1. Introduction

Drought is one of the costliest natural hazards, severely affecting the vegetation productivity of agriculture and forests (Vicente-Serrano et al., 2022). Drought affects vegetation productivity by decreasing photosynthesis, weakening plant defenses against pathogens, and causing widespread mortality (Allen et al., 2015; Vicca et al., 2016). However, predicting vegetation responses to drought severity remains challenging, as sensitivity can vary significantly depending on several factors (e.g., timing of the drought, vegetation type, and regional characteristics) (Cosgrove and Loucks, 2015; Li et al., 2023). One key goal of drought research is identifying additional factors (e.g., bedrock type and dry spell length) that affect the drought sensitivity of vegetation. In addition, a better understanding of how vegetation responds to drought requires quantifying the interactions between these factors.

Bedrock properties exert an important influence on the drought sensitivity of vegetation. Bedrock determines the physical and chemical properties of a regolith, which in turn, control moisture dynamics such as water holding capacity, as well as infiltration and percolation (Brantley et al., 2023; Hahm et al., 2014). This inherent geological heterogeneity creates a fundamental template for spatial variation in drought sensitivity, as vegetation in areas with different bedrock properties will experience distinct water limitations during dry periods (White et al., 2001; Fityus et al., 2015). For instance, karst landscapes are underlain by diverse carbonate bedrock types, primarily limestone and dolomite, which exhibit markedly different weathering properties and hydrogeological behaviors (Ford and Williams, 2007). Limestone and dolomite are highly soluble in water, leaving only 1%–3% insoluble residues as weathering byproducts (Liu et al., 2005). This process strongly restricts the water storage capacity. Moreover, carbonate bedrock dissolution generates extensive secondary porosity via fractures and conduits, producing a highly permeable subsurface environment, with studies revealing up to 75% of rainfall rapidly percolating to the depth to contribute to subsurface runoff (Veihe, 2002; Dai et al., 2018; Fang et al., 2023). In limestone, the higher solubility and conduit porosity lead to lower regolith water retention compared to dolomite and clastic rocks.

Karst vegetation is sensitive to even short-term precipitation deficits. The thin regolith of the karst regions cannot retain much soil water during inter-rainfall periods, unlike non-karst regions in which soils buffer irregular rainfall. Thus, temporal rainfall clustering (e.g., 10-day dry spells) can trigger disproportionate drought stress. Previous studies have found that in a variety of ecosystem types sensitivity to rainfall timing depends not only on precipitation totals but also on their distribution. In karst vegetation regions, vegetation activity has been shown to decline in regions after a short dry spell (i.e., 5 days) (Jiang et al., 2020). However, few studies have focused on the impact of rainfall distribution on vegetation at this timescale.

The karst regions of Guizhou Province in Southwest China are some of the largest in the world. These regions consist of a diverse range of bedrock types, including dolomite, limestone, clastic rocks, and igneous rocks. The vegetation in these karst regions is highly sensitive to drought stress, as carbonate bedrock properties amplify drought sensitivity (Wang et al., 2021). For instance, in 2010, a drought in Southwest China decreased vegetation productivity nearly three times more in karst regions than in adjacent non-karst regions (Lin et al., 2015). However, it remains poorly understood whether and how vegetation drought sensitivity varies across different bedrock types within karst regions. Furthermore, it is unclear if this bedrock-specific sensitivity is consistent across vegetation types (e.g., forest vs. grassland) and throughout the growing season.

This study, therefore, aims to address this gap by investigating the spatial heterogeneity of vegetation drought sensitivity at a regional scale across Guizhou Province, and specifically, to quantify the role of bedrock type in driving this spatial variability. Furthermore, we examine how this bedrock-induced spatial heterogeneity manifests during different phases of the growing season (early vs. late) to understand its temporal dynamics. This study investigates the role of bedrock-associated features that influence vegetation–drought relationships in the Guizhou Province, Southwest China, which represents the largest karst region. This region is characterized by a regionally similar climate that occurs over a diverse range of bedrock types, including dolomite, limestone, clastic rocks, and igneous rocks. Based on the principle that limestone's higher solubility and conduit porosity lead to lower regolith water retention compared to dolomite and clastic rocks, we hypothesized that vegetation drought sensitivity would be highest on limestone. To test this hypothesis and evaluate the broader utility of bedrock information, we first quantified and compared vegetation drought sensitivity across bedrock types (limestone, dolomite, and clastic rocks) to identify differential responses, and then employed hierarchical models to evaluate the predictive utility of incorporating bedrock information. Our study was designed to answer two specific questions: 1) Is vegetation drought sensitivity significantly higher in limestone regions than in dolomite and clastic regions? and 2) Can the incorporation of bedrock types enhance the predictive power of models relating vegetation activity to drought?

