Street-Scale Urban Air Temperatures Predicted by Simple High-Resolution Cover- and Shade-Weighted Surface Temperature Mosaics in a Variety of Residential Neighborhoods

Abstract

A simple statistical model capturing the degree to which different patterns of urban development intensify urban heat islands (UHIs) and stress human health would be useful but has remained elusive. Accurately predicting street-level urban air temperatures from land cover and thermal data is difficult due to (1) the coarse scale of common remote sensing data, which do not observe the key environments beneath urban tree canopies, and, (2) conversely, the immense labor of intense, location-specific, ground-based survey campaigns. This work tested whether remotely sensed urban heat merged with land cover heterogeneity and shade/sun fractions, if combined at a sufficiently fine scale so as to be linearly additive, would enable simple and accurate statistical modeling of street-scale urban air temperatures with minimal empirical fitting. We used ground-based thermography of a sample of 12 residential streetscapes in Portland, Oregon, to characterize the land surface temperatures (LST_g) of eleven common urban surface cover types when sun-exposed and in shade. Surfaces were cooler in shade than sun, but with surface-specific differences not explained by greenery nor (im)perviousness. Also, surfaces on streetscapes with more canopy cover, even when sun-exposed at midday, remained significantly cooler than comparable sun-exposed surfaces on streets with less canopy cover, indicating the key significance of partial diurnal shading, not typically accounted for in urban thermal statistical models. We used high-resolution orthoimagery to quantify the area of each surface cover type within each streetscape and computed an area-weighted average surface temperature (T_s), accounting for sun/shade heterogeneity. The data revealed a significant, nearly 1:1 relationship between calculated T_s values and sun-shielded air temperatures (T_a). In contrast, relationships of T_a to tree coverage, impervious area, or the LST_g of dominant surface cover types were all statistically insignificant. These results suggest that statistical models may more reliably bridge the gap between remote sensing urban surface temperatures and reliable predictions of street-scale air temperatures if (1) analysis is at a sufficiently high resolution (e.g., <10 m) to avoid some of the known scale-dependence of urban thermal environments and enable simple weighted linear models, and (2) distinctions between thermal contributions of sunlit and shaded surfaces are included along with the influence of diurnal shading. Such models may provide effective and low-cost predictions of local UHIs and help inform effective street-level approaches to mitigating urban heat.

1. Introduction

Sustainable urban futures require that development, which has historically fragmented landscapes and increased urban air temperatures, imperviousness, water demands, water losses, and human stress and discomfort, instead begin to mitigate or reverse these negative trends. Although only 55% of global population lived in urban areas in 2018, this figure is projected to be 68% by 2050 [1]. Growing urban populations will be accommodated by continued urban expansion and by redevelopment-intensification of existing urban areas. Large increases in pavement, roof, and other impervious surface cover associated with modern urban development greatly increase urban temperatures and increase resident discomfort and health risk, especially in the summer [2,3,4]. High temperatures and solar radiation near the ground can also stimulate the production of photochemical smog, a human respiratory and eye irritant [5,6]. Relations between urban heat and land cover remain an active area of research because a simple model that could accurately predict the degree to which different patterns of urban development would also intensify urban heat islands (UHIs) and stress human health would be immensely useful for urban planning and urban climate mitigation yet has remained elusive for decades [7,8,9]. Although highly-resolved urban climate assessments and predictions can be made with complex physics-based numerical models, the parameterization and computational requirements of such models make them impractical for simple, routine urban heatscape monitoring and assessment over large areas [10]. Therefore, while micro-scale models provide opportunities for improved process-based insights, there remains a need for more computationally simple approaches for the direct management of urban environments [11,12,13]. For example, exposed pavement land surface temperature (LST) can be successfully modeled using solar radiation and air temperature data [14], but the desired inverse model to accurately predict urban air temperatures from LST across a wide range of conditions has remained a challenge. Statistically relating urban heat to urban landscape composition (surface cover types) and form (building sizes, building and greenspace configurations) would not only be useful for understanding human thermal comfort [15,16], but also for building energy use [17,18], air quality [19], water use [20,21], stormwater runoff [22,23,24], and other water, energy, carbon, and resource-exchange processes in urban environments [25,26].

Taking an empirical approach, studies have statistically examined urban heat-land cover relations, especially using satellite remote sensing data at scales of tens of meters to kilometers since Rao (7.4 km ITROS-1; 1972) [27], Matson et al. (0.9 km NOAA-5; 1978) [28], Price (0.5 km HCMM; 1979) [29], and Roth et al. (1989) [30], among others. However, the plot- to street-scale provides the most appropriate and attainable scale for change to urban form within current policy frameworks yet is less commonly assessed [31,32]. Satellite remote sensing cannot elucidate the impacts of shading or under-canopy dynamics necessary to understand street-scale urban microclimate dynamics without additional data on the below-canopy surface covers and the effects of canopy shade [33]. The configuration of contrasting urban land cover types at very fine-scale levels may also exert a significant effect on LST [34,35]. The question of scale is both unavoidable and consequential in studies of the intensely heterogenous urban environment. Mirzaei summarized the differences between building, micro-scale or microclimate, and city-scale models of UHIs and argued that city-scale models that are built upon the remote sensing evaluations of that time, which were at low spatial resolution, inherently cannot account for the complex geometric structure of cities [10]. Because a single remote sensing pixel typically contains many different surface cover types including roofs, streets, multiple types of vegetation, and even oddities such as parked cars, most remote sensing efforts to characterize the air temperature variability of UHIs are not really able to account for the fine-scale complexities of urban land cover, its differential surface heating, and its non-linear and heterogeneous surface-air heat flux feedback processes [36]. The detailed quantification of surface temperature variations, surface cover types, and their co-variation with urban air temperatures at the street scale, parcel scale, and sub-parcel scale that is needed to support improved development of theory and models is also largely lacking. Such finer-scale evaluations are necessary to help illuminate the biophysical processes underlying the creation and distribution of urban heat, to document and understand scaling issues of UHIs, to reduce scale discrepancies between data sources and models used, and to provide policy-relevant guidance into cooling potential of urban form manipulation [37,38,39]. Disconnects between the location-smoothing but spatiotemporally-extensive data available from remote sensing and the highly resolved but comparatively spatially sparse ground-based sensor observations of urban heat dynamics therefore persist in part due to this scale-mismatch [7,8]. For example, in Wisconsin, USA, although both remote sensing and sensor-based methods could detect the impact of urban heat on extending the growing season of urban vegetation, with consequences for plant water use and local energy balances, there was no relationship between the two methods at the individual sensor level [40].

