Phenology-Optimized Drought Index Reveals the Spatio-Temporal Patterns of Vegetation Health and Its Attribution on the Loess Plateau

Abstract

Frequent droughts pose a severe threat to the ecological health and sustainable development of the Loess Plateau (LP). The accurate assessment of the impact of drought on vegetation is crucial for diagnosing ecological health. Traditional drought assessment methods often rely on coarse estimations based on averages of vegetation drought indices, overlooking the spatial differentiation of complex vegetation phenology. This study proposes a vegetative drought assessment method that considers vegetation phenological characteristics using MODIS EVI and LST data products. First, the start and end of the growing season timepoints were extracted from the Enhanced Vegetation Index (EVI) using Savitzky–Golay (S–G) filtering and the dynamic threshold method, determining the growing-time window for each pixel. Next, the Vegetation Health Index (VHI) series was calculated and extracted for each pixel within the growing season. The mean value of the VHI series was then used to construct the Growing Season Health Index (GSHI). Based on the GSHI, the long-term vegetation drought characteristics at LP were revealed. Finally, we integrated the Optimal Parameters-based Geographical Detector (OPGD) to identify and quantify the multiple driving forces of vegetation drought. The results showed that: (1) the spatio-temporal difference of vegetation phenology on the LP was significant, exhibiting distinct zonal characteristics; (2) the spatial distribution of growing season drought on the LP presented a “humid southeast, arid northwest” pattern, with the early 21st century being a period of high drought occurrence; (3) drought has been alleviated in large-scale natural areas, but the local drought effect under urbanization is intensifying; and (4) meteorology and topography influence vegetation drought by regulating water redistribution, while the drought effect of human activities is intensifying.

1. Introduction

Drought is an extreme climatic event that has seen an increase in frequency, duration, and extent in the context of global climate change [1]. It impacts vegetation growth, physiological functions, and ecological processes by influencing soil water availability, ultimately leading to reduced vegetation productivity, altered species composition, and ecosystem degradation [2]. Frequent droughts in recent decades have emerged as a major threat to the Loess Plateau (LP)’s ecological health and sustainable development. The LP is acknowledged as a crucial ecological barrier and a base for agricultural production in China, essential for sustaining the region’s ecological balance and securing the nation’s food supply [3]. However, due to its unique geographical environment and climatic conditions, the LP is highly vulnerable to drought. In recent decades, frequent drought events in this region have become a serious threat to the sustainable development of both its fragile ecosystems and agriculture [4,5]. Frequent droughts have led to a decline in vegetation productivity, changes in species composition, and ecosystem degradation on the LP, posing a severe threat to regional ecological health and sustainable development. Accurate assessment of the impact of drought on vegetation is crucial for diagnosing the health status of semi-arid ecosystems [6].

Traditional drought assessment mainly relies on meteorological observation data from discrete sites to construct indicators such as the Standardized Precipitation Index (SPI) and the Palmer Drought Severity Index (PDSI) [7,8]. While these indicators can reflect regional drought conditions to a certain extent, they are limited by the coverage of the duration of continuous observation at monitoring sites. With advancements in remote-sensing technology, satellite-based monitoring has become an effective tool for assessing vegetation drought due to its wide coverage, multi-temporal data, and low cost [9,10]. It provides spatially explicit and temporally continuous information on vegetation conditions and temperature stress [11]. Scholars have developed various drought indices via remote-sensing parameters, such as the Vegetation Condition Index (VCI) and the Temperature Condition Index (TCI) [12,13]. These indices have been widely used to characterize the dynamics of regional vegetation health under drought stress, providing a powerful tool for diagnosing ecological health conditions [14,15,16,17,18]. Among these, the Vegetation Health Index (VHI), which integrates information on vegetation conditions and surface temperature, has garnered more attention [19,20]. Therefore, it is of great significance to use remote-sensing methods to assess the impact of drought on vegetation health.

Prior research has utilized meteorological data and remote-sensing imagery to explore the spatio-temporal characteristics of drought at LP. Gao et al. investigated the meteorological drought evolution at LP using the Standardized Precipitation Evapotranspiration Index (SPEI) [21]; Guo et al. used two forms of the PDSI to analyze meteorological drought patterns at LP and its four subregions [22]; and Liu et al. explored the drought trends and intensity using SPEI and SPI at different time scales [4]. In recent years, the remote-sensing method has been extensively applied in drought capturing at LP. Li et al. assessed the drought conditions using the VCI, confirming the applicability of this index at LP [23]; Wang et al. analyzed the interannual, seasonal drought dynamics using the Temperature Vegetation Dryness Index (TVDI) [24]. Liu et al. explored the spatio-temporal distribution of drought during the growing season (predefined from April to October) at LP using the Crop Water Stress Index (CWSI) [25]. These studies have revealed the spatial–temporal differentiation patterns of drought at LP, indicating that meteorological and remote-sensing drought indices are suitable for capturing the dynamics of drought.

Despite this, the existing research has certain limitations. First, many studies evaluate vegetation drought conditions based on calendar years or empirically defined growing seasons, which may not accurately reflect the actual impact of drought on critical growth stages of vegetation. The LP is vast, with complex topography, diverse vegetation types, and significant climatic variation, resulting in pronounced spatial differentiation of vegetation phenology [26]. Neglecting the critical periods in the phenological development process of different vegetation types and regions may lead to biased estimates of drought impacts on vegetation growth stages [27]. Moreover, previous studies have mostly focused on the processes of drought, while the processes and mechanisms by which drought threatens regional ecological health through influencing vegetation growth remain to be further explored. However, the understanding of the mechanisms underlying the formation of the spatial–temporal patterns of vegetation drought at LP remains relatively weak, necessitating analysis from the perspective of multi-factor effects.

Given these considerations, this study proposes a method for assessing growing season drought that takes into account vegetation phenological characteristics and explores the patterns and drivers of vegetation drought at LP, which is of great value for revealing the evolution patterns of the Loess Plateau ecosystem health under drought threats. The specific research objectives include the following: (1) using S–G filtering and MODIS-EVI to identify the green-up and senescence dates of vegetation, thereby determining the growing season time window for each pixel; (2) calculating the mean-VHI (GSHI, Growing Season Health Index) for each pixel within the growing season, and validating the reliability of the GSHI through correlation with the Net Primary Productivity deviation (NPPd); and (3) revealing the spatial characteristics and evolution trends of vegetation drought at LP from 2003 to 2022, and quantitatively analyzing the driving factors of growing-season drought using the OPGD model, considering meteorological, topographical, soil, and human activity influences. Additionally, this research also provides replicable approaches for assessing growing-season vegetation drought in other ecological regions.

