• Balling, R. C., and R. S. Cerveny. 1987. Long-term associations between wind speeds and the urban heat island of Phoenix, Arizona. J. Climate Appl. Meteor. 26:712716.

    • Search Google Scholar
    • Export Citation
  • Balling, R. C., and R. S. Cerveny. 1988. Reply. J. Appl. Meteor. 27:881.

  • Bejaràn, R. A., and I. A. Camilloni. 2003. Objective method for classifying air masses and application to the analysis of Buenos Aires' (Argentina) urban heat island intensity. Theor. Appl. Climatol. 74:93103.

    • Search Google Scholar
    • Export Citation
  • Brazel, A. J., , N. Selover, , R. Vose, , and G. Heisler. 2000. The tale of two climates—Baltimore and Phoenix urban LTER sites. Climate Res. 15:123135.

    • Search Google Scholar
    • Export Citation
  • Cherry, N. J. 1988. Comment on “Long-term associations between wind speeds and the urban heat island of Phoenix, Arizona.”. J. Appl. Meteor. 27:878880.

    • Search Google Scholar
    • Export Citation
  • Comrie, A. C. 2000. Mapping a wind-modified urban heat island in Tucson, Arizona (with comments on integrating research and undergraduate learning). Bull. Amer. Meteor. Soc. 81:24172431.

    • Search Google Scholar
    • Export Citation
  • Ellis, A. W., , M. L. Hildebrandt, , W. H. Thomas, , and H. J. S. Fernando. 2000. Analysis of the climatic mechanisms contributing to the summertime transport of lower atmospheric ozone across metropolitan Phoenix, Arizona, USA. Climate Res. 15:1331.

    • Search Google Scholar
    • Export Citation
  • Fiebrich, C. A., and K. C. Crawford. 2001. The impact of unique meteorological phenomena detected by the Oklahoma Mesonet and ARS Micronet on automated quality control. Bull. Amer. Meteor. Soc. 82:21732187.

    • Search Google Scholar
    • Export Citation
  • Grell, G. A., , J. Dudhia, , and D. R. Stauffer. 1994. A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note NCAR/TN-398+STR, 117 pp.

    • Search Google Scholar
    • Export Citation
  • Hansen, J., , R. Ruedy, , J. Glascoe, , and M. Sato. 1999. GISS analysis of surface temperature change. J. Geophys. Res. 104:3099731022.

  • Jenerette, G. D., and J. Wu. 2001. Analysis and simulation of land-use change in central Arizona–Phoenix region, USA. Landscape Ecol. 16:611626.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E. Coauthors, 1996. The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc. 77:437471.

  • Klysik, K., and K. Fortuniak. 1999. Temporal and spatial characteristics of the urban heat island of Lodz, Poland. Atmos. Environ. 33:38853895.

    • Search Google Scholar
    • Export Citation
  • Leroy, M. 1998. Meteorological measurements representativity, nearby obstacles influence. Preprints, 10th Symp. on Meteorological Observations and Instrumentation, Phoenix, AZ, Amer. Meteor. Soc., 233–236.

    • Search Google Scholar
    • Export Citation
  • Livada, I., , M. Santamouris, , K. Niachou, , N. Papanikolaou, , and G. Mihalakakou. 2002. Determination of places in the great Athens area where the heat island effect is observed. Theor. Appl. Climatol. 71:219230.

    • Search Google Scholar
    • Export Citation
  • Lowry, W. P. 1977. Empirical estimation of urban effects on climate: A problem analysis. J. Appl. Meteor. 16:129135.

  • Morris, C. J. G., , I. Simmonds, , and N. Plummer. 2001. Quantification of the influences of wind and cloud on the nocturnal urban heat island of a large city. J. Appl. Meteor. 40:169182.

    • Search Google Scholar
    • Export Citation
  • Oke, T. R. 1987. Boundary Layer Climates. Routledge, 435 pp.

  • Oke, T. R. 1988. The urban energy balance. Phys. Geogr. 12:471508.

  • Oke, T. R. 1998. Observing weather and climate. Proc. Technical Conf. on Meteorogical and Environmental Instruments and Methods of Observation, Geneva, Switzerland, WMO, Rep. 70, WMO/TD-No. 877, 1–8.

    • Search Google Scholar
    • Export Citation
  • Oke, T. R. 1999. Observing urban weather and climate using “standard” stations. Preprints, Biometeorology and Urban Climatology at the Turn of the Millennium: Selected Papers from the Conf. ICB-ICUC'99, Sydney, Australia, WMO, 443–448.

    • Search Google Scholar
    • Export Citation
  • Oke, T. R., , R. A. Spronken-Smith, , E. Jauregui, , and C. S. B. Grimmond. 1999. The energy balance of central Mexico City during the dry season. Atmos. Environ. 33:39193930.

    • Search Google Scholar
    • Export Citation
  • Polonio, D., and M. R. Soler. 2000. Surface fluxes estimation over agricultural areas: Comparison of methods and the effects of land surface inhomogeneity. Theor. Appl. Climatol. 67:6579.

    • Search Google Scholar
    • Export Citation
  • Stefanov, W. L. cited 2000. 1985, 1990, 1993, 1998 land cover maps of the Phoenix, Arizona metropolitan area. Geological Remote Sensing Laboratory, Department of Geological Sciences, Arizona State University, Tempe, AZ, 4 plates. [Available online at http://elwood.la.asu.edu/grsl/Her/land_cover_phy.html.].

    • Search Google Scholar
    • Export Citation
  • Stefanov, W. L., , M. S. Ramsey, , and P. R. Christensen. 2001. Monitoring urban land cover change: An expert system approach to land cover classification of semiarid to arid urban centers. Remote Sens. Environ. 77:173185.

