Not All Green Is Equal: Growth Form Is a Key Driver of Urban Vegetation Sensitivity to Climate in Chicago

Abstract

Urban green spaces are important nature-based solutions to mitigate climate change. While the distribution of green spaces within cities is well documented, few studies assess whether inequities in green space quantity (i.e., percent cover) are mirrored by inequities in green space quality (i.e., vegetation health or sensitivity to stressors). Green space quality is important to measure alongside green space quantity because vegetation that is healthier and less sensitive to stressors such as climatic fluctuations sustain critical ecosystem services through stressful environmental conditions, especially as the climate changes. We use a 40-year remote sensing dataset to examine the spatial patterns and underlying drivers of vegetation sensitivity to short-term (monthly) climate fluctuations in Chicago. Our results show that although vegetation cover was not equitably distributed between racially and ethnically segregated census tracts, socio-demographic composition was not a key driver of spatial variation in short-term vegetation sensitivity to climate. Instead, we found that vegetation growth form was a strong predictor of differences in vegetation sensitivity among communities. At the census tract level, higher herbaceous/shrub cover was associated with increased sensitivity to climate, while higher tree cover was associated with decreased sensitivity. These results suggest that urban green spaces comprising trees will be less sensitive (i.e., more resistant) to short-term climate fluctuations than those comprising predominately herbaceous or shrub cover. Our findings highlight that urban green space quality can vary spatially within cities; however, more work is needed to understand how the drivers of vegetation sensitivity vary among cities, especially those experiencing different climatic regimes. This work is key to planning and planting high-quality, climate change-resilient and equitable urban green spaces.

1. Introduction

Urban green spaces provide essential ecosystem services like cooling, carbon sequestration, and pollution removal to communities living in cities [1,2]. However, the distribution of and access to urban green space is not always equitable, especially in terms of tree canopy cover distribution [1,3,4]. Additionally, many studies have demonstrated that urban green space is consistently smaller or harder to access in minority and/or low-income communities, leading to measurable environmental consequences for these communities such as higher land surface temperatures and lower health outcomes [5,6]. However, existing studies rarely assess how the quality of green space varies among communities, which modulates the services provided, in addition to quantifying green coverage [7].

Green space quality, though often less studied than quantity, can be more influential in determining the actual services received by communities per area [8,9,10,11,12]. However, the metrics and characteristics used to assess green space quality vary widely across the literature, encompassing dimensions such as accessibility, safety, health-related outcomes of nearby communities, ecological characteristics, and the availability of infrastructure and amenities, depending on the outcomes of interest [13,14]. In this study, we define green space quality in ecological terms, focusing on vegetation health, composition, and resistance to stressors. These characteristics affect key services such as cooling, air purification, and resilience to climate extremes [8,15,16]. For example, denser tree canopies, as measured by leaf area per basal area, provide more surface temperature cooling capacity than sparse canopies [15]. In addition, healthy vegetation can better maintain processes important for cooling like transpiration and produce fewer irritating biogenic volatile organic compounds under stressful conditions (BVOCs; [17,18,19]). Healthy vegetation can also resist or recover more quickly from pests and pathogens, sustaining these ecosystem services over time [20]. Thus, two communities with similar green space quantity may still differ in the benefits received due to differences in vegetation quality, with potential implications for environmental equity [21,22].

Vegetation sensitivity to climate is an important component of vegetation quality because vegetation that is more resistant or resilient to climatic fluctuations can continue to provide ecosystem services like cooling during drought or heatwaves [23]. Urban vegetation health is sensitive to both short-term and long-term fluctuations in climatic conditions [24,25]. Heat and drought can negatively affect vegetation health; however, the sensitivity of vegetation to these stressors may depend on species composition [24,26], vegetation growth form [27], or extrinsic factors like presence of irrigation, application of soil amendments, or impervious surfaces [23,28,29,30]. Maintaining vegetation health will be critical as these stressors are expected to become more frequent as the climate changes [31]. Whether and how the sensitivity of vegetation health to climate differs among communities has not been widely studied, but it has important implications for reducing local heat stress and air pollution in cities [23,32]. If disadvantaged communities have vegetation that is more sensitive to climate fluctuations, then the negative consequences of climate change may be exacerbated in these neighborhoods even if the amount of green space is similar. With the expansion of publicly available, analysis-ready remote sensing data, assessing temporal vegetation dynamics over large urban areas is now attainable, making it possible to assess these key patterns in vegetation quality.

We may expect vegetation sensitivity to stressful climatic conditions to vary at intra-city scales for several reasons. First, we might expect that wealthier communities have the additional resources necessary to provide supplemental irrigation to trees and gardens, potentially buffering them against heatwaves and drought [23,29]. Supplemental irrigation may be especially important in mitigating the acute effects of short-term climatic events on vegetation health [23,33]. Second, decades of racially discriminatory housing policies and practices, structural disinvestment in historically marginalized neighborhoods, and exclusionary zoning have shaped contemporary patterns of residential segregation and perpetuated an uneven distribution of environmental disservices such as soil contamination, air pollution, and high impervious surface cover, especially in Black and brown communities [34,35,36]. These inequities underpin the landscapes in which urban vegetation grows and thus can ultimately affect urban biological processes, including vegetation quality [7]. Finally, vegetation growth form, structure, and species composition are heterogeneously distributed within cities, which could generate differences in sensitivity to climate [37,38,39].