2. Materials

2.1. Study Area

The LP is situated in the northwest of China (33°43′–41°16′N, 100°54′–114°33′E), and covers a total area of about 640,000 km², with an elevation ranging from 78 to 5225 m (Figure 1). With the Taihang Mountains as its eastern boundary and the Wushaoling peaks of Qilian range marking its western limit, the area spans between the Great Wall and Qinling Mountains along its north–south axis [28,29]. This area includes several provinces and regions, namely Qinghai, Gansu, Ningxia, Inner Mongolia, Shaanxi, Shanxi, and Henan. The climate of LP is characterized by distinct seasonal variations, with wet summers and dry winters. Influenced by the summer monsoon, over 60% of the annual precipitation occurs from June to September [30].

Vegetation within the LP exhibits distinct zonal characteristics, gradually transitioning from forest to grassland from the south-east to the north-west. The main vegetation includes deciduous broad-leaved forests, temperate grasslands, and crops [31]. Deciduous broad-leaved forests are mostly distributed in the SE regions with more abundant precipitation; grasslands are distributed in the semi-arid areas of the north-west; crops are widely distributed in the central and south-eastern parts of the LP, with wheat and corn being the primary food crops [32]. In recent years, drought has emerged as a significant factor affecting vegetation growth and threatening the stability of ecosystems at LP.

2.2. Data Sources

To ensure spatial consistency across data from various sources, all datasets used in this study were unified to the WGS84 coordinate system with a resolution of 1 km. This section outlines the datasets employed and their preprocessing.

2.2.1. EVI and LST Data

The EVI data (16-day, 500 m) and LST data (8-day, 1 km) were sourced from the MOD13Q1 and MOD11A1 products [33,34]. Data were preprocessed using the GEE platform, including quality control, image mosaicking, and clipping. Pixels associated with MODIS products were classified as high quality if they processed a QC value of 00, signifying they are free from significant cloud contamination, atmospheric noise, and other factors that may degrade data quality [35]. The proportion of high-quality pixels (QC = 00) surpassed 90% in this study area.

2.2.2. NPP Data

NPP is defined as the amount of carbon that vegetation binds through photosynthesis minus the carbon used in autotrophic respiration [36]. Prior research indicates that NPP has good consistency with crop biomass and yield, making it an important benchmark for evaluating the severity of vegetation drought and its loss [37]. The NPP data (annual, 500 m) were sourced from MOD17A3, which were used to validate the effectiveness of the Growing Season Health Index [33].

2.2.3. Topographic Data

The dem data (30 m) were sourced from the NASA global elevation product [38]. This study extracted elevation, slope, and aspect information of the LP region from the NASADEM_HGT/001 dataset using the GEE platform. This information was utilized for subsequent analyses, including the examination of the influence of topographic factors on vegetation drought.

2.2.4. Climate Data

Precipitation, temperature, and sunshine data (annual, 1 km) were sourced from Earth System Science Data Center of China [39]. The GDAL tool was used for format conversion and processing of the original NetCDF(nc) format files, facilitating the analysis of climatic factors’ impact on vegetation drought during the growing season.

2.2.5. Land Use Data

Land use data (annual, 500 m) were sourced from the MCD12Q1 [40]. The original data were reclassified into forest, shrubland, grassland, cropland and construction land via International Geosphere–Biosphere Programme [41]. These data were used to assess the characteristics of the vegetation drought across different types.

2.2.6. Soil Texture Data

Soil data (1 km) including soil type and soil texture information were sourced from the HWSD 2.0 [42], constructed by the FAO. The soil sand content and soil type data were extracted to assess soil texture’s influence on growing season vegetation drought.

2.2.7. Human Activity Data

Human footprint data (annual, 1 km) were sourced from the UEMM [43], which integrated eight variables representing human pressure. Carbon emissions data (annual, 1 km) were obtained from the Emissions Database for Global Atmospheric Research [44]. Distance to cities was calculated using the Euclidean distance method based on the 1:250,000 basic geographic data obtained from the Earth System Science Data Center [39]. These datasets were used to assess the impact of human activities on vegetation drought.

3. Methods

The research methodology is presented in Figure 2, highlighting the key steps: data acquisition and preprocessing, phenology parameter extraction, GSHI calculation and validation, and the analysis of drought patterns and drivers at LP.

3.1. Data Smoothing and Phenology Extraction

Vegetation phenology refers to the periodic patterns of plants that change with the seasons, which is usually extracted from time series of remote-sensing-based vegetation indices, and the phenological parameters obtained in this way are called land surface phenology [45,46]. In this study, MATLAB (version: R2023b) was utilized to apply the Savitzky–Golay filter along with the dynamic-threshold method to extract pixel-scale vegetation growing-season phenology information from the MODIS-EVI [47]. The main steps are as follows:

(1)

Time series smoothing: Original MODIS-EVI time series data contain inherent noise and necessitate smoothing. The S–G filter is a data-smoothing algorithm that employs local polynomial least squares fitting, effectively removing high-frequency noise in time series while preserving the waveform characteristics of the data [48]. The algorithm for the S–G filter is described as follows:

$$Y_j^{*}={\displaystyle \frac{\sum _{i=-m}^{i=m}{C}_i{Y}_{j+1}}{N}}$$

(1)

where $Y$ is the original EVI value, $Y_j^{*}$ is the filtered EVI value, $C_i$ is the coefficient of the $i$-th EVI value of the filter, $j$ is the index of the original EVI series, $N$ is the size of the smoothing window ($2m+1$), and $m$ is the half-width of the window. This study used a third-order polynomial S–G filter algorithm with a window width of 6 to smooth the original EVI time series. Ultimately, while maintaining the overall morphology of the vegetation growth process, daily temporal continuity was achieved.

(2)

Extraction of the key time-points of phenology: The start of season (SOS) and the end of season (EOS) are two key nodes describing the vegetation growing season (Figure 3). This study used the dynamic threshold method to obtain the annual SOS and EOS of each pixel. Specifically, based on the maximum and minimum values of the smoothed EVI time series, the time point when the EVI exceeds 20% of the current amplitude (the difference between EVI_max and EVI_min) is the SOS, and the time point when it drops to 80% of the current amplitude is the EOS.

(3)

Defining the vegetation growing season: The period between SOS and EOS is defined as the vegetation growing season. Considering the differences in phenological periods of different regions and vegetation types, this study determined the growing season time window for each year on a pixel-wise basis.

3.2. Growing Season Health Index (GSHI)

This study proposed a Growing Season Health Index (GSHI) to characterize the yearly vegetation drought conditions at LP. GSHI comprehensively considers EVI and LST information and introduces pixel-scale vegetation phenology parameters to depict the impact of drought on critical growth periods of vegetation. The calculation steps of GSHI are as follows:

(1)

Based on the extracted information of the SOS and EOS, the vegetation growing season time range of each pixel for each year is determined.

