Contributions of roads to surface temperature: evidence from Southern California

Planners often regard streets as targets for mitigating urban heat across cities by virtue of being abundant, publicly-owned, low-albedo, low-vegetation surfaces. Few studies, however, have assessed the role streets play in contributing to urban heat, and the scale of their effect relative to the built environment around them. We examine the relationship between road area and land surface temperature across a variety of biophysical regions through the urban areas of Los Angeles and San Bernardino Counties in Southern California. Our results show that wide streets have no consistent, detectable effect on urban heat. Rather, vegetation is the primary cooling mechanism for urban areas. In the absence of trees, concrete highways are the coolest surfaces, though particular hot or cool pockets (e.g., airports, industrial centers, parks) can dominate neighborhood temperature signatures. In considering LST mitigation strategies, these hotspots might outweigh the cumulative effects of road surface changes.


Introduction
Cities are hot in part because of impervious surfaces like buildings, roads, and parking lots. In most regions, urbanization replaces vegetated land with impervious surfaces, which decreases two key cooling factors: albedo and evapotranspiration. Urban heat planning, therefore, focuses on ways to increase albedo and vegetation to mitigate the effects of impervious surfaces on urban land. Heat produced by urban land cover is typically characterized as the surface urban heat island (SUHI), a regional phenomenon that causes cities to be, on average 1.5°C warmer than surrounding undeveloped areas and is most pronounced at night (Oke 1982, Peng et al 2012. While the relative contributions of different surface materials are frequently studied through land cover analysis (Shiflett et al 2017), municipal planning decisions and policies are more typically organized around land use categories. The relative contributions of different land use categories to surface heat are less well understood.
One potential land use source of surface heat in urbanized regions is streets-large-scale, mostly low-albedo impervious surfaces that lack vegetation and shade (Taleghani et al 2016). The contribution of streets to urban heat is amplified by the large amount of land that they occupy-up to 30% in US cities such as New York (Manvel 1968, summarized in Meyer & Gómez-Ibáñez 2013. Streets can influence the SUHI because of the effects of aspect ratio, width, and orientation on solar reflectivity and ventilation performance. At the same time, however, streets offer prime opportunities to mitigate urban heat-they are publicly owned resources upon which cities can site interventions without the need to incentivize private developers or land holders (Pomerantz et al 2003, Gago et al 2013, Lee et al 2018. For this reason, several cities have begun experimenting with cool pavement and urban tree planting programs to leverage streets as a public resource for mitigating surface heat (Maxwell et al 2018, Turner et al 2021, Ko et al 2022. These intervention programs focus on street trees and cool pavements as methods for increasing albedo and vegetation.
This study examines the relationship between street width and Land Surface Temperature (LST) across a variety of urban forms to examine how and where mitigation strategies might be best applied. We examine LST across urban areas and evaluate the role of streets at both a regional and a neighborhood scale. We focus on communities in Los Angeles and San Bernardino Counties, California, which contain a mix of background biophysical conditions and built forms, and where policymakers have piloted the use of road surfaces to reduce urban heat through changes to vegetation or albedo (US EPA 2012, Garcetti 2021. These proposals follow a global trend of investment in cool pavement: Western Europe is pushing cool pavements, and pilot programs can be found in Tokyo, Athens, and Rome (Santamouris 2013, Municipality of Athens2017, Moretti et al 2021). The United States government is also incentivizing cool pavement programs and several cities have followed suit (Wiltshire-Gordon 2020, FHA 2021, Garcetti 2021, SmartCities Connect 2021. We test the hypotheses that roads contribute to urban heat and that wider roads amplify those contributions. We further hypothesize that the impact of streets will vary based on the local physical context; hotter, more arid conditions will suppress the contribution of road widths to urban heat as will more intensively developed areas.
2. Mitigating urban heat on roads: vegetation, impervious surfaces, and land morphology As conventionally described, a vegetation disparity across the urban environment creates a SUHI by producing hotter conditions in urban areas, especially developed downtown sectors, than surrounding undeveloped areas. The temperature disparity is largely driven by impervious surfaces like roads and buildings that absorb and slowly release heat throughout the day (Oke 1982, Rizwan et al 2008 and, depending on regional conditions, usually peaks during the early afternoon due to dependence on shortwave radiation (Shastri et al 2017, Lai et al 2018. Planning strategies often focus on mitigating the SUHI by reducing LST within urban cores.
The traditional SUHI has several limitations for urban planning. One shortcoming is that the direction of the relationship between urban land and surface temperature depends on background land and climate conditions. In hot desert climates, the typical SUHI pattern can be inverted: urbanized areas are often cooler than the background reference desert during the daytime, but warm up at night (Lazzarini et al 2015). Desert cities have substantially less dense vegetation and higher LST than temperate or forested cities, and a shallower diurnal heat cycle (Imhoff et al 2010). The inverse urban heat island is largely attributable to differences in vegetation and canopy cover: bare soil or sand in undeveloped desert areas is warmer than shaded or irrigated landscape in city centers (Shastri et al 2017, Mohamed et al 2021. Moreover, the relationship between urban land and temperature is heterogeneous within cities. This is problematic because developed areas, irrigated landscaping, and indigenous vegetation have distinct diurnal and annual NDVI, heat, and evapotranspiration cycles (Mini et al 2014, Hall et al 2016. These studies demonstrate that the direction and characteristics of the SUHI depend on the background land and climate conditions of the reference system.
