breaks. However, because mechanically generated turbulence is more significant at night (relative to thermally generated turbulence), the higher roughness of the windbreaks leads to increased mixing at night, and therefore reduces surface cooling. Windbreaks, or other sheltering features are common in the vicinity of climate stations. Therefore the location of the temperature sensor within either the quiet or wake zones could introduce significant biases to the observed temperatures. Furthermore, as shelterbelts grow in height, or change in density and porosity over time, the size of the 59 affected areas will change, further complicating the nature of the resulting bias on long-term temperature records. In addition to the mechanical turbulent effects caused by roughness elements, differential heating of the elements can lead to thermal convection at the microscale as well. Ishida et al. (1996) found warm, moist rising currents of air on the sunlit side of trees, whereas cool currents descended on the shaded side. Again, depending on the specific configuration of the measurement site with respect to potential heating of surfaces, micro-thermal convection currents may bring relatively warm or cool air to the instrument area. One final point regarding wind shelter/enhanced roughness relates to radiation errors of the instrument and screen. Adequate ventilation is required to eliminate radiation errors caused by radiative heating of the instrument and/or instrument shelter. Thus, if wind shelter increases without an accompanying increase of shade, radiation errors in the air temperature measurements may increase. For example, a building erected to the north of the instrument if the prevailing wind direction is also northerly, would reduce screen ventilation without providing any shade to offset the radiative heating, (iii) Advection While surface characteristics in the immediate vicinity of the observation instruments have been the focus of the discussion to this point, the influence of horizontal advection from some distance away from the instruments also has the potential to affect screen-level air temperatures when relatively warm or cool air is advected over the site. In addition, relatively dry or moist air advected to the instrument site has the potential to enhance or suppress evaporation , thereby altering the influence of evaporative cooling on the observed temperatures. Numerous theoretical treatments of local advection (e.g. Philip, 1959; Rider et 60 al. 1963; Taylor 1969; Taylor 1970; Novak, 1990) have provided methods to calculate the downwind temperature and humidity anomalies resulting from surface inhomogeneities. The downwind vertical and horizontal effects depend on the relative changes in surface properties such as surface roughness, moisture status, and temperature differences specific to each situation. Surfaces upwind of climate stations such as roads, tarmac, runways, fields of different vegetation cover and moisture status (irrigated or unirrigated), lakes, ponds or reservoirs, all provide potential thermal biases to observed screen-level air temperatures. For example, as noted earlier, De Vries (1959) found that the effect of an irrigated surface could be detected downwind to a horizontal distance equal to the along-wind dimension of the irrigated surface. Spronken-Smith et al. (2000) studied advection across an irrigated urban park in Sacramento, California. The most significant advective effect was the enhancement of evapotranspiration at the leading edge of the park, as warm dry air was advected over the moist park surface. Evapotranspiration from the park was more than 130% greater than an irrigated, rural control site, not affected by advection. This situation is similar to many typical climate stations, particularly airport stations, that are situated on grass surfaces downwind of extensive warm and dry asphalt surfaces. While increased evaporation should have a cooling effect that partially offsets the increased warmth of the advected air itself, the magnitudes of the two opposing effects are unlikely to balance. The results of Spronken-Smith et al. (2000) are also relevant to many 'urban' climate stations that are often located in parks or grassy areas. The representativity of such urban climate stations is therefore questionable. 61 While extensive urban surfaces, runways and taxiways at airports represent extreme examples of a potential warming influence at typical climate stations, observations reported in Appendix 1 of this thesis demonstrate that even seemingly minor surface variations (a gravel road, in this case) can lead to observable down-wind temperature anomalies. Figure A 1.4 shows differences in 1.5 m air temperatures from identically exposed sensors above a short alfalfa crop in a clay soil for a period after sunset. A narrow gravel road passed between the two sensors, lying 15 m upwind of the second sensor. Surface temperatures indicated that the road remained warmer than the fields throughout the night. When winds were calm, there were no measurable differences between the two air temperature observations. With light southerly winds (< 2 m s\"1), the downwind temperatures increased by 1 \u00C2\u00B0C relative to the upwind sensor, with maximum differences of 3.5 \u00C2\u00B0C recorded. The fact that temperature differences occurred only when the wind direction was perpendicular to the road indicates that the road was the most likely source for the increased down wind temperatures. Source-area modeling results (Appendix 1) show that the largest temperature differences occurred when the road occupied a larger fraction of the instrument's source area. The preceding discussion concerns surface-level inhomogeneities that are capable of producing advective effects to downwind temperatures. Three-dimensional objects such as vegetation or buildings may also provide advective heat sources, either due to different radiative and thermal properties resulting in surface temperature anomalies or, for the case of buildings, losses from interior space heating or venting of warm air from air conditioning can also heat the nearby air. These elements also increase the surface roughness, so heat losses around the building may become mixed back down to the surface as a result of increased turbulence down wind of the object. 62 Quintela and Viegas (1995) performed wind tunnel studies to examine convective heat losses from cubic models with differentially heated facets. Their results indicate that the amount of convection and therefore the potential downwind temperature effect varies with the atmospheric stability, as well as with the orientation of the heated facet with respect to the wind direction. Also, the position of the temperature sensor with respect to the building and the prevailing wind direction will determine the resulting effect on observed air temperatures. Field observations of nocturnal air temperature variations in the vicinity of a rectangular metal wagon (4.3 wide by 2.8 m high) are reported in Appendix 1. The potential thermal influences were caused only by excess surface temperatures resulting from increased absorption of solar radiation during the day, and reduced cooling at night. Results indicate an envelope of warm air adjacent to the walls in a pattern consistent with the typical \"horseshoe\" pattern of turbulence generated by individual roughness elements (e.g. Oke, 1987). A few hours after sunset, temperature excesses of between 0.6 and 0.7 \u00C2\u00B0C were observed 1.5 m downwind of the object. Temperature deficits of 0.3 \u00C2\u00B0C occurred on the upwind side of the object (a distance of Vi H in front of the object, where H is the height of the object), consistent with the lifting of colder surface air ahead of the obstruction. This object represents a relatively small surface modification compared to houses, buildings, airport terminals or hangars that are often constructed in the vicinity of climate stations. The fact that a thermal effect was detectable in this case implies that constructions around climate stations are likely to exert significantly greater influences on the observed temperatures, (iv) Cold air drainage 63 Gravity-induced flows of cold dense air, especially at night, can lead to significant spatial variations in temperature as a result of 'cold air pools' accumulating in low-lying areas. While these cold air flows are largely determined by the local topography, site specific controls can arise if the downslope flow is hindered by obstructions (or enhanced if obstructions are removed). Thus, vegetation growth on a sloping surface would increase the surface friction and reduce the downslope flow of surface-cooled air. Conversely, if vegetation is removed, cold air drainage would be enhanced. The effect of site controls on cold air drainage will therefore depend on whether the site was originally subject to the removal or receipt of the down-slope flows. As noted in section 3.3.1 (i), Bogren et al. (2000) found that cold air drainage in the complex topography of their study area complicated their interpretation of road-surface temperature differences caused by screening influences. Soderstrom and Magnusson (1995) conducted mobile temperature surveys at night in southwestern Sweden to assess frost risk in an agricultural area. Although the study area was relatively flat, they found temperature variations closely followed topography, with temperature differences on the order of 2\u00C2\u00B0C occurring with elevation changes of a few metres. The authors noted that the presence of obstacles such as trees and forests can significantly disrupt the down-slope flow