(2)

Using the original MODIS EVI 16-day and LST 8-day composite product data, we first averaged two 8-day LST composites to match the temporal resolution of EVI. Then, the VCI, TCI, and VHI of each 16-day period are calculated, respectively. The specific calculation methods are as follows:

$${\mathit{TCI}}_{ijk}={\displaystyle \frac{{\mathit{LST}}_{ij\mathit{max}}-{\mathit{LST}}_{ijk}}{{\mathit{LST}}_{ij\mathit{max}}-{\mathit{LST}}_{ij\mathit{min}}}}$$

(2)

$${VCI}_{ijk}={\displaystyle \frac{{EVI}_{ijk}-{EVI}_{ij\mathit{min}}}{{EVI}_{ij\mathit{max}}-{EVI}_{ij\mathit{min}}}}$$

(3)

$${VHI}_{ijk}=\alpha \times {VCI}_{ijk}+\left(1-\alpha \right)\times {\mathit{TCI}}_{ijk}$$

(4)

where $i$, $j$, and $k$ represent the $i$-th pixel, the $j$-th 16-day time period, and the $k$-th year, respectively; max and min represent the multi-year maximum and minimum values of the corresponding pixel; $\alpha$ represents the weight factor, which is generally fixed at 0.5 in practical applications [20].

(3)

For each pixel, the day of year (DOY) interval corresponding to the vegetation growing window (SOS to EOS) is matched, the VHI subset during the growing season is extracted from the VHI time series within the DOY interval, and its average value is calculated to obtain the GSHI of the pixel:

$$GSH{I}_{y,j}={\displaystyle \frac{\sum _{i=1}^{n}VH{I}_{i,j}}{n}}\hspace{1em}DOY\left(SO{S}_j\right)\le DOY\left(t_i\right)\le DOY\left(EO{S}_j\right)$$

(5)

where $GSH{I}_{y,j}$ represents the $GSHI$ value of pixel $j$ in year $y$; $VH{I}_{i,j}$ represents the $i$-th $VHI$ value within the growing season time range of pixel $j$; $DOY\left(SO{S}_j\right)$ and $DOY\left(EO{S}_j\right)$ are the Julian days corresponding to the start and end dates of the growing season of pixel $j$, respectively; $DOY\left(t_i\right)$ is the Julian day corresponding to the $j$-th $VHI$ value; $n$ is the total number of $VHI$ series within the growing season of pixel $j$.

(4)

Referring to the drought level classification standards of VCI, TCI, and VHI [49,50], and considering that vegetation is more sensitive to water conditions during the growth stage and drought stress will have a greater impact, this study divides GSHI into 5 levels, as shown in

Table 1. Classification of GSHI into five levels.

Level	Type	GSHI Value
D0	No drought	≥0.45
D1	Mild drought	[0.35, 0.45)
D2	Moderate drought	[0.25, 0.35)
D3	Severe drought	[0.15, 0.25)
D4	Extreme drought	<0.15

.

3.3. Validation of GSHI for Vegetation Drought Representation

To objectively evaluate the effectiveness of GSHI in depicting the yearly vegetation drought conditions, it is necessary to select appropriate reference indicators for comparative validation. Drought affects NPP by influencing soil water availability, inhibiting vegetation photosynthesis and material accumulation, ultimately leading to a decrease in NPP. Thus, NPP can well indicate the degree of drought impact on vegetation growth [51]. In view of this, we selected NPP as vegetation drought references. In order to eliminate the influence of differences in vegetation productivity across different regions, we calculated the NPP deviation (NPPd) as a standardized indicator. The formula for NPPd is as follows:

$$NPP{d}_{ij}=\left(NP{P}_{ij}-{\overline{NPP}}_j\right)/{\sigma }_{NP{P}_j}$$

(6)

where $NP{P}_{ij}$ is the $NPP$ of the $j$-th pixel in the $i$-th year, and ${\overline{NPP}}_j$ and ${\sigma }_{NP{P}_j}$ are the multi-year average of $NPP$ and standard deviation of pixel j, respectively. $NPPd$ characterizes the degree of deviation of the $NPP$ of each pixel relative to the multi-year variability, reflecting the intensity of the impact of drought on vegetation. Pearson correlation method was employed to quantitatively evaluate GSHI performance by comparing it with NPPd. The coefficient quantifies a relationship strength between two variables. It varies from −1 to 1, where values closer to the extremes indicate a stronger correlation. The calculation formula is presented as follows:

$$R_{xy}={\displaystyle \frac{\sum _{i=1}^n\left[\left(x_i-\overline{x}\right)\left(y_i-\overline{y}\right)\right]}{\sqrt{\sum _{i=1}^n{\left(x_i-\overline{x}\right)}^2}\sqrt{\sum _{i=1}^n{\left(y_i-\overline{y}\right)}^2}}}$$

(7)

where $x_i$ is the drought index of the $i$-th year, $y_i$ is the NPPd value of the $i$-th year, and n is the number of years.

3.4. Frequency Analysis

Drought frequency reflects the probability of drought events occurring over a certain period and is an important indicator for assessing the degree of regional drought risk. Based on the GSHI drought levels, this study counted the frequency of mild droughts, moderate and above droughts at LP in 5-year intervals from 2003 to 2022. The calculation formula for the frequency of each drought category is as follows:

$$f={\displaystyle \frac{n}{N}}\times 100\mathrm{\%}$$

(8)

where $n$ represents the number of years corresponding to a specific drought level during the entire study period, and $N$ represents the total number of years during the study period.

3.5. Sen and Mann–Kendall

To analyze the changing trends of growing season drought, the Theil–Sen slope estimator and the Mann–Kendall test were employed to determine the magnitude of change and its significance for GSHI at the pixel scale. The Sen estimator was used to determine the magnitude of change for GSHI at the pixel scale. Concurrently, the MK test analyzed the significance level of these changes. The calculation formulas are as follows:

$${\beta }_{GDSI}=\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}\left({\displaystyle \frac{GSH{I}_i-GSH{I}_k}{i-k}}\right)$$

(9)

where $GSH{I}_i$ and $GSH{I}_k$ correspond to the $i$-th and $k$-th year data, respectively; $\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}\left(\right)$ is the function of taking the median value; $\beta$ is the Sen’s slope of the time series. When $\beta$ > 0, $GSHI$ shows an increasing trend; when $\beta$ < 0, $GSHI$ shows a decreasing trend.