Chicago, IL, is one of the most segregated cities in the USA, and previous work has demonstrated that vegetation cover and green space access are not equitably distributed among racial/ethnic or socioeconomic groups (henceforth referred to as socio-demographic groups), with Latino and poor communities experiencing lower canopy cover and less access to green space [8,34,40,41]. However, little is known about how the quality of green space varies among these communities. In this study, we use a 40-year remotely sensed time series of vegetation greenness in the city of Chicago to understand whether vegetation sensitivity to short-term climatic fluctuations varies spatially among communities. Specifically, we wanted to assess whether disparities in green space quality mirror disparities in green space quantity. To approach this overarching goal, our study was based on three aims:

• To reassess the association between vegetation cover (i.e., vegetation quantity) and socio-demographic variables;
• To characterize the magnitude and spatial distribution of vegetation sensitivity to short-term climate fluctuation (i.e., vegetation quality);
• To identify socio-demographic and ecological drivers of spatial variation in vegetation sensitivity to short-term climate fluctuation.

We expect to find lower vegetation cover and higher vegetation sensitivity in lower-income and non-White communities, given that these communities are less likely to have the excess resources necessary to mitigate the acute effects of heat and drought on vegetation health via irrigation and other management practices [23,29]. We expect that vegetation growth form (i.e., tree versus herbaceous/shrub cover) will be a key driver of vegetation sensitivity. Our study represents a novel application of remotely sensed data in urban areas. While many studies have used remotely sensed data to assess spatial variation in green space and cover [41], few have utilized the temporal component of these data to assess variations in urban vegetation dynamics through time [23,42,43,44]. This work contributes to our understanding of the importance of green space quality, its drivers, and how it varies spatially across communities.

2. Materials and Methods

2.1. Study Area

Chicago, IL, USA, is the third largest city by population in the USA, with a total population of 2.66 million in 2023. The city covers an area of 235 km² and is located on the southwestern shore of Lake Michigan. The population of Chicago is approximately equally divided among White, Black, and Latino racial/ethnic groups. These communities tend to be highly segregated within the city, with the south and west sides dominated by Black communities, the north side dominated by White communities, and the northwest and southwest sides dominated by Latino communities (Figure 1). Chicago is considered one of the most segregated cities in the country [45].

Urban vegetation in Chicago is distributed across private and public land and consists of human-cultivated plants and small areas of remnant wild or restored natural ecosystems and vacant lots. The largest contiguous patches of urban green space are public parks and historic graveyards scattered within the city and along the lake front. These largely represent cultivated and managed areas. Urban green space covers approximately 55 km² [1]. Most trees in the city are deciduous broad-leaved trees, and in 2020, the top three most common species by leaf area were silver maple (Acer saccharinum L.), Norway maple (Acer platanoides L.), and white ash (Fraxinus americana L.; Chicago Region Trees Initiative (CRTI), 2021) [46]. The distribution of vegetation cover, especially in historical parks, has been relatively stable in Chicago in the past 40 years. For example, Jackson Park, a large urban green space on the south side of Chicago consisting of turf fields and wooded areas, was constructed for the World’s Columbian Exposition in 1893.

The climate in Chicago is characterized as a hot summer, humid continental climate typified by four distinct seasons, with hot summers and cold winters. During the 1991–2020 period in Chicago, the mean annual cumulative precipitation was 960 mm, with most of the precipitation falling between April and October. The mean temperature during the same period was 15.3 degrees Celsius. The region has been characterized by frequent fluctuations between drought and non-drought periods, with the most severe recent droughts occurring in 1988, 2005, and 2012 [31]. As the climate changes, extreme temperatures are expected to increase, especially during the summer growing season. Hotter summer temperatures are expected to increase the severity of both long- and short-term droughts [31]. Currently, the peak vegetation growing season occurs between May and September (Figure S2), although this may lengthen as increased temperatures are expected to lengthen the growing season [31].

2.2. Data Sources and Data Pre-Processing

2.2.1. Remote Sensing Data Pre-Processing with Google Earth Engine

For this study, we used Landsat 5, 7, 8, and 9 Level 2, Collection 2, Tier 1 surface reflectance data at a 30 m spatial resolution available through Google Earth Engine (GEE). To pre-process the images in the Chicago region, we removed scenes with greater than 30% cloud cover over land, which resulted in 975 scenes between April 1984 and December 2022. We masked clouds and oversaturated pixels and then calculated the normalized difference vegetation index (NDVI). For each pixel, we calculated the median monthly NDVI value across all scenes captured during a given month–year (e.g., all scenes captured in July, 2005) [47]. This resulted in 404 median month–year scenes, which were then exported from GEE for continued pre-processing in R (Figure 2 and Figure 3B).

For each pixel in each median month–year scene, we calculated the NDVI anomaly z-score, a standardized metric of NDVI anomalies commonly used in studies assessing the effects of climate or weather on vegetation [23,48,49]. The anomaly z-scores were calculated using the following equation:

$$Z_{it}={\displaystyle \frac{NDV{I}_{it}-\overline{NDV{I}_i}}{{\mathsf{\sigma }}_{NDV{I}_i}}}$$

(1)

where Z_it is the z-score at time t and location i, $\overline{NDV{I}_i}$ is the mean NDVI value for all pixels at location i, and σ_NDVIi is the standard deviation of all pixels at location i.