2. Methods and materials 2.1. Study area

Guizhou Province in Southwest China (Fig. 1; 24°30′–29°13′N, 103°01′–109°30′E) experiences a typical sub-tropical humid monsoon climate. Approximately 38% of the province's total area comprises clastic rock exposure, whereas approximately 62% is the karst bedrock (17.4% homogenous limestone, 13.1% homogenous dolomite, and 31.5% other carbonate rock assemblages) (Wang et al., 2021). The distribution of forest and shrubland are relatively dispersed and widespread across the entire province; however, species composition does not substantially differ among lithological regions. Grasslands are mostly concentrated in the eastern and southern parts of Guizhou Province (Huang and Tu, 1983; Zhang et al., 2015).

2.2. Data sources and preprocessing

We selected (MODIS) Normalized Difference Vegetation Index (NDVI) (MOD13Q1, v.006) to examine the vegetation activity affected by drought. The NDVI dataset is available on the Geospatial Data Cloud (https://www.gscloud.cn/). These MODIS datasets have a 250-m spatial resolution and a 16-day temporal resolution from 2000 to 2020. We used the Self-calibrating Palmer Drought Severity Index (scPDSI) to quantify drought severity (Palmer, 1965; Wells et al., 2004; Zhang et al., 2019), because scPDSI can capture not only meteorological drought but also reflect soil moisture conditions effectively (Mika et al., 2005), making it particularly suitable for detecting drought intensity in karst regions We created scPDSI datasets using daily meteorological data from Guizhou Province rather than from publicly available data (e.g., SPEIbase and Climatic Research Unit Time Series, CRU TS), whose temporal resolution (i.e., one month) does not match that of the vegetation indices. Meteorological data was obtained from 2000 to 2020 from the National Meteorological Science Data Center of China (http://data.cma.cn/), which included daily temperature, precipitation, and solar radiation measures. Data was retained from 33 meteorological stations evenly distributed across the Guizhou Province. Data with missing observations were excluded.

We derived the land cover map for Guizhou Province from the 1:1, 000, 000 digital vegetation map of China (Editorial Board of Vegetation Map of China, 2007) and identified three primary land cover categories: forests (24%), shrublands (25%), and grasslands (20%). The land cover map was resampled to a 250-m spatial resolution to match NDVI. The spatial distribution of these main land cover types is shown in Fig. 1.

To evaluate topsoil water hold capacity, we utilized five soil properties sourced from the Harmonized World Soil Database (Nachtergaele et al., 2010), which integrates soil information from several sources (e.g., the 1:1, 000, 000 Soil Map of China): soil water regime, soil organic matter, soil clay content, soil sand content, and soil silt content. These variables have been widely validated for their ability to reflect soil water-holding capacity (Hudson, 1994; Emerson, 1995; Verheijen et al., 2019), as they capture distinct water retention mechanisms. A statistical check for multicollinearity was performed by computing pairwise Pearson correlation coefficients. Results confirmed that for all variable pairs |r| < 0.7 (Dormann et al., 2013), supporting their treatment as independent predictors in our hierarchical models. This soil database was compiled and mapped using interpolation methods to achieve a spatial resolution of 1 km, which was resampled to achieve a 250-m spatial resolution to match other datasets.