of cold air as a result of damming, re-directing the flow around the edges, or retarding the throughflow of cold air, depending on the density and height of the vegetation. Reference to one final study of temperature characteristics in an area of cold air drainage is included because it fittingly illustrates the inherent difficulties in attributing temperature anomalies to a particular cause. Hawke (1944) analyzed thirteen years of temperature observations from a standard climate station in Hertfordshire, U K . This 64 particular station was noted to have the most extreme diurnal range of temperatures in the U K , including frequent summertime frosts. Although the station was considered to have 'typical exposure' for a normal climatological station, it was located on the slope of a narrow valley wall. Anomalous cold temperatures in the summer were attributed to down-slope drainage of cold air on nights with strong radiative cooling. A railway embankment at the base of the slope was thought to act as a barrier to the down-slope flow, causing the cold air to pool in the valley. Subsequently, a pedestrian tunnel was cut through the embankment which could have provided an outlet and reduced the cold air ponding, but the author could not find evidence that the opening in the embankment altered the cold air damming effect. Causes of extreme daytime temperatures were attributed to possible radiation errors of the Stevenson screen and increased reflection of solar radiation from the sloped surface near the screen (although increased shading as a result of the slope was also recognized). The nature of the soil (porous sand and gravel; i.e. low thermal admittance) was also considered to contribute to large diurnal temperature ranges. Furthermore, Hawke (1944) observed short periods of increased temperatures on relatively calm nights, presumably the result of the down-mixing of warmer air into the cold-air 'lake' around the climate station due to mechanical turbulence generated by the surrounding hills. Thus, this particular climate station, although meeting the standard requirements for siting and exposure of instruments, was clearly subject to many of the potential thermal biases discussed in this chapter. This example clearly illustrates that interpretation of air temperature records is meaningless unless due consideration is given to the role of local influences. 3.6 Summary of typical climate station siting biases 65 The following table summarizes the foregoing discussion by listing common climate station characteristics, and indicating whether warming (+) or cooling (-) biases are likely to affect the daily maximum and minimum temperature records. Table 3.3 Summary of common climate station siting biases and their effect on daily temperatures. Physical feature changed Processes affected Potential thermal influence T max Tmin Compaction of soil from trampling increased albedo increased thermal admittance Grass planted on bare soil increased albedo decreased thermal admittance increased evaporation + Unchecked growth of grass or weeds Change in soil moisture e.g. increased irrigation decreased albedo + decreased thermal admittance + increased evaporation decreased albedo + increased thermal admittance increased evaporation + + Addition of sidewalks, roads, Parking lots, runways decreased albedo increased thermal Admittance decreased evaporation + + + + + 66 Table 3.3 continued Physical feature changed Processes affected Potential thermal influence T max Tmin Decreased snow cover due to clearing of roads etc., and/or earlier melting due to dirtying Growth of trees around site Removal of trees Construction of buildings additional sources of heat (vehicle exhaust) + higher surface temperature advective source of heat + decreased albedo + increased thermal admittance decreased albedo + Increased shade Decreased skyview Factor 0 Increased roughness: (enhanced turbulence) (increased shelter) + Increased evapotranspiration -increased albedo decreased shade + Increased skyview Factor 0 Decreased turbulence + Decreased shelter Decreased evaporation + Increased shade Decreased albedo + Decreased skyview 0 Enhanced turbulence Enhanced shelter + Source of waste heat + + + + + + + + + This chapter has demonstrated the physical basis of microscale siting biases that are capable of influencing daily temperature records. Whether or not a systematic warming or cooling bias occurs in any particular temperature record depends on the exact history of the individual station, as well as on the nature of the environment in which the station is located. It is clear from the variety of potential influences, that 'non-climatic' warming or cooling 67 trends of several degrees Celsius can potentially occur at any station\u00E2\u0080\u0094that is, such inhomogeneities are not simply an 'urban' problem. Therefore, attempts to identify suspect climate stations by such measures as surrounding population, regional land use, or remotely sensed indicators such as NDVI or night-light intensities clearly do not have the necessary resolution to identify microscale problems. In the absence of detailed knowledge of individual station histories, there is clearly a need for the development of statistical techniques specifically designed to identify biases of this type. The remainder of this thesis deals with the problem of identifying microscale inhomogeneities in historical temperature records, and presents a technique for their detection. 68 Chapter 4 COOLING RATIOS AS A TOOL FOR HOMOGENEITY ASSESSMENT: RATIONALE AND DEVELOPMENT OF A TECHNIQUE 4.1 Introduction: It is clear from the discussions in Chapters 2 and 3 that the effect of microclimatic change on air temperature records is not well recognized. It is also apparent that existing homogeneity assessment techniques are not suited to the detection of these 'micro' effects for a number of reasons including the following: the magnitude of microclimatic biases may be small relative to the large variability of daily temperature values; microclimatic influences are best expressed under certain weather conditions or in specific seasons, and can therefore be obscured in monthly, seasonal, or annual averages; and, thirdly, microclimatic changes may produce temporal trends which are often difficult to distinguish from regional or global climate trends present in temperature series. This chapter describes the rationale and development of a new homogeneity assessment technique designed specifically to identify site-specific microclimatic effects in temperature series, through examination of ratios of night-time cooling at neighbouring climate stations. An understanding of the nature and characteristics of cooling ratios is developed through examination of several case studies. In Chapter 5, the technique is developed further, analyzing time series of cooling ratios to identify change points that may be indicative of microclimatic change. In Chapter 6, the suitability of the technique for analyzing urban temperature records is examined 69 4.2 Definitions Throughout this thesis, cooling magnitude refers to the decrease of temperature from the daytime maximum (T m a x ) to the minimum temperature that occurs during the subsequent night or early the next morning (T m i n ) , and is calculated as follows: AT = T max(j-l) - Tmjn(j) (4.1) where j is the day of the year. Historically, most climate stations have used standard maximum and minimum thermometers which are observed once or twice per day. It is acknowledged that certain limitations of standard max- and min- thermometers and/or observational procedures, may lead to occasional errors in the observed temperatures and therefore the calculated cooling magnitude. Because there is no way of determining the actual time that the observed maximum and minimum temperatures occurred, it is possible in some cases, that (T m a x (j.j) - Tmin w ) may not represent the actual nocturnal temperature decrease. For example, sudden changes in air masses or cloud may disrupt the normal diurnal temperature pattern, causing the maximum and minimum temperatures to occur at unexpected times. Also, Bootsma (1976) demonstrated that if the daily minimum temperature is observed, and the thermometer reset, close to the time that minimum temperature occurs, there is an increased probability that the same temperature would be recorded on two consecutive days if the second day is warmer. For the purposes of this study, it is assumed that such errors are relatively rare and that, in the majority of cases, (Jmcu-Tmin) is a reasonable estimate of nocturnal cooling. In the literature, daily maximum-minimum temperature differences are commonly referred to as the diurnal temperature range, or DTR. However, the term cooling is considered to be more meaningful in the context of this thesis, because the process of 70 nocturnal cooling is known to be influenced by the microclimatic characteristics of a site (e.g. soil moisture and associated thermal properties, sky-view factor, etc.) Daily cooling ratios of neighbouring stations are calculated by dividing the cooling magnitude, AT, at a 'test' station (A) by AT of a nearby 'reference' station (B), and is denoted by RA:B = ATA / ATB (4.2) Although R could theoretically assume a wide range of values, for neighbouring stations almost identical in every respect, R should equal one. If one of the stations' environments changes relative to the other, values increasingly different than one would be expected, particularly when meteorological conditions enhance spatial variations in temperatures and nocturnal cooling. Stations may be considered neighbours if they are both subject to the same local, meso- and synoptic scale influences. Actual acceptable distances between stations will vary with the particular setting. As with all relative homogeneity assessment techniques, selecting a reliable reference station is paramount to the success of this technique. An 'ideal' reference station would be one which is known to have had a high quality observation programme, as well as experiencing no changes in its siting or exposure over the period of interest, e.g. in an open rural field. Unfortunately, it is rare to find a station that has not experienced some change in its siting or exposure. Therefore, as results in Chapter 5 will show, it is acceptable to use a reference station which has good historic meta-data which indicate the quality of the observations and a record of any changes in the surroundings of the station. 