Calculating z-scores normalizes NDVI deviations against typical seasonal phenological change and greenness magnitude at each pixel [50]. In this study, z-scores were used as a metric of deviation in vegetation health from typical values during a given month–year period. Positive z-scores represent greener and healthier vegetation than average, and negative z-scores represent browner and less healthy vegetation than average. NDVI z-scores were calculated using the function app() in the package terra in R [51,52].

2.2.2. Socio-Demographic Data

We downloaded 2017–2021 United States Census Bureau’s American Community Survey 5-year summary data from the IPUMS National Historical Geographic Information System [53] at the census tract level with 2020 tract boundaries. Specifically, we downloaded race and median household income data for the city of Chicago (n = 791 tracts; Figure S5). We removed census tracts representing major airports because they contained a very small population relative to the large area. We then categorized each census tract as majority (>50%) Black, White, or Latino (Figure 1) and removed census tracts with no racial/ethnic majority (n = 76 tracts). This resulted in a total of 714 census tracts that were used for further analysis.

2.2.3. Landcover and Elevation Data

We downloaded 2017 high-resolution Chicago region land cover data at a 1ft resolution [54]. Land cover data were created using 2017 LiDAR and aerial imagery. Two classes of vegetation growth form, tree canopy and herbaceous/shrub cover, were summarized as percent cover at the census tract level using the function zonal() in the R package terra (Figure S3). We used these percentages to calculate total vegetation cover (tree canopy plus herbaceous/shrub cover). We also downloaded a high-resolution (1ft) digital elevation model (DEM) constructed using 2022 LiDAR data [55]. We summarized the mean elevation of each census tract in meters using the function zonal() (Figure S7).

2.2.4. Climate Data

Gridded monthly maximum temperature (TMAX), cumulative precipitation (PRCP), and vapor pressure data for the Chicago region during the study period (1984–2022) were downloaded from daymet using the R package daymetr at a spatial resolution of 1 km [56]. Vapor pressure and maximum temperature were used to calculate gridded maximum vapor pressure deficit (VPD), a metric of atmospheric evaporative demand. We used the Clausius–Clapeyron equation for calculating saturation vapor pressure using the SVP.ClaCla() function in the R package humidity [57].

For this study, drought conditions were characterized as the monthly standardized precipitation evapotranspiration index (SPEI), which can be used to quantify the severity of drought conditions relative to normal as measured by cumulative climate water balance (precipitation minus potential evapotranspiration) during a given period of time [58]. SPEI is particularly well-suited for assessing plant response to drought because it includes the influence of temperature on atmospheric evaporative demand [58]. SPEI values of zero represent near normal conditions, while positive values represent wetter-than-normal conditions and negative values represent drier-than-normal conditions. Monthly SPEI was calculated using daymet PRCP and potential evapotranspiration (calculated using TMAX and latitude with the Hargreaves equation) at a monthly timescale with the function spei() in the package SPEI [59]. All gridded climate and drought data were cropped and reprojected to match spatial attributes of Landsat data using the R package terra.

2.3. Calculating Vegetation Sensitivity to Climate and Drought Conditions

In this study, we define vegetation sensitivity to short-term, interannual variation in climate as the association between monthly NDVI z-scores and climatic conditions within that current month as measured by the Pearson correlation coefficient [23,24,60,61]. This represents acute responses of vegetation greenness to month-to-month fluctuations in climate. Specifically, we calculated pixel-wise Pearson correlation coefficients between NDVI z-scores during the peak growing season (May–September) and climate (TMAX, PRCP, and VPD) as well as drought (SPEI; herein collectively referred to as climate variables) during the study period on a monthly timescale. Pearson correlation coefficients represent the direction and strength of the association between NDVI z-scores and climate. For example, a negative value of vegetation sensitivity to VPD indicates browner-than-normal vegetation, which is associated with high values of VPD, demonstrating that vegetation at that pixel is sensitive to short-term fluctuations in atmospheric evaporative demand. On the other hand, positive sensitivity to SPEI indicates browner-than-normal vegetation, which is associated with low SPEI values (e.g., drought conditions; see Figure 3E,F for examples of pixel-wise relationships). The larger the absolute value of the correlation, the stronger the influence of climate on NDVI z-scores. While regression slopes have also been used to quantify the sensitivity of vegetation to climate and drought conditions, we chose to focus our analysis on Pearson correlation coefficients because slope can be easily modulated by outliers and the idiosyncrasies of the underlying data [23,32]. Pixel-wise Pearson correlation coefficients were calculated using the cor() functions within the app() framework in the R package terra. This resulted in four gridded rasters of sensitivity values (i.e., sensitivity measured by correlation coefficient for each climate variable; Figure 4A,C and Figure S6).

The resulting four sensitivity rasters were then masked to exclude cells that were not vegetated during the study period (Figure 3C,D, Figure 4A,C, and Figure S1). This is an important step because cells representing non-vegetated surfaces like roads and buildings will register as insensitive to climate because NDVI z-scores do not vary predictably with climate, potentially generating bias in our results. We masked cells that had a median growing season (May–September) NDVI value of <0.2 during the entire study period (1984–2023) as values of 0.2 to 1 indicate a moderate to high density of vegetation. After masking non-vegetated pixels, the mean sensitivity to climate was summarized within census tracts to facilitate statistical analysis using the function zonal() in the package terra (Figure 4B).

