Sensing Vegetation Resistance and Recovery Along Urban–Rural Gradients

Abstract

Understanding vegetation-mediated mitigation of urban heat islands (UHI) is essential for sustainable urban adaptation strategies. Although vegetation responses to extreme heat events have been widely explored using satellite remote sensing and statistical methods, evidence remains limited regarding how these responses vary along urban–rural gradients, particularly in terms of resistance and recovery dynamics. This study focuses on the North Tianshan Slope Urban Agglomeration (TNSUA) in Xinjiang, China. Based on Enhanced Vegetation Index (EVI) data from 2000 to 2022, an urban–rural gradient was delineated using impervious surface fraction. Vegetation resistance and recovery during extreme heat events were quantified to reveal spatiotemporal response patterns. Generalized additive models (GAMs) and Random Forest (RF) models were applied to identify key driving factors and to evaluate their relative importance across multiple spatial scales. The results indicate that rural land cover along the gradient provides a strong cooling effect, particularly in areas with an urban development intensity (UDI) of 70–85%. Vegetation responses show pronounced seasonal differences, with urban vegetation generally exhibiting lower resistance and recovery than rural vegetation. At the county scale, local UHI intensity is the dominant driver of vegetation responses, whereas at the pixel scale, precipitation and vapor pressure deficit (VPD) play the most critical roles. Overall, this study improves the understanding of vegetation responses to extreme heat events in arid regions and provides scientific support for nature-based urban heat adaptation strategies.

1. Introduction

The Intergovernmental Panel on Climate Change (IPCC) has stated that continued global warming is projected to lead to more frequent and intense extreme weather events in the future [1]. Extreme weather events impose immediate and severe pressures on ecosystems, resulting in abrupt ecological responses [2,3,4]. A growing body of evidence shows that extreme weather events can cause substantial alterations in terrestrial vegetation at regional to global scales [5,6,7,8]. Terrestrial vegetation not only bears the brunt of extreme heat but also functions as a crucial ecological buffer [9,10]. Investigating the vegetation’s responses to extreme weather events can offer crucial insights for formulating science-based ecological management strategies and promoting sustainable and resilient regional development [11,12].

Urban and rural vegetation is usually managed through planning and green infrastructure; however, urban expansion, heat island effects, and extreme weather events can together influence vegetal growth and further diminish ecosystem services [13,14,15]. The responses of urban and rural vegetation to extreme heat events are various across scales [16]. Rural land cover can effectively alleviate heat stress in urban centers by forming a “cooling belt” [17]. Comparative analyses of urban vegetation show that the cooling capacity of vegetation (CCV) is substantially stronger in humid–hot cities than in arid–hot cities [18,19,20]. However, existing studies often focus either on urban ecosystems or rural landscapes separately, with relatively limited attention paid to the transitional zones along urban–rural gradients, where land use intensity, vegetal structure, and human interventions change rapidly [21]. This limitation creates an important conceptual gap in understanding how vegetation responses to extreme heat vary continuously across urban–rural systems.

Vegetation resistance and recovery are critical functional indicators that reveal how plants respond to extreme heat [22,23]. Vegetation growth differs markedly between urban and rural areas, due to variations in local microclimates and human interventions [24]. Large-scale studies indicate that vegetation responses to extreme heat events are modulated by both temperature intensity and moisture stress, and may exhibit seasonal lag effects [25,26]. Vegetation in arid and semi-arid regions generally exhibits more negative responses [27,28,29]. Wang [24] reported that during extreme heat months, urban vegetation responds more negatively than rural vegetation, with temperature, humidity, vegetation type, and impervious surface fraction being the main factors driving the differences between urban and rural vegetation responses. Back [30] suggested that human interventions, such as irrigation, can enhance the resistance and recovery capacity of urban vegetation. In contrast, Mehmood [26] found that natural vegetation responds more directly to climatic conditions and moisture stress during extreme heat, with its resistance and recovery capacity exhibiting stronger spatial heterogeneity. Hossain [31] noted that under water-abundant conditions, highly stress-tolerant vegetation tends to have lower recovery capacity, whereas under water-limited conditions, less stress-tolerant vegetation may exhibit higher recovery capacity. These studies significantly enhanced the understanding of vegetation responses to extreme weather events. Nevertheless, inconsistencies remain among the existing findings regarding the relative importance of climatic drivers, land use characteristics, and anthropogenic influences in shaping vegetation resistance and recovery under extreme heat conditions. Although existing studies have examined vegetation responses to extreme heat events, most have overlooked the critical spatial dimension of urban–rural gradients. In particular, comparative analyses of vegetation resistance and recovery dynamics under extreme heat conditions in semi-arid regions remain limited.