71 4.3 The Nature of the Cooling Ratio : characteristics and behaviour 4.3.1 Premises underlying the usefulness of cooling ratios The usefulness of R in demonstrating microclimatic effects in temperature records is based on the following premises: \u00E2\u0080\u00A2 In most cases, neighboring climate stations are subject to the same local-, meso-, regional- and synoptic-scale controls, so differences in cooling between two such stations (as expressed by R ^ 1) must reflect site-specific microclimate differences. \u00E2\u0080\u00A2 If the microclimates of two stations do not change over time, then R will remain constant. Conversely, temporal changes of R indicate changes to the microclimatic environment of at least one of the stations. \u00E2\u0080\u00A2 Microclimatic controls on nocturnal cooling are most significant during weather conditions that are conducive to cooling (namely calm, cloudless conditions). \u00E2\u0080\u00A2 Microclimatic influences on air temperature differences are most evident at night when stable conditions allow surface controls to be concentrated in the surface layer. 4.3.2 Illustrative examples of the nature of cooling ratios In order to establish whether cooling ratios are appropriate indicators of site microclimates, and particularly whether temporal changes in R can be used to identify temperature inhomogeneities as a result of microclimatic change, it is necessary to develop an understanding of the nature of R. To this end, exploratory analyses of R for several Canadian climate station pairs have been undertaken in order to understand their behaviour 72 and characteristics, and to establish an algorithm for the technique of testing homogeneity of temperature records. Results of these analyses are described in the following sections. 4.3.3 Effect of weather on R variability i) Temperature Data The following discussion is based on cooling magnitudes calculated from temperatures observed at three climate stations near Vancouver, British Columbia: Vancouver International Airport (YVR), Steveston, and Ladner. (These records are analyzed further in Chapter 5 and site photographs are shown in Figures 5.1 and 5.2). Steveston is the reference station in both cases, because it is recognized as a good quality, long-running station that has been largely unaffected by urbanization or development over the period of its record, although there were two minor station moves (Schaefer, 1974). Y V R is a principal synoptic station, with both hourly and daily records available. Y V R is approximately 5 km north of Steveston, but both are located at almost equal distances from the coastline and at similar elevations (2 m and 0 m asl respectively). Y V R has experienced considerable change due to the growth of the international airport and construction of runways, taxiways, hangars, terminals, and parking lots. The period of overlap for Y V R and Steveston temperature records is 1937 to 1970. Ladner is approximately 12 km east-south-east of Steveston, at an elevation of 1 m asl, and has a common period of temperature observations with Steveston for 1953-1970. The Ladner station was located at a Canadian Forces Base, and it's surroundings remained essentially unchanged over the period of the record. Both Steveston and Ladner are ordinary climate stations, having daily maximum and minimum temperatures and precipitation records. 73 ii) The weather factor, & w Microclimatic controls on nocturnal cooling are most pronounced during calm, cloudless conditions when surface radiative losses are maximized. Therefore temperature and cooling differences between two sites are also greatest under such conditions, and one might reasonably expect that cooling ratios would also show a significant variation with weather. The effects of cloud amount and type, along with wind speed, on the amount of nocturnal cooling can be combined into a single weather factor, W leads to greater cooling in September than January, but both show relatively greater cooling as weather conditions improve (O w increases). This correlation indicates that & w is an appropriate descriptor of weather conditions relevant to nocturnal cooling. Seasonal variations are explored in the subsequent section. Figure 4.2 shows daily R values for an entire year (1955) as a function of 0.5 indicate that, while both series contain seasonal variations, the relationship between the two is antiphase, (i.e. maximum R LadnenSteveston occurs with minimum R YVR:steveston, and vice versa). This feature can be explained by the seasonality of soil moisture differences between the sites. Seasonality of the R YVR:Steveston is consistent with typical seasonal variations in soil moisture in the area, and the associated thermal properties (Runnalls and Oke, 2000). One of the main differences between Y V R and Steveston is the presence of extensive tarmac surfaces at Y V R (runways, taxiways, parking lots.) Greater differentiation of microclimatic environments is expected in summer when soils are dry, since moist soils are thermally similar to tarmac. August 1955 was one of the driest on record and therefore soil moisture was low. Thus, it is not surprising to find cooling at Y V R was only 73% of that at Steveston at that time. Values of R LadnenSteveston close to unity imply very similar cooling regimes at both sites, consistent with their open, rural exposure, and similar soil moisture (and soil thermal properties). The seasonality of R is likely to vary with the specific climatic regime of an area e.g. whether seasonal variations are mainly in soil moisture, as they are near Vancouver, or whether snow cover, vegetation, or artificial heating, etc. are important. 80 1.2 1.1 1 ct 0.9 0.8 0.7 0.6 x x o LadnenSteveston x YVR:Steveston o o 1 2 3 4 5 6 7 8 9 10 11 12 Month (1955) Figure 4.4 Seasonal variation of R LadnenSteveston a n d R YVRiSteveston. Monthly mean values for the year 1955. Figure 4.5 Seasonal variation of R Regina Airport: Regina CDA illustrating that seasonality of R may change over time. Data are monthly mean R for three different time periods. It is possible that the seasonal R pattern could change over time, possibly indicating microclimatic change, or other inhomogeneities at one of the stations. Figure 4.5 shows a clear example of this possibility. In this graph, monthly median R for Regina International Airport: Regina C D A are plotted for different time periods. The seasonal pattern for the later period (1984-93) is the mirror image of the earlier period (1932-41), while the average calculated for the entire series (1932-93) displays no seasonal variation, i.e., the two patterns cancel each other. In this particular case, the change in the seasonal pattern is caused by a discontinuity in winter minimum temperatures at Regina C D A , which is apparent when examining the temperatures alone (Gullett et al., 1990). The station history file does not provide any indication of the cause of this step change, and therefore no conclusions can be drawn from standard evidence as to whether it is indeed caused by a feature of the microclimatic environment, or perhaps as a result of a change in instrumentation or observation procedure. However, because the absolute value of the summer R also decreased, without a corresponding discontinuity in either the maximum or minimum temperature series, there may be more than one cause of the noted changes in R. This example is included only to demonstrate the importance of the time period selected for analysis. If the entire length of record had been used to examine seasonality in this case, one might have erroneously concluded that R for these two stations has no seasonality. 4.3.5 Role of soil moisture changes in R variability As noted in the preceding section, seasonal variations of the i?YVR:steveston are probably linked to soil moisture variations and the resulting surface thermal properties. Therefore, it is also important to consider the possibility that soil moisture variations from year to year in a 82 given season could also induce changes to R, i.e. do exceptionally wet summers lead to more winter-like R values? This question was examined by comparing R YVR:steveston for the three driest Augusts (average monthly precipitation 7.8 mm) and three wettest Augusts (average monthly precipitation 77 mm). Median R for both groups were 0.81. Figure 4.6, a plot of median monthly R for both station pairs against total monthly precipitation for the period 1953-1960 shows no strong relation. Analysis of daily R values as a function of antecedent precipitation, similarly showed no relation. Therefore it appears that the range of summertime soil moisture in Vancouver does not significantly affect the estimation of the summertime R. This independence may not exist in other regions, but because most climate stations also record precipitation, the possibility of soil moisture variations causing temporal variations of R could be investigated on a site-specific basis. 