Lagged Responses of Vegetation Growth to Hydrometeorological Drivers Across Complex Terrain in Southwest China

Abstract

Vegetation is an important component of ecosystems and plays an important role in carbon balance, water balance, and energy conversion. The spatial and temporal changes in the normalized difference vegetation index (NDVI), water resources, and hydrometeorological factors in southwest China between 2003 and 2020 were investigated using multisource remote sensing data. Correlation analyses were performed to assess the correlation among NDVI, water resource changes, and hydrometeorological factors with different time lags. A stepwise regression model with different lag times was constructed to clarify the effects of four topographic factors and eight climatic factors on NDVI, and the following conclusions were obtained: (1) NDVI increased from 2003 to 2020, and the increase became obvious after 2012. (2) NDVI was considerably affected by alterations in the soil water content caused by natural changes. The correlation of NDVI with evapotranspiration and precipitation was high, followed by NDVI’s correlation with surface temperature. The spatial distribution of the positive correlation between NDVI and evapotranspiration and NDVI and precipitation was relatively consistent, and a positive correlation was observed in most parts of Southwest China. (3) The hydrometeorological factors mainly affected NDVI with a lag of 0–1 month, and the correlation was high in western Sichuan and most of Yunnan. In Yunnan, Available Water Capacity (AWC) affected NDVI with a lag of 0–2 months; the lag was 0–1 month in western Yunnan and 1–2 months in eastern Yunnan. (4) In terms of different vertical heights, the NDVI in the regions with altitudes higher than 3000 m was affected by climate change, especially evapotranspiration and precipitation. (5) Digital Elevation Model (DEM), Latitude (Lat), Evapotranspiration (ET), Precipitation (PRCP), Land Surface Temperature (LST), and NDVI were closely related in the construction of stepwise regression models with different lag times.

1. Introduction

Vegetation, a major component of the Earth’s terrestrial ecosystems, is directly connected to the material cycle and energy flow among the soil, hydrosphere, and atmosphere, and it serves as an indicator in research on global change [1]. Vegetation growth in complex mountains is a complex process influenced by multiple factors. Climatic conditions, topographical factors, soil conditions, anthropogenic activities, global climate change, biological factors, ecosystem dynamics, and natural disasters all influence the growth of mountain vegetation to varying degrees. These factors interact and together determine the distribution, structure, and diversity of plant communities [2]. Climatic conditions mainly include temperature, precipitation, and evapotranspiration, among which temperature is one of the important climatic factors affecting vegetation growth. At lower altitudes, where temperatures are relatively high, vegetation growth may be more limited by moisture. Whereas at higher altitudes, lower temperatures become the main limiting factor for vegetation growth [3,4,5]. The seasonal distribution and intensity of precipitation also affect vegetation growth. The influence of topographic factors mainly includes elevation and topographic relief, with the level of elevation directly affecting the growth of vegetation. Changes in altitude lead to changes in environmental factors such as temperature, barometric pressure, and light, which in turn affect the growth of vegetation. The complexity of plant communities is higher in mountainous areas with high topographic relief [6]. Human activities mainly include the impact of human activities on land use, and changes in the distribution of water resources will affect the growth of vegetation [7,8]. Studies have shown that temperature and precipitation control vegetation greening in humid and arid regions, respectively [4,9,10].

To assess the correlation between various hydrometeorological factors and vegetation growth, previous studies have employed methods such as Pearson correlation analysis, partial correlation analysis, and multiple correlation analysis. Additionally, previous studies have used regression models to quantify the extent to which hydrometeorological factors influence changes in vegetation growth. Wang et al. used the correlation analysis method in the Yellow River Basin to show that effective precipitation is a limiting factor for terrestrial ecosystem change [11]. Sidi Almouctar et al. analyzed the changes in vegetation cover in the Korama Basin by using the least squares method to match the slopes of the variance trend of NDVI from 2000 to 2018 [12]. Satellite remote sensing technology, with its advantages of high spatial and temporal resolutions, wide coverage, and large amount of information, has become the first choice for sustained, large-scale, dynamic detection of vegetation conditions. Piedallu et al. modeled 20 environmental predictors using multi-source remote sensing data; the results showed that average temperature, rainfall, and soil moisture all have a strong influence on NDVI [13]. Lai et al. analyzed the relative contributions of climate change to the interannual variability of NDVI using multi-source remote sensing data [14]. Yang et al. also used multi-source remote sensing data to distinguish the characteristics of vegetation changes under the influence of climate change [15].