To address these gaps, this study develops a multi-scale analytical framework based on urban–rural gradients to systematically investigate vegetation response mechanisms under persistent extreme heat in a semi-arid urban agglomeration. By delineating urban ladders (${\mathrm{UL}}_{\mathrm{s}}$) and rural ladders (${\mathrm{RL}}_{\mathrm{s}}$), we characterize variations in development intensity and peri-urban buffer structures along the gradient. A percentile-based method for identifying “hot months” is employed to capture sustained heat stress events rather than isolated temperature extremes. Vegetation resistance and recovery indices are further used to quantify immediate and lagged ecological responses. The study integrates multiple datasets, including satellite-derived vegetation indices, climatic observations, and socio-environmental indicators, to construct a spatially explicit analytical framework. Machine learning approaches and spatial regression techniques are combined to capture nonlinear relationships and spatial heterogeneity in vegetation responses. By integrating nonlinear modeling with spatially explicit analytical approaches, this study reveals seasonal- and scale-dependent variations in the drivers of vegetation response and seeks to address the following research questions:

• Can neighboring rural land cover effectively mitigate the UHI?
• Do urban and rural vegetation show significant differences in terms of their resistance and recovery?
• If present, what environmental drivers explain these differences?

By addressing these questions, this study provides new insights into how urban–rural landscape structures regulate vegetation responses to extreme heat events. The findings contribute to improving urban green infrastructure planning, optimizing ecological buffer zones along urban–rural gradients, and supporting climate adaptation strategies in semi-arid urban regions.

2. Materials and Methods

2.1. Study Area

The study area is the TNSUA (79°53′17″ E–96°23′10″ E, 40°52′31″ N–47°14′10″ N) in Xinjiang, China (Figure 1). It is located along the northern foothills of the Tianshan Mountains and the southern edge of the Junggar Basin and is the most densely populated and highly urbanized region in Xinjiang [32].

It is a semi-arid region with a typical temperate continental climate, characterized by a mean annual temperature of approximately 7 °C and mean annual precipitation generally below 250 mm [33,34]. Grasslands constitute the principal land cover types, accompanied by scattered forest patches along the mountain area. Urban and rural settlements demonstrate a pronouncedly uneven spatial distribution, characterized by a multiple-nuclei urban system in which densely populated areas are predominantly confined to oases. The rapid expansion of built-up urban areas has intensified the UHI effect, and this thermal stress has profoundly altered vegetation’s physiological responses and ecological recovery processes [35].

2.2. Data Sources

This study integrates multi-source datasets spanning from 2000 to 2022, covering Land and Vegetation, Meteorological, Climate Indicators, and Anthropogenic factors. All the spatial datasets were harmonized to a unified spatial resolution and coordinate system (WGS 1984) prior to the analysis. Raster resampling and vector-raster conversion were performed where necessary to ensure spatial consistency. The data sources and corresponding description are provided in

Table 1. Decryption of data and its source.

Data Type	Data Name	Time	Accuracy	Source
Land and Vegetation	EVI Time Series	2000–2022	250 m	Google Earth Engine
	GLC_FCS30D	2000–2022	30 m	OpenLandMap [36]
	DEM	2000	30 m	Google Earth Engine
Meteorological	mean precipitation dataset	2000–2022	1 km	The National Tibetan Plateau Data Center [37,38]
	mean temperature dataset	2000–2022	1 km
	daily air temperature datasets (Tmax)	2003–2012 2013–2022	0.01°	Zenodo [39,40,41,42]
	daily air temperature datasets (Tmin)	2003–2012 2013–2022	0.01°
	Land surface temperature (LST)	2000–2022	1 km	Google Earth Engine
Climate Indicators	Standardized Precipitation–Evapotranspiration Index (SPEI)	2000–2022	0.1°	Zenodo [43]
	VPD	2000–2022	0.1°
Anthropogenic factors	PM2.5	2000–2022	1 km	The National Tibetan Plateau Data Center [44]
	population density	2000–2022	1 km	LandScan
	GDP	2000, 2020	1 km	The Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences [45]

.

2.3. Methods

This study integrates multi-source remote sensing data and statistical models to study vegetation resistance and recovery during “hot month” events across urban–rural gradients in the TNSUA. The workflow is composed of four main parts (Figure 2).

First, we constructed urban–rural gradients for quantifying differences between anthropogenic activities and the natural environment. Urban areas were divided into five ${\mathrm{UL}}_{\mathrm{s}}$ according to the impervious surface percentage, while ${\mathrm{RL}}_{\mathrm{s}}$ were defined as concentric buffer areas extending outward from city boundaries based on the urban area’s size and distance ladders [17]. Subsequently, we calculated the UHI intensity.

Second, we characterized the intensity and persistence of extreme heat events based on “hot months” events. Based on a 1 km monthly mean temperature dataset [38], we calculated the 90th percentile of the monthly mean temperature for each calendar month and used it as the threshold. Consecutive months exceeding this threshold were then classified as one “hot months” event. If consecutive months exceed the threshold, they were considered a single “hot months” event.

Third, vegetation growth dynamics were characterized using deseasonalized 250 m EVI time series. Vegetation resistance was defined as the relative change in EVI during a “hot months” event, reflecting the immediate vegetation response to extreme heat events, whereas vegetation recovery represented the delayed response following the events [46]. These indicators were further aggregated to a 1 km spatial resolution.