    • Search Google Scholar
    • Export Citation
  • Zehnder, J. A. 2002. Simple modifications to improve fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model performance for the Phoenix, Arizona, metropolitan area. J. Appl. Meteor. 41:971979.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    The Phoenix metropolitan area. Sky Harbor Airport is represented as a square, and Schnepf Farms is represented as a circle

  • View in gallery

    Schematic diagram of the study area used at Schnepf Farms. Locations of the field HOBO sensors are given as black circles. Here, n, c, and s refer to north, central, and south. For scale, the farm is approximately 400 m wide

  • View in gallery

    Land cover classification of the Phoenix metropolitan area for 1998 (Stefanov et al. 2001). The inner buffer limit was selected as the boundary between built land cover types and either agricultural or undisturbed land cover types. The outer buffer limit was then placed at an average distance of 4 km. The land cover classification map has been reduced from 12 to 4 land cover categories for display purposes: white—undisturbed; light gray—cultivated vegetation (active), cultivated grass, compacted soil (prior agricultural use), vegetation; dark gray—fluvial and lacustrine sediments (canals), disturbed (commercial/industrial), disturbed (asphalt and concrete), compacted soil, disturbed (mesic residential), disturbed (xeric residential); and black—water. Black polygons are the inner and outer buffer limits

  • View in gallery

    Time series of temperature and dewpoint records for all 24 HOBOs (gray lines). Dewpoint traces are the lower set of lines. The two thick black lines represent the average temperature and dewpoint traces. Values of the x axis are days of the month for Apr

  • View in gallery

    Sea level pressure fields (shading; hPa) and 500-hPa geopotential height fields (dark lines; gpm) for (a) 5, (b) 7, and (c) 10 Apr

  • View in gallery

    Contour maps of air temperature (°C) values for (a) 2000 LST 3 Apr, the largest range over the farm, and (b) 0400 LST 7 Apr, the smallest range over the farm

  • View in gallery

    Matrix of t statistics for air temperature between the 24 HOBO station z-score records. Black cells indicate that the row station was significantly warmer than the column station averaged over the study period. White cells indicate that the row station was significantly colder than the column station. Gray cells indicate that the t statistic was not significant at the 99% confidence level

  • View in gallery

    Maps created using the t-test matrix. Of the 24 maps that were created, only maps for the (a) coldest station 6n, the (b) warmest station 2c, and (c) a moderate station 3s are shown. Black circles indicate that the reference station was warmer. White circles indicate that the reference station was colder. Gray circles indicate that the station was not significantly different than the reference station

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 194 194 26
PDF Downloads 95 95 29

The Role of Rural Variability in Urban Heat Island Determination for Phoenix, Arizona

View More View Less
  • a Department of Geography, Arizona State University, Tempe, Arizona
  • | b Department of Geography and Center for Environmental Studies, Arizona State University, Tempe, Arizona
  • | c Department of Geological Sciences and Center for Environmental Studies, Arizona State University, Tempe, Arizona
  • | d Department of Geography and Center for Environmental Studies, Arizona State University, Tempe, Arizona
  • | e Department of Geography, Arizona State University, Tempe, Arizona
© Get Permissions
Full access

Abstract

The effect of rural variability in calculating the urban heat island effect for Phoenix, Arizona, was examined. A dense network of temperature and humidity sensors was deployed across different land uses on an agricultural farm southeast of Phoenix for a 10-day period in April 2002. Temperature data from these sensors were compared with data from Sky Harbor Airport in Phoenix (an urban station) to assess the urban heat island effect using different rural baselines. The smallest and largest temperature differences between locations on the farm at a given time were 0.8° and 5.4°C, respectively. A t test revealed significant temperature differences between stations on the farm over the entire study period. Depending on the choice of rural baselines, the average and maximum urban heat island effects ranged from 9.4° to 12.9°C and from 10.7° to 14.6°C, respectively. Comparison of land cover types of the agricultural farm and land cover percentages in the Phoenix urban fringe was performed with satellite imagery. Classification of the entire urban fringe by using satellite imagery allowed for the local farm data to be scaled to a regional level.

Corresponding author address: Timothy W. Hawkins, Department of Geography, Arizona State University, SCOB Bld., Rm. 330, Tempe, AZ 85287-0104. thawkins@asu.edu

Abstract

The effect of rural variability in calculating the urban heat island effect for Phoenix, Arizona, was examined. A dense network of temperature and humidity sensors was deployed across different land uses on an agricultural farm southeast of Phoenix for a 10-day period in April 2002. Temperature data from these sensors were compared with data from Sky Harbor Airport in Phoenix (an urban station) to assess the urban heat island effect using different rural baselines. The smallest and largest temperature differences between locations on the farm at a given time were 0.8° and 5.4°C, respectively. A t test revealed significant temperature differences between stations on the farm over the entire study period. Depending on the choice of rural baselines, the average and maximum urban heat island effects ranged from 9.4° to 12.9°C and from 10.7° to 14.6°C, respectively. Comparison of land cover types of the agricultural farm and land cover percentages in the Phoenix urban fringe was performed with satellite imagery. Classification of the entire urban fringe by using satellite imagery allowed for the local farm data to be scaled to a regional level.

Corresponding author address: Timothy W. Hawkins, Department of Geography, Arizona State University, SCOB Bld., Rm. 330, Tempe, AZ 85287-0104. thawkins@asu.edu

Introduction

Urban heat island studies form the core of research in the field of urban climatology (Oke 1988). The heat island refers to warmer nighttime temperatures occurring in the core of the built environment when compared with the surrounding rural environment. It is not possible to perform the ideal experiment that would capture the urban effect (i.e., measure without the city and then immediately measure with the city at any stage of its growth; Lowry 1977). Thus, several common approaches are used to characterize the magnitude of the urban effect. These approaches include employing mobile traverses across the city with automobile-mounted sensors, accessing remotely sensed thermal images, developing weather-network spatial interpolations, conducting time-trend analyses of key sites, calculating energy balances, and calculating urban and rural site differences.

The climate components influenced by urban areas are as varied as the methods used to assess urban effects. Modified climate effects include roughness length, airflow, albedo, emissivity, humidity, and precipitation, among others (Oke 1987). Because of the large amount of variation that an urban area can impart, it is difficult to compare the effects that various cities have on their respective climates. In a typical study, the magnitude of the urban heat island is compared between cities. Examples include Melbourne, Australia (1.13°C; Morris et al. 2001), Athens, Greece (4°C; Livada et al. 2002), Mexico City, Mexico (3°C; Oke et al. 1999), Tucson, Arizona (3°C; Comrie 2000), Buenos Aires, Argentina (2.8°C; Bejaràn and Camilloni 2003), and Lodz, Poland (3°C; Klysik and Fortuniak 1999).

