The Relationship between Land Cover and Sociodemographic Factors

: Multiple social and environmental justice concerns are linked to the urban form such as the distribution of socioeconomic class populations, healthcare spending, air pollution exposure, and human mobility. Because of this, the implications of the relationships between built urban form, sociodemographic factors, and air quality warrant analysis at a high spatial resolution. This study used 1m resolved LiDAR data to characterize land use in Salt Lake County, Utah, and associate it with sociodemographic and air quality data at the census block group and zip code levels. We found that increasing tree cover was associated with higher per capita income and lower minority populations while increasing built cover was linked to lower per capita income and higher minority populations. Air quality showed less strong correlations, however, decreased non-irrigated cover, increased built cover, and higher amounts of households living under poverty were related to higher long-term PM 2.5 exposure. Due to regional air pollution concerns, several policy e ﬀ orts have been undertaken to improve air quality and reduce negative health outcomes in Utah which are being informed by regulatory and research-grade air quality sensors.


Introduction
Land cover at the urban level has often been studied using LiDAR data due to its thoroughness and availability [1,2]. The urban form influences behavior ranging from transit use [3] to how children go to school [4]. Increased tree cover has been shown to promote healthier communities with highly-built areas showing the opposite effect [5][6][7]. As cities continue to increase in spatial extent and population, there is a growing need to better characterize urban emissions, particularly of criteria air pollutants (CAPs) for metropolitan planning [8,9], health [10], and emissions reduction purposes, from global to local scales [11,12].
Multiple social and environmental justice concerns are linked to the urban form such as the distribution of socioeconomic class populations [13], healthcare spending [14], and human mobility [15]. Urban areas composed of lower socioeconomic populations are disproportionately affected by hotter temperatures [16] which has been shown to have adverse health impacts [17][18][19]. Furthermore, land cover disparities have been associated with air pollution [20] and several health outcomes [21] including asthma hospitalizations [22].
The need for a thorough analysis of the implications of urban ecology on social justice using remote sensing tools, including LiDAR, has been highlighted in the literature [23]. While the relationship between urban tree cover and political economics is well-established [24], a remaining knowledge

Land Use Data
Five land cover classes were mapped for the urban extent of Salt Lake County (approximately 840 km 2 ) using two high spatial resolution datasets. Four-band, 1m resolution orthoimagery was collected under the National Agriculture Imagery Program (NAIP) in summer 2014. Orthoimage brightness values were used to calculate the normalized difference vegetation index [38], green-red vegetation index [39,40], and metrics of spatial variability of each band and index. First return, last return, and bare earth digital surface models derived from LiDAR data collected in late 2013 and early 2014 were resampled from their original 0.5m resolution to match the 1m resolution of the orthoimagery. The surface models were differenced, and metrics of spatial variability were also calculated. Orthoimage and LiDAR variables were extracted from polygons for deciduous trees, coniferous trees, irrigated low-stature vegetation, non-irrigated vegetation/soil, and impervious surface/water classes and then used to train a random forests classifier [41]. Classification results were assessed using an independent set of polygons and yielded an overall accuracy of 96.2% and a 0.94 kappa value [42]. The land cover categories are listed in Table 1.

Sociodemographic Data
Zip-code-level sociodemographic data were retrieved from the Healthy Salt Lake dataset [43], and census block group data were obtained from the United States Census Bureau American Community Survey [44]. The relevant data fields extracted were Percent Minority Population, Per Capita Income, and Percentage of Households Living Below Poverty. The shapefiles for each dataset provided centroid latitude and longitude coordinates. Salt Lake County has 612 block groups, and 34 zip codes were used in this analysis.