Finally, statistical models and machine learning models were combined to quantitatively assess the relative importance of climatic factors, moisture stress conditions, and land cover factors in shaping vegetation resistance and recovery.

2.3.1. Definition of Urban and Rural Ladders

To assess the effects of rural land cover on the mitigation of UHI, we delineated both urban and rural areas into five ${\mathrm{UL}}_{\mathrm{s}}$ [47] and four ${\mathrm{RL}}_{\mathrm{s}}$ [48].

To obtain ${\mathrm{RL}}_{\mathrm{s}}$, we extracted impervious surface pixels based on the 30 m land cover dataset [36]. For each 3 km × 3 km grid [17,49], the ratio of impervious surface area to total area was calculated with

$${\mathrm{UDI}}_{\mathrm{i}}={\displaystyle \frac{{\mathrm{S}}_{\mathrm{IP},\mathrm{i}}}{{\mathrm{S}}_{\mathrm{i}}}}\times 100\%.$$

(1)

${\mathrm{UDI}}_{\mathrm{i}}$ is the urban development intensity, ${\mathrm{S}}_{\mathrm{IP},\mathrm{i}}$ is the impervious surface area, and ${\mathrm{S}}_{\mathrm{i}}$ is the total area. Based on UDI, the grid cells were reclassified into five strata with 15% intervals: ${\mathrm{UL}}_1$ (85–100%), ${\mathrm{UL}}_2$ (70–85%), ${\mathrm{UL}}_3$ (55–70%), ${\mathrm{UL}}_4$ (40–55%), and ${\mathrm{UL}}_5$ (25–40%), following the urban–rural gradients delineation framework proposed by Yang [17].

${\mathrm{RL}}_{\mathrm{s}}$ were defined as concentric buffer areas extending outward from the urban boundary up to 20 km, segmented into four rings of increasing radius: ${\mathrm{RL}}_1$ (2.5–5 km), ${\mathrm{RL}}_2$ (5–10 km), ${\mathrm{RL}}_3$ (10–15 km), and ${\mathrm{RL}}_4$ (15–20 km), consistent with the spatial influence range identified by Yang [17]. The RL radius for each city was normalized with

$$\mathrm{D}={\mathrm{D}}_{\min }+\left({\mathrm{D}}_{\max }-{\mathrm{D}}_{\min }\right)\times {\displaystyle \frac{\mathrm{S}-{\mathrm{S}}_{\min }}{\left({\mathrm{S}}_{\max }-{\mathrm{S}}_{\min }\right)}},$$

(2)

where $\mathrm{D}$ is the radius of RL, $\mathrm{S}$ is the built-up area of the target city, and ${\mathrm{S}}_{\min }$ and ${\mathrm{S}}_{\max }$ are the minimum and maximum built-up areas among all cities.

2.3.2. “Hot Months” Event

“Hot months” are defined relative to historical climatological thresholds, serving as robust indicators of persistent heat extremes and capturing the cumulative intensity and duration of high-temperature stress [50]. We defined the “hot months” using month-specific thresholds. A month was classified as “hot” if its mean temperature exceeded the historical 90th percentile for the same month of the year, a threshold selected following sensitivity tests using the 85th, 90th, and 95th percentiles, and consistent with the findings of Wang [24]. Consecutive hot months were merged into a “hot months “event, which typically lasted for 1–3 months [51].

To characterize vegetation responses to high temperatures across seasons, we defined the seasons: DJF (December to February), MAM (March to May), JJA (June to August), and SON (September to November).

The “hot months” event was assigned to the seasons it overlapped, with the events spanning multiple seasons counted in each. Seasonal variations in vegetation resistance and recovery were quantified as the median values of the ${\mathrm{RL}}_{\mathrm{s}}$ for all events within each season.

2.3.3. Vegetation Resistance and Recovery

To quantify the concurrent and delayed responses of vegetation to extreme heat events, we calculated vegetation resistance and recovery at 250 m. Resistance represents the maximum deviation of the EVI from its pre-event level during a “hot months” event, while recovery reflects the mean deviation of EVI from the pre-event level during the three months immediately following the event.

$$\mathrm{Resistance}={\displaystyle \frac{2\times \left({\mathrm{EVI}}_{\mathrm{in}}-{\mathrm{EVI}}_{\mathrm{pre}}\right)}{\left({\mathrm{EVI}}_{\mathrm{pre}}+\left|{\mathrm{EVI}}_{\mathrm{in}}-{\mathrm{EVI}}_{\mathrm{pre}}\right|\right)}}$$

(3)

$$\mathrm{Recovery}={\displaystyle \frac{2\times \left({\mathrm{EVI}}_{\mathrm{post}}-{\mathrm{EVI}}_{\mathrm{pre}}\right)}{\left({\mathrm{EVI}}_{\mathrm{pre}}+\left|{\mathrm{EVI}}_{\mathrm{post}}-{\mathrm{EVI}}_{\mathrm{pre}}\right|\right)}}$$