Accurately determining the lithological distribution is critical for understanding the role of bedrock on vegetation, however various classification techniques often yield discrepancies in determining bedrock type. For example, estimates based on the mudstone content within carbonate rock formations suggested clastic rock substrates in southwestern Guizhou Province (Wang et al., 2004), whereas estimates based on the chemical compositional analysis of the minerals classified the same region as dolomite interbedded with shale (China Geological Survey; https://www.cgs.gov.cn/). This discrepancy arises from the region's complex lithology, where dolomite, shale, and clastic rocks are intermixed, making precise lithological identification difficult. To address this issue, we classified regions where estimates from these two approaches were inconsistent as "uncertain bedrock" (Fig. 1) and excluded them from further analysis. This approach aims to improve the reliability of the bedrock distribution data by minimizing the impact of classification biases. The lithological data from Wang et al. (2004) were originally in vector format. To ensure compatibility with raster datasets such as NDVI, scPDSI, and land cover, we converted the data using ArcGIS as follows: the vector bedrock map was converted to a raster format with a 250-m spatial resolution (consistent with MODIS NDVI data), and appropriate resampling was performed to assign bedrock types to each grid cell. Additionally, all raster datasets were spatially aligned, adopting the same coordinate system, cell size, and projection parameters, ensuring perfect spatial registration for subsequent pixel-level analyses of all variables.

2.3. Data analysis 2.3.1. Quantifying drought using the Self-Calibrating Palmer Drought Severity Index (scPDSI)

Drought severity was quantified using the Self-Calibrating Palmer Drought Severity Index (scPDSI). The computation of scPDSI follows a structured multi-stage calibration process (Wells et al., 2004), utilizing monthly precipitation (P), monthly mean air temperature (T), and soil available water capacity (AWC) as primary input data. The calculation begins with a water balance model to determine monthly moisture deficits. The procedure then advances through dual calibration pathways: one derives localized duration factors by analyzing moisture anomaly indices from historical extreme dry and wet periods, while the other establishes calibration references for climate characteristic parameters using percentile distributions from a preliminary drought index series. The fully calibrated parameters are subsequently integrated into a recursive computational framework to generate the final scPDSI values. This comprehensive calibration methodology ensures spatial comparability of the index across diverse climatic regions. The scPDSI was computed using the SPEI R package (Beguería and Vicente-Serrano, 2023). Calculations are provided in Supplementary Text S3. scPDSI values below −2 are interpreted as moderate to severe drought, whereas values above −0.1 indicate predominantly normal to wet conditions (Mika et al., 2005).

2.3.2. Quantifying vegetation activity during the growing season

We specifically focused on the drought sensitivity of the vegetation activity during the growing season, where drought can severely impact vegetation productivity (Ji and Peters, 2003). We first defined the range of the growing season in Guizhou Province to provide a more meaningful understanding of the impact of drought on vegetation dynamics. We utilized the Harmonic Analysis of Time Series method (Menenti et al., 1993) to reconstruct the NDVI sequence, and the growing season range was derived using the maximum ratio change (Celis-Hernandez et al., 2022). We considered the early growing season from the start of the growing season to the month nearest to the summer peak (April to July) and the late growing season as the months from the end of the summer peak to the end of the growing season (August to November, see supplementary materials for additional details).

Subsequently, we quantified the vegetation activity during the entire growing season, as well as the early and late growing seasons, using a remotely sensed indicator of vegetation activity that was derived from NDVI (Piedallu et al., 2019). In particular, we calculated the vegetation activity as the sum of the NDVIs calculated for a fixed period of synthesized NDVI from the beginning of the growing season phase to its end using the following equation:

\mathrm{y}_{\mathrm{i}}=\sum\limits_{\mathrm{i}}^{\mathrm{n}} \mathrm{NDVI}_{\mathrm{i}}

(1)

where i indicates the month of the growing season and n indicates the duration (the number of months calculated for vegetation activity). Higher values indicate higher vegetation activity. The spatial resolution of vegetation activity indices are the same as the monthly NDVI. To ensure temporal consistency in the subsequent correlation analysis, the scPDSI values were averaged over the corresponding growing season periods (early, late, and whole) to match the integrated nature of the vegetation activity index. This approach ensures that each data point in the correlation represents a comparable integrated seasonal condition. Previous work has shown that a period of five days without rainfall substantially affects vegetation growth in karst regions (Jiang et al., 2020). Thus, we defined dry spell duration as five consecutive days without precipitation during the growing season.

2.3.3. Evaluating drought sensitivity of vegetation activity

To investigate the vegetation response to drought, this study quantified it using the Slope_scPDSI-VA between vegetation activity and the scPDSI. This slope directly reflects the rate of change in vegetation activity per unit change in drought intensity. By focusing on this metric, the spatiotemporal variation patterns of vegetation drought sensitivity can be effectively revealed (Yang et al., 2016).