2.4. Statistical Analyses

2.4.1. Assessing Drivers of Spatial Variation in Vegetation Cover

We constructed three spatial linear mixed-effect models (LMMs) to assess how vegetation cover varied among racial/ethnic community groups and income levels in Chicago at the census tract level. Spatial LMMs are well-suited for this analysis because they account for spatial autocorrelation among neighboring census tracts in the residual error term, improving the accuracy and interpretability of model estimates. For our study, each model included either percent tree cover, herbaceous/shrub cover, or total vegetation cover within census tracts as the response variable. Both models included the following fixed effects: majority racial/ethnic population (Black, White, or Latino as a categorical variable) and median household income (log-transformed to improve normality). These covariates were chosen to understand how vegetation cover classes vary among socio-demographic groups in the city. We included community “side” (e.g., South Side and West Side) as a random intercept to account for community-level attributes, including underlying environmental factors (e.g., soil type and distance to lake) and community identity (Figure S4). To account for spatial autocorrelation in the residuals, we included an exponential spatial correlation structure with distance defined by tract centroids (X and Y coordinates). The correlation structure models a decline in residual correlation with increasing distance between tracts. This ensures that parameter estimates take into account the spatial structure of the data [62]. Each model took the following form:

$$\mathrm{P}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{C}\mathrm{o}\mathrm{v}\mathrm{e}{r}_{ij}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\cdot M{\mathrm{a}\mathrm{j}\mathrm{o}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{y}\mathrm{P}\mathrm{o}pulation}_i+{\mathsf{\beta }}_2\cdot {\log }_{10}\left(Incom{e}_i\right)+b_j+{\mathsf{\epsilon }}_{ij}$$

(2)

where $\mathrm{P}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{C}\mathrm{o}\mathrm{v}\mathrm{e}{r}_{i\mathrm{j}}$ is percent vegetation cover (herbaceous/shrub, tree, or total) in census tract i in community side j, ${\mathsf{\beta }}_1-{\mathsf{\beta }}_2$ are the fixed-effect coefficients for majority racial/ethnic group and log-transformed median household income, $b_j$ is the random intercept for community side, and ${\mathsf{\epsilon }}_{ij}$ is the spatially autocorrelated residual error modeled using exponential decay.

2.4.2. Assessing Drivers of Spatial Variation in Vegetation Sensitivity

We constructed four spatial LMMs (one for each sensitivity metric: SPEI, VPD, PPT, and TMAX) to evaluate how vegetation sensitivity to climate varied in response to a suite of socio-demographic and environmental drivers at the census tract level. The response variable in each model was vegetation sensitivity to climate, measured as the correlation coefficient (r) between NDVI z-scores and each climate metric at a monthly timescale. All models included the following fixed effects: majority racial/ethnic population, log-transformed median household income, percent tree cover, percent herbaceous/shrub cover, and mean elevation. These covariates were selected to assess how social and ecological factors influence the climate sensitivity of urban vegetation. We included both percent tree and percent herb/shrub cover to evaluate whether differences in growth form contribute to climate sensitivity. Elevation was included as a proxy for site hydrology, potentially reflecting depth to the water table. We included an exponential spatial correlation structure with distance defined by tract centroids as well as “side” as a random intercept to account for broad-scale regional differences within the study area. Each model took the following form:

$$\begin{gathered} \mathrm{VegetationSensitivity}{\mathrm{y}}_{\mathrm{i}\mathrm{j}} \\ ={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\cdot \mathrm{M}\mathrm{a}\mathrm{j}\mathrm{o}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{y}\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}{\mathrm{n}}_{\mathrm{i}}+{\mathsf{\beta }}_2\cdot {\log }_{10}\left(\mathrm{I}\mathrm{n}\mathrm{c}\mathrm{o}\mathrm{m}{\mathrm{e}}_{\mathrm{i}}\right)+{\mathsf{\beta }}_3\cdot \mathrm{T}\mathrm{r}\mathrm{e}{\mathrm{e}}_{\mathrm{i}}+{\mathsf{\beta }}_4\cdot \mathrm{H}\mathrm{e}\mathrm{r}\mathrm{b}/\mathrm{s}\mathrm{h}\mathrm{r}\mathrm{u}{\mathrm{b}}_{\mathrm{i}}+{\mathsf{\beta }}_5 \\ \cdot \mathrm{E}\mathrm{l}\mathrm{e}\mathrm{v}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}{\mathrm{n}}_{\mathrm{i}}+{\mathrm{b}}_{\mathrm{j}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{j}} \end{gathered}$$

(3)

where $VegetationSensitivit{y}_{ij}$ is the vegetation sensitivity (as measured by the Pearson correlation coefficient) to a climate variable in tract i in community side j; ${\mathsf{\beta }}_0$ is the fixed intercept; ${\mathsf{\beta }}_1-{\mathsf{\beta }}_5$ are the fixed effect coefficients for majority racial/ethnic group, log-transformed median household income, percent tree cover, percent herbaceous/shrub cover, and elevation, respectively; $b_j$ is the random intercept for community side; and ${\mathsf{\epsilon }}_{ij}$ is the spatially autocorrelated residual error modeled using exponential decay.

All models were fit using the lme() function in the R package nlme [63] with restricted maximum likelihood (REML). The significance of fixed effects was assessed with likelihood ratio tests (LRTs) conducted using the drop1() function. LRTs determine the effect of removing one fixed effect from the full model, allowing for an assessment of the individual contribution of each fixed effect. Significant p-values indicate that removing the corresponding variable significantly worsens model fit, suggesting that the variable contributes meaningfully to explaining variation in vegetation sensitivity to climate.