4.3.6 Relationship between absolute temperature and R As previously discussed, one of the difficulties in detecting inhomogeneities in temperature series arises when site-specific trends are superimposed on regional temperature trends. It therefore becomes difficult to determine how much of the observed temperature trend is due to regional warming or cooling, and how much may be a site-specific effect. Given the relative stability of R with variations in weather, it seems likely that R values would also remain constant even if regional trends of temperature were occurring, since regional effects would be common to both stations. On the other hand, it also seems possible that microclimatic differences could be accentuated during warmer periods if differences of surface thermal properties are important to the cooling ratio. That is, during warmer periods, daytime storage might be enhanced and site specific storage differences 83 20 40 60 Monthly precipitation (mm) x YVR:Stev \u00E2\u0080\u00A2 LadnerStev Figure 4.6 Variation of August mean R with precipitation for the period 1953 to 1960. This period was selected to show the greatest range of precipitation and soil moisture variation. The fact that 7? does not show a strong dependence on precipitation implies that R is likely to remain constant if regional climatic trends in precipitation occur. See also Figure 4.8. could lead to greater nocturnal cooling differences. Therefore, the question of whether temporal trends in R would occur as a result of regional warming or cooling must be considered. Zhang et al. (2000) reported that Southwestern British Columbia experienced a statistically significant increase in minimum daily temperatures of approximately 1\u00C2\u00B0C to 1.5\u00C2\u00B0C in all seasons between 1900 and 1998. Over the same period, the region also experienced a statistically significant decrease in diurnal temperature range of approximately 0.5\u00C2\u00B0C to 1.0\u00C2\u00B0C. (Zhang et al. (2000) also showed that similar, but slightly smaller, trends were evident over the period 1950 to 1998 as well, although the trends in diurnal temperature range were not statistically significant during this period). Figure 4.7b shows that the time series of monthly R Ladnensteveston (1950-1970) does not exhibit any trends. The fact that the R remain remarkably constant during this period of regional warming, indicates that the use of a ratio eliminates the regional influence common to both series, making the R a useful indicator of microclimatic differences between sites. The possibility that regional warming trends might affect R time series was examined further by examining the relation between median August R and monthly mean maximum temperatures observed at Steveston, to determine if R values were significantly different in warmer years than in colder years. As Figure 4.8 shows, although mean maximum temperatures range from 20\u00C2\u00B0C to 24\u00C2\u00B0C, R remains constant. Again, these results suggest that regional warming or cooling trends are unlikely to introduce temporal trends in R time series, thus making R a useful indicator of microclimatic differences. 85 (a) 2.5 2 c o to . _ S 1.5 to 6; 1 0.5 liUiAilUlii \u00E2\u0080\u0094 L \u00E2\u0080\u0094i\u00E2\u0080\u0094*\u00E2\u0080\u0094 i\u00E2\u0080\u0094 '\u00E2\u0080\u0094 i\u00E2\u0080\u0094 '\u00E2\u0080\u0094 i\u00E2\u0080\u0094 1 \u00E2\u0080\u0094i\u00E2\u0080\u0094 '\u00E2\u0080\u0094 i\u00E2\u0080\u0094'\u00E2\u0080\u0094 i\u00E2\u0080\u0094'\u00E2\u0080\u0094 i\u00E2\u0080\u0094'\u00E2\u0080\u0094 i\u00E2\u0080\u0094 '\u00E2\u0080\u0094 r\u00E2\u0080\u0094 '\u00E2\u0080\u0094 i\u00E2\u0080\u0094'\u00E2\u0080\u0094 i\u00E2\u0080\u0094 1 \u00E2\u0080\u0094 i\u00E2\u0080\u0094 '\u00E2\u0080\u0094 i\u00E2\u0080\u0094 1 \u00E2\u0080\u0094i\u00E2\u0080\u0094 ' i ' i ' i ' i ' r 1953 1956 1959 1963 Year 1966 1969 (b) 0 1953 - i \u00E2\u0080\u0094 \u00E2\u0080\u0094 i \u00E2\u0080\u0094 ^ \u00E2\u0080\u0094 \u00E2\u0080\u0094 \u00E2\u0080\u0094 i i i \u00E2\u0080\u0094 i \u00E2\u0080\u0094 ^ \u00E2\u0080\u0094 \u00E2\u0080\u0094 \u00E2\u0080\u0094 i \u00E2\u0080\u0094 \u00E2\u0080\u0094 r 1956 1959 1963 1966 1969 Year Figure 4.7 Time series of dairy i? (1953 -1970) for (a) YVR:Steveston and (b) Ladner: Steveston. A step change in the mean is evident in (a) while the long term mean in (b) remains constant. See text for explanation. ( 3 < 1.2 1.1 1 0.9 0.8 0.7 0.6 8 o o o o o X x X * X X X 20 21 22 T\u00E2\u0080\u009E 23 24 x YVR:Stev o Ladner.Stev 25 :(\u00C2\u00B0C) Figure 4.8 Relation between August R and August mean maximum temperatures (1953-1960) for YVR:Steveston and LadnenSteveston. R does not show strong dependence on mean temperatures, implying that R would remain constant if regional temperature changes occurred. See also Figure 4.6. 87 4.4 Optimal determination of R The foregoing analyses indicate that daily R for a climate station pair exhibits apparently random scatter that cannot be explained by weather, temperature, or soil moisture variations. Nevertheless, the central tendency of R appears to be constant over time, except for seasonal cycles. It is reasonable to conclude that a measure of the central tendency of R for a given pair of climate stations is a characteristic feature of the station pair at a given point in time that represents microclimatically controlled cooling differences between the sites. This \"characteristic\" remains constant over time, even with regional climate changes (i.e. trends in weather patterns, temperatures, or precipitation). Given the relative stability of R for a station pair, this measure seems potentially useful to identify site-specific microclimatic biases in temperature. However, given the day to day variability, it is necessary to first identify an appropriate method for calculating R. As previously noted, monthly or seasonal averages can obscure microclimatic biases that may be more pronounced on certain days than on others. Therefore, it is desirable to retain the information present in daily data, yet some measure of central tendency is clearly needed because of the scatter of daily data. Figure 4.2 shows that the scatter decreases with improved weather conditions, and that the 'characteristic' R becomes apparent at higher values of 0.5 50th percentile (above average cooling at Steveston) Maximum R 2 2 1.08 1.04 Minimum R 0.45 0.45 0.59 0.45 MeanR 0.79 0.82 0.74 0.74 Median R 0.77 0.81 0.73 0.72 Standard 0.16 0.18 0.10 0.09 deviation of R N 248 179 69 122 95% Confidence 0.02 0.03 0.02 0.02 limits for mean, median In addition to statistics for the entire data set (1953-60), the above table also includes three subsets of the data: poor weather conditions only (4>w < 0.5); favorable weather conditions only (W > 0.5); and the 50 t h percentile of cooling events (i.e., occasions when cooling at Steveston was large). This last subset was included to determine if it would be possible to select good cooling 'events' without the need to calculate & w for each day. This 89 alternative would make the technique more widely applicable, since many climate stations do not have nearby hourly synoptic data available with which to calculate 0.5 set. There is good agreement of the means and medians estimated from the 0.5 set, and the 50 t h percentile cooling data. Use of the latter increased the sample size by 75%, while ensuring that nights that did not experience significant cooling were excluded. Considering the above, estimating statistical properties from 'best' cooling events in a month (the 50 t h percentile), appears to be the optimal approach. This approach has the added considerable benefit that a reliable determination of R can be obtained without the use of detailed ancillary weather data. Figure 4.7a and 4.7b show time series of daily August R values for 1953 to 1970, for YVR:Steveston and LadnenSteveston. Day-to-day variability is apparent, but the mean value appears constant throughout the whole period in the LadnenSteveston case, while a step change in the mean is evident around 1962 for YVR:Steveston. The timing of this step change corresponds to station moves at both Steveston and the Airport around this time. These series are examined in more detail in Chapter 5. 4.5 Summary of potential advantages of cooling ratios for homogeneity analysis The above analyses suggest that cooling ratios are potentially useful to identify microclimatic biases in temperature records, offering several advantages over existing homogeneity analysis methods. In summary, 90 \u00E2\u0080\u00A2 Nocturnal cooling is a physically relevant process related to the microclimatic environment of climate stations. Cooling ratios are therefore likely to be a more physically meaningful measure of the degree of thermal modification of a climate station's surroundings than surrogate measures such as population, or satellite derived night-light intensities that are presently used to designate stations as either \"rural\" or \"urban.