Vegetation is also a primary mechanism for cooling spaces within cities: impervious surfaces and bare land are significantly warmer than vegetated areas, particularly in hot areas (He et al 2019). Shaded surfaces, and the air above those surfaces, are cooler than nearby unshaded impervious surfaces (Taleghani et al 2016). Tree cover, in particular, is effective in reducing surface temperature, but the mere presence of vegetation is sufficient to reduce localized surface and air temperature in comparison to a paved surface (Susca et al 2011, Adams andSmith 2014). In addition to shade, vegetation provides heat reduction through transpiration: as water is released into the atmosphere, sensible heat is converted to latent heat, reducing overall air temperature (Ballinas and Barradas 2016). This effect is heightened in arid climates with a high vapor pressure deficit that increases transpiration. A final pitfall in considering moderation of LST is the temptation to conflate the SUHI with the UHI, which reflects air temperature and is therefore affected by other phenomena, notably building morphology and wind (He et al 2020a(He et al , 2020b.
Urban heat varies with race and income as well as physical characteristics, in ways that exacerbate environmental injustices. For example, historically redlined neighborhoods show elevated surface temperatures of ∼2.6°C across the U.S. (Hoffman et al 2020, Wilson 2020. This trend holds for predominantly Black or Latinx neighborhoods, primarily due to a lack of vegetation (Harlan et al 2006, Dialesandro et al 2021. The need for LST intervention is unevenly distributed across urban areas.
Many cities have considered street trees as a strategy to mitigate urban heat given their large impacts on LST and their other benefits such as visual amenity and improved air quality (Mullaney et al 2015). Though tree planting is shown to be effective in reducing LST, many urban trees in Southern California are non-native and require constant irrigation, increasing water imports (Roman et al 2021). Indeed, angiosperm (broad-leaf) species comprise only 71% of trees in Los Angeles but contribute over 90% of tree transpiration (Litvak et al 2017). Landscaping comprises ∼54% of residential water use in Southern California, and irrigation is both more prevalent and less responsive to mandatory drought-related water rationing in wealthy neighborhoods, which are shadier and cooler than low-income areas (Clarke et al 2013, Mini et al 2014. Because trees-particularly leafy, traditionally 'shady' trees-come with a cost, planners seek out alternate strategies for reducing LST.
Given these challenges, planners have also turned to road surface paint as a low-cost, easily-implemented method for mitigating LST. Roads, which are often constructed from dark asphalt, are logical targets for intervention: as continuous, often wide, impervious areas, they logically increase LST and radiant air temperature (Cheela et al 2021, Pomerantz et al 2003). Studies have tested potential reductions in LST from cool pavements, and found that higher-albedo surfaces are indeed cooler (Sodoudi et al 2014, Sen et al 2019. Citywide effects of cool pavement are mixed, however: as with any urban heat strategy that relies on increasing albedo, radiant temperature and midday temperature can increase even as incoming solar radiation decreases (Middell et al 2020, Erell et al 2014. Cool pavement pilot programs are assessing the effects of different surfaces within cities. Cool pavements are often introduced inadvertently without regard to their albedo-increasing properties. For example, in Southern California, many of the widest roads are already concrete, which is classified as a cool pavement -studies have found that concrete with a higher cement content increases albedo regardless of the remaining composition (Lee et al 2002, Levinson and Akbari 2002, Sen et al 2019. This practice is likely to continue: because concrete sets rapidly and can include recycled tires for little expense, it continues to be the primary surface used in newly-repaired freeways (Caltrans 2002, CalRecycle 2020). Concrete is less prone to cracking than painted asphalt, and its construction can incorporate reflective materials to increase albedo (Cheela et al 2021). Concrete surfaces in Southern California can be used to understand existing cool pavements in context.
While most research has focused on the potential benefits and pitfalls of increasing cool pavement presence throughout cities (see for instance Santamouris 2013), several studies have examined specific aspects of the relationships between streets and heat. Microscale studies have compared road surfaces to one another, finding evidence of LST reduction with reflective pavements and shaded surfaces (Sodoudi et al 2014, Lee et al 2018. Others have evaluated the effects of building morphology, demonstrating that canyons and airflow can improve cooling at block and citywide scales (Johansson 2006, Giridharan et al 2007. Hoehne et al (2020) found increased sensible heat from combined car emissions and road surfaces across Phoenix. However, their LST readings seemed to correlate with imperviousness or bare ground, as opposed to irrigation. Yamazaki et al (2009) used very high resolution imagery (2 m) to examine LST, and found higher temperatures on impervious and road surfaces than in vegetated areas or water. However, they did not evaluate the effects of roads across a neighborhood or city scale, or in areas that are either highly vegetated or impervious.
There is substantial literature examining possible LST mitigation strategies across particular cities or neighborhoods (Deilami et al 2018, Mohammed et al 2020. However, few studies examine the relative impacts of vegetation or cool pavement strategies in distinct neighborhoods, rather than as a citywide panacea. Urban morphology can change at a neighborhood, or even a block scale within a city, affecting localized and citywide temperatures (Yuan et al 2020). Sodoudi et al (2014) examined a hybrid cool pavement and vegetation cooling model in Tehran, and found it to be more effective than either strategy in isolation. Middell et al (2020) found that cool pavement was not appropriate as a one-size-fits-all model, and should be applied with consideration of local context. To our knowledge, no studies have considered which areas might benefit from varying forms of LST mitigation.