We conducted post-hoc tests to assess differences in vegetation cover type and sensitivity among racial/ethnic community groups using the Tukey method with the emmeans() function in the R package emmeans [64]. We calculated pseudo-marginal, conditional, and partial R² values for each model using the function r.squaredGLMM() in the MuMIn package [65]. All statistical analyses were conducted in R version 4.4.1 [52].

2.4.3. Assessing the Robustness of Community Composition Effects on Vegetation Sensitivity

Although Chicago has a long history of racial segregation, migration or displacement of racial/ethnic groups and gentrification have changed the socio-demographic composition of some census tracts during the 40-year period of this study (Figure S5). This means that our metrics of sensitivity, which were calculated using a composite of Landsat data during the past 40 years (Figure 3), coincide with changes in underlying community composition (i.e., majority racial/ethnic group and median household income). These underlying changes could potentially limit our ability to detect the effect of community composition on vegetation sensitivity among census tracts.

To assess the robustness of our results, we constructed an additional model to assess the drivers of spatial variation in vegetation sensitivity using only census tracts that had stable racial/ethnic and socioeconomic composition during the 40-year period. These tracts were identified using the Longitudinal Tract Data Base (LTDB), which provides harmonized estimates of tract-level decadal census data within 2010 boundaries from 1970 to 2020 [66]. Using this data, we identified 488 tracts with stable majority racial/ethnic population and 246 tracts with stable median household income quartile (e.g., a tract that remained in the first quartile of median household income relative to all tracts during each decadal census period). Of these, 198 tracts exhibited both stable racial/ethnic majority and income quartile. This was further narrowed down to 188 tracts that had the same boundaries in 2010 and 2020, which could thus be used for our robustness analysis (Figure S5). These 188 tracts represent a 26% subset of our full dataset (714 tracts). Using these tracts, we were able to assess the effect of racial/ethnic majority and socio-economic status on vegetation sensitivity in areas of the city where these demographics have remained stable during the study period. We constructed four spatial linear mixed-effect models (one for each sensitivity metric) using the same structure described above.

3. Results

3.1. Drivers of Spatial Variation in Vegetation Cover

Herbaceous/shrub cover vegetation differed among racial/ethnic communities as defined by the majority racial/ethnic population in each census tract (majority population: χ² = 44.4, p < 0.001;

Table 1. Results from the models designed to test the effect of demographic and socioeconomic variables on herbaceous/shrub cover, tree canopy cover, and total vegetation cover among census tracts. Values represent chi-squared statistics from the analysis of deviance. Significant values at p < 0.05 are bolded, and levels of significance of fixed effects are indicated by asterisks. Conditional, marginal, and partial R² values are also reported.

Response Variable	Majority Population	Income	Marginal R2	Conditional R2
Herbaceous/shrub cover	44.38 ***	1.56	0.13	0.29
Partial R2	0.06	<0.01	-	-
Tree canopy cover	53.74 ***	2.2	0.11	0.21
Partial R2	0.1	<0.01	-	-
Total vegetation cover	95.23 ***	0.02	0.14	0.30
Partial R2	0.13	<0.01	-	-

*** p ≤ 0.001.

; Figure 5A). Black communities had higher herbaceous/shrub cover (16.4 ± 1.2%) than either White or Latino communities (11.8 ± 1.2% and 10.7 ± 1.2%, respectively). The conditional R² value of the full model, which includes variance explained by the random effect community side, was 0.29, and the marginal R² value of the full model was 0.13.

Canopy cover differed among Black, White, and Latino communities (majority population: χ² = 53.7, p < 0.001; Table 1; Figure 5B). Latino communities had lower canopy cover (mean ± SE: 16.5 ± 0.9%) than either Black or White communities (21.7 ± 0.9% and 21.7 ± 0.9%, respectively). Total vegetation cover (tree canopy plus herbaceous/shrub cover) differed among all three racial/ethnic communities, where Black communities had the highest (38.4 ± 1.7%) and Latino communities had the lowest (38.4 ± 1.7%; Table 1; Figure 5C). Median household income did not explain the variation in any of the three vegetation cover variables (Supplemental Tables S1–S6).

3.2. Vegetation Sensitivity

3.2.1. Pixel-Wise Patterns of Vegetation Sensitivity

Most vegetated pixels in Chicago showed that NDVI z-scores are sensitive to short-term climatic fluctuation, where environmental stress (i.e., high VPD, drought, low precipitation, and hot temperatures) is associated with low NDVI z-score values (i.e., values representing browner-than-normal vegetation). Specifically, 82.4% and 72.4% of pixels demonstrated negative sensitivity values to VPD and TMAX, respectively (Figure S6B,D). This indicates browner-than-normal vegetation under high VPD and hot temperatures. Similarly, 81.1% and 79.4% of pixels demonstrated positive sensitivity values to SPEI and PRCP, respectively (Figure S6A,C). This indicates browner-than-normal vegetation under drought and dry conditions.