So the influence of terrain undulation degree on vegetation cannot be ignored. Currently, there are fewer studies on vegetation changes at different elevations, so the paper investigated the effects of both elevation and meteorological and hydrological factors on vegetation growth. Conventional land surface models, such as the Global Land Data Assimilation System (GLDAS), primarily simulate vertical soil moisture dynamics but often fail to capture the complex lateral flows and deep groundwater storage typical of karst terrains. To address this limitation, this study utilizes Available Water Change (AWC) [16] to characterize the actual water resources accessible to the ecosystem. Derived from the residual between Gravity Recovery and Climate Experiment (GRACE) observations and GLDAS simulations, AWC serves as a proxy for deep groundwater and surface water components that are not explicitly represented in standard models. This metric allows us to isolate the specific water reservoirs sustaining vegetation during dry spells in this ecologically fragile region, providing a more accurate representation of the hydrological constraints on vegetation dynamics. At the same time, current research on vegetation dynamics in Southwest China focuses on analyzing the spatial and temporal characteristics of vegetation, and analyses of the driving factors concentrate on meteorological factors. The relationship between vegetation and climate factors, such as topography, hydrology, and surface cover, needs to be further investigated.

Southwest China is a typical karst area in China with a fragile ecological environment. Therefore, studying the changes in vegetation and the relationship between vegetation and various factors in Southwest China is crucial. Several scholars have analyzed the effect of meteorological conditions on NDVI [6,17,18,19]. Other studies have examined long-term land-use changes in southwestern China, as well as the effects of these changes on soil erosion and water resources [20,21]. To address the NDVI changes and influencing factors in Southwest China under climate change, this study investigated the spatial and temporal changes in NDVI, equivalent water thickness change (EWTC), terrestrial water storage change (TWSC), and AWC in Southwest China between 2003 and 2020 by using multisource remote sensing data and the method of correlation analysis. The correlation among NDVI, EWTC, TWSC, and AWC was analyzed. The correlation between hydrometeorological factors and NDVI and the correlation between human activities and NDVI changes were assessed at different time lags, and the effects of hydrometeorological factors and human activities on NDVI under different digital elevation models (DEMs) were explored to identify the factors that influence the vertical changes in vegetation in the complex terrain area of Southwest China. The research findings provide empirical support for understanding vegetation growth patterns in complex terrain and offer a theoretical framework for the sustainable development of ecologically fragile regions.

2. Materials and Methods

2.1. Study Area

The main research area of the study is located in Southwest China (Figure 1), including Sichuan and Chongqing, and the Guizhou region. Southwest China has large altitude differences and fragile climatic conditions at high altitudes. The Sichuan Basin in this region has an elevation of approximately 500 m, while the Yunnan Plateau and Guizhou Plateau have elevations of 2000 m and 1000 m, respectively. The western Sichuan Plateau, meanwhile, generally lies at elevations above 3500 m. In terms of elevation gradients, the study area exhibits distinct vertical vegetation zonation. Areas below 800 m consist of subtropical evergreen broadleaf forests, many of which have been degraded into secondary coniferous forests and bamboo groves. The 800–1800 m elevation zone transitions into mixed evergreen-deciduous broadleaf forests and mixed coniferous-broadleaf forests. The 1800–3500 m elevation zone consists of subalpine dark coniferous forests. Areas above 3500 m are characterized by shrubland, meadows, and scree vegetation. This vertical zonation is complete, and the elevation gradient and water and heat conditions are the primary drivers of the aforementioned differentiation.

Long-term excessive human intervention, soil erosion, degradation of surface vegetation, and rocky desertification are serious in southwest karst areas and restrict the socio-economic development of Southwest China [22]. Vegetation restoration is an effective means to control soil erosion and restore the ecological environment of rocky desertification in Southwest China [23]. The Southwest China region is located between 96–11° E and 21–35° N, spanning China’s first, second and third terraces. The southwest region is the upper reaches of the Yangtze River, the Yellow River and many other rivers, and has a complex topographic structure with large differences in elevation, covering a wide range of landform types such as plateaus, plains, mountains and hills.

The region is characterized by a subtropical monsoon climate, featuring abundant precipitation and distinct dry and wet seasons. The complex topography, ranging from towering mountains to deep valleys, has fostered a diverse vegetation landscape that includes broadleaf forests, coniferous forests, shrublands, and grasslands. Notably, the fragile karst ecosystem in this area is highly sensitive to climate change.

2.2. Materials

2.2.1. Moderate-Resolution Imaging Spectroradiometer

The Moderate Resolution Imaging Spectroradiometer (MODIS) data used in this study included an NDVI product with a time resolution of 16 days (MOD13C2) and an eight-day surface temperature dataset (MOD11A2) from 2003 to 2020 (

Table 1. Variables and data sources.