The MK method is applied to test the significance of the GSHI trend over a time series of $n$ (20 years in this study). The equation used to compute the statistic S is presented below:

$$S=\sum _{i=1}^{n-1}\sum _{j=i+1}^n\mathrm{s}\mathrm{g}\mathrm{n}\left(x_j-x_i\right)$$

(10)

where $x_i$ and $x_j$ are the $i$-th and $j$-th items of the time series, respectively, and $\mathrm{s}\mathrm{g}\mathrm{n}$ is the sign function, defined as:

$$\mathrm{s}\mathrm{g}\mathrm{n}\left(\theta \right)=\left\{\begin{array}{ll}+1,& \mathrm{if} \theta >0\\ 0,& \mathrm{if} \theta =0\\ -1,& \mathrm{if} \theta <0\end{array}\right.$$

(11)

When n ≥ 10, the test statistic $Z$, which is the standardized $S$, can be calculated using the following formula:

$$Z=\left\{\begin{array}{ll}{\displaystyle \frac{S-1}{\sqrt{Var\left(S\right)}}},& \mathrm{if} S>0\\ 0,& \mathrm{if} S=0\\ {\displaystyle \frac{S+1}{\sqrt{Var\left(S\right)}}},& \mathrm{if} S<0\end{array}\right.$$

(12)

$$Var\left(S\right)={\displaystyle \frac{n\left(n-1\right)\left(2n+5\right)}{20}}$$

(13)

The value range of the statistic $Z$ is $(-\mathrm{\infty },+\mathrm{\infty })$. At a given significance level $\alpha$, when $\left|Z\right|>u_{1-\alpha /2}$, it indicates that the study sequence has significant changes at the $\alpha$ level.

3.6. OPGD Model

Vegetation drought is the result of the multi-factors. We selected indicators that may affect drought from four dimensions: meteorology, topography, underlying surface, and human activities, and OPGD was used to quantitative identify the effects of various factors on vegetation drought during growing season. We set a 5-year sliding time window to select relatively dry years, thereby screening out the dominant driving factors.

(1)

Optimal parameter selection. The discretization processing of continuous data by the geographical detector needs to be set manually, which is influenced by inaccurate discretization and human subjective factors. Based on this, the OPGD model combines different discretization methods and levels for continuous variables and uses the discretization parameter corresponding to the highest explanatory power as the optimum [52].

(2)

Factor and interaction detection. The single-factor detection and two-factor interaction detection in the geographical detector are employed to detect spatial stratified heterogeneity of influencing factors and discover the underlying drivers of drought. The principle is as follows:

$$q=1-{\displaystyle \frac{\sum _{h=1}^L{N}_h{\sigma }_h^2}{N{\sigma }^2}}=1-{\displaystyle \frac{SSW}{SST}}$$

(14)

$$SSW=\sum _{h=1}^L{N}_h{\sigma }_h^2, SST=N{\sigma }^2$$

(15)

where $q$ indicates the relative contribution of the examined factor, bounded between 0 and 1, with higher values suggesting stronger contributions; $h$ stands for the predictor variable or its categorical level; $N_h$ and $N$ denote the sample count in stratum $h$ and total sample size, respectively; ${\sigma }^2$ and ${\sigma }_h$ reflect the variance within stratum $h$ and across the entire dataset, respectively; $SSW$ and $SST$ represent the aggregated within-strata variance and total variance sum, respectively.

Table 2. Interaction types of detection factors.

Interaction Type	Judgment Criteria
Non-linear Weakening	$q\left(X_1\cap X_2\right)<Min\left[q\left(X_1\right),q\left(X_2\right)\right]$
Non-linear Attenuation	$Min\left[q\right(X_1),q(X_2\left)\right]<q(X_1\cap X_2)<Max\left[q\right(X_1),q(X_2\left)\right]$
Bi-factor Enhancement	$q\left(X_1\cap X_2\right)>Max\left[q\left(X_1\right),q\left(X_2\right)\right]$
Mutually Independent	$q\left(X_1\cap X_2\right)=q\left(X_1\right)+q\left(X_2\right)$
Non-linear Enhancement	$q\left(X_1\cap X_2\right)>q\left(X_1\right)+q\left(X_2\right)$

outlines the criteria for classifying interaction patterns.

4. Results

4.1. Spatio-Temporal Pattern of Vegetation Phenology

The spatial differentiation of LP vegetation phenology is significant, exhibiting zonal regularities that change with climatic gradients. The spatial distribution of SOS generally shows a trend of gradual delay from the south-east to the north-west (Figure 4a). The SOS in the southeastern region is concentrated in early March to early April, while the SOS in the northwestern region is delayed to late April to early May. Due to the differences in regional water and heat conditions, the spring temperature rises earlier in the southeast, with abundant precipitation, promoting vegetation to green up earlier. The spring temperature in the north-west is lower, with scarce precipitation, inhibiting the vegetation growth process. The interannual variation of SOS shows large fluctuations from 2003 to 2008, relatively stable from 2009 to 2017, and a trend of earlier green-up after 2018. Among them, the interannual fluctuation amplitude in the north is greater than that in the south.

The EOS shows varying trends across different geographic regions (Figure 4b). Specifically, most regions experience EOS concentrated between October and November. In the northeastern region, the vegetation withering period occurs earlier, while it is later in the SW region. This may be because the autumn temperature drops earlier in the north, with reduced precipitation, causing the vegetation to enter dormancy earlier, whereas the autumn climatic conditions are relatively better in the south. The interannual variability of EOS is more stable than that of SOS, considering that the autumn climate fluctuations are relatively small and have limited impact on vegetation growth. However, the EOS in some years, such as 2008, 2012 and 2013, is generally earlier, which may be related to extreme climatic events in those years.

Comparing the phenological characteristics of the four main vegetation types at LP reveals significant differences (Figure 4c). Forests exhibit relatively stable phenological changes, with smaller interannual fluctuations in SOS and EOS and generally earlier timing, likely due to their well-developed root systems and stronger drought resistance. Grasslands and shrublands, however, are more heavily influenced by drought, showing greater interannual fluctuations. Recently, grasslands and shrublands have experienced a trend of earlier phenology, possibly indicating improved regional water and heat conditions. Cropland vegetation phenology is significantly affected by human management, with a shortening growing season characterized by delayed SOS and advanced EOS, likely resulting from advancements in agricultural technology and optimized management measures.

4.2. Reliability Evaluation of the GSHI

To validate the effectiveness of the GSHI in characterizing vegetation drought at LP, we used the NPPd as a reference indicator. Figure 5c,d plots the interannual variation curves of the GSHI and the NPPd for comparison. Overall, the changes in the GSHI show high synchronicity and consistency with the NPPd, effectively capturing the evolution of vegetation drought at LP over the past 20 years. For example, both indices exhibit troughs around 2005 and 2009, indicating that the LP experienced more severe droughts during these two periods, which is consistent with the actual situation.