A recent explosion of studies has sought to overcome these issues of insufficient urban air temperature station density or remote sensing pixel measurement specificity by using high-density mapping by mobile air temperature sensors, e.g., mounted to bicycles or cars. These studies collect streaming-transects of air temperature data along pre-defined routes spanning a variety of land use areas and during specified time windows of intensive city-wide data collection campaigns [8]. The studies then use these intensive ground-based data to calibrate statistical models ingesting remote sensing-based urban vegetation, canopy cover, building volume, impervious fraction, LIDAR, and/or satellite spectral signatures to interpolate UHIs at high spatial resolution [41,42,43,44,45,46,47,48,49,50]. The geometry and thermal dynamics of urban canyons, where streets are bordered by buildings on both sides, complicate the relationships between surface and air temperature in urban environments [19,51], but can be calibrated using the intensive data of such mobile-mapping field campaigns. Representing significant advances in high-resolution, context-aware, empirical mapping of urban air temperatures, such campaign-based approaches represent extraordinarily useful UHI interpolation methods but are not able to represent simple yet generalizable UHI modeling and prediction methods. The campaign-based UHI heat map interpolations are laborious to produce and unavoidably specifically tied to the range of land cover and microclimate conditions captured by the survey routes during the defined temporal sampling windows. Yet, these mobile air temperature mapping campaign studies do highlight the utility of fine-scale data for considering spatially differentiated impacts of UHIs on urban ecosystems and residents.

Because statistically relating remotely sensed landscape composition, configuration, and thermal imaging directly to the air temperatures of the UHI has remained challenging, many remote sensing studies have pivoted to instead study surface urban heat islands (SUHIs). SUHIs are distinguished by the elevation of urban land surface temperatures (LST) above the LST of the peri-urban or rural surroundings, rather than by the difference in urban-rural air temperature that defines a UHI. Many city-scale studies have documented the spatiotemporal variability of LST within an urban SUHI area, i.e., the city’s “microclimate thermal extremes” [52]. Research since the first satellite SUHI observations by Rao [27] has sought to understand SUHI covariance with urban land use and generalized land cover categories, spatial patterns of imperviousness, and tree canopy cover, among other remote sensing-based metrics. This field of study was reviewed by Voogt & Oke and Arnfield in 2003 [7,8] and has since continued in the same vein with many more studies [52,53,54,55,56,57,58]. Fractional imperviousness and tree canopy coverage are, by far, the most common variables employed to try to explain SUHI variations both within and between cities, but the coefficients of determination between SUHI values and imperviousness or canopy cover values are highly variable and often weak. Because a single remote sensing pixel typically contains many different surface cover types including roofs, streets, multiple types of vegetation, and even bodies of water, remote sensing-based assessments of SUHIs still suffer from the challenges of coarse pixel resolutions, sub-pixel mixing and smoothing, and inability to view beneath canopy cover or other obstructions.

To provide new insights on urban warming as related to the fine scale surface cover variations of plots, parcels, and residential streets, the aims of this study were to: (1) identify the fine-scale variation in summer surface temperatures across many urban surface covers in situ, (2) evaluate the influence of shading on surface temperatures at the sub-parcel scale, and (3) determine the utility of using fine-scale surface cover, shading, and thermal data to model the relationship between surface temperature measurements and overall streetscape air temperatures.

2. Methods

2.1. Field Sites

The study was conducted using a sample of 12 distinct segments of residential urban streets in Portland, Oregon, USA (Figure 1). Each site consisted of one longitudinal half of a street canyon, spanning the length of the city block between cross-streets and spanning from the peak of the residential houses’ rooflines to the street centerline. The 12 sites represented triplicate sampling across two key factors of urban canyon form and function: tree canopy overhanging the street and dominantly coniferous vs. broadleaf deciduous street tree types. High-resolution (1-m) orthoimagery from 2008 or 2012 (DOI:10.5066/F73X84W6, https://earthexplorer.usgs.gov/ (accessed on 8 June 2016))) was used to digitize and quantify total site area, street area, and street-overhanging tree canopy areas. The six streets designated low tree cover averaged 11% street-overhanging canopy (range 6–23%) and the six streets designated high tree cover averaged 45% overhanging canopy (range 35–51%). All 12 study streets ran in a north-south direction to minimize differential impacts of diurnal heating or regional winds [59].

2.2. Thermography and Air Temperature Data Collection

Sites were visited on 27–29 June 2016, with each site visit occurring on one day to ensure site-level consistency of all data collected. These days had similar meteorological conditions at Portland International Airport (sunny, no precipitation, daily maximum temperature 25.5–29.4 °C, daily minimum temperature 13.5–15.5 °C, average daily wind speed 2.9–3.6 m/s, and average wind direction 315–330°) [60]. Thermal imaging of each site was conducted between 1–5 pm (

Table 1. Study site locations and characteristics.

ID	Street Tree Type	Canopy Density	Location	Date Imaged	Times Imaged	Site Area (m2)
DL1	Deciduous	Low	East of N Exeter Ave between N Newark St and N Hudson St	27 June	3:46–3:55 p.m.	4316
DL2	Deciduous	Low	East of N Chautauqua Blvd between N Winchell St and N Baldwin St	28 June	2:16–2:28 p.m.	14,645
DL3	Deciduous	Low	East of SE 73rd Ave between SE Raymond St and SE Schiller St	29 June	2:23–2:32 p.m.	4348
DH1	Deciduous	High	West of N Exeter Ave between N Newark St and N Hudson St	27 June	3:29–3:45 p.m.	4484
DH2	Deciduous	High	West of N Chautauqua Blvd between N Winchell St and N Baldwin St	28 June	2:00–2:14 p.m.	5689
DH3	Deciduous	High	Both sides of SE 72nd Ave between SE Schiller St and SE Foster Rd	29 June	2:44–2:56 p.m.	4573
EL1	Evergreen	Low	East of 135th Ave between SE Yamhill St and SE Salmon St	28 June	4:35–4:39 p.m.	2301
EL2	Evergreen	Low	West of N Exeter Ave between N Fessenden St and N Cecelia St	28 June	3:56–4:04 p.m.	6635
EL3	Evergreen	Low	East of N Exeter Ave between N Fessenden St and N Cecelia St	28 June	3:13–3:19 p.m.	904
EH1	Evergreen	High	West of NE 22nd Ave between NE Irving St and NE Hoyt St	29 June	1:39–2:20 p.m.	3198
EH2	Evergreen	High	West of SE 50th Ave between SE Steele St and SE Insley St	27 June	1:08–2:49 p.m.	3522
EH3	Evergreen	High	East of SW 45th Ave between SW Vesta Dr and SW Vacuna St	27 June	1:52–2:21 p.m.	4065

). Climatologically, Portland experiences a warm-summer Mediterranean climate (Csb) [61]. Average June high temperature is 23.0 °C, mean 17.9 °C, low 12.3 °C, and precipitation 41 mm. Only July and August are warmer and drier, and only 11% of average annual precipitation for Portland occurs during July-September (1991–2020 climate normal) [62]. On average, 28-June air temperatures peak at about 25 °C, with less than 0.2 °C cooling by 5 pm and only about 1.6 °C cooler temperatures at 1 pm (1991–2020 climate normal) [63].