Air Pollution Data
Salt Lake County area is home to a dense criteria air pollution observational network facilitating a wide range of observation studies [45][46][47]. We used data from the Utah Division of Air Quality (UDAQ) regulatory observational network and the University of Utah (UofU) stationary and mobile platform network. The mobile network consists of air quality sensors that measure fine particulate matter (PM 2.5 ) and ozone [48,49] mounted on top of electric Utah Transit Authority (UTA) light-rail (TRAX) trains ( Figure 1). We estimated mean PM 2.5 concentrations for three years (2017-2019) at each zip code's centroid using the inverse distance square weighting (IDW) method [50,51] applied to data from the complete air pollution observation network.

Sociodemographic Data
Zip-code-level sociodemographic data were retrieved from the Healthy Salt Lake dataset [43], and census block group data were obtained from the United States Census Bureau American Community Survey [44]. The relevant data fields extracted were Percent Minority Population, Per Capita Income, and Percentage of Households Living Below Poverty. The shapefiles for each dataset provided centroid latitude and longitude coordinates. Salt Lake County has 612 block groups, and 34 zip codes were used in this analysis.

Air Pollution Data
Salt Lake County area is home to a dense criteria air pollution observational network facilitating a wide range of observation studies [45][46][47]. We used data from the Utah Division of Air Quality (UDAQ) regulatory observational network and the University of Utah (UofU) stationary and mobile platform network. The mobile network consists of air quality sensors that measure fine particulate matter (PM2.5) and ozone [48,49] mounted on top of electric Utah Transit Authority (UTA) light-rail (TRAX) trains ( Figure 1). We estimated mean PM2.5 concentrations for three years (2017-2019) at each zip code's centroid using the inverse distance square weighting (IDW) method [50,51] applied to data from the complete air pollution observation network. Figure 1. Salt Lake County study area. The TRAX light-rail observation sensors are on train cars traveling the Red, Green, and Blue lines that overlap along the central part of the network. The TRAX Observation Project "HAWTH" sensor (black X) is co-located with the Utah Division of Air Quality regulatory "HW" sensor (yellow circle) near the center of the map. Reproduced from [49]. UDAQ: Utah Division of Air Quality; UofU: University of Utah.

Geographical and Sociodemographic Comparisons
The 1m land use cover data were intersected with the zip code and block group shapefiles to calculate areal land cover by category using QGIS version 3.14.15. These values were converted to percent land cover at each geographical resolution. In this study, coniferous and deciduous trees were combined to form the "Tree" category. Similarly, low impervious surface and roof were combined into the "Built" category. A linear model was used to compare the different land use, sociodemographic, and pollution variables. Salt Lake County study area. The TRAX light-rail observation sensors are on train cars traveling the Red, Green, and Blue lines that overlap along the central part of the network. The TRAX Observation Project "HAWTH" sensor (black X) is co-located with the Utah Division of Air Quality regulatory "HW" sensor (yellow circle) near the center of the map. Reproduced from [49]. UDAQ: Utah Division of Air Quality; UofU: University of Utah.

Geographical and Sociodemographic Comparisons
The 1m land use cover data were intersected with the zip code and block group shapefiles to calculate areal land cover by category using QGIS version 3.14.15. These values were converted to percent land cover at each geographical resolution. In this study, coniferous and deciduous trees were combined to form the "Tree" category. Similarly, low impervious surface and roof were combined into the "Built" category. A linear model was used to compare the different land use, sociodemographic, and pollution variables.

Geographical Distribution of Land Cover Categories
Salt Lake County has a marked East-West sociodemographic divide demarcated by U.S. Interstate 15 which runs North-South ( Figure 1). The West Side is developing at a rapid pace and is home to a higher-percent minority and lower-income population. The land cover implications of these differences are shown in Figure 2, along with R 2 values using a linear fit. Figure 2a clearly demonstrates the longitudinal distribution of tree cover, with the older and wealthier East Side home to more trees. The built cover (Figure 2b) is highest in the central part of the county as this is the commercial and industrial center along Interstate 15. The built cover drops rapidly on the East Side and is comparatively higher on the West Side due to the siting of factories and commercial buildings in that part of the county. The distribution of water is relatively constant with the exception of the Jordan River which runs near to the center of the county (Figure 2c). Lastly, irrigated cover is highly concentrated in the central part of the county as it generally belongs to commercial and institutional buildings, as well as homes, and is less present on the East Side ( Figure 2d).