To quantitatively evaluate the performance of GSHI in characterizing vegetation drought during the growing season, we determined the long-term correlation coefficients between the GSHI and the NPPd for each pixel. Figure 5a,b illustrates the distribution of correlation coefficients and their significance across the LP region. The results demonstrate that the GSHI exhibits strong correlations in the northwest region, with correlation coefficients consistently above 0.8, and most areas pass the 0.05 significance test. This is mainly because the northwest region primarily features natural vegetation and experiences minimal human activities, allowing the GSHI to effectively reflect the impact of drought on vegetation. However, in the southeast region, especially in agricultural areas such as the Guanzhong Plain, the correlation between the two indices is slightly lower. This may be due to the high proportion of agricultural land in this region, where crop growth is influenced by human management measures.

To further compare the performance of the GSHI across different land cover types, we extracted areas with unchanged land use at LP from 2003 to 2022 and conducted zonal statistics on the correlation coefficients (Figure 5e). The results show that the GSHI has the highest correlation with the NPPd in grassland and shrubland regions, with median correlation coefficients around 0.9, possibly due to their higher sensitivity to temperature stress. Although the correlation coefficients in forest regions have a wider distribution range, the median value remains around 0.8. The correlation coefficients in croplands are generally slightly lower than those in natural vegetation, with a median value of around 0.7 and a wider distribution range, likely due to the influence of human management measures, such as irrigation, fertilization, and crop rotation. Built-up areas have the lowest correlation coefficients, with a median value below 0.5, and are mainly distributed in regions that do not pass the 0.05 significance test in Figure 5b.

4.3. Spatio-Temporal Pattern of Growing Season Drought

4.3.1. Spatial Distribution Characteristics

The spatial distribution of the GSHI, as illustrated in Figure 6a, reveals significant regional disparities in vegetative drought conditions across the LP during the growing season. Generally, the intensity of drought conditions increases from the south-east to the north-west, forming a spatial distribution of “humid southeast, semi-humid central, and semi-arid northwest”. The southeastern region, including the Weihe Plain and the Fenhe Plain, experiences abundant precipitation and favorable thermal conditions, resulting in an elevated GSHI and lower drought risk; the central region, such as the Yinchuan Plain and the Hetao Plain, exhibits a relatively dry climate, characterized by a slight decrease in the GSHI. In contrast, the northwestern region, particularly the Ordos Plateau, falls within the semi-arid climate zone, marked by scarce precipitation and intense evaporation, resulting in generally low GSHI values. This area is particularly susceptible to vegetation drought. Locally, the Loess Hilly and Gully Region in northern Shaanxi Province and the agricultural region in the central part of the LP represent concentrated areas of drought occurrence. These regions are characterized by fragmented topology, significant soil erosion, and a vulnerable ecological environment, rendering vegetation growth particularly sensitive to drought.

4.3.2. Interannual Variation Characteristics

The growing season drought affecting vegetation at LP demonstrates notable interannual differences and distinct phased characteristics (Figure 6b). The period between 2005 and 2009 was a high-incidence period of drought, with the proportion of extreme and severe drought areas reaching 5.6% and 19.1%, respectively, in 2005. This marked the highest levels observed throughout the study period. After 2010, the GSHI exhibited a fluctuating upward trend, suggesting that the drought risk during the growing season at LP has generally eased. However, in some individual years such as 2013 and 2015, due to the occurrence of local droughts, the proportion of severe and extreme drought areas rebounded. In 2013, the drought primarily affected the cropland in the south-eastern region, while forest land remained largely unaffected. In contrast, the 2015 drought extensively impacted the grassland in the north-western region. These variations in drought impact across different spatial and land types may be attributed to the regional climate variability and the inherent resilience of different ecosystems. Overall, the early 21st century was a high-incidence period of extreme and severe droughts at LP, and the frequency and intensity of droughts during the growing season at LP have weakened in the past 10 years (2012–2022) compared to the early 21st century.

From the interannual variation of the GSHI for different land use types, there are significant differences in the growing season drought conditions of LP vegetation among different land types. As shown in Figure 6c, the GSHI of forest is generally higher than that of other land types, and the interannual fluctuations are relatively small, indicating that forest vegetation has stronger resilience to drought. In contrast, the GSHI of shrubland and grassland fluctuates more drastically, with higher drought sensitivity (Figure 6d,e). The GSHI of cropland is more stable, likely benefiting from human interventions such as irrigation (Figure 6f). The increasing trend in the GSHI values of various vegetation types in recent years may be closely related to the implementation of ecological restoration projects, such as the Grain for Green Program, and the adoption of drought-resistant agricultural practices. The conversion of cropland to forest and grassland, along with improved agricultural drought mitigation measures, could have contributed to the overall enhancement of vegetation health and drought resilience in the LP region.

4.3.3. Drought Frequency

The frequency of vegetation drought at LP from 2003 to 2022, calculated using the GSHI, is shown in Figure 7a,b. Overall, the frequency of drought occurrence during the growing season at LP region exhibits distinct spatial heterogeneity, with drought frequency gradually decreasing in recent years. During the period from 2003 to 2012, mild drought showed a widespread spatial distribution throughout the LP region (Figure 7a). The average frequency of mild drought was 25% from 2003 to 2007 and 16% from 2008 to 2012. Subsequently, during the period from 2013 to 2022, mild drought at LP region was alleviated over a wide-scale area (most notably in the central region), with the overall frequency of mild drought decreasing to 5% by 2018–2022. This may be related to the large-scale ecological restoration and soil and water conservation efforts at LP region in recent years.

The distribution of moderate and above drought exhibits similar spatio-temporal patterns to mild drought (Figure 7b). In the early 21st century, areas with frequent drought occurrences were predominantly found in the north-western part of the LP. This region is part of the arid and semi-arid climate zones, characterized by limited precipitation, high evaporation, and a fragile ecological environment. The frequency of moderate and above drought often exceeds 40%. In contrast, the southeastern part of the LP region has abundant precipitation and better vegetation coverage, with the frequency of moderate and above drought being relatively low, mostly below 20%. In recent years, the frequency of moderate and above drought has significantly improved, decreasing from an average frequency of 17% in 2003–2007 to 2% in 2017–2022. However, it should be highlighted that in areas around towns and cities at LP, the frequency of drought occurrences remains relatively high. During the growing season, these areas still face a considerable risk of vegetation water shortage.