Source	Variable	Abbr.	Temporal Resolution	Spatial Resolution
	Latitude	Lat
	Longitude	Lon
	Digital elevation model	DEM		90 m
	Slope	Slope
MODIS	Normalized Difference Vegetation Index	NDVI	16-day	1 km × 1 km
	land surface temperature	LST	8-day	1 km × 1 km
GRACE	Changes in equivalent water thickness	EWTC	Monthly	1° × 1°
	Changes in available water	AWC	Monthly	1° × 1°
GLDAS	Precipitation	PRCP	Monthly	0.25° × 0.25°
	Evapotranspiration	ET
	Surface runoff	Qs
	Subsurface runoff	Qsb
	Changes in terrestrial water storage	TWSC

). These data have undergone preprocessing procedures, such as radiometric calibration, to enhance their sensitivity to vegetation.

2.2.2. GRACE

GRACE is the most widely used satellite for evaluating global water storage changes [15,24]. The data provided by GRACE can convert the water quality changes caused by the Earth’s surface gravity into equivalent water thickness forms of terrestrial water reserves. Three processing centers, namely, the German Research Center, the Center for Space Research at the University of Texas, and the Jet Propulsion Laboratory, provide three datasets each. In this study, the monthly EWTCs were inverted using the provided GRACE Level-2 RL06 data from 2003 to 2020, and the average of the three datasets was calculated. A DDK5 filter with a radius of 160 km was used to eliminate north–south stripe errors and smooth them into the data to be used with a spatial resolution of 1° × 1°.

2.2.3. GLDAS

GLDAS is a NASA land surface modeling system developed in the Hydrological Sciences Laboratory of NASA Goddard. It integrates global satellite-based measurements to facilitate advanced simulations of climate and hydrological surveys. Data assimilation is utilized to incorporate satellite and observational data into advanced surface models, including Noah, Variable Infiltration Capacity, Mosaic, and Common Land Model, to provide surface state and flux [16].

On the basis of the monthly precipitation (PRCP), evapotranspiration (ET), and surface and groundwater runoff (Q_s and Q_sb, respectively) products provided by GLDAS and with the corresponding data from 2003 to 2020, TWSCs are estimated using water balance theory. The formula is as follows:

TWSC = P – ET – Q_s – Q_sb.

(1)

The surface water and groundwater reserves are not provided by the GLDAS model [25], so the TWSC based on GLDAS simulation ignores the changes in surface and groundwater. In the ecologically sensitive and fragile southwestern region of China, surface water and shallow groundwater have a considerable effect on the changes in local water resources. Therefore, human activities obtained with TWSC data simulated based on water balance theory and those derived with EWTC data inverted based on GRACE data exhibit differences, resulting in variations in the data simulation results. To some extent, these differences can be due to the variation in groundwater, which is consistent with the observed data on groundwater variation [26]. The differences between ETWC and TWSC show good agreement with observed changes in groundwater levels within the basin, and numerous studies have used these metrics to characterize changes in available water in the study area [27,28,29]. Therefore, the difference between EWTC and TWSC can be used as an indicator of AWC for analysis. The formula is as follows:

AWC = EWTC − TWSC.

(2)

2.3. Methods

Multisource remote sensing data from GRACE, GLDAS, and MODIS were employed to study the effects of hydrological and meteorological factors and human activities on vegetation changes in the southwestern region under different DEMs (Figure 2). The detailed information of these multisource remote sensing data is shown in Table 1. The article investigates the vegetation change from 2003 to 2020, in which NDVI and LST data are from MODIS, with a spatial resolution of 1 km × 1 km, a temporal resolution of 16 days for NDVI and 8 days for LST. EWTC data are from GRACE, with a spatial resolution of 1° × 1°, and a temporal resolution of monthly. The data of PRCP, ET, Q_s, Q_sb, and TWSC are from GLDAS, with a spatial resolution of 0.25° × 0.25°, and a temporal resolution of monthly. ET, Q_s, Q_sb, and TWSC data are from GLDAS, with a spatial resolution of 0.25° × 0.25° and a temporal resolution of monthly. The resolution of DEM is 90 m. The above data are resampled to a spatial resolution of 0.25° × 0.25°and a temporal resolution of monthly. To align the spatial resolutions of the multi-source datasets, this study standardized the data to a resolution of 0.25° × 0.25° with a monthly temporal resolution. Since GRACE has a resolution of 1° × 1°, resampling will introduce some uncertainty into the analysis, primarily due to a potential smoothing effect on extreme values. MODIS has a resolution of 1 km × 1 km, which is resampled to 0.25° × 0.25°. This process introduces significant spatial sampling uncertainties due to mixed-pixel effects, sampling bias caused by cloud cover, and error amplification from complex terrain, thereby smoothing out local-scale surface heterogeneity and extreme ecological events.

Although the AWC index used in this study can reflect water resource changes at the regional scale, it still has certain limitations. First, changes in equivalent water thickness (EWTC) derived from GRACE gravity satellite data may, after filtering, smooth out the rapid dynamic responses of groundwater in karst regions. Second, as a residual measure, AWC combines signals from deep karst aquifers, surface reservoirs, and human extraction activities. The physical mechanisms underlying AWC are complex, making it difficult to directly characterize ecological availability in the same way as soil water. Furthermore, the GLDAS model has limitations in simulating lateral flow under complex karst topography, which may result in model structural errors being incorporated into the AWC residuals.