Geographical Distribution of Land Cover Categories
Salt Lake County has a marked East-West sociodemographic divide demarcated by U.S. Interstate 15 which runs North-South ( Figure 1). The West Side is developing at a rapid pace and is home to a higher-percent minority and lower-income population. The land cover implications of these differences are shown in Figure 2, along with R 2 values using a linear fit. Figure 2a clearly demonstrates the longitudinal distribution of tree cover, with the older and wealthier East Side home to more trees. The built cover (Figure 2b) is highest in the central part of the county as this is the commercial and industrial center along Interstate 15. The built cover drops rapidly on the East Side and is comparatively higher on the West Side due to the siting of factories and commercial buildings in that part of the county. The distribution of water is relatively constant with the exception of the Jordan River which runs near to the center of the county (Figure 2c). Lastly, irrigated cover is highly concentrated in the central part of the county as it generally belongs to commercial and institutional buildings, as well as homes, and is less present on the East Side (Figure 2d).

Sociodemographic Distribution of Land Cover Categories
Spatially-resolved sociodemographic variables and land cover categories are shown in Figure 3. The more affluent East Side is clearly highlighted in Figure 3a with only one census block group west of I-15 having an average per capita income of over USD 40,035/yr, and the majority averaging less than USD 28,366/yr. The tree cover (Figure 3b) displays a similar pattern, with few census block groups west of I-15 having over 31.9% of tree cover and most averaging less than 24.1% tree cover. The minority population is most concentrated in the northwest part of the county (Figure 3c) reaching values as high as 64.7%, with the east side of the county generally having less than 13.5% minority population. The built cover is highest in the central part of the county and along the I-15 and I-80 highways (Figure 3d). The spatial distribution pattern consistently follows the population density map shown in Figure 1.

Sociodemographic Distribution of Land Cover Categories
Spatially-resolved sociodemographic variables and land cover categories are shown in Figure 3. The more affluent East Side is clearly highlighted in Figure 3a with only one census block group west of I-15 having an average per capita income of over USD 40,035/yr, and the majority averaging less than USD 28,366/yr. The tree cover (Figure 3b) displays a similar pattern, with few census block groups west of I-15 having over 31.9% of tree cover and most averaging less than 24.1% tree cover. The minority population is most concentrated in the northwest part of the county (Figure 3c) reaching values as high as 64.7%, with the east side of the county generally having less than 13.5% minority population. The built cover is highest in the central part of the county and along the I-15 and I-80 highways (Figure 3d). The spatial distribution pattern consistently follows the population density map shown in Figure 1.  Figure 4 shows relationships between sociodemographic variables and land cover, along with R 2 values using a linear fit. Tree cover increases with increasing per capita income at the census block group as seen in Figure 4a, however, it decreases with an increasing minority population (Figure 4b).
The built environment displays the opposite pattern with decreasing cover associated with higher per capita income (Figure 4c) and increasing cover related to a higher minority population ( Figure  4d).
Tree cover and per capita income (Figure 4a) show the strongest association of the four comparisons depicted in this analysis. It can be surmised that higher-income populations are able to  Figure 4 shows relationships between sociodemographic variables and land cover, along with R 2 values using a linear fit. Tree cover increases with increasing per capita income at the census block group as seen in Figure 4a, however, it decreases with an increasing minority population (Figure 4b).
The built environment displays the opposite pattern with decreasing cover associated with higher per capita income (Figure 4c) and increasing cover related to a higher minority population (Figure 4d).
Tree cover and per capita income (Figure 4a) show the strongest association of the four comparisons depicted in this analysis. It can be surmised that higher-income populations are able to afford to either own property with a proportionately larger amount of trees (either by living near an untouched forest or growing trees on their property) or invest in neighborhood park development. Built cover (Figure 4c,d) shows less strong associations with income and minority populations because the central areas contain both higher-cost and more affordable living spaces. Furthermore, a low amount of built cover can be associated with either a more undeveloped rural location on the fringes of the county or an area with a large amount of purposely placed trees.
Urban Sci. 2020, 4, x FOR PEER REVIEW 6 of 12 afford to either own property with a proportionately larger amount of trees (either by living near an untouched forest or growing trees on their property) or invest in neighborhood park development. Built cover (Figure 4c,d) shows less strong associations with income and minority populations because the central areas contain both higher-cost and more affordable living spaces. Furthermore, a low amount of built cover can be associated with either a more undeveloped rural location on the fringes of the county or an area with a large amount of purposely placed trees.