4.3.4. Drought Trend

Based on the Sen+MK method, this study further analyzed the spatial trend of the GSHI (Figure 7c,d). From the GSHI change rate, the vegetation growing-season drought in the LP region has generally shown a mitigating trend (β > 0) over the past two decades. Approximately 65% of the region has shown an increasing trend in the GSHI, indicating that the vegetation growth conditions have improved in most parts of the LP. The areas with higher GSHI increase rates are mainly concentrated in the SW, NE, and central regions of the LP, especially in the northern Shaanxi Plain, where the drought mitigation trend is more significant. In contrast, the areas with a decreasing trend in the GSHI account for about 10% of the total study area (Figure 7d), mainly distributed in the Guanzhong Plain and the Hetao Plain. These areas are mainly used for crop cultivation, and the decrease in the GSHI may be related to regional land use changes and unreasonable agricultural activities [53].

In addition, a decreasing trend in the GSHI was also noted in areas surrounding cities including Yinchuan, Taiyuan, and Xining. This is mainly due to the increase in impervious surfaces such as roads and buildings during urban expansion, which leads to a decrease in groundwater resources in the surrounding areas and may cause long-term drought. Moreover, urban activities (such as transportation and industrial emissions) can lead to an increase in greenhouse gases, which can affect local climate patterns and may result in reduced rainfall or uneven precipitation distribution, making drought phenomena more frequent.

In terms of change types, about 32% of the LP region did not pass the significance test for the GSHI change trend. These areas are scattered around the northwestern and southeastern parts of the LP, surrounding towns and cities. This may be related to the large interannual fluctuations of drought, indicating that high-intensity human activities and urban heat island effects in urban areas may also be important sources of uncertainty in regional drought change trends [54].

4.4. Driving Mechanisms of Drought During the Growing Season

4.4.1. Optimal Discretization Results of Detection Factors

Using a five-year sliding time window, the GSHI of relatively dry years (2005, 2009, 2015, and 2019) within each window was selected as the dependent variable. Meteorological factors, topography, underlying surface, and human activities were selected as detection factors to explore the driving mechanisms of drought at different time scales. In this study, continuous variables were discretized using three methods: natural breaks (nb), quantile (qt), and standard deviation (sd), with the number of groups ranging from 5 to 9 to calculate the q-value. The study area was divided into a regular grid with a cell size of 5 km × 5 km, and zonal statistics were performed. The optimal discretization methods and classification numbers for all continuous factors are shown in

Table 3. Explanation power of different factors on the GSHI.

Category	Detection Factor	2005		2009		2015		2019
		Method	Classes	Method	Classes	Method	Classes	Method	Classes
Meteorology	Precipitation	sd	6	sd	6	sd	6	sd	6
	Temperature	qt	9	sd	8	nb	8	sd	9
	Sunshine hours	sd	9	sd	8	sd	9	sd	9
Topography	Elevation	sd	6	sd	6	sd	6	sd	6
	Slope	sd	9	sd	9	sd	9	sd	9
	Aspect	-	-	-	-	-	-	-	-
Underlying surface	Soil type	-	-	-	-	-	-	-	-
	Sand content	sd	9	sd	9	sd	9	qt	9
	Clay content	nb	9	nb	9	nb	9	nb	8
Human activities	Human footprint	sd	6	sd	6	sd	6	sd	6
	Distance to city	sd	9	sd	9	sd	9	sd	9
	Carbon emissions	qt	7	qt	8	qt	8	qt	9

.

4.4.2. Single-Factor Detection

The detection results (Figure 8) show that precipitation, soil type, and elevation are the three dominant factors influencing growing season drought on the LP, with their explanatory power generally higher than other factors. Among the meteorological factors, precipitation has the strongest explanatory power, with an average q-value reaching 0.36, indicating that moisture conditions dominate the regional drought pattern. The influence of sunshine hours and temperature is secondary, with q-values of around 0.20. Among the topographic factors, the impact of elevation is more significant, with q-values between 0.2 and 0.3, followed by slope, with q-values of around 0.17; the impact of aspect on drought is relatively weak. For the underlying surface factors, the influence of soil type is relatively significant, with an average q-value around 0.30. The average q-values for soil texture are 0.12 (sand content) and 0.17 (clay content), respectively. In addition, the explanatory power of human activity factors is relatively weak in the early period but shows an increasing trend year by year. For example, the q-value of distance to cities increased from 0.05 in 2005 to 0.18 in 2019, indicating that the influence of human activities on local drought processes is becoming increasingly prominent, which may be related to the accelerated urbanization and industrialization in the LP region in recent years.

Overall, the interaction q-values of most factor combinations are higher than the single factor q-values, and there are significant enhancement effects among different factors. Among them, the interaction between precipitation and other factors is the most significant. The combinations of precipitation with temperature, elevation, and land use type show non-linear enhancement effects across most years, with interaction q-values of around 0.5. This indicates that the matching of water and heat conditions, the synergistic distribution of precipitation and topography, and the coupling of hydrothermal combinations with underlying surface types are the key factors leading to regional drought. In addition, the explanatory power of the combinations of sunshine radiation duration with temperature and precipitation is also high, with q-values of around 0.4. This may be because solar radiation affects regional water evaporation and vegetation transpiration by influencing the surface energy balance, acting on the drought process alongside other meteorological elements.

From the perspective of temporal dynamics, the interaction effects of some factor combinations differ over the years. For example, the interaction between temperature and land use type and between precipitation and elevation was significantly enhanced from 2005 to 2015 but weakened in 2020. This may be related to human activity interventions such as land use changes and the implementation of vegetation restoration projects in different periods. In contrast, the interaction effects among meteorological factors such as precipitation with temperature and wind speed are relatively stable, indicating that the impact of climate change on regional drought patterns is long-term and continuous. It is worth noting that the interaction effects of socio-economic factors such as population density and nighttime light composite with natural factors are generally weak, which may be because socio-economic activities are mainly limited to local areas such as towns, while meteorological and topographical factors dominate the regional-scale drought pattern.

4.4.3. Multi-Factor Interactions

The interaction detection results (Figure 9) show that most factors exhibit non-linear enhancement or bi-factor enhancement. Among them, the interaction between meteorological factors is the strongest, especially the interaction q-value between precipitation and temperature, which is generally above 0.6, indicating that the combination of water and heat conditions is the key to influencing drought. The interaction between precipitation and sunshine duration is also relatively significant, with q-values of around 0.5, reflecting the influence of radiation conditions on evapotranspiration and water deficit. The interaction between topographic factors is secondary, with q-values of around 0.48 for elevation and slope, indicating that topography plays an important role in regulating water redistribution. The interaction between elevation and aspect is also relatively evident in local regions, with q-values of around 0.3. There is a widespread non-linear enhancement effect between the underlying surface factors and meteorological factors, with a higher interaction intensity between soil type and precipitation and temperature, with q-values of around 0.58. This may be because soil texture affects soil infiltration and water storage processes, thereby regulating the utilization efficiency of precipitation and evapotranspiration conditions. The overall interaction between human activity factors and other factors is relatively weak but shows an increasing trend year by year, indicating that the influence of human activities on the drought pattern is becoming increasingly prominent.