There are many site factors and criteria involved in locating weather stations in any area (Leroy 1998). The process is especially complex for urban sites (Oke 1999). Although understanding the urban signal from weather sites in the city is critical and the main interest in urban climatology, relatively little research addresses the rural signal. Many urban effect studies employing a rural site for comparison do not specify the details of the rural variability or conditions surrounding the rural site from which urban effects are to be calculated (Oke 1999).

The purpose in this paper is to assess, using Phoenix, Arizona, as an example, the impact of rural variability in calculating the urban heat island effect. It is suggested that a more cautious approach is needed in specifying the urban effect for a city (i.e., presenting possible ranges of the effect). The authors themselves are guilty of proclaiming heat island effects without more cautionary stances in analyzing several urban sites in relation to one rural site (e.g., Brazel et al. 2000).

Past urban heat island effect analyses for Phoenix primarily included time-trend analyses of urban temperature changes. Missing from these analyses is explicit consideration of rural temperature variability caused by land cover differences. Urban effects depend not only on the land cover changes within the city over time, but also on the land cover specifics of rural environs with which urban sites are compared. Oke (1998) demonstrated energy budget variability between urban and rural areas that depended greatly on the rural base-site land characteristics and wetness. Polonio and Soler (2000) also demonstrated land surface heterogeneity in agricultural areas and the controls on surface energy budget regimes. These factors have implications for weather station placements in rural areas to represent regional conditions. Fiebrich and Crawford (2001) noted in the dense Oklahoma Mesonet considerable variability of meso- and microclimate conditions around rural sites in that network that were said to be a function of vegetation variability. Rural variability is important and is at the heart of debates on locating new sites in a national climate network to assure adequate representation of a region's climatic behavior for climate applications and in assessing global change (Leroy 1998).

In the Phoenix metropolitan area, there is ample opportunity to address the issues of rural and urban spatial and temporal variability relative to the standard approach of comparing urban versus rural temperature trends and magnitudes. The city's surroundings in the mid-twentieth century (population <100 000) were dominated by agricultural landscapes with much surface wetness, with desert lands located considerably away from the burgeoning city urban fringe. Today, the city (population 3 million) has sprawled into an “archipelago” of cities in the expanding urban metropolitan region with new developments leapfrogging into the surrounding desert. The previous agricultural fringe is being altered, producing a patchwork fringe composed of urban, residential, remnant agricultural, sprawl oases, and desert landscapes (Jenerette and Wu 2001).

The growth of the metropolitan area is quickly encroaching on and, in some cases, passing the series of mountain ranges that define the Phoenix valley. These mountains produce the dominant circulation regime for the valley. During the day, solar heating of the mountains to the east induces a thermal low that produces westerly winds. At night, the process is reversed, with cold-air drainage producing easterly winds (Ellis et al. 2000).

Because the circulation is primarily dominated by the larger-scale mountain–valley system, there has not been any demonstration of an urban–rural breeze effect for Phoenix, other than an empirical reference to it by Balling and Cerveny (1987). Sky Harbor Airport had shown increases in wind speed over time that might be due to urban heat island development. The results of this study were debated because of issues regarding instrumentation (Cherry 1988; Balling and Cerveny 1988).

Brazel et al. (2000) examined several urban sites for Phoenix and noted intraurban variability in maximum and minimum temperatures on the order of 1°–2°C and 6°C, respectively, for a dry month—May. It was hypothesized that there would be significant rural temperature variability, based on inspections of aerial photographs and remotely sensed imagery of rural lands around Phoenix. The purpose of this paper is to examine the temperature variability of rural land conditions and to reexamine the contrast of this variability against the urban area for Phoenix.

Methods

A fine-mesh weather station network of dewpoint and temperature sensors was deployed over a small area (approximately 400 m × 300 m) on a local farm in the East Valley of the Phoenix metropolitan area to investigate rural variability. Wind speed and direction were also measured over the grass field on the farm. Field work was conducted locally at Schnepf Farms in Queen Creek, Arizona, for the period from 1300 LST 3 April 2002 to 0600 LST 12 April 2002. The study period ideally would have included the entire period of April and May when Phoenix's urban effect is likely to be largest. Schnepf Farms, however, is a working farm, and the owner could only allow access to his fields for this limited time period without interrupting the farm's ongoing activities. Schnepf Farms is located approximately 30 km southeast of Sky Harbor Airport in the southeast portion of the valley that encompasses Phoenix (Fig. 1). The first-order National Weather Service (NWS) Automated Surface Observing System (ASOS) station for Phoenix is located at Sky Harbor Airport at an elevation of 346 m, 80 m below Schnepf Farms. With the assumption of a dry adiabatic lapse rate of approximately 10°C (1000 m)−1, the maximum temperature difference due to elevation differences is 0.8°C. As will be shown later, 0.8°C is an order of magnitude smaller in relation to the overall urban heat island effect of Phoenix. Therefore, no correction was made to the data for elevation.

Figure 2 shows a map of the farm. For scale, the farm is approximately 400 m wide. Land cover types included a peach orchard, a cornfield, a vegetable garden, a barren field, a pine tree grove, cauliflower and squash fields, compacted soil, and a grass field. Across the farm, in a 3 by 8 array, 24 temperature and humidity sensors were placed at 1.5 m above the local surface (Fig. 2). The columns were labeled from east to west as 1–8. The rows were labeled n (north), c (center), and s (south). Table 1 gives the specific characteristics of the immediate surrounding area for each sensor on 12 April, the last day of the fieldwork.

The sensors used for this study were “HOBO Pro RH/Temp” sensors manufactured by Onset Computer Corporation. HOBO sensors have previously been compared with temperature readings obtained from the ASOS station at Sky Harbor Airport to ensure comparability. The correlation coefficient between the two datasets was 0.98. The 24 HOBOs used on the farm were calibrated to a master sensor. Each HOBO was bolted to an arm that extended from a pole that was 1.5 m above the surface. Aluminum roasting pans shielded the HOBOs from direct radiation. Numerous holes punched in the outer margins allowed for ventilation. There was a 1-cm gap between the top of the HOBO and the shield.