Long-Term PM 2.5 Concentrations and Land Cover and Household Poverty
The relationships between fine particulate matter concentrations at zip code centroids and study variables are shown in Figure 5, along with R 2 values using a linear fit. Higher long-term pollutant concentrations are associated with increasing built cover (Figure 5a) and lower concentrations are linked with increasing non-irrigated cover (Figure 5b). There is no clearly visible relationship between PM 2.5 concentrations and tree cover (Figure 5c). A comparatively weaker association between increased pollutant concentrations and percent of households living below poverty is shown in Figure 5d.

Long-Term PM2.5 Concentrations and Land Cover and Household Poverty
The relationships between fine particulate matter concentrations at zip code centroids and study variables are shown in Figure 5, along with R 2 values using a linear fit. Higher long-term pollutant concentrations are associated with increasing built cover (Figure 5a) and lower concentrations are linked with increasing non-irrigated cover (Figure 5b). There is no clearly visible relationship between PM2.5 concentrations and tree cover (Figure 5c). A comparatively weaker association between increased pollutant concentrations and percent of households living below poverty is shown in Figure 5d.

Discussion
The geographical distribution of land cover in Salt Lake County shown in Figure 1 demonstrates the natural and built environment. The Jordan River is a major body of water and runs North-South in the center of the county. The East Side of the county is more mature and composed of more green spaces, including parks, and thus has had more time to develop urban forests. The built and irrigated areas are centrally located because that is where commercial and institutional buildings are found.
The irrigated areas correspond to lawns and low shrubs, commonly associated with commercial and some residential buildings.
The relationships between land cover and sociodemographic variables shown in Figure 4 demonstrate a well-established and known pattern that links wealthier and lower-minority areas with larger amounts of green space. This finding is akin to results from the existing literature [32,33]. Conversely, lower-income and higher-minority areas generally have a higher fraction of its surface as built cover. However, these associations are not strong, nor statistically significant, echoing findings from previous studies [27]. A visual representation of the East-West divide is shown in Appendix A, Figure A1.
Localized air pollution is generally most strongly linked to nearby emission sources compared to transport from outside sources. The built environment is associated with both mobile (vehicle) and stationary (buildings, industrial, etc.) sources. Therefore, the relationship between increased pollution and increased built cover ( Figure 5) is to be expected as has been found in multiple studies [48,52,53]. Non-irrigated cover refers to untouched land cover and is inversely proportional to built cover, thus showing the opposite pattern. Tree cover is either urban forest or undisturbed natural forest. The urban forests are generally located in areas that have higher pollution due to being in more built-up areas while undisturbed forests tend to be on the edges of cities and would be associated with less pollution. Elevated periods of PM 2.5 are generally found in the winter when most trees are generally inactive thus reducing their ability to filter pollutants. However, with increased incidences of wildfires, particularly in the summer when trees generally have their full array of leaves, vegetation may prove to be a more important factor in ameliorating air quality concerns. Finally, while the relationship is less obvious, deeply impoverished areas may be more impacted by higher levels of pollution.