Study area and case study selection
We examined the urbanized portion of Los Angeles and southwest San Bernardino Counties, California. We chose these counties because of their size, variety of climatic conditions and urban forms, and growing urban heat island (Dialesandro et al 2019, Ladochy et al 2021. Together, the counties are home to ∼12 million people in over 100 incorporated cities, with substantial income disparities and a legacy of environmental injustice (Su et al 2009, US Census Bureau 2020. Our study area spans an east-west transect of California, covers elevation from sea level to >1000 m, and encompasses mediterranean and desert Koppen climate zones (Kesseli 1942). Coastal areas have a summertime marine layer, providing an overall cooling effect (Edinger 1959). The built form encompasses single-family homes, apartments, high-rise residential and office buildings, and industrial uses, and streets that range from large arterials to narrower streets built before the private car became dominant. While some of our predictive variables are hyperlocal, we examine LST at a neighborhood and citywide scale because it is well-suited for guiding interventions at regional, rather than block-level, scale (Turner et al 2022).
The primary urban area in our study region is the City of Los Angeles, a global megacity which contains a dense urban core, sprawling residential areas, and low-rise industrial zones. Notably, Los Angeles enforces a highway dedication ordinance requiring developers to physically widen streets to accommodate more traffic in exchange for building permits (Manville 2017). At the same time, the city considers streets to be a primary avenue for mitigating LST; in 2021 the Mayor's office announced an initiative to bring 200 blocks of cool pavement and 2,000 new trees to eight residential areas (Garcetti 2021).
In Southern California, we examine areas with both traditional and inverted SUHI. In coastal Los Angeles, heavily impervious urban areas (i.e. South and East LA) are warmer than the more vegetated mountainous or coastal neighborhoods (Dousset 1989, Hulley et al 2019. However, the eastern part of the state shows a reverse urban heat effect consistent with other hot desert cities (Shiflett et al 2017).