The aluminum roasting-pan shields were used for economic reasons to provide shelter from direct radiation. The shields do not prevent radiation reflected from the surface from being absorbed by the sensor. To assess the impact of reflected radiation, temperature data were collected using HOBOs located side by side. One HOBO was shielded by the aluminum roasting pan, and the other was shielded by an Onset Corporation shield. This process was replicated over both a grass and dirt surface. Discrepancies in daytime temperatures between the two different types of shields were on the order of 5°C over both surfaces because of reflected radiation. However, nighttime temperatures were in good agreement. As a consequence, only nighttime temperature values were used in this study. This method seemed to be a reasonable cost–benefit tradeoff because urban heat island effects are greatest at nighttime. Dewpoint temperatures for the two types of shields were in agreement at all times of day.

Five-minute air temperature Ta and dewpoint temperature Td data were collected for 210 h beginning 1300 LST 3 April, and ending 0600 LST 12 April. The 5-min data were averaged into hourly values for each HOBO such that the 0000 LST data value was the average of the 0000–0055 values. For all 24 HOBOs, the hourly temperature and humidity time series were plotted to gain insight into the overall variation on the farm over the period of study as well as insight into the variation among HOBOs. The entire record was plotted, but only the nighttime values were used for the rest of the analysis.

With the hourly data, a series of contour maps was created to examine the spatial variation in temperature over the whole farm as opposed to a single station. These maps gave insight into the timing and causes of temperature variation over the farm.

A standardized procedure was necessary to examine the entire farm for the entire period of study. For each of the 108 nighttime hours, an hourly air and dewpoint temperature z score for each HOBO was calculated as
i1520-0450-43-3-476-e1
where x is the hourly air or dewpoint temperature value for one HOBO, x is the average air or dewpoint temperature for all of the sensors for that hour, and s is the standard deviation of the air or dewpoint temperature for all of the sensors for that hour. Calculating hourly z scores allowed for each HOBO to be examined for its departures from the mean value over the entire farm. Using the z scores, it could be determined whether some locations on the farm were warmer or cooler, or more or less humid, than other locations over the entire study period. This method allowed for the evaluation of rural areas as baselines in the assessment of the urban heat island effect of Phoenix.
To assess quantitatively the differences among the 24 HOBOs, a two-tailed t test was run for all of the station pairs. The t statistic was calculated as
i1520-0450-43-3-476-e2
where X1 and X2 are the means for two of the HOBO z-score records and
i1520-0450-43-3-476-e3
where n1 and n2 are the sample sizes for the same two HOBOs (108) and
i1520-0450-43-3-476-e4
where s1 and s2 are the standard deviations for each HOBO.

The input data for the t tests were the nighttime hourly air and dewpoint temperature z scores for each of the HOBOs. The calculated t statistic indicated whether there were any significant differences in the mean z score between stations. Thus, the t test indicated whether stations were above or below the mean, and the significance of those departures from the mean. A 24 × 24 matrix of t statistics between all of the HOBO pairs was created for both air and dewpoint temperature. The significantly positive and negative statistics were color coded to reveal a visual pattern in the differences between air and dewpoint temperature departures from the mean for each of the HOBOs. Furthermore, a series of 24 maps (one for each HOBO) was created for both air and dewpoint temperatures using the same color coding so that the spatial pattern of t-test values could be examined. The maps allowed for visual inspection of one station, and the matrix allowed for visual inspection of the entire study area.

The ultimate goal of this study was to assess the differences in the urban heat island effect associated with the use of different rural baselines. Therefore, the maximum and average urban heat island were calculated for the study period for each of the 24 HOBOs. The Arizona State University Office of Climatology supplied hourly Sky Harbor Airport observations that were used to represent the urban temperature. Sky Harbor data were optimal because of their fine time resolution (hourly).

Although the goal of this study was to demonstrate the importance of rural variability in urban climate studies, intraurban climate variability must also be addressed. The Phoenix landscape is composed of two relatively localized central business districts in close proximity to each other. Outside of these areas, the landscape is dominated by huge sprawling tracks of urban/suburban environments. Sky Harbor Airport is centrally located in this dominant part of the city landscape (Fig. 1). To ensure that Sky Harbor was representative of urban conditions, daily average temperatures for Sky Harbor were compared with the Phoenix City station (National Climatic Data Center Cooperative station) for 1948 to 2002. The Phoenix City station is located between the two central business districts and consequently is fairly representative of the heart of the urban area. The correlation coefficient between the daily average temperatures at the two sites was 0.99. The mean absolute error between the two stations was 0.9°C. This variation is an order of magnitude smaller than the typical size of the urban heat island. Graphical comparisons of Sky Harbor Airport temperatures with Phoenix City temperatures revealed a distribution centered on a line with a slope of 1. That is, Sky Harbor temperatures were not skewed in a given direction away from Phoenix City temperatures.

It is certain that there is important intraurban climate variation, and comparison of two urban stations is not sufficient to characterize this variation completely. However, because the goal of this study was to investigate rural variability, Sky Harbor data were deemed suitable for use as an urban baseline from which to compare rural data.

Temperature differences were calculated by subtracting the nighttime rural temperature at each of the 24 HOBOs from the urban temperature at Sky Harbor Airport. Positive differences mean that the urban area was warmer. The maximum urban heat island was simply the largest positive temperature difference over the period of study. The average urban heat island was calculated as the average of the maximum daily urban heat island. In other words, the maximum hourly urban heat island effect from each of the 8 days was averaged.

A 1998 land cover classification of the Phoenix metropolitan region derived using Landsat Thematic Mapper and ancillary data (Stefanov 2000; Stefanov et al. 2001) provided a regional-scale context for the Schnepf field study. An urban–rural fringe buffer zone with an average 4-km width (actual buffer width ranged from approximately 2.5 to approximately 8.5 km) was delineated by defining the urban core as composed of built or disturbed land cover types and the rural fringe as composed of agricultural or native desert land cover types (Fig. 3). This buffer location and width adequately captured the variability of land cover types in the rural areas adjacent to the urban core area. The land cover classification map in the figure has been reduced from 12 to 4 land cover categories for display purposes. Table 2 presents the land cover percentages within the buffer zone.