(4)

${\mathrm{EVI}}_{\mathrm{pre}}$ was defined as the mean EVI of the same months throughout the study period, excluding the year in which the “hot months” event occurred, thereby representing the typical baseline vegetation condition under non-extreme conditions. Specifically, for a hot month event occurring in a given calendar month, ${\mathrm{EVI}}_{\mathrm{pre}}$ represents the multi-year mean EVI of that same month across all non-event years at each pixel. This month-matched and pixel-wise approach ensures that the reference condition reflects the typical vegetation state for that location and month, thereby minimizing the biases caused by seasonal variability and spatial heterogeneity [52,53]. ${\mathrm{EVI}}_{\mathrm{in}}$ corresponds to the EVI of the month during the “hot months” event that exhibited the greatest deviation from ${\mathrm{EVI}}_{\mathrm{pre}}$. ${\mathrm{EVI}}_{\mathrm{post}}$ was calculated as the mean EVI during the three months immediately following the termination of the “hot months” event. The three-month recovery window was selected because vegetation responses to extreme heat often exhibit short-term lag effects, and a three-month period can capture the typical recovery process while reducing the influence of short-term fluctuations or subsequent seasonal changes [54,55].

2.3.4. Statistical Analysis

To analyze urban–rural differences, we introduced factors such as climate factors, environmental context, and land cover types in rural areas, which are summarized in

Table 2. Possible factors that influence urban–rural differences in resistance and recovery.

Type	Factor Name	Description
Non-land cover factors	city_size_log	Logarithm of the size of the city (km2).
	impervious_frac	Average fraction of impervious area over all the urban pixels of the city (unit: %).
	elevation_diff	Average elevation difference between the urban and rural pixels of the city (unit: m).
	dtmax_in/post_event	Average daytime UHI within each administrative city boundary during (in) or during the three months after (post) the “hot months” event (unit: °C).
	dtmin_in/post_event	Average nighttime UHI within each administrative city boundary during (in) or during the three months after (post) the “hot months” event (unit: °C).
	background_prcp	Climatological mean precipitation for the month of the “hot months” event (unit: mm). Calculated as the average of monthly mean precipitation over the years 2000–2022 excluding years with hot months in the same month and city, spatially averaged over the city area.
	background_tmean	Climatological mean temperature for the month of the “hot months” event (unit: °C). Calculated as the average of monthly mean temperature over the years 2000–2022 excluding years with hot months in the same month and city, spatially averaged over the city area.
	spei_in/post_event	Mean SPEI within each administrative city boundary during (in) or during the three months following (post) the “hot months” event.
	vpd_in/post_event	Mean VPD within each administrative city boundary during (in) or during the three months following (post) the “hot months” event.
	corr_spei_diff	Average difference in the Pearson’s correlation between deseasonalized EVI and SPEI between the urban and the rural pixels, calculated separately for each season of the year.
	corr_vpd_diff	Average difference in Pearson’s correlation between the deseasonalized EVI and vapor pressure deficit between the urban and the rural pixels, calculated separately for each season of the year.
	optimal_tmax_diff	Difference in the average optimal monthly mean maximum temperature for EVI between urban and rural pixels during the growing season (April to October).
	optimal_tmin_diff	Difference in the average optimal monthly mean minimum temperature for EVI between urban and rural pixels during the growing season (April to October).
land cover factors	Crop	Fraction of each land cover type in the rural area of the city.
	Deciduous_forest
	Evergreen_forest
	Grass
	Mixed_forest
	Shrub
	Wetland

, and quantitatively assessing their explanatory power. Land cover factors include the composition of the rural areas, thereby reflecting the background biome context of the city rather than differences in internal urban vegetation types. The inclusion of these factors helps not only to reveal the spatial heterogeneity of urban–rural differences among cities, but also to identify key ecological and environmental factors affecting urban and rural vegetation responses.

To identify the primary factor that lead to urban–rural differences in vegetation resistance and vegetal recovery from “hot months” events, we conducted city-scale analyses using Spearman correlation and GAM. The model performance was evaluated using the coefficient of determination (R²). We implemented leave-one-city-out cross-validation for all single-variable GAMs and the multivariate land cover GAM to evaluate the model’s performance on unseen cities and reduce potential spatial dependence. For individual non-land-cover variables (e.g., VPD, SPEI, and UHI metrics), single-variable GAMs were fitted using B-splines to quantify each predictor’s explanatory power. For the seven rural land cover variables, a multivariate GAM was fitted simultaneously to obtain the overall R², avoiding the interpretation of individual coefficients due to multicollinearity. Partial Spearman correlations were also calculated to assess the independent contributions of each land cover type. To account for the multicollinearity among predictors, we interpret variable importance as the strength of association with the response rather than implying independent causal effects. The Spearman correlation and GAM were formulated as follows:

$$\mathsf{\rho }=1-{\displaystyle \frac{6\sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{d}}_{\mathrm{i}}^2}{\mathrm{n}\left({\mathrm{n}}^2-1\right)}}$$