Data sources and calculations
We calculated LST at 30 m resolution using Landsat 8, parameterized with water vapor and emissivity from NCEP/ NCAR reanalysis and ASTER imagery, for a 3-month composite of June, July, and August 2020 (Ermida et al 2020).
To check our calculations, we validated our LST data using an alternative algorithm (Landsat Provisional LST) and data source (NASA ECOSTRESS LST readings). All three approaches are widely used in the literature, and we found agreement among the datasets. Thus, the remainder of the analysis uses the Landsat 8 imagery, so as to use a standard Landsat base for all variables and because of the ease of calculation in Google Earth Engine. Because we observed thermal LST, we were in effect observing the radiant temperature of tree canopy, shrubs, and grass in vegetated areas, rather than the temperature of the shaded pavement. Studies show, however, that shaded surfaces are significantly cooler than those in direct Sunlight (Barbierato et al 2019, Middell et al 2020. Our data on street area, width, and class (highway, arterial, and residential street) use a novel method derived by Millard-Ball (2022), which derives street area and width from the voids between tax assessment parcels, and matches each void to OpenStreetMap (OSM) ways (maps available at [REDACTED]). Because right-of-way boundaries did not overlap exactly with 30 m pixels, we used two distinct measurements to assess street concentration within pixels. The first was street area, which we calculated by rasterizing street polygons at 1 m and summing the resulting 1 m pixels within each 30 m pixel. The second was street width, which we defined as the maximum width of any street that ran through each 30 m pixel.
We integrated additional data sources on urban form, vegetation, and demographics in order to incorporate other factors that prior studies show to have a strong influence on LST. We used building footprint polygons (Microsoft 2021) to calculate the largest building footprint within each 30 m pixel. While we examined clusters of 3 × 3 and 5 × 5 pixels to examine whether hotspots made surrounding areas warmer, we did not find a significant spatial spillover effect from streets. This null result may be due to the limited hyperlocal utility of LST as an indicator (Turner et al 2021).
We also examined Local Climate Zones (LCZ), a product created to show categories of land use, vegetation, and development for urban temperature studies (Stewart and Oke 2012). To explore questions of environmental justice, we classified pixels as a Disadvantaged Community or not according to the California Environmental Protection Agency's (EPA's) designation, which considers pollution burden, health outcomes, and vulnerability (CalEPA 2015).
To assess vegetation and land use, we examined Soil Adjusted Vegetation Index (SAVI) and albedo at 30 m using Landsat 8 data in Google Earth Engine (Roy et al 2014). A higher SAVI value indicates more greenness, with middling values corresponding to low vegetation and high values corresponding to forest. For each pixel's centroid, we calculated latitude, longitude, elevation, and distance from the Pacific Ocean (IHO 1953, Farr et al 2007.
For more specific information on data sources and calculations, see appendix A.

Regression
We use a linear regression model to test the association between LST (our dependent variable) and street area, while controlling for other predictors that may confound the relationship. These control variables consist of largest building footprint, SAVI, and albedo, all of which we standardize to mean zero and standard deviation one in order to be able to compare the magnitudes of the coefficients; disadvantaged community status as a binary variable; Local Climate Zone; and elevation and distance from the Pacific Ocean which we discretize into 10 bins in order to allow for nonlinear relationships. For an overview of other potential models, see appendix B.

Case studies
Our primary results are based on the urbanized areas of the two counties in our dataset. We complement these results with a more focused analysis of 14 case study communities (table 1; figure 1), in order to better understand the mechanisms that link street widths with urban heat. To do so, we compiled total street area and median LST, albedo, and SAVI in each neighborhood. We then selected each case study based on an extreme value strategy, choosing the areas with the maximum and minimum values for each variable. There was some overlap: the Pacific Palisades had the lowest median albedo and LST; Vernon had the lowest SAVI and highest LST; Colton had the highest albedo and lowest SAVI; and Grand Terrace had the least street area and lowest LST (figure 2). We also included in our case study selection the three California Transformative Climate Communities (TCCs) within our study area: Ontario, Watts, and the San Fernando Valley (Transformative Climate  Communities 2021). For the purposes of this study, the San Fernando Valley area is called Pacoima, as almost all of the TCC zone is within that neighborhood. The TCCs are part of a California State initiative to reduce the legacy of redlining and environmental racism on underserved communities throughout the state via community-led action plans (Transformative Climate Communities 2021). We include the three TCCs because they represent long-marginalized areas with substantial environmental disadvantages.