Based on percentages within the buffer zone, an average value for the urban heat island was calculated for Phoenix. A land cover classification was determined for each of the 24 HOBOs at Schnepf Farms using the same nomenclature as the 1998 satellite imagery classification (Table 1). The vast majority of stations at Schnepf Farms (19 of 24) were cultivated vegetation, whereas cultivated vegetation accounted for only 9.6% of the rural buffer surrounding Phoenix based on the satellite imagery. Other classifications represented at the farm included cultivated grass, compacted soil (prior agriculture use), vegetation, and compacted soil. The sites on the farm represented 30.0% of the land area within the urban buffer of the image. The largest component missing from the farm was undisturbed desert, which accounted for 56.5% of the buffer zone. The next three largest components of the buffer region [compacted soil (prior agricultural use), cultivated vegetation (active), and vegetation] were represented on the farm.

For each of the five classifications represented on the farm, the average and maximum urban heat island were calculated based on the HOBO data that were representative of the classifications. The heat island values were then multiplied by the percent representation within the buffer zone (rescaled to account for classifications not represented on the farm) and summed over the classes to give one number for the maximum and average urban heat island. This one number accounted for many of the rural environments surrounding Phoenix. This approach would be improved by including data from all land classifications (especially undisturbed desert), and increasing the spatial distribution of sampling sites around Phoenix. However, the process does illustrate some important points that are discussed later in the paper.

Results and discussion

Figure 4 shows the air and dewpoint temperature plots for the 24 HOBOs over the period of study. Superimposed over the individual station plots are the average temperature and dewpoint traces of the 24 stations. Note that daytime temperatures are slightly skewed toward higher values because of the radiation shield. The most obvious feature of the graph is the diurnal temperature and dewpoint cycles. Fluctuations for any given hour in the dewpoint for the 24 HOBOs were generally larger than fluctuations in the air temperature. This difference was primarily a function of the differential irrigation practices over the farm (M. Schnepf 2002, personal communication). For the temperature curves, the hourly fluctuations between all 24 HOBOs were generally larger during the day as opposed to at night. This pattern is likely due in part to the effects of the shielding apparatus.

The other obvious feature of the overall graph is the response of the temperatures and dewpoints to the frontal passage during the early part of 6 April. Daytime high temperatures dropped by about 10°C and dewpoints increased by about 5°C. Over the remaining period of the fieldwork, temperatures gradually increased and dewpoints gradually decreased. The average wind speed before and after the frontal passage was 1.7 m s−1. During the frontal passage, the average wind speed was 3.4 m s−1.

Figure 5 shows the synoptic conditions prior to the frontal passage (5 April), during the frontal passage (7 April), and following the frontal passage (10 April). Data were obtained from the National Centers for Environmental Prediction reanalysis (Kalnay et al. 1996). On 5 April, a 500-hPa ridge existed over Arizona. High sea level pressure dominated the central part of the country and stretched into Arizona. At Schnepf Farms, air temperatures were relatively high and dewpoint temperatures were relatively low (Fig. 4). Off of the coast of California, 500-hPa troughing was occurring and sea level pressures were dropping. By 6 April, and into 7 April, the trough had moved into the southwestern United States. Sea level pressures dropped. The frontal boundary can be seen in the sea level pressure field stretching through eastern New Mexico and western Texas. At Schnepf Farms, temperatures dropped and dewpoints rose during this period (Fig. 4). Several days later on 10 April, the trough had pushed eastward and a zonal flow dominated the southwest and much of the United States. Temperature and dewpoint conditions at Schnepf Farms returned to values similar to those prior to the frontal passage.

With the nighttime hourly data available for the entire farm, it was possible to create contour maps of conditions on the farm for any hour over the period of study. Figure 6 shows the extremes in temperature range over the farm. The maximum range occurred at 2000 LST 3 April prior to the frontal passage. The range between the warmest and coolest HOBOs was 5.4°C. The coolest temperature was found in the grass fields in the northeast part of the farm at station 1n.

The warmest stations at this hour were 1c and 2c. These sites were hardpan dirt. The color and texture of the hardpan at 1c and 2c more closely resemble the soil present in the natural desert surrounding Phoenix. The fact that there were several degrees of separation between air temperatures overlying desertlike soil and grass—the standard NWS surface over which to take temperatures—indicates rural baseline temperature variances.

The smallest temperature range of 0.8°C occurred on 0400 LST 7 April, right after the frontal passage when temperatures were depressed and wind speeds were highest. Also, the minimum range occurred in early morning, again when temperatures were coldest. When comparing wind speeds during the times of the smallest and largest temperature ranges, low wind speeds should result in larger temperature ranges because of a lack of mixing. Wind speeds during the high range were 0.1 m s−1; during the low range, wind speeds were 3.2 m s−1.

To assess quantitatively the variation in rural HOBO sites over the whole farm for the entire study period, a t test was run between all the stations using hourly z-score data for air and dewpoint temperatures. The computed t statistic was a representation of how different the stations were to each other relative to the mean of all the stations. Figure 7 shows the 24 × 24 matrix of air temperature t statistics between all stations. The table is meant to be read by looking across the rows as opposed to down the columns. Black cells indicate that the average temperature z score of the row station was significantly warmer than the average temperature z score of the column station. White cells indicate that the row station was significantly colder than the column station. Gray cells indicate that there was no significant relationship. The actual numbers associated with each t test are not nearly as important as the pattern of shading over the entire matrix. Examining the rows across the matrix illustrates this point.

Row 1n, the grass field, has the coldest nighttime temperatures. This is the standard NWS observing surface but is not representative of rural agricultural and desert areas. Rows 5n through 6c are for the most part colder than all other stations. These stations correspond to the eastern part of the peach orchard. Stations 6s and 8s, the southwestern part of the orchard, also were colder than the rest of the farm except for the eastern orchard. The northwestern part of the orchard, stations 7c, 8n, and 8c, was the warmest part of the orchard. The trees in the northwestern part of the orchard had previously been killed by frost, and thus the microclimate associated with 8n and 8c was dramatically different from the rest of the orchard.