It is worth noting that due to the limited operational lifespan of the GRACE satellites, their standard monthly gravity field products are typically available only through the end of 2016. The follow-on mission, GRACE-FO, was launched in May 2018 to continue the observations. Since there is a data discontinuity between the GRACE and GRACE-FO missions, to avoid introducing errors due to data splicing discrepancies, the TWSC, EWTC, and AWC datasets used in this study are uniformly restricted to the period covered by the original GRACE mission, ending in 2016.

2.3.1. Correlation Analysis of Hydrometeorological Factors and NDVI

This study used Pearson’s correlation coefficient (r) to analyze the correlation between vegetation and hydrological and meteorological factors, namely, EWTC, TWSC, and AWC, and tested the correlation via a two-tailed Student’s t-test. All calculations were performed in ArcGIS 10.4, Python 3.9, and SPSS 22.0. The monthly AWC, EWTC, and TWSC from 2003 to 2020 were matched with the hydrological and meteorological factors with different lag times, and linear interpolation was used to fill in the outliers and missing values in the data to ensure the accuracy of the analysis.

Vegetation growth typically exhibits a delayed response to antecedent climatic and hydrological conditions, with optimal correlations often occurring at lags of 0–2 months in arid and semi-arid regions [27,28,29]. Therefore, Pearson correlation coefficients between NDVI and each hydroclimatic variable were calculated at lags of 0, 1, and 2 months to capture the most representative coupling.

2.3.2. Building a Multiple Regression Model

Stepwise multiple linear regression models were constructed based on four topographic factors and eight climate factors with different lag scales. Considering the problem of inconsistency in the magnitude of the independent variables, these data were standardized with a Z-score. Four of the topographic factors are longitude, latitude, elevation, and slope. The eight climate factors are LST, EWTC, TWSC, AWC, PRCP, ET, Q_s, and Q_sb. The time lag scale mainly analyzes results from 0–2 months, e.g., LST_1 represents the LST value with a one-month lag.

Using Significance value, determine the significance of the statistical results and apply VIF (Variance Inflation Factor) to detect multicollinearity problems in the data. The model with the best prediction accuracy is selected in terms of meeting the requirements of Significance value and VIF. Model accuracy was tested using the determination coefficient ($R^2$) and Standard Error of the Estimate ($SEE$).

$$R^2={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^N}{\left(P_i-\overline{P}\right)}^2{\left(A_i-\overline{A}\right)}^2}{{\displaystyle {\sum }_{i=1}^N}{\left(P_i-\overline{P}\right)}^2\times {\displaystyle {\sum }_{i=1}^N}{\left(A_i-\overline{A}\right)}^2}}$$

(3)

$$SEE=\sqrt{{\displaystyle \frac{{\displaystyle {\sum }_{i=1}^N}{\left(P_i-A_i\right)}^2}{N-k}}}$$

(4)

where $N$ is the sample size and $k$ is the number of independent variables in the model.

3. Results

3.1. Data Overview

Variations in NDVI

NDVI showed an overall increasing trend during 2003–2020, with a notable increase after 2012. The monthly mean NDVI exhibited a single peak, with the highest values occurring in July–August and the lowest values occurring in January–March (Figure 3a).

AWC presented an overall increasing trend from 2003 to 2020 (Figure 4). Specifically, the change in AWC was positive after 2008, and the increase became obvious from 2010 to 2015. EWTC showed an overall fluctuating upward trend from 2003 to 2020, with a positive value in 2008 and an obvious upward trend after that. TWSC exhibited an overall small trend between −20 and 20 mm in the period of 2003–2020.

The annual variation in EWTC had positive values from July to November and negative values from December to June (Figure 5). The annual TWSC variation was similar to the annual EWTC variation and mainly influenced by PRCP, ET, Q_s, and Q_sb. The southwest region entered the flood season in May when water resources are abundant. The AWC variation lagged behind the EWTC and TWSC variations, with positive values from August to January and negative values in the other months.

The spatial distributions of AWC, EWTC, and NDVI indicate that the regional average water resources under the influence of the hydrometeorological factors exhibited minimal change in 2003–2020 in the southwestern part of the region under natural influences (Figure 6). EWTC increased in most parts of the southwest region. The most obvious increase was found in eastern Sichuan, followed by southern Chongqing, most of Guizhou, and western Yunnan. EWTC decreased in Ganzi, Sichuan, and eastern Yunnan. The overall spatial change trend of AWC was similar to that of EWTC. Increasing trends were found in eastern Sichuan, followed by southern Chongqing, most of Guizhou, and western Yunnan. The NDVI spatial distribution map shows that vegetation grew well in eastern Sichuan, Chongqing, and Guizhou, which are rich in water resources and have appropriate elevations for good NDVI growth conditions.