To ensure the ecological relevance and statistical robustness of our calculations, we filtered Slope_scPDSI-VA scores based on their significance and direction. Pixels with that did not have a statistically significant (p > 0.05) Slope_scPDSI-VA were excluded from further analysis. Because the study region is predominantly limited by drought rather than waterlogging, pixels exhibiting a statistically significant negative Slope_scPDSI-VA (p < 0.05) were also considered potential artifacts (e.g., from cloud contamination, topographic effects, or model residuals) and were consequently masked out (assigned a value of zero). Thus, the final trend maps and subsequent analyses only represent pixels with statistically significant positive slopes, indicating a credible greening response to increased moisture availability.

Statistical comparisons of Slope_scPDSI-VA among different groups (i.e., bedrock types, vegetation types, and growing season stages) were performed using the non-parametric Wilcoxon rank-sum test, as the data did not meet the assumptions of normality.

Slope_scPDSI-VA is characterized by a large number of zeros and a bounded range. To accommodate these data characteristics, we employed zero-inflated beta regression models for statistical analysis (Ospina and Ferrari, 2011). To systematically evaluate how the interactive effects of bedrock type, vegetation type, and growing season stage influence the slope, we constructed and compared four different models. All models were fitted using maximum likelihood estimation via the glmmTMB package, directly testing whether the interactive effects of the predictors provide a statistically significant improvement in explaining the variation in slope values.

To verify the potential spatial association between bedrock types and vegetation types across the study area, a Chi-square test of independence was conducted. Specifically, a contingency table of their co-occurrence frequencies was constructed, and the significance of the association was determined by comparing observed and expected frequencies.

2.3.4. Hierarchical model for drought sensitivity of vegetation activity

To compare the relative importance of bedrock type in explaining the observed spatial variation in vegetation activity with other variables, we used a hierarchical modeling approach to model the modulation of vegetation's drought sensitivity by bedrock types, allowing slopes and intercepts to differ among the regions of bedrock types, which can capture the various drought sensitivities. This model consists of two levels; the first level is a linear model used to estimate the vegetation activity response to drought.

y=β0i+β1i∗xi

(2)

where y_i is vegetation activity, x_i represents other predictor variables, β_0i is the intercept for bedrock region i, and β_1i is the slope for bedrock region i. At the second level, we model the coefficients from the first level as functions of the bedrock types:

β0i=γ01+γ01∗BTi

(3)

β1i=γ10+γ11∗BTi

(4)

where BT_i represents the bedrock type i, γ01 and γ10 are the intercepts of the second level model, and γ01 and γ11 stand for the slopes of bedrock type. We emphasize that the goal of our hierarchical modeling was inferential, aimed at quantifying the relative contribution of bedrock type and dry spell duration to vegetation activity, rather than predictive. Consequently, we adopted a model comparison approach using the entire dataset. We compared the performance of this hierarchical model against that of the model only including scPDSI, models combining scPDSI with dry spell duration, and models integrating scPDSI with soil variables to further illustrate the importance of bedrock. The selected soil variables are widely recognized to amplify or mitigate the impact of drought on vegetation activity (Shao et al., 2018), while dry spell duration is a key factor characterizing drought properties. We based model selection on the Akaike Information Criterion (AIC), considering the model with the lowest AIC value as the most parsimonious and optimal for predicting vegetation activity (Letten et al., 2013; Piedallu et al., 2019).

3. Results 3.1. Temporal and spatial patterns of scPDSI and vegetation activity

Drought severity did not differ significantly across bedrock types (Figs. 2 and 3). Moreover, drought severity did not differ significantly in the study region from early to late growing seasons. In Guizhou Province, moderate to severe drought conditions occurred around 12% of the entire study period. Approximately 65% of the recorded scPDSI values exceeded −0.1, indicating predominantly normal to wet conditions.

Vegetation activity differed significantly between bedrock types (Fig. 3). Specifically, vegetation activity was significantly higher in clastic regions than in limestone and dolomite regions during the early, late, and whole growing seasons, with this difference being particularly pronounced during the late growing season. Vegetation activity was lowest across the entire growing season and during the early growing season in limestone regions, but was slightly higher in dolomite regions in the late growing season.