(5)

$\mathsf{\rho }$ is the Spearman rank correlation coefficient, n is the number of observations, and ${\mathrm{d}}_{\mathrm{i}}$ denotes the difference between the ranks of the two variables for the i-th observation.

$$\mathrm{g}(\mathrm{E}(\mathrm{Y}))={\mathsf{\beta }}_0+\sum _{\mathrm{j}=1}^{\mathrm{p}}{\mathrm{f}}_{\mathrm{j}}\left({\mathrm{X}}_{\mathrm{j}}\right)$$

(6)

$\mathrm{Y}$ represents the response variable, g( ) is the link function, ${\mathsf{\beta }}_0$ is the intercept, ${\mathrm{X}}_{\mathrm{j}}$ denotes the j-th explanatory variable, and ${\mathrm{f}}_{\mathrm{j}}$( ) is a smooth function describing the potentially nonlinear effect of ${\mathrm{X}}_{\mathrm{j}}$ on Y.

This study used the geographically weighted regression (GWR) model to analyze the spatial effects of natural and human factors on temperature. The GWR model was implemented in ArcMap 10.8, using a fixed kernel type with the optimal bandwidth determined by the corrected Akaike Information Criterion. The dependent variable was the annual mean air temperature at the city level. Eight explanatory variables were included in the GWR model, representing both natural and anthropogenic influences, including impervious fraction, precipitation, DEM, EVI, SPEI, GDP, population density, and PM2.5 concentration. At the pixel scale, RF combined with SHAP (SHapley Additive exPlanations) were employed to quantify the relative importance of each factor and identify potential interaction effects, ensuring statistical robustness while capturing local spatial heterogeneity.

3. Results

3.1. Effects of Rural Land Cover Composition on Mitigating UHI

${\mathrm{RL}}_4$ exhibited the most pronounced cooling effect on ${\mathrm{UL}}_2$ (Figure 3). Land cover patterns exerted significant explanatory power over UHI intensity across ${\mathrm{UL}}_{\mathrm{s}}$, with neighboring rural land cover (NRLC) providing cooling effects for all ${\mathrm{UL}}_{\mathrm{s}}$. Considering specific land cover types, woodland showed the peak explanatory power for UHI within the ${\mathrm{UL}}_4$ layer, but its effect on the urban core was relatively weak, highlighting the spatial limitation of woodland-based UHI mitigation. In contrast, impervious surfaces exhibit a broader adaptation range across urban areas, affecting multiple layers; however, their ecological function is also significantly constrained by distance ladders, preventing effective regulation in areas immediately adjacent to the city. The ability of different rural land cover types to regulate UHI depends on specific combinations of distance ladders and urban areas.

3.2. Vegetation Resistance and Recovery: Urban vs. Rural

3.2.1. Spatiotemporal Patterns of Resistance and Recovery

This study further revealed the actual responses and differences in vegetation across different cities, showing that urban vegetation generally exhibits more negative resistance and recovery compared to rural vegetation (Figure 4). Vegetation resistance was predominantly negative in both urban and rural areas across the four seasons in the TNSUA.

Urban vegetation exhibited the highest positive resistance in summer, while rural vegetation peaked in winter and spring (Figure 4a–h). Increased positive resistance was observed in the western region than in the eastern region (Figure 4a–h). In the three months following the high-temperature months, rural vegetation mainly exhibited its positive recovery in winter, whereas urban vegetation only showed weak positive recovery in summer, with negative recovery dominating the other seasons (Figure 4i–p). The largest magnitude of negative recovery occurred in spring (Figure 4i–p).

Regarding the disparity between urban and rural vegetation, the differences were the most pronounced in winter, with rural areas exhibiting significantly higher resistance and, especially, recovery than urban areas at a significance level of p < 0.05. Seasonally, urban vegetation resistance and recovery peaked in summer, whereas rural vegetation showed a gradual increase in resistance and recovery from summer to winter, reaching its maximum in winter. Spatially, the western and central cities displayed the most pronounced advantages in resistance and recovery.

The “hot months” in spring and summer had the greatest impact on vegetation across the TNSUA. Urban vegetation was more vulnerable than rural vegetation and exhibited more negative responses, highlighting the regulatory role of regional differentiation on vegetation growth under “hot months” events.