3.2. Relationships Between NDVI and Hydro-Climatic Factors

The relationships between NDVI and the hydro-climatic variables at lags of 0–2 months were analyzed for each year (Figure 7). NDVI in western Yunnan was positively correlated with AWC in the current month, and this correlation was weak in one month lagged by NDVI. AWC in eastern Yunnan was negatively correlated with NDVI in the current month and in 0–2 months lagged by NDVI. At the pixel scale, across 70% of western Sichuan and most of Yunnan, EWTC was positively correlated with the NDVI lagged by 0–1 months, and this positive correlation was most significant in the current month. At the pixel scale, TWSC was positively correlated with NDVI lagged by 0–1 months across 82% of Southwest China, with this positive correlation being most significant in the current month. The strongest correlations occurred in eastern Sichuan and most of Yunnan. NDVI was positively correlated with the hydrometeorological factors, and the correlation was high in the western part of Sichuan and most of Yunnan (0–1 month in western Yunnan and 1–2 months in eastern Yunnan).

The spatial distribution of the positive correlation between NDVI and ET and PRCP was relatively consistent and showed positive correlation in most parts of Southwest China (Figure 8). The largest areas with a positive correlation were Aba and Ganzi Prefectures in West Sichuan, followed by the eastern part of Sichuan, Chongqing, Guizhou, and the eastern part of Yunnan. The positive correlation between NDVI and LST was dominated by the correlation in Sichuan, Chongqing, and Guizhou, and weak negative correlations were found in Yunnan. NDVI was positively correlated with surface runoff in most places, except southwest Yunnan. It was positively correlated with surface runoff at lags of 1–2 months in Ganzi and Sichuan and at lags of 0–1 months in the rest of the study area. NDVI was positively correlated with subsurface runoff before January in Southwest China and negatively correlated with subsurface runoff at lags of 0–2 months in Sichuan.

EWTC and TWSC-SH showed a positive correlation with NDVI at lags of 0–2 months, where the correlation coefficients of EWTC and TWSC-SH in the current month were high (Figure 9). This finding shows a weak positive correlation with AWC, mainly below 0.25. With regard to the hydrometeorological indicators, NDVI showed a strong positive correlation with ET, PRCP, and LST, among which ET had the highest positive correlation of 0.96, PRCP had 0.89, and LST had 0.86. The median positive correlation of ET was 0.63, that of PRCP was 0.52, and that of LST was 0.47. The positive correlation of the meteorological factors with NDVI became increasingly obvious after being lagged by 0–1 month, and the correlation decreased after being lagged by 2 months. NDVI was positively correlated with Q_s and Q_sb (median correlation coefficient of 0.37), and NDVI had the highest correlation with the hydrological factors of the month.

The stepwise multiple linear regression results (

Table 2. Optimal models with different lag times.

Model	Inputs	R2	SEE
ML-0	ET, DEM, Lat, EWTC	0.41	0.77
ML-1	ET_1, DEM, Lat, TWSC_1	0.37	0.80
ML-2	ET_2, DEM, TWSC_2, Lat	0.25	0.86
ML-3	LST_3, Lat, TWSC_3, DEM	0.15	0.92

Note: All models are significant at p < 0.001.

) reveal the temporal dynamics of vegetation response to environmental drivers. The model without a time lag (ML-0) achieved the highest predictive accuracy (R² = 0.41, SEE = 0.77), identifying Evapotranspiration (ET) as the primary climatic driver alongside topographic factors (DEM, Latitude) and Equivalent Water Thickness Change (EWTC). This indicates that vegetation growth in the study area responds almost instantaneously to current water and energy conditions.

As the lag time increased from 1 to 3 months (ML-1 to ML-3), a consistent decline in model performance was observed. The coefficient of determination dropped from 0.37 at a 1-month lag to 0.15 at a 3-month lag, accompanied by an increase in the Standard Error of the Estimate (SEE) from 0.80 to 0.92. This trend suggests that the memory effect of climate factors on vegetation growth is relatively short, with the explanatory power of antecedent climate conditions diminishing significantly beyond a one-month timeframe. Notably, Terrestrial Water Storage Change (TWSC) replaced EWTC as a significant predictor in the 1 to 3 month lag models, implying that deeper soil water or groundwater storage variations exert a more delayed influence on vegetation compared to surface water changes.