3.2. The difference of Slope_scPDSI-VA across bedrock types and sub-seasons

When all vegetation types were analyzed together, the drought sensitivity of vegetation (i.e., a positive scPDSI-vegetation activity correlation) was greater in limestone regions than in other bedrock types (Fig. 4). When vegetation types were considered separately, patterns of drought sensitivity were more complex. In grasslands, the response of vegetation activity to drought did not differ significantly between limestone and other bedrock types in the late growing season. However, drought sensitivity of grasslands was greater at dolomite than at clastic regions. The proportion of land that showed elevated drought sensitivity differed significantly among growing seasons, bedrock types, and vegetation types (Fig. 5 and Table S2). A key finding was that the bedrock type with the fewest areas sensitive to drought was dolomite bedrock.

All three bedrock types demonstrated higher Slope_scPDSI-VA in the early growing season than in the later part (Fig. 4). Notably, the most pronounced decline in sensitivity was observed for limestone grasslands, where the early-season Slope_scPDSI-VA was 2.83 times that of the late season (Fig. 4). This was followed by limestone shrublands and dolomite grasslands, with early-season values 2.27 and 2.60 times higher than their respective late-season values. We verified the distribution relationship between bedrock and vegetation types to rule out potential confounding effects.

Zero-inflated beta regression analysis confirm that both the second-order and third-order interactions among bedrock type, vegetation type, and growing season exert significant effects on the draught sensitivity. All models incorporating different predictor combinations were statistically significant, demonstrating that the interactive effects jointly contribute in a statistically significant and ecologically meaningful way to the variation in drought sensitivity (Table 1). A Chi-square test of independence showed no significant association between bedrock type and vegetation type distribution (χ² = 2.026, df = 4, p = 0.731).

3.3. Performance of the hierarchical model for predicting vegetation activity

Hierarchical model comparisons consistently indicated that changes in vegetation activity in response to drought can largely be explained by bedrock type (Table 2). The integrated model, which combined both bedrock type and dry spell duration, was unequivocally identified as the optimal model (ΔAIC = 0) across all vegetation types and growing seasons.

Critically, the model containing only bedrock type uniformly ranked as the second-best performing model across all scenarios—forests, grasslands, shrublands, and for the entire landscape (All)—in every growing season period (early, late, and whole). This model consistently and significantly outperformed the model containing only dry spell duration. For instance, in the late growing season across all vegetation, the AIC for the bedrock model (−3, 486, 842) was substantially lower than that (−3, 351, 488) for the dry spell duration model. This confirms that bedrock type is primary spatial determinant of drought sensitivity patterns.

The incorporation of dry spell duration, while secondary to bedrock type, provided a significant and systematic improvement in model fit. The integrated model consistently achieved a lower AIC than the bedrock-only model. The magnitude of this improvement (ΔAIC) was notably larger during the late growing season (e.g., ΔAIC = 63, 692 for 'All' vegetation) compared to the early season (ΔAIC = 37, 760), demonstrating that the explanatory power of dry spell duration is more pronounced as the growing season advances.

4. Discussion

Our results demonstrate that the sensitivity of vegetation to drought varies across bedrock types in the extensive karst landscape of Guizhou Province. We found that the vegetation most sensitive to drought was that of limestone regions, particularly during the early-growing season. This finding challenges the adequacy of commonly used surface-level variables, such as climate and topsoil attributes, in capturing the complexity of drought impacts in karst-dominated landscapes. Notably, we significantly improved the predictive power for vegetation activity under drought conditions by integrating bedrock type and dry spell duration into a hierarchical model, underscoring the necessity of accounting for lithological controls in ecological drought assessment.

4.1. Bedrock effects

Our finding that vegetation is more sensitive to drought in limestone regions than in dolomite regions can be attributed to geochemical and structural factors. Limestone undergoes rapid dissolution, forming highly interconnected conduit-like fractures (e.g., caves and underground rivers), leading to inefficient water storage and faster drainage. Consequently, vegetation in limestone terrains experiences more severe drought stress due to limited water availability during dry periods. In dolomite, chemical weathering (i.e., Mg²⁺ released during dolomite hydrolysis) promotes the formation of secondary clay minerals (e.g., montmorillonite and chlorite), which in turn improve soil water retention through their high specific surface area and cation exchange capacity (Righi and Meunier, 1995). In addition, dolomite generally develops short, poorly connected fractures, whereas weathered clay minerals fill these fractures, producing nano-to micro-scale pores that retain water through capillary action (Leeder, 2012). These insights highlight the crucial role of bedrock geochemistry and fracture networks in mediating ecosystem resilience to aridity.