\" (i.e. interpreted to be little or unaffected, or adversely affected, by human site disturbance). \u00E2\u0080\u00A2 Cooling ratios derived from occasions when strong cooling occurs, (and therefore when microclimatic biases should be most pronounced) reduce the possibility of these subtle biases being obscured as they would be in monthly, seasonal, or annual averages. \u00E2\u0080\u00A2 While temperatures themselves are inherently variable because they respond to regional or global climate trends, cooling ratios, on the other hand, appear to be remarkably stable (i.e. invariant with absolute temperature or soil moisture), making them better suited to the detection of small, site-specific trends. \u00E2\u0080\u00A2 R is dimensionless, which simplifies interpretation and permits comparisons between sites. \u00E2\u0080\u00A2 Cooling ratios are calculated from daily maximum and minimum temperatures, which are the basic observations made at almost every climate station so that the technique is widely applicable. Analysis of cooling ratio time series therefore appears to be potentially useful for the identification of site specific microclimatic biases in temperature series. This possibility is explored in the following chapter, where a technique is presented to aid in the interpretation of time series of cooling ratios. 91 Chapter 5 ANALYSIS OF TIME SERIES OF COOLING RATIOS: APPLICATION OF HURST RESCALING 5.1 Introduction Chapter 4 described the derivation of cooling ratio time series from climatological records of daily maximum and minimum temperatures, and demonstrated the suitability of the cooling ratio as a measure of microclimate differences between neighbouring climate stations. Change points in the time series are assumed to indicate changes to the micro-environment that influence site-specific controls on the nocturnal cooling process, and that introduce inhomogeneities into the temperature record. As discussed in Chapter 2, inhomogeneities of this type have proven difficult to detect in conventional analyses based on the temperature series alone. The current chapter presents examples of cooling ratio time series from several pairs of Canadian climate stations, examined in conjunction with station history information. Application of the technique of Hurst Rescaling (Hurst, 1951; Outcalt et al., 1997) is investigated, and found to be useful in the interpretation of the time series. These results are then compared to other homogeneity analyses of Canadian temperature records (Gullett et al. 1990, Gullett et al. 1991 and Vincent 1998) in order to illustrate the advantages of the present methods in the detection of microclimate biases in climate records. 5.2 Hurst Rescaling: background While analyzing historical hydrological records in relation to flooding of the Nile River, Harold Edwin Hurst developed a time series transformation technique to identify long-92 term persistence in geophysical time series (Hurst, 1951). Since then, Hurst rescaling has been successfully used to identify distinct physical regimes within a variety of geophysical time series (e.g. precipitation: Potter 1976 and Potter 1979; river flow levels, sunspot frequency, earth-quake frequency, lake sediment varves, bedding thickness and tree ring thickness: Mandelbrot and Wallis, 1982 and Outcalt etal. 1997). As outlined by Outcalt et al. (1997) (also see Appendix 2), Hurst rescaling entails subtracting the mean of the time series from each of the original observations. These deviations are then summed cumulatively to create a transformed time series, Qt.. Thus for the case of time series of monthly median cooling ratios, Rh for n months, as defined in Chapter 4 (and Appendix 2), Q,= I(RrR) (5.1) Then, Fn ~ Qmax ~ Qmin (^-2) Where r\u00E2\u0080\u009E is defined as the adjusted range of the series, and Qmax and Qmi\u00E2\u0080\u009E are the maximum and minimum values of Qit respectively. The rescaled range is r\u00E2\u0080\u009E/s, where s is the standard deviation of the original time series. The rescaled range is expected to increase asymptotically with the square root of the number of observations, n. That is, r\u00E2\u0080\u009E/s ~ n H, where H, the Hurst exponent, has an expected value of 0.5. (See Appendix 2 for the calculation of H). However, it has been shown that many geophysical time series have a value of H greater than 0.5 - an occurrence termed the Hurst phenomenon. Elevated values of H are attributed to persistence, non-stationarity in the record mean, or pooling samples with different distribution characteristics (Outcalt et al. 1997). 93 Outcalt et al. (1997) also showed that if the accumulated deviations from the mean are normalized by the adjusted range of the series, (QrQmin)/rn, the normalized series can be plotted and visually inspected to identify distinct physical regimes within the series. Regime transitions are easily identified at inflection points in the normalized series. For example, a persistent period of above average conditions becomes transformed as an ascending trace, because positive differences from the mean are accumulating. A transition to below average values is marked by an inflection point, (i.e. a change from a positive to a negative slope) Steeper slopes result from larger deviations from the mean, and therefore small changes in the slope (i.e. not just changes of sign) may also signal regime transitions. There are several advantages to this rescaling technique, the most obvious being the ease of interpretation compared to original, often highly variable time series. Also, Outcalt et al. (1997) point out that \"Hurst rescaling liberates investigators from using arbitrary periods or the informal method of visual decomposition.\" Hurst rescaling is potentially useful in the analysis of time series of climate station cooling ratios (as defined in Chapter 4) as a method of identifying microclimatic regimes which could affect temperature observations. Changes to the microclimatic environment constitute changes to the physical controls on nocturnal cooling, and should therefore become evident as 'regime transitions' when the normalized transformed series is plotted. 5.3 Analysis and interpretation of rescaled cooling ratios Hurst rescaling was applied to time series of R ratios from several climate station pairs to determine if this data transformation technique is useful for identifying inhomogeneities in R time series. In the following examples, the original time series are monthly median R values 94 for the 50 percentile of cooling events (i.e. only those nights which experienced above average cooling are used to calculate the median R value; see Chapter 4 for an explanation of the calculation of R). These series were then rescaled and the Hurst exponent was calculated according to the procedure outlined in Appendix 2. As the first example below illustrates, if H= 0.5, then inflection points on the rescaled trace are not significant; that is, the series can be considered homogeneous. If the Hurst exponent exceeds 0.5, then the rescaled trace can be inspected for trace inflections. Station history files were examined in conjunction with the trace to determine if inflection points correspond to documented physical changes to either station's micro-environment. Although the correspondence of trace inflection points to the timing of documented site changes does not prove causality, Gullett et al. (1991) state that \"whenever the available historical information supports [these] mathematical changepoints with evidence of human-induced influences on the dataset, confidence is assured that the inhomogeneities are indeed unrelated to natural climate variations.\" Given the relative invariability of cooling ratios over time as outlined in the previous chapter, and the fact that the use of a ratio eliminates climatic variability common to both stations, even greater confidence can be assumed in attributing the cause of trace inflection points to documented non-climatic changes. In the case where historical records are not available to indicate the cause of the inflection point, it is still reasonable to interpret change points in the R series as inhomogeneities capable of biasing the record. The station history files used in the following analyses consist of periodic reports by inspectors, photographs, site plans, maps and miscellaneous information held in the archives of the Meteorological Service of Canada in Downsview, Ontario. Station records were 95 selected on the basis of several criteria: thoroughness of the station history file; length and completeness of temperature records; proximity to other stations in order to calculate meaningful cooling ratios; uniformity of regional characteristics to eliminate other sources of between-station differences; and the variety of site-specific features or changes at different stations which might aid illustration of the technique. 5.3.1 Example 1. Ladner : Steveston The first example of Hurst rescaled R is from the station pair of Ladner and Steveston, two rural stations near Vancouver, BC. Statistics for the original and rescaled series are listed in Table 5.8. These sites were introduced in Chapter 4, and summaries of the station histories are given in Tables 5.1 and 5.2. Temperature data are available for 1953 to 1970. Site photographs are shown in Figures 5.1 and 5.2. This example illustrates the importance of calculating the Hurst exponent prior to interpreting the rescaled trace. In this case, the calculated value of H is 0.5, indicating the absence of mixed regimes, or essentially a random series. The apparent stationarity of the mean R values (Figure 5.3) supports this conclusion, yet the Hurst rescaled trace shows a number of inflection points which could have been misinterpreted as significant regime transitions if the Hurst exponent were not known. Table 5.1 Summary of station history for Ladner, BC. Date Excerpts from station history file 24-05-1953 Excellent exposure. Open land in all directions, some shelter from a narrow windbreak to the west. 23-05-1958 A good open exposure. Screen is on a 3'3' concrete block surrounded by grass. 