Rural vegetation exhibits significant temporal trends in relation to both resistance and recovery. The regulatory effect of neighboring rural land cover on UHI intensity is not only constrained by spatial patterns but also shows significant temporal dynamics. In addition to spatial differences, rural vegetation responses to “hot months” events also change seasonally. Therefore, this study further explores the temporal dimension by analyzing the median values of vegetation resistance and recovery corresponding to each rural ladder within the study area. The results show that during the spring and summer growing seasons, resistance sharply increases, whereas in autumn and winter, it steadily decreases. Recovery continuously increases in spring, decreases initially and then stabilizes in summer and winter, and remains steady with increasing distance, while in autumn recovery decreases, indicating a weakening of the distance-dependent effect. At the regional scale (Figure 5), rural ecosystem recovery shows clear seasonal and ladder variations: it generally increases in spring but decreases in autumn and winter. This seasonal variation suggests that climate conditions and the vegetation growth cycle jointly drive the temporal responses of rural ecosystems under extreme heat, highlighting the seasonal sensitivity of rural vegetation to external disturbances. ${\mathrm{RL}}_4$ consistently exhibits a higher resistance and recovery compared to the other three rural ladders, indicating that vegetation in rural fringe areas demonstrates stronger recovery to “hot months” events.

3.2.2. Seasonal and Regional Urban–Rural Response Differences

The essence of the urban–rural differences in vegetation responses lies in the variation in the vegetation’s growth conditions. To investigate the causes of the urban–rural differences in vegetation resistance and recovery, we first compared vegetation’s growth status under normal conditions for different land cover types in both urban and rural areas. As shown in Figure 6, except for dryland cropland and forest, the median EVI for all other land cover types in urban areas was lower than that for the same types in rural areas. This pattern indicates that the urban environment exerted a persistent inhibitory effect on vegetation’s growth, resulting in urban vegetation having a relatively disadvantaged “growth baseline” even during non-stress periods. This inherent growth disadvantage may have been one of the fundamental reasons for its weaker resistance and slower recovery when facing “hot months” events.

Seasonally, in summer and autumn, the fraction of positive resistance in urban areas was similar to that in rural areas. However, in winter and spring, the fraction of positive resistance in urban areas was significantly lower than that in rural areas (Figure 7a). Regardless of the sign of resistance, the magnitude of responses in urban areas during winter and spring was consistently smaller than in rural areas (Figure 7b–d). This suggests that during “hot months” events, the urban environment further weakened the vegetation’s ability to cope with high temperatures.

When resistance and recovery were negative, the negative response magnitude in urban areas during summer was significantly higher than that in rural areas (Figure 7d). In other words, urban vegetation experienced more intense stress than rural vegetation when “hot months” events occurred during the cold season (Figure 7c,d).

Finally, the difference in the fraction of positive responses between urban and rural areas peaked in autumn, while in other seasons, it was below zero, indicating that rural vegetation had a higher fraction of positive responses (Figure 7a). The urban–rural difference in vegetation responses decreased from summer to winter (Figure 7d). Therefore, the overall resistance and recovery of urban vegetation were more negative than those of rural vegetation.

3.3. Driving Factors of Urban–Rural Vegetation Response Differences: Multi-Scale Analysis

To explore the driving mechanisms behind urban–rural vegetation response differences, this study integrated correlation analysis, GAM, and RF methods to assess the impact of various potential factors at different scales.

3.3.1. Identification of Key Driving Factors

Through the correlation analysis of urban–rural vegetation response differences (Figure 8a–d), this study preliminarily identified the key influencing factors. The analysis revealed that UHI intensity is a key climactic factor that dominates the sign of the difference: it shows a positive correlation with both resistance and recovery differences in summer, while in winter the correlation with the recovery difference becomes negative. At the same time, moisture stress (SPEI and VPD) significantly modulates the strength of the recovery, showing a negative correlation with the difference direction in both summer and winter.

At the landscape level, land cover type factors exhibited the strongest linear explanatory power (R²: 20–60%), suggesting that land cover might be a key factor influencing the difference patterns. In terms of the magnitude of the difference, the driving factors showed stronger seasonal consistency: urban development intensity (urban size and impervious surface fraction) was negatively correlated with the response magnitude in spring, while rural landscape composition and terrain features (elevation difference) showed seasonal reversal effects. In addition, regional climate backgrounds (precipitation and temperature) and vegetation moisture stress conditions also played important roles in modulating the magnitude of the differences.

To identify the potential nonlinear effects between factors and quantify their independent contributions, nonlinear models were further applied (Figure 8e–h). The results confirmed the previous conclusions: although the simple linear correlation between land cover factors and difference magnitude was weak, the nonlinear models showed a high explanatory variance (R² significantly increased). This indicates that the effect of land use on the difference magnitude mainly operates through nonlinear mechanisms, and its complex effects cannot be captured by linear relationships. This analysis further clarified the existence of specific nonlinear relationships between factors such as UHI intensity and response differences.

The results show significant spatial variation in relation to the impacts of SPEI, EVI, and impervious fraction (Figure 9). In 2000, impervious surface expansion in the northwest had a strong positive effect on temperature, but by 2020, its influence had become negative, suggesting that suitable policies could enhance urban vegetation’s recovery to extreme heat. The SPEI’s coefficients transitioned from slightly negative in 2000 to slightly negative in the west by 2020, indicating that intensified droughts raised the temperatures. The EVI’s positive coefficients were concentrated in the north in 2000, while in 2020, negative values dominated. Natural factors, especially moisture stress, showed spatial heterogeneity, with precipitation decreasing eastward and the SPEI increasing. Human factors, linked to urbanization, showed high values near urban areas, expanding outward over time. The GWR model showed strong explanatory power, with local R² values ranging from 0.097 to 0.833 in 2000 and from 0.220 to 0.876 in 2020, with most cities exhibiting values above 0.7. This indicates that the selected natural and anthropogenic variables effectively explain the spatial variability of temperature patterns, and highlights that the strong spatial heterogeneity captured by the GWR model.