The warmest locations overall, as evidenced by Fig. 7, were stations 1c and 2c, over the hardpan dirt surface. The next warmest stations were 4n and 4c, the well-watered crop field. With ample moisture, these surfaces did not cool as much during the night, thus producing large departures from the average.

Figure 8 shows 3 of the 24 maps that were created for each HOBO as examples. The maps displayed were created for (a) the coldest station, (b) the warmest station, and (c) a moderate station. These maps further facilitated interpretation of the t-test matrix.

The same method was used to create a t-test matrix for dewpoint temperatures (not shown). A pattern very similar to the temperature matrix existed. The most humid part of the farm was the east side of the orchard. The next most humid area was the tiny grove of pine trees. For both the orchard and the pines, the canopy probably served to keep humidity values relatively high. The stations in the cornfield, although not possessing a canopy, had recently been flood irrigated and consequently had higher humidity values.

The driest locations over the study period were the mowed grass field and the hardpan dirt surfaces on the eastern edge of the farm. These surfaces did not have a canopy, were not irrigated, and were located farthest away from the most humid locations.

Because the main goal of this project was to assess the rural baseline when evaluating the urban heat island effect, the average and maximum urban heat island were calculated for each HOBO over the period of study. Table 3 shows the results of the urban heat island calculations for each HOBO, assuming it was the only site to represent a “rural case.” Depending on which HOBO was used to calculate the heat island, the average heat island over the study period ranged from 9.4° to 12.9°C, a difference of 3.4°C. The maximum heat island ranged from 10.7° to 14.6°C, a difference of 3.9°C. For both the average and maximum heat islands, the smaller heat island value occurred using station 2c, the hardpan dirt surface, and the larger heat island value occurred using 1n, the grass surface. The maximum heat island for each HOBO occurred on 10 April between 2000 and 2100 LST, after sundown and when urban heat island evolution is expected to peak during the cooling phase (Oke 1988). High pressure and the associated stable conditions had been developing for 4 days. Also, the sun had been set long enough for the rural farm to cool but not long enough for the urban core to cool, thus producing the large temperature differential.

The smallest urban heat island values occurred during the passage of the frontal system beginning on 6 April. Using the average heat island effect for the entire farm, 20 out of the 25 smallest heat island effects occurred during, or just prior to, the frontal passage. During this time, relative humidities and wind speeds were higher throughout the Phoenix area. As a consequence, temperature variations were more moderate, thus suppressing the heat island.

There is clearly enough significant temperature and dewpoint variation observed at the scale of a single small farm to question the validity of using a single baseline value to represent rural conditions in assessment of urban heat island effect. These variations arise from differences in land cover (mainly presence/absence of vegetation and differences in soil characteristics). However, the question remains as to whether this variation of land cover is a local effect (and perhaps not significant at the regional scale) or is present at larger spatial scales (i.e., the urban/rural fringe) that would be significant in assessment of urban heat island effects.

Recasting of the Schnepf land cover observations to match the land cover classification of Stefanov et al. (2001) allows for comparison between the farm-scale and regional urban–rural fringe land cover percentages (Tables 1 and 2). Although attempts to answer the validity of scaling the Schnepf Farms data to the entire Phoenix urban fringe are well beyond the scope of this study, it is still interesting to assess the Phoenix urban heat island on a regional level using land classifications from satellite imagery and temperature values for the classifications from Schnepf Farms. Temperature data from the farm were scaled based on percent area represented in the buffer zone of the satellite imagery. This process yielded one number for the Phoenix urban heat island that accounted for differences in the rural environment. This process produced a maximum and average urban heat island of 12.6° and 11.0°C, respectively.

Strong caution should be taken not to use these numbers as a definitive assessment of Phoenix's heat island. The values do not account for large portions of the rural area (especially undisturbed desert) or variation that may occur at different geographical locations around Phoenix. The maximum and average urban heat island temperatures derived using the rural buffer values fall well within the ranges obtained from the local Schnepf Farms data. This is a result of the analytical approach used wherein undisturbed desert is not considered when scaling up the Schnepf Farms data. In other words, the portion of rural buffer area land cover considered in the analysis (∼30%) is sufficiently similar to the Schnepf Farms land cover percentages that general agreement of the average and maximum urban heat island values is obtained. Inclusion of the undisturbed desert class in the analysis would be expected to move the calculated urban heat island values toward the lower end of the Schnepf Farms ranges, because the undisturbed Sonoran desert surface is more physically similar (in general) to the farm hardpan dirt surfaces than the farm grass surfaces. As stated previously, however, quantitative assessment of the effect of inclusion of undisturbed desert in the urban–rural comparison will require collection of temperatures at rural sites with a significant percentage of undisturbed desert land cover.

Demonstrated variations in land cover characteristics at both the rural farm scale and the regional scale exert significant control on urban heat island values. Another source of variability to consider when comparing urban and rural air temperatures is the effect of topography. The urban–rural buffer region we have defined for the Phoenix metropolitan area contains significant topographic and elevational variation (∼300–3000 m): it includes wash and river channels, level agricultural areas (e.g., the Schnepf Farms), alluvial fans, pediments, and a mountain range. This topographic variability will influence the regional to local wind patterns and associated homogenization of diurnal air temperatures measured at a given location and could therefore introduce significant variance to calculated urban heat island values. Regional airflow and temperature simulation models constructed using digital topographic and remotely sensed biogeophysical data [e.g., fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5); Grell et al. 1994; Zehnder 2002] are necessary to address adequately the potential degree of variability introduced by topography.

The point of this exercise is to illustrate the complex nature of assessing the impact of a built environment. The results presented here partially account for rural variation. Most other heat island studies do not take rural differences into account. Despite the intent in this study to capture the rural variability, there are still large gaps in the assessment. An ideal situation to employ the methods discussed in this paper would necessitate the establishment of numerous similarly diverse study sites around the Phoenix core that allowed for the collection of data over all of the land use classifications at various geographical locations. At that point, much greater confidence could be had in using a single value to assess the impact of the built environment.