96 29-08-1963 Exposure: A heavy clay soil covered with tall wild grass which is kept trimmed short in the instrument area. A completely open area with a row of trees 30 feet to the west and a tree and two small buildings 40 feet to the north; several maximum thermometers being broken each year 19-08-1969 Maximum thermometer replaced; reading 4 degrees Fahrenheit too high 6-11-1969 Station relocated 200 feet north; no change in elevation or exposure 8-12-1969 New screen and stand 6-10-1971 Station closed Table 5.2 Summary of station history information for Steveston, BC. Date Excerpts from station history file 1-12-1950 The instruments are located in the large back yard of the observer's property. His residence is to the south and there are a few fruit trees and small buildings to the east and west. Unobstructed to the north. The Station is located on Lulu Island in the flat lands of the delta of the Fraser River. 1959 Screen painted 13-7-1962 Observer requested to take two observations a day 11-03-1970 Station moved and downgraded to precipitation only As the station histories imply, there is every reason to expect the individual temperature series to be homogenous, and thus the cooling ratio to remain constant over time. Both stations were well exposed in similar rural settings, with no station moves or changes to exposure throughout this period (Ladner did move a short distance, but probably too close to the end of the record to have any effect). Other than a few screen paintings, several broken maximum thermometers at Ladner, a possible change from one to two observations per day at Steveston (indicated by symbols on trace in Figure 5.3), there are no apparent sources of inhomogeneities at either site. Therefore, a stationary mean R, and a Hurst exponent of 0.5 appear to be good indicators of unchanging microclimate conditions, and homogeneity. Figure 5.1 Site photographs for Ladner, B.C., August 1963. S76>J\u00C2\u00A3S/ ON, B.C. ^ ST\u00C2\u00A3^\u00C2\u00A3SToaJ, B.C. 11- \u00E2\u0080\u00A2 M Figure 5.2 Site photographs for Steveston, B.C., February 1959. 99 Figure 5.3 R Ladner: steveston* a n d Hurst rescaled R (bold line). H = 0.5 Symbols (+) denote events documented in station history files (see Tables 5.1 and 5.2). 100 5.3.2 Example 2: Edmonton Municipal Airport: Edmonton International Airport Edmonton Municipal Airport and Edmonton International Airport are both first order synoptic stations near Edmonton, Alberta. The first (EMA) lies within the City of Edmonton, (3 km northwest of the city centre), and is documented to be influenced by the urban heat island effect of the city (Hage, 1972). Also, Gullett et al. (1991) found the daily minimum temperature to exhibit an inhomogeneity in the form of a warming trend from 1968 to 1983, consistent with the expected influence of the urban heat island effect. Edmonton International Airport (EIA) is 26 km south-southwest of the city centre, and is generally believed to be outside of the city's thermal influence (Hage, 1972). The period of overlap for daily temperature records is 1961 to 1998. Edmonton International was not tested for inhomogeneities by Gullett et al. (1991). These two stations were selected to explore the applicability of Hurst rescaling because of the extensive station history information that is available for both stations and is necessary to interpret the significance of inflection points in the rescaled time series. The Hurst exponent calculated for the R series was 0.76, (Table 5.8) indicating the presence of mixed cooling regimes which should be apparent in the rescaled trace. (i) Effect of seasonal variations of R It is necessary to address the effect the seasonality of R time series (demonstrated in Chapter 4) on the interpretation of the Hurst rescaled time series. Outcalt, et al, (1997) showed that seasonal cycles need not be removed before applying this technique, because they do not change the overall pattern of the rescaled trace. This also appears to be the case for rescaled R time series, as shown by the similarities of the two traces in Figure 5.4, one 101 1975 1985 1995 Figure 5.4 Comparison of Hurst rescaled traces using monthly and annual data for Edmonton Municipal : Edmonton International cooling ratios. H = 0.76 (monthly) and H = 0.73 (annual). Major inflection points are evident in both series. being determined from annual mean values, the other from monthly means. Using monthly data adds only small irregularities onto the trace, but does not change the location of major inflection points. H - 0.73 for the annual series, compared to 0.76 for the monthly series (Table 5.8). Therefore, there does not appear to be any significant advantage to using either annual or monthly means. While the former may be slightly easier to interpret, the latter can be used to identify a specific month in which a regime transition occurred. Also, the sample size is reduced by a factor of 12 for the annual average, which may be problematic if the original time series is short. This problem becomes apparent in the calculation of the Hurst exponent, where values of H < 0.5 result.from an insufficient sample size. Thus, a smaller sample size will make the significance of the rescaled trace difficult to evaluate. Throughout this chapter, time series of monthly R values are used in order to maintain a larger sample size and therefore calculate a meaningful value of the Hurst exponent. (ii) Identification of cooling regimes in the rescaled trace Figure 5.5 shows the R time series and the rescaled trace for Edmonton Municipal: Edmonton International for the period 1961-1991. Timing of events noted in the station history files (Table 5.3) that could possibly affect temperature observations are indicated by symbols on the trace. Changes to the immediate environment of both sites are evident in the series of site photographs shown in Figures 5.6 (Edmonton Municipal) and 5.7 (Edmonton International). 103 Figure 5.5 R Edmonton Municipal: Edmonton International and Hurst rescaled trace (bold) H = 0.76. Symbols denote events documented in station history files (Tables 5.3, 5.4): + Edmonton Municipal; O Edmonton International. 104 Timtmh.iL A , Alfa IV , j U ( V ( ) L . i m \u00E2\u0080\u009E n . r ^ n . A , A i l \u00C2\u00AB i . I0.25), and therefore there is the possibility that non-turbulent processes are present, the results indicate that the road was at times potentially influencing the down-wind temperature observation. When Ri was small (least stable) the downwind-upwind temperature difference was at its maximum of approximately 3\u00C2\u00B0C. S A M results indicate that at this time, the road crossed both the 90% and 50% source areas. When Ri was 0.1, the source area shifted slightly upwind, so that the road crossed only the 90% source area. The upwind-downwind air temperature difference at the time was 0.8\u00C2\u00B0C. During the most stable conditions (Ri=0.16) the readjust barely intersected the 90% isopleth, and therefore did not affect the downwind temperature observation (i.e. upwind-downwind temperature difference of 0 \u00C2\u00B0C) Thus, model results indicate that the road was within the source area contributing to the downwind temperature observations, and the observations indicate that temperature differences between 1 and 3\u00C2\u00B0C persisted for close to 1 hour. Had the meteorological conditions remained favourable for continued cooling throughout the night, it is likely that the thermal influence of the road would have persisted as well. 200 B C \"3 T3 a o Ri = 0.03 ATair=3\u00C2\u00B0C :ance ( 30 -, 15 -\u00E2\u0080\u00A23 0 -wind -15 -Cross -30 -0 Ri = 0.1 ATair = 0.8\u00C2\u00B0C 0% 20 40 60 80 100 120 140 8 3 0 i 15 '\u00E2\u0080\u00A2B o I -30 t-i o Ri = 0.16 A Tair=0\u00C2\u00B0C 90% .50% 50 100 Distance upwind of sensor (m) 150 Figure A1.7 90% and 50% scaler concentration source areas for temperature sensor 1.5 m above the ground. Conditions for three stabilities are given: Ri= 0.03, 0.1, 0.16. Position of road is indicated by shading. This greater than 3\u00C2\u00B0C increase in downwind air temperature is larger than what would be commonly expected for such an apparently minor surface feature. The road is only 7 m wide and surrounded by considerable expanses of vegetation-covered fields; indeed if a climate station had been located there, its siting would have been considered \"ideal\" in most respects. In light of these findings, not only does the potential exist for substantially greater influences on air temperatures observed at climate stations in the vicinity of large parking lots and airport runways for example, but even the small concrete or gravel walkways often constructed to provide a pathway up to Stevenson screens are capable of introducing significant warmth to the air temperatures observed nearby if the wind is aligned appropriately. A1.5 Effect of a small building on surrounding air temperatures In contrast to the road which is essentially a one-dimensional case, the influence of a building on the surrounding air temperature is substantially more complex because of the three-dimensional geometry. Not only does the building have more surface area, as well as different thermal properties compared to the surrounding ground, it is also a roughness element that disrupts the air flow that would normally be present over expanses of open flat fields. Therefore, as well as providing an additional source of heat, the turbulence generated by the rough object under stable nocturnal conditions could result in the down-mixing of warmer air aloft, and therefore increase the air temperature observed at screen level. In order to determine the potential influence on the surrounding air temperature, a small building