5. Discussion

5.1. Analysis of the GSHI’s Performance and Applicability

The GSHI proposed in this study takes into account the spatio-temporal matching characteristics of vegetation growth rhythms and extracts the average VHI within the growing season window for each pixel. We selected two relatively severe drought years (2005 and 2015) and compared different VHI and VCI weighting schemes (fixed growing season and annual) to investigate the differences in the correlation between various indices and the NPPd, which directly measures vegetation growth status (Figure 10). The results showed that, in both 2005 and 2015, the correlation between the GSHI based on pixel phenology and the NPPd was higher than that of the VHI and the VCI based on fixed growing seasons and annual averages, indicating that the GSHI can more accurately reflect the impact of drought on vegetation. At the same time, by integrating EVI and LST information, the VHI can reflect vegetation drought conditions more timely and comprehensively than a single vegetation index (such as VCI), offsetting to some extent the lag effect of vegetation response to drought. The GSHI further introduces vegetation phenology information on the basis of the VHI, compensating for the shortcomings of fixed time windows and making drought assessment more targeted and ecologically meaningful.

Furthermore, we conducted a correlation analysis using the meteorological drought index SPEI, a widely used multi-scalar drought index. The SPEI’s ability to capture the impact of both precipitation deficits and potential evapotranspiration on drought conditions makes it a suitable reference for evaluating the performance of vegetation drought indices in detecting drought events [55]. Considering that this study focuses on growing season drought, we selected the mean SPEI-6 values averaged over April to October for comparison, as they capture the moisture conditions during the main growth period of vegetation in the LP region. To ensure consistency, we also calculated the mean value of vegetation drought indices (VCI, VHI, TCI) for the same period. We then computed the pixel-wise Pearson’s correlation coefficients between these vegetation drought indices and SPEI-6 for each year from 2003 to 2022 and analyzed their spatial patterns of correlation strength (Figure 11).

The results show that the GSHI exhibits a strong correlation with SPEI-6 in most regions of LP overall. The correlation is particularly high in the southwestern and central parts of LP, where the vegetation is mainly natural and contiguous. In contrast, the correlation is relatively poor in the southeastern region, where crops are planted. This spatial differentiation of correlation may be related to the composition of the GSHI. One of the component indicators of the GSHI is the VCI, which compares vegetation greenness with the same period in previous years. The VCI is influenced by crop rotation, artificial irrigation, and other factors, resulting in a lower correlation in agricultural regions. This is also reflected in the comparison of VCI-SPEI and VHI-SPEI. However, when compared, the GSHI shows a higher correlation than the VCI in the agricultural areas of southern LP and high-altitude mountainous regions. This may be due to the sensitivity of the GSHI component index TCI to temperature-driven evaporation stress in farmland and altitude-modulated thermal conditions. Therefore, the complementary advantages of VCI and TCI in the GSHI (where VCI dominates in natural ecosystems, while TCI enhances sensitivity in anthropogenically managed or topographically complex areas) explain its robust performance across different landscapes. Overall, the GSHI is more suitable for characterizing the drought conditions of natural vegetation, a point also reflected in Figure 5. In regions with a high intensity of human activities, the correlation between the GSHI and meteorological drought indices may be weakened due to the influence of human management measures. This highlights the importance of considering the heterogeneity of land cover types when applying vegetation drought indices.

5.2. Vegetation Drought Change Characteristics on the LP

The LP of China has experienced significant spatio-temporal variations in growing season drought from 2003 to 2022, as revealed by the GSHI. The study highlights a gradual increase in drought frequency and intensity from the southeast to the northwest, forming a moisture gradient from humid conditions in the southeast, semi-humid in the central region, to semi-arid in the northwest. This distribution is related to the deteriorating climatic conditions and the water–heat configuration within the region. The northwestern region, in particular, faces the dual challenges of scarce precipitation and high evaporation due to its location in the arid to semi-arid climate zone, resulting in an exceptionally fragile ecosystem and a high incidence of vegetation drought. In addition, different vegetation types exhibit significant differences in their response to drought. Grasslands and shrublands are the most widely affected vegetation types, while croplands benefit from artificial irrigation measures and are relatively resistant to drought impacts. In contrast, forests demonstrate stronger drought resistance and resilience due to the integrity of their ecosystem structure and function [56,57]. Therefore, when formulating drought prevention, mitigation, and ecological restoration strategies, it is essential to consider the diversity of geographical and ecological conditions and prioritize the protection and restoration of more vulnerable ecosystems [58].

From a temporal perspective, the early 21st century, particularly 2005 and 2009, was a period of high incidence of growing season drought at the LP. This is closely related to the significant precipitation reduction events that occurred during that time. Since 2010, with the warming and humidification trend of the regional climate and the success of China’s large-scale vegetation restoration projects, the vegetation conditions at LP have significantly improved, and the frequency and intensity of droughts have weakened [59]. However, localized drought phenomena in urban areas have become increasingly severe. As urbanization accelerates, the increase in common artificial surfaces and buildings enhances the heat absorption capacity of the ground, suppressing the formation of light rainfall [60]. Furthermore, the increase in impervious surfaces reduces the infiltration of surface water, leading to a decrease in groundwater resources in surrounding areas and exacerbating extreme local drought conditions. This phenomenon poses new challenges for urban planning and management, requiring measures to mitigate the urban heat island effect and improve urban ecological resilience.

5.3. Influence Mechanism of Growing Season Drought

This study quantitatively analyzed the influence mechanism of vegetation drought at LP from the aspects of meteorology, topography, soil, and human activities. The results indicate that meteorological factors are the primary factors shaping the regional drought pattern, while elevation and soil also participate in regulating the drought process, and the influence of human activities is increasing.

The mean q-values of precipitation in the single-factor detection results are consistently the highest, indicating that moisture conditions play a dominant role in the regional drought pattern. This is because precipitation is the main water input for soil and vegetation in the semi-arid LP region. The amount and temporal distribution of precipitation determine the soil water storage and water availability for plant growth, thereby directly influencing the occurrence and severity of drought [61]. In contrast, the influence of temperature and sunshine duration on drought is relatively weaker in the single-factor detection results. However, the interaction detection results reveal that the combination of water and heat conditions (precipitation and temperature) exhibits the strongest interaction effect on drought among all factor pairs, with q-values generally above 0.6. This indicates that although temperature alone may have a limited direct impact on drought, it can significantly modulate the effect of precipitation on drought severity through the water–heat coupling mechanism [62]. Higher temperatures can exacerbate drought stress by increasing evapotranspiration and accelerating soil moisture depletion, particularly when precipitation is scarce. Conversely, lower temperatures can mitigate drought severity by reducing evaporative water loss and maintaining soil moisture levels.