Effects of vegetation
We found that more vegetation, as expressed by higher SAVI, is universally correlated with lower LST. Notably, SAVI was the only tested variable without a sign change across all case study areas-that is, in each case, the regression coefficient is negative (table 2). Across the study area, SAVI also showed the strongest scaled correlations of any variable. We examined the signless magnitude of scaled correlations for street area, SAVI, albedo, and large building footprints, and found that SAVI's mean correlation with LST was 1.8% higher than the correlation of the next largest variable. SAVI was lower in environmentally-disadvantaged neighborhoods, which were >1°C warmer on average than non-disadvantaged neighborhoods.
In examinations of specific areas, we found that higher SAVI correlated strongly with lower LST, but lower SAVI did not necessarily imply higher LST. Rather, areas with lower SAVI were more diverse, with a wider range of LST (figure 3). In some coastal areas (i.e. Long Beach) the effect of SAVI on LST was less visible, likely because of the 10:30 am collection time: Los Angeles experiences a summertime morning marine layer in coastal areas that can reduce LST (Edinger 1959).

Effects of roads
As a whole, roads had no consistent effect on LST in either the full dataset or our 14 case study neighborhoods. The null effect is apparent in our regression results in examinations of street width and area, both singly and in combination with other variables (table 2; figures 2 and 3). In some neighborhoods, more land devoted to streets is associated with increased temperatures, while in others it is associated with reduced temperatures, and in all cases the magnitude of the effect is small. Different model specifications with different choices of independent variables (appendix B) also fail to establish any consistent effect.
Within residential neighborhoods, vegetation was the dominant signature, and road area was a negligible factor in affecting LST (table 2). In highly impervious neighborhoods, road surface was often dwarfed by the presence of large buildings, which had a much more substantial impact on LST, with stronger correlations visible in table 2.
Highways were consistently the coolest road class, with lower LST values than arterials or residential streets. Across the study area, the median LST of highways was >1°C cooler than other road surfaces (figure 4). In highly impervious areas with low SAVI, this effect is heightened; in Vernon, highways were 2.5°C cooler than other road surfaces. In more highly vegetated areas such as the Pacific Palisades or Chino Hills, the effect is flattened, and most road surfaces have similar LST values.
In desertified or impervious areas, highways were often cooler than all other surrounding surfaces. Median highway temperature was 0.94°C cooler than areas without roads across the study area, though this effect is skewed by the highly desertified San Bernardino areas: in the greener LA County, highways were 0.3°C warmer than non-road surfaces. In extremely unvegetated areas, though, the effect is especially pronounced; Vernon's highways were 2.4°C cooler than surrounding areas. In the neighborhood, the (concrete) highways showed a lower albedo than buildings (predominantly large warehouses) but a higher albedo than other roads. Most vegetated regions (i.e. Pacific Palisades, Rolling Hills) showed lower non-road LST than highway LST, but industrial or desertified regions (i.e. Pacoima, Colton) had lower highway LST (figure 4). Table 2. Coefficients between street area and LST tend to be small, and they do not demonstrate a particular pattern. SAVI is the most noticeable predictor of LST, with cooling effects across all areas. Scaled regression coefficients for study area and all case studies shown. Variables are normalized so as to be directly comparable with one another. Model also includes Land Cover Zone, CalEPA Environmental Disadvantaged Status, and bins for elevation and ocean proximity to account for nonlinearity. Overall R 2 = 0.767. Variables that were correlated with streets were omitted in order to maintain the model's focus on streets as a possible predictor of LST. For more regression results, see appendix B.