Conclusions
This study used 1m resolved LiDAR data to characterize land use in Salt Lake County, Utah, and associated it with sociodemographic and air quality data. This highly resolved data product allowed comparisons at the census block group and zip code level. We found that increasing tree cover was associated with higher per capita income and lower minority populations while increasing built cover was linked to lower per capita income and higher minority populations. Air quality showed less strong correlations, however, decreased non-irrigated cover, increased built cover, and higher amounts of households living under poverty were related to higher long-term PM 2.5 exposure.
The findings from this work highlight glaring differences in terms of land cover and potential air pollution exposure spatially distributed along sociodemographic lines. These disparities have been analyzed separately in the past, but this manuscript combines a detailed comparison using all three variable groups showing there are several non-negligible associations between them. This methodology facilitates an understanding of the implications of land use, particularly urban forests, with regards to air pollution. Thus, the effects of reducing pollution emitters may not be felt equally across an area since tree and built cover are likely to have an impact on potential pollution exposure. Therefore, efforts to mitigate air pollution should consider involving land cover as part of a more holistic approach, the effects of which are likely to be felt most strongly by populations spending large amounts of time outdoors [54].
Several positive advances in air quality monitoring and legislative efforts have already taken place in Utah. The TRAX Light-Rail Train Air Quality Observation Project [49] has been monitoring air quality across Salt Lake County since 2015 providing unprecedented spatial and temporal resolution. Air quality data from this project was used in a recent study associating air pollution with school absences [55]. These findings resulted in the 2019 Utah State Legislature passing Rep. Wheatley's HB 0344 "Student Asthma Relief Amendments" [56] which provides stock inhalers (albuterol) to all schools in the state. Based on research showing how the local mass transit system is a valuable tool in reducing air pollutants [57], Rep. Briscoe successfully passed HB 0353 "Reduction of Single Occupancy Vehicle Trips Pilot Program" providing free transit fares on poor air quality days to encourage ridership [58]. Additionally, in an effort to better quantify the environmental impacts of new developments, Sen. Luz Escamilla passed SB 0112 "Inland Port Amendments" [59].
Future work will increase the resolution of air quality observations-both temporally (seasonally) and spatially (census block group level). The health impacts of poor air quality have been well-studied, both in Salt Lake County and elsewhere, however, the link between urban form and health, particularly at high resolution, needs further development. The Wasatch Front Regional Council, the region's Metropolitan Planning Agency, has produced projected land use scenarios [60] which will be analyzed to understand implications for environmental justice as well as potential air quality and health impacts.

Conflicts of Interest:
The author declares no conflict of interest.

Appendix A
The geographical land cover layout of Salt Lake County can be observed in Figure A1. The East Side of the county, to the left of the dividing black line, is more mature and has a higher fraction of green areas, while the West Side of the county, to the right of the dividing black line, is developing quickly and has less green areas due to a higher fraction of industrial and commercial buildings. The downtown area (red oval) is highly built up and has few green areas outside of two parks.
Urban Sci. 2020, 4, x FOR PEER REVIEW 9 of 12 Wheatley's HB 0344 "Student Asthma Relief Amendments" [56] which provides stock inhalers (albuterol) to all schools in the state. Based on research showing how the local mass transit system is a valuable tool in reducing air pollutants [57], Rep. Briscoe successfully passed HB 0353 "Reduction of Single Occupancy Vehicle Trips Pilot Program" providing free transit fares on poor air quality days to encourage ridership [58]. Additionally, in an effort to better quantify the environmental impacts of new developments, Sen. Luz Escamilla passed SB 0112 "Inland Port Amendments" [59]. Future work will increase the resolution of air quality observations-both temporally (seasonally) and spatially (census block group level). The health impacts of poor air quality have been well-studied, both in Salt Lake County and elsewhere, however, the link between urban form and health, particularly at high resolution, needs further development. The Wasatch Front Regional Council, the region's Metropolitan Planning Agency, has produced projected land use scenarios [60] which will be analyzed to understand implications for environmental justice as well as potential air quality and health impacts.
Funding: This research received no external funding.