On a pixel-by-pixel basis, Pearson’s correlation coefficients between NDVI z-scores and SPEI and VPD showed the greatest variability and highest average magnitude (i.e., the mean absolute value of the Pearson correlation coefficient) among the four variables tested (SPEI range: −0.35 to 0.48, mean: 0.14, SD = 0.12; VPD range: −0.46 to 0.39, mean: 0.15, SD = 0.12; Figure 4A). In comparison, correlations with PRCP and TMAX were generally weaker and less variable (PRCP range: −0.32 to 0.44, mean: 0.11, SD = 0.10; TMAX range: −0.21 to 0.18, mean: 0.04, SD = 0.05; Figure S6). The mean absolute value of the correlation coefficients indicates the average strength of the relationship between NDVI z-scores and short-term climate values. These results suggest that vegetation is more sensitive to short-term variability in SPEI and VPD than to PRCP or TMAX, as indicated by the higher mean magnitude of correlations. In other words, the NDVI responds more strongly and consistently to changes in drought conditions and atmospheric evaporative demand than to temperature or precipitation alone.

3.2.2. Drivers of Spatial Variation in Vegetation Sensitivity

Vegetation sensitivity to all climate variables was best explained by the percent cover of herbaceous/shrub vegetation in census tracts (partial R² range among all models: 0.24–0.32;

Table 2. Results of the four linear mixed-effect models designed to detect the effect of majority racial/ethnic population (i.e., Black, White, or Latino), median household income, land cover, and elevation on vegetation sensitivity to climate. Values represent chi-square statistics from the analysis of deviance. Significant effects at p < 0.05 are bolded, and levels of significance of fixed effects are indicated by asterisks. Marginal and conditional R² values as well as partial R² values for each fixed effect are listed on the line below for each model. Variables are ordered from highest to lowest marginal R² values.

Response Variable	Majority Population	Income	% Herbaceous and Shrub	% Tree	Elevation (m)	Marginal R2	Conditional R2
SPEI r	6.98 *	1.32	267.50 ***	42.62 ***	27.19 ***	0.34	0.54
partial R2	0.02	<0.01	0.27	0.05	0.03	-	-
VPD r	3.18	0.48	297.49 ***	38.62 ***	41.97 ***	0.37	0.47
partial R2	<0.01	<0.01	0.32	0.04	0.05	-	-
TMAX r	1.08	0.19	239.72 ***	6.12 *	38.23 ***	0.36	0.41
partial R2	<0.01	<0.01	0.27	0.01	0.05	-	-
PRCP r	4.91	0.90	248.74 ***	44.30 ***	14.31 ***	0.31	0.56
partial R2	0.02	<0.01	0.24	0.05	0.01	-	-

* p ≤ 0.05 and *** p ≤ 0.001.

). Census tracts with higher herbaceous and shrub cover were more sensitive to all four climate variables (SPEI: χ² = 267.5, p < 0.001; VPD: χ² = 297.5, p < 0.001; TMAX: χ² = 239.7, p < 0.001; PRCP: χ² = 248.7, p < 0.001; Table 2; Figure 6A,C). In contrast, census tracts with higher tree canopy cover were less sensitive to all climate variables (SPEI: χ² = 42.6, p < 0.001; VPD: χ² = 38.6, p < 0.001; TMAX: χ² = 6.1, p = 0.01; PRCP: χ² = 44.3, p < 0.001; Table 2; Figure 6B). However, as indicated by partial R² values, this effect was weaker than the effect of herbaceous/shrub cover. Census tracts at higher elevation had vegetation that was less sensitive to climate (SPEI: χ² = 27.2, p < 0.001; VPD: χ² = 42, p < 0.001; TMAX: χ² = 38.2, p = 0.01; PRCP: χ² = 14.3, p < 0.001; Table 2; Figure S9; Tables S7–S14).

We found evidence that only vegetation sensitivity to SPEI differed among the racial/ethnic communities within Chicago (SPEI: χ² = 7, p = 0.03; Table 2); however, post hoc tests revealed that no one community had the highest vegetation sensitivity among the three racial/ethnic groups. Black communities had higher vegetation sensitivity to SPEI than Latino communities. Yet, sensitivity did not differ between White relative to Black or Latino communities (Figure 7A). Median household income did not have a significant effect on vegetation sensitivity to climate in any model tested. Among all models, the marginal R² value ranged from 0.30 to 0.42, while the conditional R² value ranged from 0.41 to 0.55. Because all models showed similar results, the figures focus on the two variables most ecologically important for plant growth and health—SPEI and VPD.

3.2.3. Robustness of Socio-Demographic Effects on Vegetation Sensitivity

When assessing the drivers of sensitivity only among census tracts that had stable racial/ethnic and income quartiles, we found similar results as when using all census tracts (i.e., the full dataset) relative to this reduced set (Tables S14–S23). Among all four models, herbaceous/shrub cover explained most of the variation in sensitivity to climate among all tracts. Similar to above, we found evidence that sensitivity to climate differed among racial/ethnic communities but not income levels. Majority population was a significant effect in only one of the four models, in this case, maximum temperature (TMAX: χ² = 7.8, p = 0.02). Again, post-hoc tests showed that Black communities had higher vegetation sensitivity to TMAX than Latino communities (Table S22).