To collect each thermal image, the camera operator stood on the street curb opposite the target streetscape and took an image at a horizontal view angle using a FLIR T650sc thermal camera (pixel array: 640 × 480, accuracy: ±1 °C, FOV: 25° × 19°). Each thermal image was saved as a numerical data array of 307,200 thermal surface temperature values, one for each pixel. Each thermal image depicted a portion of the opposing streetscape’s residential landscape and included in the images ground surfaces, parked car, yard landscaping objects, etc., at distances approximately 5–20 m from the camera (e.g., Figure 2b). The camera operator then moved laterally down the block, taking sequential, adjacent images with some overlap between images. This resulted in a tiled, longitudinal, panorama-like compilation of the and residential streetscape “heatscape”. The FLIR camera captured concurrent optical images (e.g., Figure 2a), and a GPS waypoint was recorded at the location of each picture. A final longitudinal photo looking down the street was used to capture the temperature of the street surface itself (e.g., Figure 2c), as the camera view from the opposite curb did not always include much of the street surface in the foreground.

Throughout the study period, air temperature data (T_a) were collected at each street study site at 10-min intervals using iButton temperature loggers hung from trees at least six feet above the ground and in shade, to provide natural radiation shielding. For each site, the streetscape thermal environment a pedestrian might experience was represented by the mean afternoon T_a on the day of thermal imaging, calculated as the mean of that site’s temperature readings from 12:00 p.m. until 5:00 p.m. This averaging also minimized possible effects of intermittent cloud cover. For sites DH2 and DL2 the iButton loggers unexpectedly failed during the imaging day, so recorded data from the most recent day with similar weather conditions was used. Having imaged sites DH2 and DL2 on June 28, when the Portland International Airport recorded a high temperature of 26.1 °C and no precipitation, we used the sites’ temperature data from June 25, when the Airport weather station had recorded a high of 25.6 °C and no precipitation [60].

2.3. Thermal Image Analysis

To delineate the surface cover types and quantify their temperatures within each thermal image a MATLAB code was developed (version 9.1, R2016b). The code displayed a thermal image on which the user manually digitized polygons to delineate and categorize each surface type present, using multiple polygons per surface type if needed. Polygons were intentionally drawn to end slightly before the visible edge of a surface type to avoid edge effects with adjacent surfaces. The 11 surface cover types categorized were concrete (light pavement, sidewalks, some driveways), asphalt (dark pavement, streets, some driveways), building roofs (typically asphalt shingle), parked cars, green grass, brown grass, gravel, mulch, shrubbery, bare ground (earth), and other (decking, shade umbrellas, etc.). For each surface cover type, areas in sun or in shade at the time of imaging were delineated separately, yielding 22 cover/shade combinations. For all the images comprising a single study site, all digitized polygons of the same cover/shade category were considered together, and one mean temperature calculated from all the pixels in those polygons for that street. We refer to these site-specific, cover/shade-specific, mean land surface temperatures extracted from ground-based thermography as LST_g values, e.g., one LST_g value representing the mean of all delineated thermal imagery pixels for all the concrete surfaces imaged in the sun at site DL1. For a few sites, a cover/shade category visible in the streetscape’s aerial orthoimagery was not visible in the thermal imagery; in these cases, the cover/shade category was assigned the mean LST_g of its comparable cover/shade conditions at the other two sites of the same canopy coverage and dominant street tree type classes.

2.4. Surface Cover Mapping

In addition to digitally locating and characterizing the average temperatures of each cover/shading combination in the thermography of the study sites, 1 m-resolution aerial orthoimagery (from 2008 or 2012; DOI:10.5066/F73X84W6, https://earthexplorer.usgs.gov/ (accessed on 8 June 2016)) was used to manually delineate the aerially-visible areas of the 11 key land cover types and tabulate the fractional cover of each to the total planar area of each field site (Figure 2e). As each site consisted of the street-facing half of the properties on one side of a residential street, from the houses’ roof peaks to the center of the street pavement the same 11 surface cover types were digitized from the orthoimagery as were sought within the thermal imagery. These surface covers were further classified as vegetated-pervious (green grass, brown grass, shrubbery, tree canopy), bare-pervious (mulch, bare ground, gravel), or impervious (concrete, asphalt, roofs, cars). Where surface cover was hidden in the satellite imagery due to tree canopy or other obstruction, an estimate was made of the hidden cover (e.g., grass continuation beneath the canopy of a tree surrounded by grass), and Google Streetview [64] was used as a supplementary reference to verify the surface cover types where possible. The fraction of each streetscape’s total planar area covered by each surface cover type was estimated as mapped from the orthoimagery; for example, 20% of the digitized land cover polygon areas for a site may have been observed to be dark-colored asphalt pavement.

2.5. Surface Cover Shade Apportioning and Average Streetscape Surface Temperatures

For each streetscape, the percent of each of the 11 surface cover types that was shaded during the heat of the day was estimated as the ratio of shaded area to total area of that cover type that was identified within the thermal imagery of that site; e.g., 10% of the pixels observed and delineated as asphalt in all the thermal images for one streetscape may have been categorized as shaded. The overall shaded-asphalt area for the streetscape was then calculated by applying this fraction to the total area of that surface cover for the site; e.g., if a site was estimated to have 600 m² of asphalt area (from orthoimagery digitization, as above) and 10% asphalt shade cover (from thermal imagery), then the site would be estimated to include 60 m² of shaded-asphalt (or 2% of the total site area, as above) and 540 m² of sunny-asphalt (or 18% of the total site area).

To calculate one area-weighted average streetscape surface temperature value for each site (T_s), the area fraction of each of the 22 cover/shade categories was multiplied by that category’s mean LST_g (from the thermal imagery, as above).

2.6. Statistical Hypothesis Testing and Temperature Modeling

To evaluate differences between the average surface temperatures (LST_g) of surface covers in sun and shade, on streets with a large or small amount of street-overhanging canopy cover, and on streets with primarily coniferous or broadleaf deciduous street trees, two-sample t-tests with unequal variances were used. The relationships between the temperatures of sun-exposed and shaded surface covers were also tested using ordinary least squares linear regression.

To test the relationship between urban surface temperatures and local street-scale air temperature, the statistical relationship between the set of 12 site-level T_s and T_a values was evaluated using ordinary least squares linear regression, to determine whether a simple linear conversion factor could be used to consistently and accurately translate the high-resolution, surface cover and sun/shade area-weighted T_s ‘heatscape’ values into the street-scale T_a air temperature estimates most relevant for urban climate and human health.