Model accuracy gradually deteriorated with increasing lag time. At the following 0–2 months, the input climate factors all had ET, indicating a strong relationship between ET and vegetation. Among all climate drivers, ET exhibits the strongest correlation with NDVI, primarily due to its comprehensive indicative role. Unlike individual factors such as precipitation or air temperature, ET simultaneously reflects atmospheric energy demand, water availability, and the physiological activity of vegetation. In karst regions, where soils are shallow and water drains rapidly through fissures, ET effectively represents the actual rate of water consumption by vegetation based on currently available water. However, the relationship between ET and NDVI is not a unidirectional causal one. On the one hand, an increase in NDVI implies greater leaf area and transpiration capacity, thereby driving up ET. On the other hand, excessively high ET accelerates soil moisture depletion, which in turn inhibits vegetation growth under water-limited conditions, creating a negative feedback loop.

The optimal model in Table 2 does not take into account other factors, so we try to remove the ET variables from Table 2 and again construct the optimal model with different lag months step by step (

Table 3. Optimal model with different lag times after ET removal.

Model	Inputs	R2	SEE
ML-ET-0	PRCP, LST, TWSC, Qsb, DEM	0.38	0.79
ML-ET-1	PRCP_1, LST_1, Lat, DEM	0.30	0.84
ML-ET-2	LST_2, PRCP_2, Lat, DEM	0.23	0.88
ML-ET-3	LST_3, Lat, TWSC_3, DEM	0.15	0.92

Note: All models are significant at p < 0.001.

). There is a slight reduction in the overall accuracy of the model after excluding ET (except for the best model with a 3-month lag), and the model introduces two variables, precipitation and temperature (except for the best model with a 3-month lag). The results illustrate that ET can respond to the combined changes of precipitation and temperature. In addition, DEM was still selected inside the best model at different lag scales, which shows that the effect of DEM on vegetation is not negligible.

However, stepwise multiple linear regression has several limitations. This method is prone to overfitting, particularly when there are many predictor variables and the sample size is limited. Stepwise multiple linear regression assumes a linear relationship among variables, which may overlook the nonlinear characteristics of vegetation responses. The screening process relies on arbitrarily set statistical thresholds, which may exclude factors that are ecologically significant but exhibit multicollinearity with other variables.

3.3. Influence of Hydrometeorological Factors on NDVI Under the Influence of Different Elevations

The distributions of the correlation coefficients of NDVI, ET, and PRCP were similar, with high values of probability density distributed between 0.6 and 0.75 (Figure 10). The correlation coefficients at high altitudes (above 4000 m) were mainly between 0.75 and 1, the correlation coefficients within 1000–2000 m were between −0.25 to 0.5, and the correlation coefficients within 0–1000 m altitudes were between the correlation coefficients of NDVI and LST in the range of 0.6–0.75.

The high probability density of the correlation distribution between NDVI and Q_sb was distributed around 0.4, in which the high value of the correlation coefficient for DEM below 3000 m was mainly around 0.4. The high value of the correlation coefficient for DEM above 3000 m was between 0.5 and 0.6. The high probability density of the correlation distribution between NDVI and surface runoff was distributed around 0.4. Specifically, the high value of the correlation coefficient for DEM below 3000 was mainly around 0.4, and the high value of the correlation coefficient for DEM above 3000 m was lower, that is, between 0.2 and 0.4. The high values of the correlation coefficients with DEM below 3000 were mainly around 0.4, and the correlation coefficients in the region with DEM above 3000 m were low, ranging from 0.2–0.4.

The probability density values for the correlation distribution between NDVI and EWTC are primarily concentrated between 0.2 and 0.4. In low-altitude areas with a DEM below 2000 m, the highest correlation coefficients range from 0.2 to 0.4. In contrast, in areas with a DEM above 3000 m, the highest correlation coefficients range from 0.4 to 0.6.

The high probability density of the correlation distribution between NDVI and TWSC was around 0.5. The high value was around 0.5 for the low-altitude areas with DEM lower than 2000 m, and the high value of the correlation coefficient for the high-altitude areas with DEM higher than 3000 m was in the range of 0.5–0.75. The high value of the correlation distribution between NDVI and AWC was in the range of 0.2–0.4, and the high value was concentrated in the range of 0.2–0.3 for the high-altitude areas with DEM higher than 3000 m. The correlation coefficients for the high-altitude areas with DEM higher than 3000 m were concentrated as long as they were in the range of 0.2–0.3.

4. Discussion

4.1. The Influence of Hydrometeorological Factors on NDVI in the Complex Terrain of Southwest China

In southwestern China, the distribution of NDVI exhibits significant spatial heterogeneity, with values generally higher in the northeast and lower in the northwest. This study indicates that there are significant spatial variations in the correlation between annual average NDVI and hydrometeorological factors in Southwest China, and that different hydrometeorological factors exert their influence on NDVI with varying time lags. The response of NDVI to temperature also showed a strong time lag effect, except in winter [30].