4. Discussion

4.1. Overview

This study uses a 40-year time series of remote sensing data to assess patterns of vegetation sensitivity to climate in the city of Chicago and represents a novel use of remotely sensed data in urban ecosystems [23,24]. Our study is one of the first to assess how both green space quantity (as measured by percent cover) and ecological quality (as measured by vegetation sensitivity) jointly vary among communities within a city. With respect to green space quality, we found that vegetation growth form was a key driver of differences in vegetation sensitivity among communities. At the census tract level, higher herbaceous/shrub cover was associated with increased sensitivity to climate, while higher tree cover was associated with decreased sensitivity to climate. While we re-affirmed inequity in vegetation cover among Latino relative to Black or White communities in Chicago, we found weak evidence that inequities in green space cover were mirrored by inequities in green space quality, in this case, as measured by vegetation sensitivity to climate. These findings have implications for climate-resistant and -resilient urban green space planning as well as for environmental equity and justice in Chicago.

4.2. Vegetation Cover Is Not Equitably Distributed Among Racial/Ethnic Communities in Chicago

This study confirms previous findings demonstrating lower tree canopy and total vegetation cover among Latino communities in Chicago relative to Black or White communities [8,34,40,41] (Figure 4B; Table 1). This disparity directly translates to the magnitude of ecosystem services provided to Latino-majority communities [7,67]. Extensive work has demonstrated that tree canopy cover in particular reduces land and air surface temperature in urban landscapes through direct shading and evaporative cooling [2,68,69,70]. Lee et al. [71] demonstrated that, for Latino communities in Chicago, this canopy cover inequity translates to higher land surface temperatures, especially during the hottest summer days. Their work also highlighted that health-related impacts associated with heat exposure could be exacerbated by climate change [72]. In contrast to racial/ethnicity-based inequalities, we found no evidence of income-based inequalities in either herbaceous/shrub or canopy cover, which is consistent with previous research that found that neighborhood racial/ethnic composition is a stronger predictor of environmental disparities than wealth [7] (Table 1).

4.3. Vegetation Is Sensitive to Short-Term Climate Fluctuation in Urban Landscapes

Despite the challenges associated with using remote sensing to detect urban vegetation dynamics (e.g., mixed pixel problems, heterogenous landcover, etc.), we were able to detect signals of vegetation sensitivity to climate in these highly managed landscapes. We found that, on a monthly timescale, more stressful conditions (drought, high VPD, high temperature, and low precipitation) resulted in lower NDVI z-score values, and the magnitude of this effect varied among census tracts (Figure 4B). The sensitivity to water-related climate variables detected here is similar to other studies that have used a remote sensing approach to assess urban vegetation sensitivity to climate and, in particular, to drought [23,24,27]. For example, Leisenheimer et al. (2024) [24] used 10 and 20 m Sentinel-2 imagery to assess how NDVI sensitivity at various timescales varied among seven common street tree species in Leipzig, Germany. They also found that drought negatively affects the NDVI but that the magnitude varies among species and timescales, with some species showing a time-lagged effect. This body of work, including the current study, demonstrates the utility of remote sensing time series data in assessing the drivers and distribution of urban tree health and can help advance strategies for planning climate-resilient urban forests.

In the current study as well as in the studies mentioned above, correlation coefficient (r) values between the NDVI and climate varied widely among vegetated cells, indicating that factors other than climate explain deviations in the NDVI. For example, heavy pruning or removal of old trees with declining health could result in reductions in the NDVI that are not associated with climate. In Chicago, an outbreak of emerald ash borer significantly impacted the health of urban ash trees (Fraxinus spp.) starting around 2012 [46]. This type of damage could be unrelated to climate or could worsen under stressful climatic conditions, resulting in compounded reductions in the NDVI. Understanding how vegetation sensitivity to climate interacts with insect infestations will help us better predict how large-scale disturbances will affect urban vegetation health [73,74].

4.4. Vegetation Growth Form Drives Spatial Variation in Vegetation Sensitivity to Climate

In support of our prediction, we found that growth form is a key driver of vegetation sensitivity to climate. Census tracts with higher herbaceous/shrub cover tended to be more sensitive to short-term, interannual climate fluctuations (Figure 6A; Table 2), suggesting that herbaceous/shrub vegetation is more sensitive to climate than established tree cover. In Chicago, the herbaceous/shrub land cover class consists of mostly herbaceous vegetation. This includes large areas of turf grass for recreation as well as small areas of wetland and restored prairie. This difference in sensitivity between herbaceous/shrub and tree cover is likely due to differences in resistance versus resilience of vegetation types to short-term fluctuations in climate which, in this study, was measured on a monthly timescale. In managed landscapes, the canopy health of large trees is likely more resistant to monthly fluctuations in climate relative to herbaceous vegetation due to deeper root systems. For example, Miller et al. [27] used high-resolution airborne imagery to assess urban vegetation response to drought in Los Angeles, California and found that turf grass senesced quickly due to drought relative to trees. However, when drought conditions abated, turf grass recovered more quickly than urban trees. These results suggest that at short timescales, turf grass is resilient, while urban trees are resistant to drought.