To compare the capability of predicting street-scale T_a from the surface cover and sun/shade area-weighted T_s values to predicting T_a from common variables often invoked in the literature, ordinary least squares linear regressions were also used on the latter. Regressions tested the relations of the 12 sites’ street-scale T_a to each streetscape’s overall fraction of impervious cover, tree canopy cover, and to the LST_g values of individual spatially dominant surface covers (green grass, concrete, asphalt).

3. Results

3.1. Surface Cover Types, Shading, and Temperature Variations

The twelve residential study sites exhibited large variations in surface cover, which provided a useful distribution for analysis of variations in surface and air temperatures (Figure 3). Roofs, concrete, and asphalt comprised the largest portions of impervious cover among the sites, and green grass and bare ground the largest portions of pervious cover. Ten of the twelve study sites had greater impervious than pervious cover (approximately 55–85% impervious). The two study sites with greater pervious cover than impervious cover, DH2 and EH2, had large areas of green grass (>60%) because the blocks were partly residential and partly neighborhood city parks. Still, the inclusion of park area is representative of the mixed land use of residential neighborhoods in many cities, including Portland, and so is a useful inclusion relevant to conversations about the expansion of urban green spaces to mitigate urban heat. Among the expected heterogeneity of urban impervious and vegetated surfaces, there were also new insights about urban surface cover types not as frequently recognized as asphalt, trees, and grass. Parked cars covered non-negligible areas of the streetscapes (avg. 2%), even in these residential neighborhoods. Bare ground covered non-negligible areas (avg. 3%), which was particularly significant (up to 17%) on streets dominated by coniferous evergreen street tree canopy cover. Most streets exhibited significant portions of light-colored concrete pavement (sidewalks, curbs, many driveways), with concrete pavement 81% as abundant as dark-colored asphalt pavement on average across the 12 sites (range 0–126%). With the study occurring in the early summer (i.e., late June), before pronounced summer drought and heat settle on the area during July-September, most grass areas were still green at the time of the study (avg. 96%). However, on streets with at least some brown grass, the fraction of total grass area that was already brown was still significant this early in the season (avg. 11%). As grass lawns in most residential areas of the city are not irrigated, the proportion of brown (temporarily senesced) grass cover would be much greater later in the season.

The observed mean LST_g varied significantly among surface types (Figure 4). Surface temperatures ranged from roof surfaces in the sun averaging 61.0 °C across the six low tree cover sites, to green grass in the shade averaging just 23.8 °C across all twelve sites (Table 1). In general and on average across all the study sites, impervious surfaces were warmer than pervious surfaces, surfaces on streets with much street tree canopy coverage were cooler than on streets with low canopy cover, and the dominant street tree functional type (coniferous evergreen vs. broadleaf deciduous) did not have an independent significant effect on surface temperature variations.

An interesting finding was that mulched surfaces, typically covered in dark-colored bark chips, were substantially warmer than expected when in the sun, even slightly hotter than asphalt. In the shade, however, mulched surface temperatures were more moderate, between shaded concrete and shaded brown grass. On average across both sun and shade exposures, mulch surfaces were most similar in temperature to dark asphalt. Although few of the study sites exhibited gravel areas, for those that were observed, the gravel temperatures were most like the mulch and bare ground. In contrast, parked cars, which were more spatially abundant than expected, were also significantly cooler in the sun than other impervious surfaces, more similar to the temperature of brown grass, on average in the sun, than to pavements. In the shade, however, parked cars were as warm as asphalt on average, suggesting the dual function of shiny car body surfaces in the sun but energy absorption through car glass surfaces and consequent car thermal radiation even when parked in the shade.

Among pervious surfaces, the presence and greenery of vegetation were key factors, especially in the sun. Sun-exposed green grass, brown grass, and shrubs were substantially cooler than other pervious and impervious cover types. Brown grass surfaces were notably warmer than green grass and shrubbery but still substantially cooler than the other mulch, bare ground, and gravel pervious surfaces, especially in the sun. Bare earth was the surface type apparently most sensitive to both sun exposure and shading type. In the shade, average bare earth temperatures were essentially equivalent among all the sites, but bare earth surfaces in the sun exhibited the greatest temperature variability of any surface type. Bare earth surfaces were substantially cooler on streets with mainly broadleaf deciduous street trees or much street tree canopy cover than on streets with dominantly coniferous street trees or less canopy cover.

Overall, the largest and most significant influences on surface temperatures across all surface types were due to shading. The temperatures of sun-exposed surfaces were significantly greater than shaded surfaces within each of the surface cover types, by about 15 °C, on average overall (Figure 4). Whereas mean shade temperatures differed by less than 5 °C among all surface types in our study, mean sun-exposed surface temperatures differed over a range of more than 31 °C (Figure 5a). Excluding the less abundant cover categories due to small sample sizes (gravel, G) and extreme surface variety (other, O), the contrast in temperatures between a surface in the sun and in the shade ranged from a low of sun/shade ratio of 1.13 (3.2 °C mean difference) for shrubs to a high of 1.89 (25.2 °C mean difference) for roofs, on average across all sites. Similar ratios between sun-exposed and shaded surfaces were also apparent individually at each site (Figure 5b). In this respect the data suggested two clusters of sun/shade temperature relationships, one closer to the 1:1 line and another closer to the 1:2 line. We separated and classified these two clusters as “cool-type” and “hot-type” surfaces (Figure 5c). Cool-type surfaces were green grass, brown grass, and shrubbery, selected as exhibiting a sun-exposed LST_g less than 35 °C on average among all the sites (as in Figure 5a). Hot-type surfaces were roof, concrete, asphalt, gravel, mulch, and bare ground, selected as exhibiting a mean sun-exposed LST_g greater than 40 °C on average among all the sites. (Parked cars and the “other” surface category were excluded due to their intermediate and highly variable values.) Comparing these groupings, the hot-type surfaces increased in temperature approximately twice as much after sun-exposure compared to cool-type surfaces (Figure 5c; hot-type sun LST_g = 1.9 × shade LST_g, cool-type sun LST_g = 1.2 × shade LST_g).

Aside from direct shading vs. sun-exposure at the time of thermal imaging and temperature assessment, the data also revealed that the relative magnitude of street tree canopy cover was a key factor affecting the temperatures of surfaces, in general. Overall, the temperatures of sun-exposed surfaces on streets with low tree cover were significantly greater than the temperatures of the same sun-exposed surfaces on streets with high canopy cover (Figure 4;

Table 2. Urban surface cover average temperatures measured by ground-based thermal image analysis (LST_g, °C), by site and sun/shade exposure. Averages provided among sites with primarily broadleaf deciduous (D) or coniferous evergreen (E) street trees and for sites with relatively high (H) or low (L) coverage of the street by overhanging tree canopy, and all sites (all). Blanks: not all cover types appeared in thermal images at all sites. Background color is used to visually distinguish between the four tree type-canopy density groups, with evergreen sites lightly shaded. Data are presented in regular font with summary statistics for groups in italics.