Vegetation and albedo
The primary cooling factor in most case study areas was vegetation, not albedo. Areas that lacked vegetation had the greatest variation in LST and correlations with albedo. This pattern was visible on road surfaces,  Chino Hills, Hidden Hills, Lancaster, and Pacific Palisades) the residential, high-SAVI areas are cooler than streets, and arterials with tree-lined medians can be especially cool. However, highways are the coolest streets in neighborhoods with diverse ground cover. In particularly impervious areas (i.e. Colton, Pacoima, and Vernon) highways are often the coolest areas overall.
where higher-albedo concretized highways showed lower LST than arterials and residential streets ( figure 4). In some areas, specific large-footprint industrial buildings with strong SAVI or albedo signatures dominated neighborhood effects (table 2). In homogeneous residential areas without major parks or barren sites (i.e. Upland, Watts, Pacoima), the relationship between albedo and LST was negative, as logic dictates. In case study areas with hotspots, some lower-albedo vegetation reduced overall LST (as in parks) or higher-albedo warehouses increased LST (figure 2).

Discussion
We found that SAVI was the strongest predictor of lower LST at pixel and neighborhood scales. This finding supports previous studies pointing to the dominant role of vegetation in mitigating urban heat, even in highly developed areas (Ballinas and Barradas 2016, Deilami et al 2018, Feng et al 2021. While urban vegetation does reduce LST, it is not always possible to rely on urban greening as a strategy for moderating SUHI. In Southern California's mediterranean and desert climates, for instance, increasing urban vegetation involves considering tradeoffs like water demand for irrigation and context factors such as species suitability (Mini et al 2014). In our study area specifically, many neighborhoods are dry and hot; grass and tree cover would not be feasible without a heavy investment in irrigation (Gober et al 2009). In Los Angeles, only 14% of the city's water is sourced locally (LADWP 2018). San Bernardino's primary aquifer is at a historic low and still losing water (SBVWCD 2022). Regional sustainability plans at both the city and county levels aim for reductions in water imports (Garcetti 2021). Our study region cannot rely on increased vegetation to reduce LST, so it is important to consider other options.
Our central question concerned whether streets contribute to urban heat and, accordingly, whether narrowing streets or introducing cool pavements would be a useful mitigation strategy. We found no consistent evidence that road surfaces in the study area increased LST relative to their surroundings. In accordance with a previous study conducted at the city-scale, we instead found that large, continuous surfaces, such as warehouses or parking lots, explained more variation in neighborhood-scale LST than streets (Liu and Zhang 2011). Urban context, therefore, appears to moderate the contributions of albedo and vegetation to surface temperature. For instance, parking lots in commercial and industrial areas occupy more land than streets, and they are typically surfaced in low-albedo black asphalt, likely amplifying their contributions to urban heat.
Within individual neighborhoods, overall morphology was important in the consideration of individual features (e.g. albedo or SAVI), echoing previous studies that emphasize the need for aligning LST mitigation strategies with local conditions (Sodoudi et al 2014, Feng et al 2021. Notably, we found that highways consistently had the lowest LST of all road surfaces, a result that confirms materials studies (Sen et al 2019, Cheela et al 2021. Because highways in our study area are primarily concrete, their higher albedo makes them cooler than asphalt. In shady neighborhoods, highways were the coolest streets, though not the coolest surfaces overall. In neighborhoods without shade, highways were the coolest of all surfaces. We also examined local morphology, and found its importance reflected in differences across land use types. In single-family residential neighborhoods with no parks, malls, or industrial sites, higher albedo was associated with lower LST. This effect was consistent regardless of climatic conditions. The presence of large warehouses, an airport, parks, or forested areas in many greener neighborhoods, however, led to a positive relationship between albedo and LST. The mechanism was imperviousness or building material rather than albedo: large white warehouses and bare ground are low LST hotspots, while parks and green spaces are cooler than their surroundings. Although we did not measure the effect of shade, we hypothesize that shade from nearby structures like buildings contributed to lower LST on some surfaces like warehouse roofs.
Between neighborhoods, localized conditions also played a role in determining LST trends. The study area was climatically and socioeconomically diverse at a neighborhood scale. Across Los Angeles County, we found that coastal areas had below-average LST and a weaker relationship between LST and SAVI likely due to a morning marine layer (figure 3). Local differences in vegetation and land use were equally important. Although the Pacific Palisades and Lancaster are both primarily classified within the 'brush, scrub' LCZ, the wealthy Palisades is heavily irrigated, while Lancaster has almost no irrigated urban canopy (Nowak et al 1996, Galvin et al 2019. LST mitigation strategies reflecting urban heterogeneity have been examined most notably in Hong Kong, where studies show that hotspots driving high LST are heterogeneously distributed throughout the city. There, proposed mitigation strategies are aimed at reducing LST in the areas of highest contribution or social vulnerability, rather than seeking to improve conditions citywide (Wong et al 2016, Hua et al 2021. Greening high-rise developments, providing shade for the elderly, and strategic additions of pocket parks in coastal areas are all potential means of addressing a regional problem with targeted, local solutions (Giridharan et al 2008, Peng and Jim 2013, Peng and Maing 2021. Additionally, Hong Kong's varied topography and unique climate have led researchers to develop locally-determined 'seasons' for examination based on highly local conditions (Giridharan et al 2007, Chan 2011. In desert climates, strong seasonal effects might also be considered in constructing neighborhood-based LST mitigation strategies.
Our results suggest that effective mitigation of LST is dependent on local context. Future research could examine several ways in which neighborhood-scale LST varies both across and within neighborhoods. Other aspects of urban form, including parking lots, industrial sites, and parks, could be evaluated to assess relative contributions to LST. In some areas, there is potential for LST reduction co-benefits of investments in native plant cover and mixed green/gray shade as options for pedestrians. Areas with the least existing shade have the highest potential for LST moderation, particularly with respect to changes in albedo. Researchers might examine whether relationships between LST and urban morphology are functions of scale, shade, evapotranspiration, vegetation's albedo, or other factors. These analyses might also consider how each cooling mechanism, particularly shade, functions across a diurnal cycle: our analysis captures late morning conditions, before peak Sunlight, and further research might consider the effects of surfaces through the late afternoon and evening. A locally-driven LST mitigation program in Los Angeles might look similar to the model currently being piloted in Athens, Greece, using large, open, impervious spaces to improve microclimates (C40 2022). Strategies for SUHI mitigation rely on a highly targeted approach; planners and scientists alike are identifying areas of high LST or low heat resilience and addressing local conditions (Skoulika et al 2014, Mavrakou et al 2018, Mavrakou and Polydoros 2021. In instances where regional changes might be impractical, a suite of targeted, local interventions may create incremental improvements in surface heat mitigation or focus on different heat-related goals such as improving thermal comfort for pedestrians.