While we chose to focus on acute, short-term responses of vegetation greenness to climate fluctuation, it is possible that the effect of monthly climatic conditions on greenness could be delayed and thus not captured in our current analysis [75]. This would result in strong associations between current NDVI z-scores and lagged conditions (e.g., climate in the previous months or years). Previous studies have demonstrated that remotely-sensed responses of vegetation to climate, and particularly to drought, in humid continental ecosystems similar to Chicago are best explained by short-term drought conditions such as in the current or preceding 1–3 months, suggesting that vegetation responds relatively quickly to drought conditions [60,75]. For example, Hua et al. [75] found that in Nebraska, USA, the greenness of grassland and deciduous forests is most strongly correlated with drought conditions during the preceding 35 and 48 days, respectively. The slightly longer lag effect of drought on tree-dominated ecosystems could help explain the differences in sensitivity detected in our study between herbaceous/shrub and tree cover. More generally, lag effects could also help explain some of the variations in NDVI z-scores in our study that are not explained by current climatic conditions. Assessing vegetation sensitivity to lagged climatic conditions like those in the preceding month could help clarify some of the differences in sensitivities we detected here and provide a more holistic understanding of vegetation sensitivity to climate at different timescales.

We expected that vegetation in low-elevation census tracts could more easily access ground water, potentially buffering vegetation health against drought and dry conditions. In contrast, vegetation at lower elevations was more sensitive to all climate variables (Table 2, Figure S9). This may be because vegetation at lower elevations experiences stress from flooding and soil water inundation. Chicago’s proximity to Lake Michigan and shallow groundwater levels make the city particularly prone to flooding, especially during intense summer rainstorms [76]. Flooding can cause soil saturation and root inundation, negatively affecting plant growth and health through reductions in soil oxygen availability [77]. Vegetation at lower elevations may be stressed due to chronic water inundation, making them more sensitive to fluctuations in climate. Our results suggest that topographical and hydrological characteristics of city landscapes may be important determinants of vegetation responses to climate independent of growth form.

The findings presented in the current study have implications for urban green space planning, especially in the context of climate change. As the temperature increases and heatwaves become more frequent and extreme, urban green infrastructure will provide essential cooling services to communities in cities [31,78,79,80]. The results from this work suggest that urban green spaces comprising trees will be less sensitive (i.e., more resistant) to short-term climate fluctuations than those comprising predominately herbaceous or shrub cover. Trees also provide more cooling services relative to herbaceous vegetation [68]. Thus, urban green spaces that prioritize tree planting may ensure that important ecosystem services like cooling that are provided by healthy vegetation will be maintained during short-term fluctuations in climate like heatwaves or flash droughts [31]. Where flooding is common, planting species that are tolerant of flooding may enhance vegetation resilience. Such species are especially important in cities like Chicago, where precipitation is expected to increase as the climate changes, increasing the severity of flooding [31].

There are, however, many important considerations and priorities when planning urban green spaces that will also need to be addressed as part of the broader planning process. This includes the arrangement of vegetation to improve nighttime cooling, plant diversity and conservation, and green space cultural value [41,81,82,83]. In addition, more work is needed to understand how to best plan urban green spaces so that planted vegetation is balanced in such a way that ecosystem services will be maintained both through short-term fluctuations in climate (e.g., heatwaves or flash droughts) as well as long-term events (e.g., prolonged drought).

4.5. Socio-Demographic Variables Are Not Key Drivers of Vegetation Sensitivity

Socio-demographic variables are not key drivers of spatial variation in vegetation sensitivity among communities in Chicago. Vegetation sensitivity only to drought (SPEI) differed among racial/ethnic communities, where we found that Black and White communities had vegetation that was more sensitive to drought than Latino communities (Figure 7A). We found no support for our prediction that sensitivity is higher among poor relative to wealthy communities (Figure 7B,D). Overall, most of the variance in sensitivity among communities was explained by the vegetation growth form and elevation rather than socio-demographic variables (Table 2). However, the differences we detected in percent cover of vegetation growth form among these communities can still manifest in real and experienced differences in environmental quality. For example, although Black communities did not have higher sensitivity to climate relative to White communities, these communities tend to have higher herbaceous/shrub cover (Figure 5A). This means that a higher proportion of their green space may be more vulnerable to short-term fluctuations in climate.

As has been implicated in previous work by Dong et al. (2023) [23] in Los Angeles, CA, USA, we expected that wealthier communities would have more resources to invest in supplemental irrigation, which could potentially mitigate the effect of climate on vegetation, resulting in lower sensitivity. The contrasting patterns detected in Los Angles relative to the current study are likely explained by the difference in underlying climate between these two cities. Los Angles is in an arid, Mediterranean region where the hottest part of the year also coincides with the driest. In contrast, Chicago’s climate is categorized as a hot summer, humid continental climate, where the peak growing season coincides with the wettest part of the year. Because of this, Chicago residents do not need to rely as heavily on outdoor irrigation during the peak growing season [84].

This study focused on vegetation responses to short-term fluctuations in climate, but assessing how vegetation responds to longer-term disturbances could provide deeper insights into spatial variation in vegetation quality among communities. For example, there may be differences in the length of vegetation recovery time after prolonged drought (i.e., resilience to drought). Remotely sensed data like those used here can be used to assess vegetation resilience and are poised to provide a better understanding of spatial variation in vegetation resilience, an important metric of vegetation quality in cities.

5. Conclusions

Urban green space is a critical nature-based solution for addressing climate change in cities. Beyond the amount of green space, its ecological quality plays a key role in how effectively it mitigates climate impacts. This study is among the first to examine spatial variation in ecological green space quality within a city, revealing that both quality and quantity can differ significantly across urban areas. Understanding this variation is essential for advancing effective and equitable climate adaptation. Further research is needed to identify other factors influencing vegetation sensitivity, particularly those linked to socio-demographic disparities, such as air pollution, soil contamination, or land surface temperature.