Sun-Exposed Urban Surface Cover Type
Site	Roof	Concrete	Asphalt	Car	Green Grass	Brown Grass	Mulch	Shrubs	Bare Ground	Gravel
DH1	53.6	48.0	54.3	35.4	35.6	39.7	47.6	32.1
DH2	45.6	40.2	49.8	32.0	27.6			26.0
DH3	50.7	42.5	44.8	30.4	31.3	31.3	43.5	22.5	24.0
EH1		46.1	43.2	32.3			49.2	27.2
EH2	46.0	45.4	47.3		27.3	32.6			38.9
EH3		48.8	49.8		26.2		41.3	28.6
DL1	65.8	51.6	53.9	44.1	34.5	37.5	51.0	32.4	47.5
DL2	59.7	47.8	48.8	38.6	28.8	31.1	52.8	25.9
DL3	62.9	46.3	48.5	46.7	28.8	33.8	54.3	25.1
EL1	54.7	38.7	42.4	33.7	27.8		46.6	22.7
EL2	62.9	49.2	49.8	40.1	32.3	35.2	50.3	29.2	53.3
EL3	59.8	50.7	53.7	37.6	31.4		54.0	29.0		55.6
D avg.	56.4	46.1	50.0	37.9	31.1	34.7	49.8	27.3	35.8
E avg.	55.9	46.5	47.7	35.9	29.0	33.9	48.3	27.4	46.1	55.6
H avg.	49.0	45.2	48.2	32.5	29.6	34.5	45.4	27.3	31.4
L avg.	61.0	47.4	49.5	40.1	30.6	34.4	51.5	27.4	50.4	55.6
all avg.	56.2	46.3	48.8	37.1	29.9	34.5	48.7	27.3	39.2	55.6
Shaded Urban Surface Cover Type
Site	Roof	Concrete	Asphalt	Car	Green Grass	Brown Grass	Mulch	Shrubs	Bare Ground	Gravel
DH1	30.8	28.1	31.3	29.8	27.3		29.1	30.4	29.1
DH2		23.1	26.4	24.5	20.8		21.4	22.1	20.4
DH3	25.8	22.8	23.5	25.4	21.2	22.5	24.1	22.2	23.2
EH1		26.3	25.4		23.8		28.0	26.4	24.8
EH2		26.3	27.6	26.8	23.8	24.8			24.8
EH3	25.1	26.3	30.1	24.9	23.7		28.0	24.6
DL1	30.1	28.6	34.9	35.5	27.8	27.8	32.5		29.5
DL2	21.9	24.4	26.6	23.6	21.1		22.5	21.7	22.3
DL3		23.9	24.4	23.8	23.9		24.2	21.7
EL1	27.4	22.5	23.4	22.5	20.5	21.7	21.9	19.7	21.1
EL2	28.8	27.2	26.7	32.7	26.7		27.1	27.1	26.9
EL3	35.9	29.8	31.4	30.4	24.8	26.1	26.1	25.8		25.2
D avg.	27.1	25.1	27.9	27.1	23.7	25.1	25.6	23.6	24.9
E avg.	29.3	26.5	27.4	27.5	23.9	24.2	26.2	24.7	24.4	25.2
H avg.	27.2	25.1	27.4	26.3	23.4	23.7	26.1	25.1	24.5
L avg.	28.8	26.1	27.9	28.1	24.1	25.2	25.7	23.2	24.9	25.2
all avg.	28.2	25.8	27.6	27.3	23.8	24.6	25.9	24.2	24.7	25.2

). For example, sun-exposed bare earth, car, and roof surfaces on low-cover streetscapes were 19.0 °C, 12.0 °C, and 7.6 °C hotter on average than the same sun-exposed surface types on high-cover streetscapes. We attributed this effect to greater tree canopy coverage being more likely to have provided shade for part of the day prior to the mid-afternoon measurement times of this study, and those shorter periods of direct sun then not only reducing the total diurnal warming of the later sun-exposed surface, but likely also reducing the average daily soil moisture loss and so preserving more evaporative cooling potential, especially for bare ground lacking deeper vegetative water uptake.

3.2. Surface-to-Air Temperature Relationships

As noted previously, a simple and accurate means to predict near-ground urban air temperatures (T_a) at a fine spatial scale would be highly desirable to enable more easily predicting, assessing, and mitigating the negative human, natural, and infrastructure impacts of urban heat islands. We tested if site-level air temperatures (T_a) could be accurately predicted by a simple translation from the area-weighted mean surface temperature of each site (T_s), as T_a ≈ T_s. The results showed a significant and highly predictive relationship with a nearly 1:1 slope (T_a = 1.07 × T_s, R² = 0.70, p = 0.0013, Figure 6a). In this ordinary least squares regression analysis, we constrained the regression intercept to equal zero to enable conversion of air to surface and surface to air temperatures in an equivalent manner. The best fit ordinary least squares regression equation with unconstrained intercept was nearly identical (T_a = 1.09 × T_s − 0.5, R² = 0.70, p = 0.0013). These results indicate that the total combination of thermal heterogeneity in streetscape surface cover—including each cover’s fractional area, fractional sun and shade exposures, and characteristic surface temperatures in sun and shade—together explained 70% of residential street-scale variation in air temperatures.

We also used the data from this study to test if site-level air temperatures could be accurately predicted, as has commonly been done in more coarse-scale remote sensing studies, by projection from the remotely sensed surface temperature (LST) of a dominant surface cover type, from site percent impervious cover, or from tree canopy coverage. The distribution of street-scale air temperatures among the eleven study sites with functioning air temperature sensors was weakly and insignificantly predicted by the site-to-site variations in the temperatures of the dominant surface types (all R² < 0.24, all p > 0.13, Figure 7a,b). The distribution of sites’ air temperatures was also not well-predicted from the sites’ variations in percent impervious area (R² = 0.2, p = 0.17, Figure 7c), and was statistically unrelated to sites’ variations in street canopy coverage (R² = 0.0, p = 1, Figure 7d).