Topography and soil further regulate the spatial heterogeneity of drought by influencing water redistribution and vegetation growth conditions [63]. The q-values of elevation and slope factors are generally above 0.2, indicating that topography plays an important role in shaping the regional drought pattern. This is because the complex terrain of LP affects the spatial distribution and retention of water resources. In high-altitude and steep areas, precipitation is more likely to form rapid surface runoff due to thin soil layers and strong erosion, resulting in limited water retention and severe water stress for vegetation growth. In contrast, low-lying and flat areas are more conducive to water accumulation and infiltration, providing relatively favorable soil moisture conditions for vegetation growth. Additionally, soil type also has a high influence on drought. Soil properties determine the water holding capacity, infiltration rate, and water conductivity of the soil, thereby regulating the utilization and dissipation of precipitation [64]. For example, sandy soils with high permeability and low water retention are more susceptible to drought under the same precipitation conditions compared to loamy soils.

In addition to natural factors, human activities are increasingly becoming a driving force influencing regional drought [65]. However, the impact of different types of human activities varies. For example, overgrazing and urban expansion lead to surface destruction, exacerbating soil erosion and land degradation, making ecosystems more fragile [28,66]. In contrast, ecological restoration projects such as the Grain for Green Program and comprehensive management of small watersheds help improve the quality of the regional ecological environment, increase vegetation coverage, and thus alleviate drought stress [67,68]. Moreover, previous studies have indicated that vegetation responses to increasing atmospheric CO₂ concentrations may alleviate the impact of drought to some extent. To enhance the drought resistance of vegetation ecosystems at LP, it is essential to coordinate the relationship between humans and nature, optimize the layout of national land space, and carry out ecological protection and restoration according to local conditions. This requires a comprehensive consideration of the regional climate background, local environmental conditions, and the impact of human activities on the drought process [64].

5.4. Limitations and Future Works

This study conducted a useful exploration of LP growing season drought and its driving factors, incorporating vegetation phenological characteristics, but we also recognize that the current methods still have some limitations, which provide new ideas and directions for further improving regional and even global-scale vegetation drought assessment methods in the future:

(1)

The temporal uncertainty of remote sensing vegetation index time series data affects the representativeness of the GSHI. To reduce the influence of clouds, atmosphere, and other factors, commonly used vegetation index (such as EVI and NDVI) time series data have undergone maximum value composite (MVC) processing, assigning the vegetation index value to a multi-day composite period (such as 16 days). However, this leads to the degradation of temporal information, causing the extracted phenological parameters (such as SOS and EOS) to deviate from the actual phenological periods. Previous studies have shown that the root mean square error of this deviation can reach 10 days [69]. Future research needs to fully consider the uncertainty of vegetation index time series and explore data sources and phenology extraction algorithms with more precise temporal information to improve the reliability of vegetation drought assessment.

(2)

The analysis of drought driving factors needs to move from correlation to causality. There are complex feedback interactions between drought and vegetation, and this causal relationship involves multiple factors, requiring the use of causal relationship discovery techniques [70]. Traditional correlation analysis methods (including the OPGD used in this study) cannot determine the direction of causal connections. Clarifying the causal relationship between drought and various influencing factors and their action paths will deepen the understanding of the occurrence and development mechanisms of regional drought. This is of great value for improving the foresight of drought prediction, early warning, and risk management.

(3)

The timing and duration of drought occurrence affect the degree of vegetation damage and recovery ability, and this impact may vary with phenological stages [71]. Based on the current research, future studies can further refine the division of vegetation growth stages and set differentiated weights for the drought vulnerability of different phenological stages to construct a more refined drought risk assessment model. This will not only help improve the sensitivity and accuracy of depicting drought but also provide important references for formulating stage-specific drought regulation measures.

(4)

The spatial resolution limits the ability of the GSHI to capture localized drought drivers. While MODIS’s 500 m–1 km resolution supports regional-scale analysis, higher-resolution imagery (e.g., 10–30 m Landsat/Sentinel-2) could better resolve localized drought. However, such datasets face challenges in maintaining long-term temporal continuity and robustness due to cloud contamination and shorter observational records. Future work should prioritize balancing spatial detail with temporal consistency through advanced gap-filling algorithms or multi-sensor fusion to enhance drought monitoring precision.

6. Conclusions

This study proposes a Growing Season Health Index (GSHI) that considers vegetation phenological characteristics. The GSHI provides a new perspective and method for regional ecological health assessment by characterizing the impact of drought on vegetation during critical growth periods, it can more accurately reflect the impact of drought stress on ecosystem health. Based on this, the spatio-temporal patterns and influencing mechanisms of vegetation drought during the growing season in the LP were analyzed. The main conclusions are as follows:

(1)

The spatio-temporal differentiation of LP vegetation phenology is significant, showing good consistency with climatic gradients. At the regional scale, phenological changes are evident, with SOS earlier in the southeast and later in the northwest, and large interannual variability; the spatial distribution in EOS is mainly manifested as later in the south than in the north.

(2)

The vegetation drought during growing season at LP exhibits significant spatial differentiation characteristics, with an overall pattern of higher in the northwest than in the southeast. The early 21st century was a high-incidence period of drought, and the frequency and intensity of drought in natural vegetation areas have generally weakened after 2010, but localized drought phenomena in urban areas have become increasingly severe due to urbanization. Different vegetation types respond differently to drought, with forest ecosystems having higher drought resistance and stability than farmlands and grasslands.

(3)

The spatial differentiation of drought during the growing season is influenced by a combination of meteorological, topographic, soil, and human activity factors. Precipitation serves as a critical factor governing regional drought patterns, with the interaction of water and heat conditions exhibiting the most pronounced effect on drought intensification. Topography and soil also play essential roles by regulating water redistribution and vegetation growth, thus impacting drought patterns. In recent years, the intensity of human activities has shown an upward trend.

(4)

Incorporating vegetation phenology into drought assessment enables a more comprehensive and accurate depiction of the interannual impact of drought on vegetation, which is of great value for improving the scientific and precise nature of regional ecological protection and health assessment.

Further work is needed in refining the vegetation growth stage division, improving the GSHI, strengthening remote-sensing data quality control, and the shift of drought driving factor analysis from correlation to causality. In addition, comparative verification in different regions needs to be strengthened to further evaluate the universality and reliability of the GSHI.