Conclusions
We examined LST across urban areas in Southern California with respect to mitigation potential along road surfaces, but found no consistent statistical relationship that would suggest that wide streets, as measured by greatest street width intersecting each pixel, are a major contributor to urban heat across urban area as a whole. Rather, we observed that vegetated areas are universally cooler than unvegetated areas, and that concrete highways, which have high albedo, can be cooler than other impervious, loweralbedo surfaces. While streets are often emphasized as the place to implement urban heat island mitigation policies such as cool surfaces, not all streets across all regions are measurably hotter than other urban uses, suggesting that policymakers might need to take a more targeted, context-specific approach, focusing on the individual neighborhoods where streets might contribute to surface heat due to a lack of vegetation and a high proportion of low-albedo asphalt surfaces. Municipalities might also focus on other features that contribute substantially to surface heat, including large parking lots or warehouses.
A more holistic approach might consider microclimates and local conditions, including shade, coastal effects, and dominant neighborhood land use. While streets comprise a large share of land use, they are often dwarfed by green spaces, parking lots, or buildings. There may be marginal gains from moving to cooler pavements, but the bigger drivers of high LST are often large, unbroken areas (i.e. parking lots or large buildings). Our local case study areas show a diverse view of LST, with various neighborhoods affected by climate, urban morphology, and land cover. As LST and SUHI mitigation become higher policy priorities, cities should avoid placing undue emphasis on publicly-owned streets without considering neighborhood context.
A.1. LST Following Ermida (Ermida et al 2020), we calculated thermal LST at 30 m in Google Earth Engine using a statistical mono-window algorithm. We derived a Normalized Difference Vegetation Index (NDVI) and fractional vegetation cover from Landsat 8 cloud-free mosaics (Roy et al 2014). We obtained total column water vapor from 2.5 degree NCEP/NCAR reanalysis (Kalnay et al 1996) and bare ground emissivity from 100 m ASTER imagery (Hulley et al 2015). We then calculated thermal infrared emissivity and LST at Landsat 8 resolution.
To validate our calculations, we compared our summer 2020 LST to Landsat Provisional LST (30 m) and ECOSTRESS LST (38 × 69 m) from similar times of day (∼6 pm UTC) (He et al 2019, Hulley et al 2019). Our data showed similar data distribution to both datasets.