4. Discussion

4.1. Interpretation and Application of Results

The magnitude of temperature difference between sun-exposed and shaded surface covers quantified in this study agreed with the limited results from other studies. Results from a set of studies in Florence, Italy, found similar magnitude differences (11–23 °C) at similar times of day as this study, across grass, asphalt, gray sandstone, and white gravel surface cover types [65,66]. A study in Munich, Germany, found that the shade of four different tree species reduced underlying grass surface temperatures by 6.8 °C and underlying impervious surfaces by 15.5 °C, on average [67]. In Florence and in our study, the degree to which sun exposure or shading modulated surface temperatures varied greatly by surface cover type.

Shade was not the only important factor, however; cooler shaded and warmer sun-exposed behaviors differed by land cover type. For example, pervious gravel and mulch surfaces were surprisingly as hot as dark asphalt pavement when in the sun. Bare soil, on the other hand, exhibited a very wide range of surface temperatures, depending substantially on both direct shade and indirect or partial-daytime shade. Bare soil temperatures were hotter beneath coniferous evergreen than deciduous broadleaf street trees, and on streets with relatively more street tree canopy. This likely reflects both a greater tendency for soil water repellency under coniferous trees reducing the overall soil moisture available in those locations for evaporative cooling, and bare ground experiencing shade for a greater portion of the day being able to retain more moisture overall to then contribute to evaporative cooling even when later exposed in the sun, whereas bare ground on streets with less canopy cover having greater total exposure, depleted soil moisture, and so reduced evaporative cooling potential.

The most impressive result of this study was that, without any calibration, the cover- and shade-weighed average surface temperatures of our set of urban residential streetscapes very nearly matched the shaded afternoon air temperatures on the same streets. This simple approach was able to explain 70% of the air temperature variance among our set of urban residential environments using a linear model slope very close to one and using a zero model constant. The remaining 30% of site-to-site variation in T_a not explained by this T_a ≈ T_s model will have been contributed in part by uncertainty in the delineated surface cover proportions, estimated overall sun-exposed vs. shaded fractions, and thermographically measured surface temperatures. We attribute most of the unexplained variance in the model, however, to the role of local urban form and potential aerodynamic differences among the study sites, which were not captured by our approach. Wind speed, wind direction, building height, surface roughness, and street orientation may all even interact to affect the advection of heat from land surface to air and the flux of heat away from the near-surface air space [51,68]. The particular importance of the height-to-width ratios of urban canyons and local wind regimes on UHI has previously been highlighted, for example by a computational numerical modeling study of Bat-Yam, Israel, wherein air temperatures were predicted by incorporating variable surface cover and street-scale urban canyon geometries [51]. Although all the sites in this study were similar residential neighborhood settings and intentionally oriented in the same cardinal direction, differences in house set-back distance from the street, height of front-lawn landscaping, tree density, and placement of the streets within the broader and locally rougher or smoother, windier or less windy urban matrix likely contributed in part to the site-to-site differences [69].

This study’s results from ground-based thermal imaging compared favorably to previous evaluations of the relationship between satellite-imaged land surface temperatures (LST) and urban air temperatures (T_a) at coarser scales. Although we did not find T_a-LST_g relationships to be significant when using the LST_g values from specific land cover types, several other studies have previously developed remarkably similar regression equations, though using coarser-scale data. For example, our (non-significant) regressions between air temperature and specific grass and pavement surface cover types in the sun were of the form T_a = (0.1 to 0.2) × LST_g + (20.2 to 22.4)°C with R² = 0.03 to 0.18 (Figure 7a). In comparison, a study of Birmingham, UK, related urban air temperatures and satellite LST (1 km MODIS) with a regression equation of approximately T_a = 0.11 × LST + 19.3 °C with R² = 0.3 [70]. A study in Shenzhen City, China, related air temperature and LST (1 km MODIS) during the day in the summer via T_a = 0.10 × LST + 27.5 °C with R² = 0.15 [71]. A study of Jakarta, Indonesia, related air temperature and LST (Landsat) and again found a similar regression equation of T_a = 0.21 × LST + 19.5 °C with R² = 0.65 [72]. A study of Hong Kong, where the setting may have been more influenced by marine winds and water that these other sites, related air temperature and LST (ASTER) and found a daytime regression equation of T_a = 0.52 × LST + 14.0 °C with R² = 0.74 [73]. The reason for this overall convergence in regressions between the T_a and LST across many studies is unclear but does not seem to have a directly biophysical or climatological basis. An analysis of surface temperature sensor to air temperature sensor relations across eight settings (grass and asphalt surfaces in sun and shade and in wetter or drier conditions) in Freising, Germany, suggested a similar regression of T_a = 0.28 × surf. sensor T + 18.1 °C with R² = 0.46 overall, but slightly lower slopes for regressions specific to grass or paved surface types and slightly higher intercepts for dry compared to wet surfaces [74].

The larger coefficients of determination of these other studies compared to this study are most likely attributable to the other studies’ larger sample sizes and to the smoothing effects of their coarser satellite LST pixels. With thermal satellite pixels at resolutions of 100 m-1 km, any local land surface temperature extremes at the sub-grid scale will unavoidably be numerically smoothed into the pixel-scale value, and these moderated and smoothed values then likely to more closely match the thermally-buffered and coarsely well-mixed air temperatures of the urban environment. In other words, we hypothesize that it is possible, although unconfirmed, that higher coefficients of determination between coarse-scale satellite LST and urban air temperature data sets such as in the above studies may obtain the ‘right answer’ (exhibited by strong correlation coefficients) for the ‘wrong reason’ (i.e., pixel-smoothing accidentally creating similar numerical anomalies as actual aerodynamic mixing). Other studies have also produced high apparent T_a-LST correlations due to other numerical issues, such as autocorrelation. For example, a study in London, UK, used similar-scale data and focused on urban street canyons and their variable façade temperatures, as did this study, but the London study’s apparently exceedingly strong air-to-surface temperature relationship was due to their sample population consisting of hourly repeated measures of two similar streetscapes, and so mainly reflecting temporal autocorrelation among consistent field conditions [5]. Still, drawn from literature, theory, and their other lines evidence, the broader conclusions by Watkins et al. regarding the importances of differences in urban surface cover albedo, warming, sensible heat flux, shading, and vegetative and soil latent heat fluxes for urban air temperatures presaged our own in this study [5].