Conclusions

The purpose of this dense field weather station deployment was to assess the relative effect of variable rural conditions in the characterization of the urban heat island effect for Phoenix. Twenty-four HOBO temperature and humidity sensors were placed in a dense 3 × 8 grid over a wide range of surfaces at a rural farm. The goal was to examine the variation in the rural nighttime temperatures associated with the different microclimatic conditions. The major results of this field project may be summarized as follows:

  • Variation in temperature over the entire farm at a given time during the study period ranged from 0.8° to 5.4°C. This wide difference was mostly due to differences in irrigation practices.
  • The t-statistic values revealed that, over the entire nighttime period, the grass field and the area of the peach orchard in which the trees were still alive were the coolest on the farm. Areas over the hardpan dirt and dry, fallow field were the warmest.
  • Depending on which HOBO was used as the rural baseline, the average urban heat island ranged from 9.4° to 12.9°C, a difference of 3.4°C. The maximum heat island ranged from 10.7° to 14.6°C, a difference of 3.9°C.
  • Regionally scaling the Schenpf Farms data to the entire Phoenix area using satellite imagery produced a maximum and average urban heat island of 12.6° and 11.0°C, respectively. These numbers do not account for large portions of the rural environment and are simply used to show that one must be cautious in assessing the impact of the built environment.

Phoenix, Arizona, has previously been shown to have one of the largest urban climate effects in the world (Hansen et al. 1999). That finding was further corroborated with this study. This study also illuminated the fact that it is critical, when assessing the urban effect on an area's climate, to consider the nature of the rural baseline. For Phoenix, depending on which rural baseline was used, a significant difference in the value of the urban heat island resulted. As urban climatologists and ecologists monitor change of the urban landscape in the future it will also be increasingly important to track the rural change in reference to urban growth in order to observe urban effects and to model change attributable to urbanization and growth.

Acknowledgments

This work resulted from a graduate seminar entitled “Urban Climatology.” Thanks are given to Mark Schnepf for allowing us access to the study site and to Chuck Saffell, Nancy Selover, Brent Hedquist, Pamela Szatanek, Brenda Koerner, and Tom Moore for field assistance. Thanks are also given to the three anonymous reviewers who helped to improve this manuscript from earlier versions.

REFERENCES

  • Balling, R. C., and R. S. Cerveny. 1987. Long-term associations between wind speeds and the urban heat island of Phoenix, Arizona. J. Climate Appl. Meteor. 26:712716.

    • Search Google Scholar
    • Export Citation
  • Balling, R. C., and R. S. Cerveny. 1988. Reply. J. Appl. Meteor. 27:881.

  • Bejaràn, R. A., and I. A. Camilloni. 2003. Objective method for classifying air masses and application to the analysis of Buenos Aires' (Argentina) urban heat island intensity. Theor. Appl. Climatol. 74:93103.

    • Search Google Scholar
    • Export Citation
  • Brazel, A. J., , N. Selover, , R. Vose, , and G. Heisler. 2000. The tale of two climates—Baltimore and Phoenix urban LTER sites. Climate Res. 15:123135.

    • Search Google Scholar
    • Export Citation
  • Cherry, N. J. 1988. Comment on “Long-term associations between wind speeds and the urban heat island of Phoenix, Arizona.”. J. Appl. Meteor. 27:878880.

    • Search Google Scholar
    • Export Citation
  • Comrie, A. C. 2000. Mapping a wind-modified urban heat island in Tucson, Arizona (with comments on integrating research and undergraduate learning). Bull. Amer. Meteor. Soc. 81:24172431.

    • Search Google Scholar
    • Export Citation
  • Ellis, A. W., , M. L. Hildebrandt, , W. H. Thomas, , and H. J. S. Fernando. 2000. Analysis of the climatic mechanisms contributing to the summertime transport of lower atmospheric ozone across metropolitan Phoenix, Arizona, USA. Climate Res. 15:1331.

    • Search Google Scholar
    • Export Citation
  • Fiebrich, C. A., and K. C. Crawford. 2001. The impact of unique meteorological phenomena detected by the Oklahoma Mesonet and ARS Micronet on automated quality control. Bull. Amer. Meteor. Soc. 82:21732187.

    • Search Google Scholar
    • Export Citation
  • Grell, G. A., , J. Dudhia, , and D. R. Stauffer. 1994. A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note NCAR/TN-398+STR, 117 pp.

    • Search Google Scholar
    • Export Citation
  • Hansen, J., , R. Ruedy, , J. Glascoe, , and M. Sato. 1999. GISS analysis of surface temperature change. J. Geophys. Res. 104:3099731022.

  • Jenerette, G. D., and J. Wu. 2001. Analysis and simulation of land-use change in central Arizona–Phoenix region, USA. Landscape Ecol. 16:611626.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E. Coauthors, 1996. The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc. 77:437471.

  • Klysik, K., and K. Fortuniak. 1999. Temporal and spatial characteristics of the urban heat island of Lodz, Poland. Atmos. Environ. 33:38853895.

    • Search Google Scholar
    • Export Citation
  • Leroy, M. 1998. Meteorological measurements representativity, nearby obstacles influence. Preprints, 10th Symp. on Meteorological Observations and Instrumentation, Phoenix, AZ, Amer. Meteor. Soc., 233–236.

    • Search Google Scholar
    • Export Citation
  • Livada, I., , M. Santamouris, , K. Niachou, , N. Papanikolaou, , and G. Mihalakakou. 2002. Determination of places in the great Athens area where the heat island effect is observed. Theor. Appl. Climatol. 71:219230.

    • Search Google Scholar
    • Export Citation
  • Lowry, W. P. 1977. Empirical estimation of urban effects on climate: A problem analysis. J. Appl. Meteor. 16:129135.

  • Morris, C. J. G., , I. Simmonds, , and N. Plummer. 2001. Quantification of the influences of wind and cloud on the nocturnal urban heat island of a large city. J. Appl. Meteor. 40:169182.

    • Search Google Scholar
    • Export Citation
  • Oke, T. R. 1987. Boundary Layer Climates. Routledge, 435 pp.