3.3.2. Nonlinear Effects and Spatial Heterogeneity at Pixel Scale

Figure 10 shows that during and in the three months following a “hot months” event, SPEI and VPD were the most influential predictive factors, indicating that moisture stress is a key driver of urban–rural vegetation response differences. Additionally, the optimal maximum temperature (“optimal_tmax”) and optimal minimum temperature (“optimal_tmin”) also had strong positive effects. Regarding vegetation types, forests and grasslands were more sensitive to meteorological factors. The SPEI during the event and after the event showed significant positive contributions to the resistance and recovery of forests and grasslands in autumn and winter, while the VPD in the event and after the event contributed significantly to the recovery magnitude in spring. These findings suggest that under the same moisture stress conditions, different land cover can lead to different urban–rural vegetation response patterns through variations in vegetal structure and function, highlighting the key role of surface properties in regulating the ecosystem’s recovery. Both “optimal_tmin” and “optimal_tmax” showed significant positive effects on recovery across all vegetation types in autumn. The spatial distributions of all factors are presented in Figure 11.

4. Discussion

4.1. Rural Land Cover Effects on UHI Mitigation

Natural land cover and relatively simple functional patterns in rural areas have significant potential to mitigate UHI effects. This study aims to explore the quantitative and qualitative impacts of rural land cover on mitigating UHI effects in the TNSUA from 2000 to 2022. Furthermore, cities are categorized based on their UDI to conduct a differentiated analysis of the impact of rural land cover on UHI mitigation across varying urbanization intensities. By doing so, we effectively constrain and quantify the influence of urban development via the research findings. The results show that NRLC can mitigate the UHI effect across the entire urban area. Specifically, the study found that NRLC significantly cools urban areas, with the most pronounced effect in areas with a UDI of 70–85%. The ${\mathrm{RL}}_4$ land cover type was found to have the most significant impact on UHI effects, similarly to the finding from Yang [17]. From a planning and management perspective, the results highlight the importance of rural ecosystems in mitigating urban thermal stress at the urban agglomeration scale. In particular, rural land cover types with higher ecological stability should be considered in climate-adaptive spatial planning. Protecting and optimizing peri-urban ecological spaces may enhance the overall resilience of regional ecosystems to extreme heat events.

4.2. Seasonal Variations in Urban–Rural Vegetation Response Differences

This study found that high-temperature months generally promoted negative responses in vegetation, with urban vegetation responding more promptly to “hot months” events compared to rural vegetation, which exhibited delayed responses (Figure 4). Thus, the results indicate that high-temperature months may promote vegetation growth in the TNSUA during cold seasons but primarily inhibit growth in warmer seasons. Eastern cities in the urban agglomeration generally exhibited more negative responses than those in the central and western regions, where built-up urban areas were larger, aligning with the UHI effect. Despite stronger heat island effects in the western and central regions, the regression results indicated this was due to the lower vegetation moisture stress and higher optimal growth temperatures in those areas. The cities in the western and central regions are located in the Ili River Valley and Oasis areas, where vegetation moisture stress is lower, and optimal growth temperatures are higher, allowing urban vegetation to continue growing at higher temperatures through stronger transpiration [56].

4.3. Factors Influencing Urban–Rural Vegetation Response Differences

The local UHI intensity is an important factor in explaining differences in vegetation’s response signs (e.g., “dtmax_in_event”, “dtmax_post_event”, “dtmin_in_event”, and “dtmin_post_event” in Figure 8). To explore the driving factors influencing temperature, this study used the GWR model to analyze natural and human factors. It was found that EVI, SPEI, and impervious surface fraction were key temperature-regulating factors, consistent with the results observed in Figure 8 and Figure 10, suggesting mutual interactions and constraints between temperature, vegetation, and moisture stress. Among the different land cover types, moisture stress was crucial in influencing urban–rural differences (e.g., corr_vpd and corr_spei in Figure 10). In urban-scale and pixel-scale regression analyses, predictive factors related to temperature were important for both the signs and magnitudes of the responses, though the specific predictive factors varied. In urban-level regressions, factors such as seasonal climate characteristics (background_pre), urban size, impervious surface fraction, and rural land cover types were significant (Figure 8). In pixel-scale regressions, the most important temperature-related predictive factors were urban–rural optimal temperature differences (optimal_tmax_diff and optimal_tmin_diff) (Figure 10). Taken together, these results suggest that temperature and moisture stress are the main drivers influencing vegetation resistance and recovery. This finding is consistent with the results reported by Wang [24] and Sui [57], who also highlighted the combined roles of temperature conditions and water availability in regulating vegetation responses to extreme heat events.