Notably, in this study the street-scale air temperatures were not at all predictable from the temperature of the dominant surface cover types, from the impervious fraction of the streetscape, nor from the street tree canopy cover of each street. Still, fractional imperviousness and tree canopy coverage are, by far, the most common variables employed to try to explain SUHI variations both within and between cities. For example, a study of 38 of the largest cities in the USA found significant positive linear relationships between percent impervious surface area (%ISA) and LST deviations across all biomes except deserts (MODIS 1 km LST, Landsat 30 m %ISA) [75]. Another study, in Minneapolis-St. Paul, Minnesota, USA demonstrated the consistency of the positive LST-%ISA relationship over the four seasons of the year [76]. A study in Shanghai, China also found LST strongly positive related to imperviousness (R² = 0.6) but only weakly related to urban greenness (R² = 0.2) and with the opposite sign as expected (Landsat 60 m LST, 30 m %ISA and NDVI) [77]. On the other hand, a study relating LST data (MASTER) to urban greenness (NDVI, AVIRIS) at the 50 m-scale in Los Angeles, California, USA, generated strong linear relationships for each of three study areas (R² = 0.4 to 0.5) [78]. Two studies, one in Boston, Massachusetts, and one in Madison, Wisconsin, USA, found the expected positive relationship between LST and %ISA and the expected negative relationship between LST and tree canopy cover, although with different shapes and ranges of their regressions [47,79]. A study of Phoenix, Arizona, USA, found mean surface greenness (NDVI) and median family income to be the strongest two predictors of daytime land surface temperatures (and mean NDVI and percent paved area to be the strongest nighttime LST predictors), after aggregating all variables to the Census block-group scale [80]. Finally, a study of LST relations to %ISA and green space density found strong and statistically significant relationships of increasing LST with increasing %ISA or with decreasing greenspace in Bangkok, Thailand, in Jakarta, Indonesia, and in Manila, Philippines [81].

4.2. Implications for Mitigating Urban Heat by Plot-Scale Landscaping Choices

The results of this study quantitatively indicated useful avenues for reducing local urban air temperatures by plot-level landscaping decisions and surface cover conversions. The most significant cooling benefits across all urban surface types would be achieved by increasing urban shading, according to our data. All surface covers were significantly cooler in the shade (Figure 4). Because shade temperatures were quite similar while sun-exposed temperatures differed greatly across surface types (Figure 5a), the benefits of direct shade were greatest for the surface cover types that were hottest in the sun. The differences between sun-exposed and shaded surface temperatures exhibited a linear response, such that typically hot roofs, gravel, mulch, and asphalt which were consistently up to 50% (more than 20 °C) cooler in the shade (Figure 7b). Shading had a more moderate cooling effect for car, grass, shrub, and bare ground surfaces but still provided a 12–40% temperature reduction benefit on average. As air temperature variations were so highly related to surface temperatures among our sites, this suggests that local UHIs might be substantially cooled by adding even modest amounts of shade. In particular, the most efficient UHI mitigation may be achieved by preferentially investing in shading over the surface cover types typically hottest when in the sun (roof, gravel, mulch, asphalt, concrete), rather than by adding shade over the types of cover that are generally cooler.

The benefits of even partial diurnal shade, such as provided by greater overall street canopy cover compared to lesser cover, were also significant in our data. The effect of increased street-level canopy cover, even if not able to shade a patch of ground during the mid-afternoon, still contributed an overall temperature reduction of 6–19 °C for some of the typically hottest cover types such as roof, car, and mulch in the sun. Encouraging greater shading through increased tree canopy or alternative shading structures provides a simple policy target for reducing surface temperatures overall. This benefit will be most pronounced for streets with low starting canopy coverage, however; once shaded, there did not appear to be a further cooling benefit due to greater canopy cover as there had been for surfaces in the sun. Also, greatly increasing tree canopy may produce the undesirable effect of trapping warm air in urban canyons and promoting higher nighttime temperatures [82]. On the other hand, an enhanced cooling benefit might be cultivated by the combination of grass vegetation beneath tree canopy shade, but this would necessarily come at the cost of the combined water use of both the trees and grass [83] and it is the shading of the tree, not its transpiration, that would provide the tree’s contribution of a cooling effect [84]. The under-canopy effects within urban microclimates remain incompletely understood, however, and may be specific to different types of urban groundcover under different conditions. Future work might usefully investigate urban below-canopy groundcover vegetation, microclimate, and moisture-competition dynamics.

Vegetated surfaces were clearly more effective at moderating surface temperatures than non-vegetated surfaces in our study, whether the latter were pervious or impervious, and whether in sun-exposed or shaded conditions (Figure 4 and Figure 5b). We attribute this effect only partly to the latent heat loss from vegetated surfaces associated with evaporation and transpiration. Although the cooling benefit of green spaces has previously been identified in both sun and shade [85], cooling requires adequate moisture [20,26]. We interpret our results showing much warmer surface temperatures for bare ground at low-canopy cover sites and for highly porous mulch and gravel surfaces in the sun to illustrate that the supposed cooler nature of pervious surfaces should be expected to be lost as they dry. As non-vegetated surfaces cannot provide ongoing hydraulic lift and exposure of deeper soil moisture to the atmosphere, cooling benefits from porous surfaces, and subsequent rapid heating over those surfaces, will be highly transient. For example, modeling experiments for an idealized Central European city indicated that the regions experiencing the largest increases in air temperature during a heat wave were those above or near areas of natural soils, which dried quickly [86].

Vegetation-covered soil, even if the vegetation was temporarily senescent, dry, brown grass, did not exhibit the same thermal behavior as unvegetated pervious surfaces in this study, but rather still behaved thermally more similar to green grass. As the dried grass could not have been transpiring at anything close to the rate of green grass, we hypothesize that the vegetative cover may still assist in maintaining higher soil moisture near the surface, such that soil evaporation may help keep the surface cooler even in the absence of substantial plant transpiration. Even if the leaves are temporarily senescent during seasonal summer drought, vegetation may still have the effect of maintaining near-surface soil moisture by passive root hydraulic redistribution or via its effects on soil structure and organic matter. In contrast, unvegetated mulch or gravel surfaces may be more effective at maintaining moisture in the ground in deeper soil layers, which is indeed one of the purposes of applying mulch to garden areas, yet their much hotter surface temperatures may damage near-surface microbial, mycorrhizal, and fine-root communities. Mitigating urban heat may be better served by replacing mulch and gravel areas with shrub cover, grass, or bare earth, despite all being equivalently considered pervious cover types.

Reducing impervious surface cover, on the other hand, is widely understood to be a useful goal if working to mitigate excessive urban heat. Yet, this study’s data suggested that the typical recommendation to replace impervious surfaces with pervious surfaces, while potentially helpful for stormwater infiltration, may not help mitigate urban heat in some cases. If replacing light-colored concrete with dark-colored mulch, bark chips, dry gravel, or stones in the sun, the conversion of impervious to pervious cover may actually increase the local heat burden and the air temperatures at the street-scale (Figure 5a). To reduce both impervious area and urban heat, the results of this study suggest replacing asphalt and concrete with bare earth or grass where possible, especially if at least part-day shade can also be provided for those areas. If irrigating grass to keep it green, and especially cool, is a challenge due to cost or limited water resources, then even unirrigated, seasonally brown grass can provide, according to our data, a much cooler influence on the street-scale thermal environment than typical pavements (Figure 5a and Figure 7b).