was simulated by placing a grain-filled steel harvest wagon of dimensions 4.3 m x 2.4 m x 2.8 m high in an open field of short alfalfa (Figure Al. l (b)) . By using a wagon 202 instead of a real building, the structure could be moved to any location and isolated from other features such as trees, roads, or buildings. Also, the block-like shape of the structure was simpler and more symmetric than most buildings. The long axis of the wagon was oriented northwest to southeast. Sheets of galvanized steel were secured at the base of the wagon to prevent airflow underneath. Instrumentation consisted of three identical aspirated fine-wire thermocouples (CSI ASPTC model) mounted 1.5 m above the surface. Two sensors were placed southwest of the wagon, one 1.5m from the base of the southwest wall (approximately Vi H, where H is the height of the object), and one at 3 m, or approximately 1 H. The predominant wind direction at night was southwesterly, and so these instruments were usually upwind of the wagon. The third sensor was the mobile one, described previously. Measurements were made on two consecutive nights, August 12-13 and 13-14, 1998. Meteorological conditions for that period (including daytime) are shown in Figure A1.8. Measurements began in late afternoon of August 12 and therefore did not capture all of the antecedent daytime conditions; however, strong solar radiation, clear skies and light winds prevailed during the day. Surface temperatures observed using a Minolta JR thermometer at 1600 h. showed that the southwest-facing wall of the wagon was 48\u00C2\u00B0C, while the surrounding ground was at 32\u00C2\u00B0C. Maximum air temperatures reached approximately 23\u00C2\u00B0C. During the night of the 12 t h-13 t h , winds were very light (less than 1.5 ms\"1, with frequent calm periods). The wind direction was consistently south-westerly. Skies remained cloudless throughout the night, and therefore Q* was strongly negative. Conditions were ideal to promote cooling, and surface and air temperatures dropped to around 5\u00C2\u00B0C at the time of the minimum on August 13th. 203 1945 405 1225 1855 25 845 August 12-14, 1998 Figure A l .8. Meteorological conditions during the 'building' observation period, (a) Wind speed and direction, (b) Incident short- and longwave, and net all-wave radiation, (c) Air and surface temperatures and relative humidity. The second day started out sunny and clear. Air temperatures reached approximately 25\u00C2\u00B0C, and ground surface temperatures rose to 38\u00C2\u00B0C. Wind speed increased to over 3 m s\"1 in mid afternoon, and the solar radiation curve indicates slight cloudiness throughout the afternoon. On the second night, except for a lull to 2 m s\"1 around sunset, winds were south-south-westerly, with speeds between 3 and 4 m s\"1. Much less cooling occurred, with minimum air temperatures remaining about 10\u00C2\u00B0C warmer than the previous night. Temperature transects using the mobile thermometer were undertaken at sunset, a few hours after sunset, and sunrise. Surface temperatures for all surfaces of the building and the ground were measured at the completion of each transect using the Minolta IR thermometer. Al.5.1 Wall and ground surface temperatures The potential thermal influences of the building can be described by examining the rate of change of the surface temperatures of the various facets of the building and the ground surrounding it, compared to cooling of the open ground surface away from the building. On the first day, maximum surface temperature on the south-west facing wall exceeded 48\u00C2\u00B0C. Throughout the observation period, the exposed ground surface remained cooler than the wall surface. The greatest difference between the ground and south-west wall occurred shortly before sunset, when the ground was 25.8\u00C2\u00B0C cooler. As cooling continued throughout the night, the wall temperatures were similar to each other, but the ground surface remained 3 to 4 \u00C2\u00B0C cooler than the walls. Ground temperatures next to the building approached those of the walls, and therefore also remained 3 to 4 \u00C2\u00B0C higher than the open ground. By OlOOh, the walls were approximately the same temperature as the reference 1.5 m air temperature. On the second night, temperature differences near sunset were substantially less than on the preceding day, with a difference of approximately 8\u00C2\u00B0C between the south-west facing wall and the exposed ground (c.f. >25\u00C2\u00B0C on the first night). Higher wind speeds on the second day were likely responsible for preventing the same degree of spatial variability in temperatures from developing as they did during the first night. Cooling following sunset was also significantly less than the first night, but by sunrise, the open ground was 3\u00C2\u00B0C cooler than the walls, comparable to the first night. Walls equilibrated with the air temperatures around sunset, except for the northwest facing wall which remained slightly warmer until just before sunrise. Thus, particularly around sunset, significant surface temperature differences existed, providing a possible source of heat to influence air temperatures in the vicinity of the building. Al.5.2 Air temperatures Figure A1.9 shows temperature excesses observed by the mobile temperature sensor compared to the sensor located 3 m (approximately \H) upwind of the building, for the post-sunset runs. Although temperature transects were made at sunset, winds were virtually calm throughout the run. As a result, the thermal influence of the relatively warm wagon was not advected downwind and therefore spatial temperature differences were not observed, and results are not shown. Observations around midnight (approximately 2.5 hours after sunset) indicate the presence of a relatively warm envelope of air surrounding the wagon (Figure A1.9a). The warmest areas were immediately adjacent to individual walls, with a differences of up to 206 0500h A u g u s t 13,1998 Plan view of air temperature variability around a rectangular obstacle. Temperature differences are relative to up-wind sensor located at position A. a) OOOOh August 13; b) 0500h, August 13; c) 2l00h August 13; d) 0500h August 14. 207 Figure A 1.9, continued. 208 0.8\u00C2\u00B0C occurring next to the north-west wall, and 0.5\u00C2\u00B0C next to the southwest wall. 0.5 m out from the wall, differences of 0.2 to 0.5\u00C2\u00B0C were observed. 1.5 m out from the walls, no effect was observed near the northwest wall, but to the south and east, differences were still significant (0.1 \u00C2\u00B0C to 0.6\u00C2\u00B0C). Air temperature differences were smaller at sunrise than they were at midnight (Figure A1.9b). At 0.5 m away from the walls, air temperatures were up to 0.3\u00C2\u00B0C warmer than upwind of the wagon. At 1.5 m from the wall, temperature differences were negative or negligible except east of the building, where warm spots were observed at 1 and 3 m away from the wall. On the second night at sunset temperatures adjacent to the southwest and northwest walls were respectively 0.7 and 0.4\u00C2\u00B0C warmer than the upwind reference temperature (Figure A1.9c). At 0.5 m from the walls, the differences were 0.1 and 0.4\u00C2\u00B0C. Downwind of the building, temperatures were slightly less than the upwind reference temperature. By sunrise, the relative warmth next to the southwest wall had been eliminated, while small negative differences were observed downwind of the northwest and northeast walls. The higher wind speeds on this night presumably increased the homogenization of air temperatures relative to the first night. Al.5.3 Discussion Much of the research concerning air flow around buildings has been concerned with the turbulent properties around the buildings with respect to wind loading, pedestrian comfort and pollutant dispersion. As a result, the nature of flow around isolated buildings characterized by the displacement, cavity, and wake zones, upwind and lateral vortices, flow separation and re-attachment is fairly well documented (e.g. Hosker, 1986). Most research on heat losses of buildings has been concerned with the rate of heat loss from the building with implications for energy use within the building itself (e.g. Quintela and Viegas, 1995). Few studies have been concerned with the downwind effects on air temperature that the excess heat would cause. One exception is the study by Ashie et al. (1998), who measured temperature variations around a rectangular block in a wind tunnel under different atmospheric stabilities. They found warm spots downwind of the block in unstable conditions, whereas cool spots were found down-wind under stable conditions. Conditions in the field were significantly different from Ashie et a/.'s (1998) more controlled wind-tunnel measurements, accounting for the differences between the findings. In Ashie et al's (1998) tests, the \"building\" was the same temperature as the surrounding horizontal surface. The situation in the field is much more complex, with each of the facets of the building at different temperatures, as well as spatial variability of ground surface temperature in the vicinity of the building. The results of the present study can be explained qualitatively in the context of the flow patterns introduced by the building to the stable upwind air (Figure ALIO). In the vertical, upward displacement begins to occur up to a distance of 3/7 upwind of the obstruction. For stable conditions, this results in cooler surface air being lifted vertically. This potentially explains the observation (not shown) that the sensor at ViH upwind of the building persistently recorded temperatures more than 1\u00C2\u00B0C below the sensor IH upwind, during the first night (differences were less than 0.4\u00C2\u00B0C on the second night). On the downwind side of the obstruction, air is recirculated in the wake of the building, bringing 210 Figure A L I O Generalized stream lines and flow patterns around a solid barrier, a) and b) vertical cross section for barrier placed normal to flow; c) plan view for building normal to the flow, and d) plan view for orientation diagonally to the flow. (After Oke 1987.) 