4.4. Limitations and Future Directions

This study has some limitations. First, the study used a traditional 3-month seasonal window, which may obscure the exact transition time between positive and negative recovery. Methods based on vegetation phenology or local temperature seasonality may yield more consistent spatial patterns. Second, the datasets used in this study mainly relied on a single data source, lacking cross-validation or uncertainty assessments using multi-source data. Future studies should combine multi-source climate data, different vegetation index products, and high-resolution urban data to systematically assess the robustness of the results and improve the reliability of the conclusions. Third, the EVI data used in this study were derived from MODIS products, publicly available from only 2000 onward. This limited the length of the time series, making it difficult to fully account for long-term climate change trends. Future studies could use longer multi-source remote sensing or climate reanalysis data to better separate the effects of long-term warming and extreme events on vegetation responses. In addition, this study focused on environmental influences on vegetation at the regional scale. The study area was classified into two main categories, urban and rural, to maintain spatial consistency between the climate and vegetation data. We acknowledge that differences in land use and vegetation composition may affect the responses, and future research should incorporate higher-resolution land cover and vegetation data to better control for heterogeneity and to refine the urban–rural comparisons. Finally, we used a Random Forest model with SHAP for feature attribution. Although Random Forest is relatively robust to multicollinearity for prediction, multicollinearity among the predictors (e.g., temperature, and moisture stress) can still influence SHAP interpretation. Therefore, our SHAP analysis aims to reveal the overall contribution patterns rather than the independent causal effects of individual predictors.

Previous studies have shown that vegetation responses to extreme heat events are closely related to multiple ecological mechanisms. High temperatures not only directly suppress photosynthetic efficiency and stomatal conductance, but also increase VPD, thereby intensifying moisture stress in ecosystems [58]. VPD is an important driver affecting vegetation water and carbon fluxes, and its influence is expected to increase further under climate warming [59]. The resulting soil moisture deficit further constrains carbon uptake processes and reduces vegetation resistance during heat events [60]. In many cases, extreme heat events also occur simultaneously with drought conditions, forming compound heat–drought events that greatly reduce the ecosystem’s productivity and prolong vegetation’s recovery time [61]. Vegetation recovery rates are mainly controlled by water replenishment, with stronger droughts generally leading to longer recovery times [62]. However, vegetation in arid regions can enhance its survival under long-term heat and water stress through adaptive strategies such as deeper root systems, a reduced leaf area, and improved water use efficiency [63]. Future research could further investigate the resistance and recovery of different vegetation types under compound climate events based on vegetation’s characteristics.

Although this study focuses on a single semi-arid region, its analytical workflow, including urban–rural gradients’ delineation, hot month identification, and vegetation resilience assessment, can be transferred to other climatic and geographic contexts. The findings also offer practical insights for urban planning and climate adaptation, such as optimizing green infrastructure placement along urban–rural gradients, selecting heat-tolerant vegetation species based on resistance and recovery patterns, and strengthening ecological buffer zones to enhance urban resilience to extreme heat events.

5. Conclusions

This study systematically examined vegetation resistance and recovery responses during persistent “hot months” within an urban–rural gradient framework in a semi-arid urban agglomeration. The results indicate that rural areas play an important ecological buffering role in mitigating the thermal stress associated with the UHI. Meanwhile, vegetation responses exhibit clear seasonal and spatial scale-dependent differences.

The main findings are summarized as follows. First, rural areas play an important ecological buffering role in mitigating UHI. Along urban–rural gradients, vegetation in rural areas shows higher stability under persistent heat conditions compared with in urban areas. This suggests that rural ecosystems play an important role in regulating regional thermal environments and alleviating UHI impacts. Second, ${\mathrm{RL}}_4$-type rural land cover exhibits the most significant buffering effect in relation to maintaining vegetation stability. The results show that ${\mathrm{RL}}_4$ areas demonstrate higher vegetation resistance and recovery during extreme heat periods, indicating that this type of rural land cover plays a key role in maintaining regional ecological stability and regulating thermal environments. Third, the coupling effects of temperature and moisture stress represent an important mechanism influencing vegetation stability. Under persistent heat conditions, moisture stress plays a critical role in regulating vegetation responses, suggesting that the interaction between temperature and moisture stress significantly influences vegetation resistance and recovery. Finally, vegetation responses show clear seasonal and scale-dependent characteristics. Across different seasons and urban–rural gradient levels, vegetation stability and its driving factors vary, indicating that both urbanization intensity and climatic conditions jointly shape vegetation responses to extreme heat events. From a planning and management perspective, the results highlight the importance of rural ecosystems in mitigating urban thermal stress at the urban agglomeration scale. In particular, rural land cover types with higher ecological stability should be considered in climate-adaptive spatial planning. Protecting and optimizing peri-urban ecological spaces may enhance the overall resilience of regional ecosystems to extreme heat events.