4.3. Implications for Predicting Urban Air Temperatures from Remote Sensing Data

Despite omitting specifics about urban form, our intentionally simple T_a ≈ T_s model provides a direct, low-cost, and well-performing assessment method, which could be used to monitor changes in street-scale urban heat dynamics or to predict the effects of land cover composition changes at the plot- to street-scale on the occurrence of hot air temperatures for residents and pedestrians. The near parity between the area-weighted average surface temperature metric (T_s) that we developed and a variety of urban residential streets’ air temperatures (T_a), suggests that this approach may indeed provide an easy, sufficiently accurate means of predicting street-scale urban air temperatures as hoped. The method requires neither extensive nor continuous monitoring and provides detailed insights that cannot be gained from satellite remote sensing due to limitations of resolution and below-canopy observation, making it a viable and low-cost option for local assessments street-scale urban heat dynamics. The method relies upon high-resolution surface cover mapping, local estimates of sun vs. shade area fractions, and characteristic values of temperature of each surface cover type in the sun and shade. We posit that one key in the success of this methodological test and preliminary model development was the use of sufficiently high-resolution data. Manual delineation of thermal imagery and orthoimagery was chosen in this demonstration study to prioritize accuracy, given the reasonable time required for delineation, and to allow for below-canopy land cover characterization. However, automated land cover delineation of orthoimagery, combined with a method to allocate surface covers and shade fractions beneath concealing canopy cover, and with high-resolution areal thermal imagery or regionally appropriate characteristic surface temperatures for each surface cover type/shade combination, may provide an opportunity to apply this study’s simple regression approach to predict and monitor land cover influence on Ts and Ta at high spatial resolution but over broader areas.

The questions of appropriate numerical form and appropriate spatial scale for modeling urban heat/landcover statistical relations are not trivial. The relations between urban temperatures and landscape features known to be key in modulating urban heat are also known to be diverse, sometimes non-linear, and sometimes scale dependent. For example, the illuminating analysis by Ziter et al. demonstrated that greater impervious cover should be expected to relate linearly to higher air temperatures across a range of analysis scales, due to the key importance of aerodynamic mixing within or between the analysis site and its surroundings over relatively more smooth impervious surfaces [47]. On the other hand, Ziter et al. demonstrated that increasing urban tree canopy cover should be expected to relate in a non-linear manner to decreasing air temperatures at a given scale; furthermore, the magnitude of this non-linearity is itself scale-dependent, with an initially more linear relation at scales below 10 m becoming more and more nonlinear with increasing scale, at least up to about 60 m [47].

The theoretical basis provided by Ziter et al. [47] should warn us that for site or pixel scales above about 10 m, we should not expect to capably predict urban air temperatures from data on tree canopy cover (or related or derivative metrics such as shade cover, greenness, roughness, optical depth, etc.), nor by linearly combining such tree canopy cover data with other metrics. Some studies have indeed failed to find significant relationships between urban air temperatures and tree cover, pavement cover, and other surface cover abundances [87], or found that the organization and underlying land cover of tree canopy, for example, have significant impacts (though hidden from satellites) on the type (linear vs. non-linear) and magnitude of canopy cooling effects [88]. One study even found relationships between LST and impervious or canopy cover of the same nature as those by Ziter et al. [47], but exhibiting much wider LST variation across their Boston, Massachusetts, USA study area [79]. In Boston, Melaas et al. also documented an approximately linear LST response to increased canopy cover a but non-linear response to imperviousness [79], the opposite of the non-linear/linear trend types found by Ziter et al. [47]. The present study therefore focused on developing a bottom-up statistical model using remotely sensed data at <10 m scale, but of types that are either already available at high-resolution or that could feasibly be made widely available from aerial or satellite platforms in the future. Further work in this area is recommended to more broadly validate the approach developed in this study and simultaneously further advance understanding of below-canopy factors and landscape scaling in remote sensing and analyzing urban areas

5. Conclusions

This study was motivated by the need to improve the characterizations and evaluations of UHIs by initiating a cross-scale link between point-based surface temperature, high-resolution land cover heterogeneity, and urban warming across scales from one to tens of meters. Our results in Portland, Oregon, showed that surface temperatures are significantly higher in the sun than in the shade for all surface types and that vegetated surfaces comprise cool surface types that remain much cooler in the sun than impervious and non-vegetated pervious covers. At the same time, dry or dark pervious materials such as bark chips and gravel could be as hot as asphalt pavement. Ultimately, area-weighted mean surface temperatures based on ground-based thermography and high-resolution surface type composition were able to predict and explain 70% of the site-to-site variation in the street-level air temperatures most relevant for urban residents, ecosystems, and infrastructure. Although this study provides essential new insights into the fine-scale variation in urban surface temperatures and their relationship to air temperature, its scope is limited to the influence of surface cover type and canopy shading on temperatures. This study did not account for variations in urban form or geography that might affect heat transfer processes. This study also was limited to residential environments in a moderate-density temperate-zone city. The relevance of impervious and grass surface covers or other landscape metrics may differ for industrial or commercial land uses [89] and other climate zones. Future work should examine the role of these street-scale characteristics and extend analysis across seasons and throughout the diurnal cycle. Additionally, new evaluations should examine the role of hydrological characteristics and processes in moderating heat transfer and ambient air temperatures within surface cover types at the street level.

The implications of this study are that, first, increasing direct shading may provide the greatest cooling benefit overall and will have the greatest cooling effect if shade additions are prioritized over the hottest land covers (e.g., gravel, mulch, pavements). Second, increasing transient or indirect shading is still quite valuable, even if not all surfaces are directly beneath a tree or shade structure. The part-day shading associated with a large amount of canopy cover at the street level appeared to still provide substantial cooling benefits, especially for the hottest land cover types. Third, air temperature at the street scale cannot be predicted from the types of landscape variables typically used in remote sensing analyses of UHIs and SUHIs, at least not for the selected study sites of this Portland, Oregon-based study. The temperature of dominant surface cover, the local percent impervious cover, and the local canopy cover fraction were all unable to support any statistically significant relationship with T_a in this study. Fourth, instead, an area- and shade-weighted streetscape surface mean temperature, which accounts for fine-scale variations in surface covers and their temperatures, does provide a simple, accurate means to predict urban near-surface air temperatures. Using thermal imagery, or even point-based measurements, of land surface temperature across surface cover types and shading conditions in combination with relative land surface area of each surface and shading type to evaluate air temperature provides a transferable approach to evaluate local UHIs and a method to devise policy-relevant recommendations for heat mitigation. These findings could next be integrated into planning processes through multidisciplinary management teams, be adapted for locally relevant guidance, and be balanced with considerations of social factors when developing urban planning policy [90].