A.2. Physical data
We calculated Soil Adjusted Vegetation Index (SAVI) and Albedo from Landsat 8 (30 m) using Google Earth Engine (Roy et al 2014). Because we examined specific, localized examplesoften highly arid onesin addition to the area as a whole, we selected SAVI to provide the most accuracy within semi-arid or arid regions (Vani et al 2017).
We derived elevation from a 30 m Shuttle Radar Topography Mission DEM (Farr et al 2007) in Google Earth Engine. Using a polygon of the Pacific Ocean's boundaries, we calculated distance from the coast (IHO 1953). We calculated latitude and longitude for each pixel centroid in Google Earth Engine.

A.3. Street and building data
We used street data from Millard-Ball (2022), obtained using GIS to derive road area from the spaces between plots of land. Each road segment corresponded to an OpenStreetMap (OSM) identifier, which was used to obtain street width and category. We aggregated OSM road categories into three: highway, arterial, and residential street. To calculate street area per 30 m pixel, we used Google Earth Engine to first rasterize the street polygons at 1 m, and then to aggregate them within each 30 m pixel. We obtained building footprints from Microsoft (2021), and identified the area of the largest building that intersected each 30 m pixel.

A.4. Land use data
To define our study area, we selected neighborhoods within city limits in San Bernardino County's Southwest corner, adjacent to Los Angeles County (SB County 2020). The majority of San Bernardino County is not urbanized, and we excluded small, individual cities (i.e. Barstow, Needles) surrounded by desert so as to examine a cohesive urban area. Within Los Angeles County, we selected incorporated neighborhoods; most of the excluded area is in the Angeles National Forest (USC 2017).
To examine land use, we used Land Cover Zones (LCZs) classified at 30 m for urban temperature analyses (Stewart and Oke 2012). The LCZs break down urban form based on vegetation and building density. In our study area, many of the less-developed urban areas are either chaparral or desert environments.
We obtained polygon data for Environmentally Disadvantaged Areas from CalEPA (2015), and rasterized them at 30 m. Environmentally Disadvantaged Areas represent the top 25% of the CalEnviroScreen 3.0 Assessment, which scored census tracts based on their economic condition as well as their climatic and pollution burdens. Several areas with very low populations but high pollution burdens are also included as Disadvantaged.

Appendix B. Regression data
In evaluating the regression model, we tested multiple scenarios to evaluate our choice of independent variables with respect to confounding effects, and to ensure that there was no substantial model calibration error. We found, no matter which independent variables were included or omitted, that Street Area and LST are not consistently correlated: the values of correlations are weak in comparison to other variables, and the models that do not account for environmental factors are weaker. Across the study area, Street Area and LST were slightly negatively correlated because of the reverse SUHI effect of the inland desert regions (table B1). However, the sign changed depending on the study area and the model specification (table 2). Table B1. No matter which set of variables are used, Street Area is not a strong predictor of LST: coefficients tend to be small, and they do not have a consistent positive/negative correlation. SAVI is the most noticeable predictor of LST, with strong negative correlations. This table shows scaled regression coefficients for the entire study area. Variables are normalized so as to be directly comparable with one another. Models shown below include Street Area alone; Street Area and each primary variable individually; Street Area and each primary variable with all context variables; and all variables together with Street Area as quadratic and cubic. Other primary variables include SAVI, Albedo, and Largest Building Footprint. Context variables include Land Cover Zone, CalEPA Environmental Disadvantaged Status, and bins for elevation and ocean proximity to account for nonlinearity. All values are significant at *** p 0.001.