  • Oke, T. R. 1988. The urban energy balance. Phys. Geogr. 12:471508.

  • Oke, T. R. 1998. Observing weather and climate. Proc. Technical Conf. on Meteorogical and Environmental Instruments and Methods of Observation, Geneva, Switzerland, WMO, Rep. 70, WMO/TD-No. 877, 1–8.

    • Search Google Scholar
    • Export Citation
  • Oke, T. R. 1999. Observing urban weather and climate using “standard” stations. Preprints, Biometeorology and Urban Climatology at the Turn of the Millennium: Selected Papers from the Conf. ICB-ICUC'99, Sydney, Australia, WMO, 443–448.

    • Search Google Scholar
    • Export Citation
  • Oke, T. R., , R. A. Spronken-Smith, , E. Jauregui, , and C. S. B. Grimmond. 1999. The energy balance of central Mexico City during the dry season. Atmos. Environ. 33:39193930.

    • Search Google Scholar
    • Export Citation
  • Polonio, D., and M. R. Soler. 2000. Surface fluxes estimation over agricultural areas: Comparison of methods and the effects of land surface inhomogeneity. Theor. Appl. Climatol. 67:6579.

    • Search Google Scholar
    • Export Citation
  • Stefanov, W. L. cited 2000. 1985, 1990, 1993, 1998 land cover maps of the Phoenix, Arizona metropolitan area. Geological Remote Sensing Laboratory, Department of Geological Sciences, Arizona State University, Tempe, AZ, 4 plates. [Available online at http://elwood.la.asu.edu/grsl/Her/land_cover_phy.html.].

    • Search Google Scholar
    • Export Citation
  • Stefanov, W. L., , M. S. Ramsey, , and P. R. Christensen. 2001. Monitoring urban land cover change: An expert system approach to land cover classification of semiarid to arid urban centers. Remote Sens. Environ. 77:173185.

    • Search Google Scholar
    • Export Citation
  • Zehnder, J. A. 2002. Simple modifications to improve fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model performance for the Phoenix, Arizona, metropolitan area. J. Appl. Meteor. 41:971979.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

The Phoenix metropolitan area. Sky Harbor Airport is represented as a square, and Schnepf Farms is represented as a circle

Citation: Journal of Applied Meteorology 43, 3; 10.1175/1520-0450(2004)043<0476:TRORVI>2.0.CO;2

Fig. 2.
Fig. 2.

Schematic diagram of the study area used at Schnepf Farms. Locations of the field HOBO sensors are given as black circles. Here, n, c, and s refer to north, central, and south. For scale, the farm is approximately 400 m wide

Citation: Journal of Applied Meteorology 43, 3; 10.1175/1520-0450(2004)043<0476:TRORVI>2.0.CO;2

Fig. 3.
Fig. 3.

Land cover classification of the Phoenix metropolitan area for 1998 (Stefanov et al. 2001). The inner buffer limit was selected as the boundary between built land cover types and either agricultural or undisturbed land cover types. The outer buffer limit was then placed at an average distance of 4 km. The land cover classification map has been reduced from 12 to 4 land cover categories for display purposes: white—undisturbed; light gray—cultivated vegetation (active), cultivated grass, compacted soil (prior agricultural use), vegetation; dark gray—fluvial and lacustrine sediments (canals), disturbed (commercial/industrial), disturbed (asphalt and concrete), compacted soil, disturbed (mesic residential), disturbed (xeric residential); and black—water. Black polygons are the inner and outer buffer limits

Citation: Journal of Applied Meteorology 43, 3; 10.1175/1520-0450(2004)043<0476:TRORVI>2.0.CO;2

Fig. 4.
Fig. 4.

Time series of temperature and dewpoint records for all 24 HOBOs (gray lines). Dewpoint traces are the lower set of lines. The two thick black lines represent the average temperature and dewpoint traces. Values of the x axis are days of the month for Apr

Citation: Journal of Applied Meteorology 43, 3; 10.1175/1520-0450(2004)043<0476:TRORVI>2.0.CO;2

Fig. 5.
Fig. 5.

Sea level pressure fields (shading; hPa) and 500-hPa geopotential height fields (dark lines; gpm) for (a) 5, (b) 7, and (c) 10 Apr

Citation: Journal of Applied Meteorology 43, 3; 10.1175/1520-0450(2004)043<0476:TRORVI>2.0.CO;2

Fig. 6.
Fig. 6.

Contour maps of air temperature (°C) values for (a) 2000 LST 3 Apr, the largest range over the farm, and (b) 0400 LST 7 Apr, the smallest range over the farm

Citation: Journal of Applied Meteorology 43, 3; 10.1175/1520-0450(2004)043<0476:TRORVI>2.0.CO;2

Fig. 7.
Fig. 7.

Matrix of t statistics for air temperature between the 24 HOBO station z-score records. Black cells indicate that the row station was significantly warmer than the column station averaged over the study period. White cells indicate that the row station was significantly colder than the column station. Gray cells indicate that the t statistic was not significant at the 99% confidence level

Citation: Journal of Applied Meteorology 43, 3; 10.1175/1520-0450(2004)043<0476:TRORVI>2.0.CO;2

Fig. 8.
Fig. 8.

Maps created using the t-test matrix. Of the 24 maps that were created, only maps for the (a) coldest station 6n, the (b) warmest station 2c, and (c) a moderate station 3s are shown. Black circles indicate that the reference station was warmer. White circles indicate that the reference station was colder. Gray circles indicate that the station was not significantly different than the reference station

Citation: Journal of Applied Meteorology 43, 3; 10.1175/1520-0450(2004)043<0476:TRORVI>2.0.CO;2

Table 1.

Surrounding ground cover conditions and satellite land use classification for each of the 24 HOBO stations (1 in. = 2.54 cm; 12 in. = 1 ft)

Table 1.
Table 2.

Percent area of the different land cover classes contained within the buffer region surrounding Phoenix

Table 2.
Table 3.

Maximum and average urban heat island effect using Sky Harbor Airport as the urban station and each of the 24 HOBOs as the rural station

Table 3.
Save