We also found that the drought sensitivity of vegetation varies greatly across dolomite regions. This pattern may be attributed to the substantial environmental heterogeneity inherent to dolomite landscapes (Hosseinzadeh and Tavakoli, 2024). Dolomite-derived soils show greater spatial variability in water-holding capacity due to differential weathering patterns and fracture distributions compared to clastic rock regions. This underlying heterogeneity results in considerable local-scale variations in the sensitivity of vegetation to drought. In contrast, the relatively uniform regolith in clastic rock areas promotes more consistent drought sensitivity responses across the landscape. These results underscore how bedrock-mediated soil heterogeneity can decouple large-scale climatic drivers from localized ecosystem responses, thereby complicating predictions of drought impacts in karst environments.

Our finding that bedrock type strongly influences vegetation sensitivity to drought extends far beyond our study region. It explains the perplexing spatial heterogeneity in drought impacts observed across many lithologically complex landscapes worldwide. For instance, the contrasting drought responses between vegetation on porous lavas versus compact ash beds in volcanic landscape (Páez-Bimos et al., 2022), or between deep-weathered granite and shallow soils on metamorphic shields, can be reinterpreted through the lens of bedrock-mediated water storage and drainage. By integrating bedrock properties into the drought vulnerability framework, we move from a largely two-dimensional (climate-soil) view to a three-dimensional one that acknowledges the critical subsurface template controlling water availability. This revised paradigm is essential for improving the predictive accuracy of ecological and hydrological models, which often fail to capture fine-scale drought responses precisely because they omit this foundational geological driver. Future efforts to map global ecosystem vulnerability to climate change must, therefore, incorporate lithological data to account for this hidden heterogeneity in water storage and drainage.

4.2. Re-evaluating vegetation-type drought sensitivity in the context of bedrock geology

It is well established that vegetation type drought sensitivity is hierarchical, typically with grasslands being most vulnerable and forests most resilient, due to differences in root depth and water-use strategies (Medrano et al., 2009; Liao et al., 2025). Our results strongly corroborate this global pattern within the karst region of Guizhou. However, the novel insight from our study is that this canonical hierarchy is not fixed but is powerfully modulated by the underlying bedrock. We found that the differences in sensitivity between, for instance, grasslands and forests were most pronounced on drought-prone limestone bedrock, but nearly negligible on water-retentive dolomite. This indicates that the intrinsic drought tolerance of a vegetation type can be superseded by the extrinsic water-supply capacity of the bedrock substrate. Consequently, models predicting vegetation shifts under climate change that rely solely on vegetation functional types may be critically inaccurate in geologically complex landscapes. Our findings argue for a context-dependent view of plant hydraulic strategies, where the bedrock template dictates the actual expression of drought vulnerability.

4.3. Season effects

Our study also highlights the critical role of dry spell duration on vegetation drought sensitivity. Specifically, the substantial influence of dry spell duration on drought severity arises from rapid drainage via karst conduits and fractures. The high permeability of karst systems results in faster depletion of shallow water reserves compared to other landscapes during dry periods, making vegetation especially sensitive to even short-term precipitation deficits. This underscores the crucial yet overlooked role of rainfall frequency in shaping vegetation drought sensitivity. Previous studies have focused on drought intensity (Chiang et al., 2021; Vicente-Serrano et al., 2022), however, the influence of changed rainfall timing—especially prolonged dry spells between precipitation events—has received far less attention. Even moderate droughts can trigger disproportionate reductions in productivity when they coincide with extended rain-free intervals during key phenological stages (e.g., leaf flushing or flowering) in karst regions, where water storage capacity is inherently limited. Furthermore, alterations in rainfall seasonality, such as delayed wet season onset, may enhance drought stress in karst ecosystems more severely than in non-karst areas, even under similar annual precipitation totals. This sensitivity underscores the importance of considering temporal distribution metrics rather than just cumulative amounts when assessing drought risks in the karst regions. Our results confirm that forests, shrublands, and grasslands across all bedrock types show the highest sensitivity to drought during the early growing season, with sensitivity up to three times higher than in the late growing season, which is similar to previous findings (Zhou et al., 2019; Wang et al., 2021). This heightened early-season sensitivity occurs because vegetation growth is most active in spring and early summer (Yan et al., 2013), where drought-induced water limitation substantially inhibits leaf development and carbon assimilation (Song et al., 2019). This effect may be amplified by legacy effects from the dry season (December–March), which accounts for only ~10% of annual precipitation.