211 relatively warm air down to surface level, and at the same time incorporating into the mix the warm air that is rising vertically from the roof and walls. Figure A 1.10c shows a plan view of the typical lateral and downwind zones affected by a building. Recirculation behind and to the sides of the building helps to mix the warm air adjacent to windward walls into the cavity zone. This pattern explains the positive temperature anomalies observed 1.5 m away from the walls parallel to the wind direction, and the relatively warm areas near the centre of the downwind wall of the building. With slight variations in wind direction and speed, the theoretical patterns shown in Figure A L I O would be constantly shifting and changing. Nevertheless, the observations indicate a similar \"horse-shoe\" pattern of thermal influence extending laterally and downwind of the building. A1.6 Summary Observations of screen-level air temperatures near a gravel road and a small building indicate that even seemingly minor surface features such as these can exert significant thermal influences on the surrounding air temperatures, particularly during stable night-time conditions. A 7 m wide gravel road was found to increase downwind temperatures by as much as 3.5\u00C2\u00B0C for short periods of time, and more than 2\u00C2\u00B0C for a 30 minute period. If meteorological conditions had remained favourable, it is possible that the downwind thermal anomalies would have persisted throughout the night. Significant temperature anomalies were also observed in the vicinity of a small building. Greater than 0.5\u00C2\u00B0C increases in temperature were observed within a zone of approximately V2 the height of the building during light winds and stable nocturnal conditions, but even during more neutral conditions with stronger winds and greater mixing, measurable warm and cool spots were observed up to 2 H downwind of the building. Implications of these findings are significant to the siting of climate stations and to the interpretation of air temperatures recorded there. If even seemingly minor surface features are capable of influencing nearby air temperatures, then the potential for excessively warm temperatures in the vicinity of larger buildings, parking lots, or airport runways must therefore be much greater than conventionally assumed. Clear skies and light winds occurring in this study were close to ideal for the development of spatial variability in air temperatures, and consequently, the observed temperature anomalies were probably close to the maximum possible for such features. In the context of typical climate station records, the net bias introduced into monthly or longer-term average temperatures would depend on the frequency of occurrence of these particular meteorological conditions, and is therefore site specific. 213 Appendix 2 ALGORITHM FOR CALCULATING COOLING RATIOS AND HURST RESCALING FOR HOMOGENEITY ANALYSIS OF TEMPERATURE RECORDS The following outline describes the steps necessary for the application of the technique developed in the body of the dissertation for homogeneity analysis of temperature records. The algorithm is general: that is, only the calculation steps are presented, not specific computer programming code. The procedure is not computationally intensive and can be easily applied using standard spreadsheet software, although a customized computer program may be desirable, particularly if many temperature records are being analyzed. Application of the algorithm presupposes that station pairs have been selected as suggested in Chapter 4, i.e. both stations are subject to the same synoptic and topographic controls and differ only in their microclimatic setting, records are reasonably long to permit changes over time, and that the station chosen as a reference has sufficient station history information available to assure the record is of high quality and possible inhomogeneities could be identified from the history information. A. Creation of a cooling ratio time series: i) Calculation of daily cooling: A time series of daily cooling (ATj) at each station is calculated from the daily maximum and minimum temperatures {Jmax and r m m ) : ATj = Tmax (j.j) - Tmin (j), (A2.1) where j refers to the day of the year. 214 ii) Selection of the >50 percentile of cooling events in each month: Cooling magnitudes from step i) that are greater than the median monthly cooling at each station are retained for further analysis. The reason for this selection, as outlined in Chapter 4, is to ensure that only occasions when sufficient cooling occurred to permit site specific controls to have an influence on the cooling process, are retained for analysis. iii) Determination of daily cooling ratios (Rj): Using the >50th percentile data, Rj ~ AT j (test station) / ATj (reference station) (A2.2) iv) Monthly median cooling ratio (Ri): Monthly median values are computed from the data in step i i i above, resulting in a time series of monthly median cooling ratios that can be analyzed for trends or change points, or subjected to further analysis. B. Hurst rescaling of time series of Ri (modified from Outcalt et al., 1997): i) The mean of the series is subtracted from each of the n values of the original series. ii) Deviations from the mean are accumulated to create a transformed series, <2;. Qi = Z (Ri-R) (A2.3) i = 1 215 The adjusted range, Fn = Qmax ~ Qmin (A2.4) where Qmax and Q m i n are the maximum and minimum values of Qt, respectively. The Hurst exponent, H, is approximated by the method given by Outcalt et al. (1997), that is H = log [rn/s]/log n, (A2.5) where s is the standard deviation of the original record . The value of H determines whether further analysis is necessary: \u00E2\u0080\u00A2 If the value of H is approximately 0.5, then variations in the series of R are random, and not indicative of regime transitions. \u00E2\u0080\u00A2 If H is greater than 0.5, this is the Hurst phenomenon, indicating that the series likely contains distinct regimes of different distributional characteristics. In the present context, values of H greater than 0.5 indicate that the microclimatic controls on nocturnal cooling at one of the stations have changed during the record. That is, elevated values of H indicate the presence of site specific inhomogeneities in the temperature record. Further graphical analysis can identify the timing of these cooling regime transitions. 216 \u00E2\u0080\u00A2 If the value of H is less than 0.5, this indicates the sample size, n, is too small to permit conclusions about the significance of any of the change points: that is, the record is not long enough. If the Hurst phenomenon is present, graphing the rescaled time series can lead to the identification of the timing of regime transitions. The transformed series, Qi, is normalized by the adjusted range as follows: Qi = (Qi-Qmin)/rn (A2.6) Qi is therefore a series of normalized, accumulated deviations from the mean. Normalization limits the values of Qi' to values between 0 and 1, and greatly simplifies the interpretation of the variability of the original series. When Qi is plotted, inflection points and changes in the slope can be interpreted as follows: \u00E2\u0080\u00A2 An ascending trace indicates a period of 'above average' conditions. The larger the deviations from the mean, the steeper the slope. \u00E2\u0080\u00A2 A descending trace indicates persistence of below average conditions. \u00E2\u0080\u00A2 Changes to the slope and/or changes to the sign of the slope indicate regime transitions. A more meaningful interpretation can be achieved if the rescaled trace is plotted on the same graph as the original cooling ratio series. Regime transitions apparent in the rescaled trace can then be compared to determine if the regime change occurred with an increase or decrease in cooling ratios, a change in variability, the beginning or end of a trend etc. As noted in Chapter 6, when the original series contains persistent trends, an inflection in the rescaled traces occurs in the middle of the trend, rather than at either end. Therefore, the importance of examining both the original and the rescaled trace is clear. The timing of regime transitions can then be compared with station histories for both records in an attempt to identify the cause. If there is insufficient history information, the analysis may be repeated with a third station as a reference series for both records used in the first analysis, provided a suitable third candidate station exists. By creating two new cooling ratio series and the associated Hurst rescaled series, it is then possible to identify which station in the pairing contained the inhomogeneity identified in the analysis. 218 "@en . "Thesis/Dissertation"@en . "2002-11"@en . "10.14288/1.0091430"@en . "eng"@en . "Geography"@en . "Vancouver : University of British Columbia Library"@en . "University of British Columbia"@en . "For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use."@en . "Graduate"@en . "Detecting microclimatic biases in historical temperature records"@en . "Text"@en . "http://hdl.handle.net/2429/14823"@en .