These observed time lags are closely linked to internal eco-hydrological processes, as soil moisture memory plays a pivotal role. Precipitation infiltrates into the root zone with temporal buffering, so wetter soils can sustain vegetation growth for weeks to months after rainfall events cease [16]. In humid regions such as eastern Sichuan and Chongqing, deep-rooted vegetation utilizes soil water stored during previous wet periods, allowing NDVI to remain elevated despite short-term meteorological variability. By contrast, in drier areas such as western Sichuan and Yunnan, constrained soil water storage shortens soil moisture memory and leads to faster NDVI responses to increasing temperatures.

Vegetation phenology further modulates these lag structures. In seasonally snow-covered highlands, spring green-up is often delayed relative to rising temperatures due to the timing of snowmelt. The gradual release of meltwater into the soil decouples early-season vegetation growth from concurrent temperature or precipitation signals and shifts apparent correlations toward earlier months [28]. In addition, interannual variations in rooting depth, which are driven by species composition, stand age, and disturbance history, influence the rate at which plants access deep soil moisture and thereby modify the temporal scale of NDVI responses to hydroclimatic variability.

NDVI shows a significant overall correlation with annual average temperature and annual cumulative precipitation in Southwest China, and the response of NDVI in Southwest China to changes in precipitation and temperature exhibits distinct spatial patterns. The greening of vegetation across most of southwestern China is primarily driven by the combined effects of temperature, precipitation, and evapotranspiration [7,15,18], with temperature having the greatest effect on NDVI in summer and precipitation exerting the greatest effect in the whole year [10]. The positive correlation between NDVI and temperature is primarily concentrated in eastern Sichuan and the Chongqing region, where rainfall is abundant, humidity is high, and the terrain is complex. Rising temperatures promote vegetation growth in these areas. In Yunnan and western Sichuan, however, NDVI and temperature exhibit a negative correlation. These regions receive less precipitation, have drier air, and lower average temperatures. As temperatures rise, surface evaporation increases, which is detrimental to vegetation growth.

In the border region between Sichuan and Chongqing, the climate is favorable, and vegetation thrives as annual precipitation increases. However, in western Yunnan and southwestern Sichuan, vegetation growth has shown a declining trend as annual precipitation has increased. Precipitation in this region is abundant and meets the basic requirements for vegetation growth. When precipitation increases, cloud cover also increases, reducing solar radiation, which is detrimental to vegetation photosynthesis. Consequently, as annual average precipitation increases, vegetation growth declines. This pattern is similar to that observed in northern Guangdong, northeastern Hunan, and eastern Guangxi, where precipitation is equally abundant [7,18].

4.2. The Effects of Hydrometeorological Factors on Vegetation Growth at Different Altitudes

Elevation is the primary factor driving spatial heterogeneity in vegetation, and the southwestern mountainous regions feature complex land surfaces and significant variations in elevation. These variations in elevation are similar to the variations in vegetation cover distribution [4,14,18]. At different vertical heights, NDVI was considerably affected by natural variability, especially evapotranspiration and precipitation, in the regions with elevations higher than 3000 m. Tao et al. suggested that elevation-dependent warming is the main driver of elevation-dependent vegetation change [4]. Changes in elevation also affect the distribution of vegetation and different soil types. As elevation increases, the local ecosystem becomes increasingly fragile and barren, which affects vegetation growth [31]. Studies have shown that in the high-altitude regions of western Sichuan, due to the constraints of the plateau terrain, increased precipitation often intensifies soil erosion, making soil and water loss more likely to occur, which in turn may have an adverse effect on vegetation growth [7]. At the same time, the intensity and scale of human activities exhibit a significant altitudinal gradient [11]. Therefore, elevation plays a significant role in the spatial differentiation of vegetation in the high-mountain canyon regions of southwestern China [14].

At high elevations, lagged vegetation responses are further shaped by snowmelt dynamics and ecosystem resilience. In the alpine zones of western Sichuan, snowmelt serves as a delayed water source that sustains soil moisture availability well into the growing season, thereby decoupling NDVI peaks from early summer precipitation [29]. Moreover, high-altitude ecosystems often exhibit low resilience, such that recovery from drought or disturbance may require multiple growing seasons. This slow recovery introduces multi-year lag effects in NDVI trajectories following extreme hydroclimatic events. Variations in rooting depth along elevation gradients also play an important role. Deeper root systems at mid-elevations buffer vegetation against short-term water deficits, whereas shallow-rooted alpine vegetation responds more directly and occasionally more abruptly to changes in snowpack and soil moisture.