Vegetation has been shown to be less sensitive to drought during summer and autumn (Foster et al., 2014; D'Orangeville et al., 2018; Van Kampen et al., 2022). This pattern can be explained by several factors: water-use efficiency increases later in the season (Wang et al., 2021) as root systems become more developed and capable of accessing deeper water sources, especially in karst regions where plants develop extensive root networks to exploit bedrock fissures and fractures (Cai et al., 2023). Furthermore, plants in these ecosystems show phenological adaptations that shift their growth peaks to early growing seasons when water is readily available, whereas they develop drought-resistant traits like thicker cuticles and decreased stomatal conductance as the season progresses (Zahedi et al., 2025). Additionally, the accumulated biomass and canopy closure by mid-season generate microclimates that decrease evapotranspiration demands. However, summer droughts may still impact long-term productivity by triggering early leaf senescence and decreasing carbon assimilation in subsequent years (Wu et al., 2018). Continued monitoring is warranted, given the projected intensification of summer droughts in this region (Fan et al., 2011).

4.4. Bedrock-vegetation-season synergistic effects

Our findings culminate in a central thesis: the drought vulnerability of karst ecosystems is not an arithmetic sum of individual factors but rather a product of their synergistic interactions. The more acute drought sensitivity emerges from a convergence of the more vulnerable bedrock (limestone), the more phenologically sensitive season (early growing season), and the more responsive vegetation type (grasslands). For instance, the exceptionally high sensitivity index we observed on limestone during the early growing season is disproportionately driven by the convergence of limited bedrock water storage, high atmospheric demand, and the shallow-rooted, rapidly growing strategy of grasslands. Conversely, the attenuated sensitivity in dolomite regions, particularly later in the season, represents a decoupling achieved through the synergy of superior bedrock water retention, increased plant water-use efficiency, and microclimatic buffering.

4.5. Impact on plant diversity

Although this study primarily focuses on vegetation productivity responses, the findings have important implications for plant diversity conservation. While our study focuses on the productivity of broad vegetation types, the observed bedrock-mediated divergence in drought sensitivity has important, albeit indirect, implications for plant diversity (Hahm et al., 2014; McCormick et al., 2021). The finding that forests, shrublands, and grasslands respond differently to drought depending on the underlying bedrock suggests that bedrock geology creates a template of environmental heterogeneity. This template likely influences plant community assembly and the distribution of species with varying drought tolerances. For example, the heightened vulnerability of vegetation on limestone may favor drought-adapted specialists, while the relatively buffered conditions on dolomite and clastic rocks could serve as refugia for more moisture-dependent species. Consequently, the spatial mosaic of bedrock types likely contributes to the maintenance of regional plant diversity (β-diversity) by supporting distinct species pools across different lithologies (Harrison and Inouye, 2002; Stein et al., 2014). This bedrock-driven microhabitat heterogeneity likely favors distinct species, particularly microhabitat-specialized herbs whose distribution can be missed by coarse sampling. Indeed, sampling intensity requirements differ significantly across plant growth forms, with herbs needing the highest intensity for accurate richness estimation (Ling et al., 2026). However, this very mechanism also implies a significant threat. If climate change leads to more frequent and severe droughts, as projected, the increased physiological stress on limestone bedrock could exceed the tolerance threshold of many current species, potentially leading to a loss of specialist species and a homogenization of plant communities in these areas (McKinney and Lockwood, 1999). Therefore, our results underscore the necessity to incorporate bedrock geology into biodiversity conservation strategies. Future research that directly links species-level compositional data with bedrock-dependent drought sensitivity will be critical to validate these predictions and to identify priority areas for the conservation of plant diversity in karst regions under climate change.