4.3. Limitations of the Analysis of Vegetation Changes in the Mountainous Regions of Southwest China

This study takes the southwestern mountainous area as a whole and does not consider zoning. Different partitions, methods, or spatial divisions may lead to different results [32]. In the future, the southwestern mountainous areas can be modeled and studied separately based on factors such as terrain, to explore whether there are general patterns of vegetation growth in different regions affected by hydroclimatic factors and terrain. Furthermore, the study period spans from 2003 to 2020, and the results are influenced by the characteristics of this timeframe [33]. The study relied on interpolation methods to standardize the resolution of all factors, which may impose certain limitations on the analysis of spatial characteristics [7]. In the future, multi-source data could be utilized to further refine the understanding of how hydrometeorological factors influence NDVI variations in the complex terrain of China’s southwestern mountainous regions. Finally, quality issues such as noise from remote sensing data sources can also affect the accuracy of the model. In summary, further research will be conducted to improve data quality, partitioning, higher spatial resolution, and scale in the future.

Future studies would explicitly quantify the contributions of soil moisture memory, snowmelt timing, rooting depth adaptation, phenological shifts, and ecosystem resilience to observed NDVI lag structures [34,35]. Integrating process-based ecohydrological models with satellite observations would help disentangle these mechanisms and improve the predictive capacity of vegetation dynamics under ongoing climate change in mountainous regions.

5. Conclusions

This study utilized multi-source remote sensing data to investigate the spatiotemporal variations in NDVI, effective water content (EWTC), total water content (TWSC), and available water content (AWC) in Southwest China during the period 2003–2020. Correlation analysis and stepwise multiple linear regression were employed to examine the relationships between NDVI and EWTC, TWSC, and AWC, as well as the correlations between hydrometeorological factors and NDVI under different time lags. The study focused on the influence of hydrometeorological factors on NDVI at different elevations. The following conclusions were drawn.

(1)

NDVI showed an overall increasing trend from 2003 to 2020, with a notable increase after 2012. The annual variation in NDVI exhibited a single peak, with the highest value occurring in July–August and the lowest value occurring in January–March. Vegetation grew well in the southwestern part of the country, except for western Sichuan. This pattern suggests that vegetation greening is jointly driven by large-scale ecological restoration programs implemented after 2012, whereas the single-peak seasonality reflects the synchronization of photosynthetic activity with warm and wet summer conditions. Furthermore, the relatively low NDVI in western Sichuan is primarily attributable to alpine cold stress, limited soil development, and prolonged snow cover rather than water scarcity alone.

(2)

In terms of annual changes, AWC and EWTC presented an overall fluctuating upward trend from 2003 to 2020, with a significant shift observed after 2008. TWSC exhibited minor fluctuations over the study period. Regarding monthly variations, AWC, EWTC, and TWSC showed a single peak and trough pattern, with AWC changes generally lagging behind EWTC and TWSC. The turning point around 2008 may be related to shifts in regional hydrological and climatic patterns, particularly changes in precipitation patterns and glacial snowpack dynamics. The persistent lag of AWC relative to EWTC and TWSC suggests that vegetation does not respond to instantaneous precipitation inputs but rather relies on soil moisture memory and delayed groundwater redistribution processes, which are of great significance for the stability of karst mountain ecosystems.

(3)

The analysis indicates that NDVI variations are closely associated with hydrometeorological factors, particularly ET and PRCP. The study found significant time-lagged correlations between NDVI and water storage variables. EWTC and TWSC showed positive correlations with NDVI at a lag of 0–2 months. This suggests a temporal response of vegetation growth to changes in water availability. Meteorological factors such as ET, PRCP, and LST showed strong positive correlations with NDVI, with the highest correlations observed in western Sichuan and eastern Yunnan. This pattern arises because an increase in NDVI enhances transpiration, while excessive evapotranspiration accelerates soil moisture depletion, thereby limiting further vegetation growth. The significant correlation observed over the 0–2 month period indicates that vegetation primarily relies on surface water and soil water, which respond rapidly, whereas the influence of deep groundwater often becomes apparent only gradually after the short-term signals have faded. The spatial differences between western Sichuan and eastern Yunnan reveal the varying sensitivities of alpine and lowland ecosystems to hydrometeorological variability.

(4)

Spatially, the influence of topography modulates these relationships. Regression models confirmed that DEM, latitude, ET, precipitation, and air temperature are key variables influencing the distribution of NDVI. The impact of these factors is not uniform. In regions above 3000 m, particularly above 4000 m, the correlation between NDVI changes and hydrometeorological factors, mainly evapotranspiration and PRCP, was more pronounced, highlighting the sensitivity of alpine vegetation to climatic variations. Altitude amplifies or attenuates vegetation responses to climate drivers by modulating the radiation balance, snowmelt timing, and soil development. The increased correlations observed at high altitudes indicate that even minor changes in water and heat conditions can trigger disproportionate ecosystem responses, thereby highlighting their vulnerability in the context of climate change.

From an environmental management perspective, these findings highlight the necessity of implementing differentiated conservation strategies that prioritize high-altitude zones due to their heightened sensitivity to climate variability, while emphasizing soil moisture retention measures to buffer against the time-lagged hydrological responses in karst ecosystems.