4.6. Limitations and future prospects

This study has limitations that arise from the resolution of available data. The use of a regional-scale bedrock classification likely obscures within-category heterogeneity, and the lack of species-specific data prevents mechanistic trait-based explanations. Importantly, such limitations would be expected to attenuate the observed contrasts between bedrock types and vegetation classes by introducing noise. The fact that we still detected strong signals suggests that the actual underlying effects of bedrock geology and vegetation type are robust and potentially even stronger than we report. Future research should focus on two key steps to build on these findings. First, using more detailed geological maps and measuring specific plant traits on the ground will help us understand the precise mechanisms behind the patterns we observed. Second, incorporating this finer-scale knowledge into ecological models will improve our ability to predict how landscapes with different bedrock will respond to future droughts.

5. Conclusion

In this study, we found that the drought sensitivity of vegetation varies substantially across bedrock types in a broad karst landscape. Our analyses indicated that vegetation in regions with limestone bedrock were especially sensitive to drought, particularly during the early part of the growing season. Although no differences were found in climatic drought stress, as measured using scPDSI, across regions with various bedrock types, the subsurface geologic features strongly modulated the response of vegetation to that stress. Our results underscore the role of lithology in shaping the spatial heterogeneity of vegetation–climate interactions. These findings are especially relevant for enhancing drought detection at regional scales. Although satellite-based soil moisture products such as GRACE (Tapley et al., 2004) provide valuable long-term data, their coarse spatial resolution limits their use for local ecosystem evaluations. In contrast, high-resolution data on bedrock types provide a more precise proxy for subsurface water retention capacity and vegetation response to stress, particularly in widespread karst landscapes that play important ecological and hydrological roles. However, broader application of this approach necessitates further regional validation. This study advances our understanding of vegetation–drought–subsurface interactions in subtropical karst landscapes. Future research should incorporate high-resolution lithological data and species-specific physiological observations to predict and assess variation in terrestrial ecosystem resilience under climate change.

In conclusion, our study highlights the need to redefine drought risk frameworks to explicitly incorporate hydroclimatic extremes and subsurface geological constraints, as their interactions fundamentally shape ecosystem resilience. Future studies should explore whether alterations in rainfall seasonality (e.g., delayed wet season onset) further exacerbate drought stress in karst ecosystems, especially given their limited water storage capacity. Continued monitoring is warranted to track how these delicate ecosystems respond to changing climate patterns.

Acknowledgments

We would like to express our sincere gratitude to Zheng Xue and Luo Chunyan for their invaluable contributions to the processing of remote sensing imagery. Additionally, we extend our thanks to Du Xinzong for his significant contributions to the analysis of meteorological data. This research was funded by The Chinese Academy of Sciences Hundred Talents Program, Category B, National Postdoctoral Programs, Yunnan Provincial General Project Fund (202401CF070060), Science and Technology Projects of Xizang Autonomous Region, China (XZ202402ZD0005), Key R & D Program of Yunnan Province (202403AC100028), and National Natural Science Foundation Regional Project (32360395, W2412149).

CRediT authorship contribution statement

Xiaona Li: Data collection, Data preprocessing, Results validation, Data analysis. Dingwu Zhang: Data collection, Data preprocessing, Results validation, Data analysis, Visualization, Chart creation, Draft writing. Yinqianxue Pan: Draft writing, Data collection and processing. Xiaogang You: Data preprocessing and analysis. Weijun Luo: Manuscript review and revision. Hongyan Liu: Manuscript review and revision. Zihan Jiang: Resource acquisition, Funding support, Project management, Final review, Methodology guidance.

Data availability statement

Our vegetation activity (NDVI), drought intensity (scPDSI), vegetation type, soil property, and bedrock type data can be accessed on the Geospatial Data Cloud (https://www.gscloud.cn/), National Meteorological Information Center of China (http://data.cma.cn/), and China Geological Survey (https://www.cgs.gov.cn/) websites, respectively. The vegetation type data are sourced from the Vegetation Map of China (2007), and the soil property data are from the Harmonized World Soil Database (Nachtergaele et al., 2010).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pld.2026.02.001.