Drought thresholds that impact vegetation reveal the divergent responses of vegetation growth to drought across China

2.1 Data

2.2 The framework for drought threshold detection

Here, we defined the drought threshold as an inflection point beyond which the response fraction of vegetation growth to droughts will change from low to high, corresponding to the abrupt loss of vegetation resistance to increasing drought stress. In our study, we used a percentile-based method to identify droughts and vegetation growth suppression, based on soil moisture data and NDVI respectively. We adopted the framework for detecting drought thresholds developed by Li et al. (2023) to assess the performances of vegetation response trajectories and evaluate the existence of drought thresholds impacting vegetation growth in China. In detail, the framework consists of four main procedures: (1) feature extraction using principal component analysis (PCA), (2) coincidence analysis between drought and vegetation response, (3) nonlinear tests for vegetation response trajectories to different levels of drought stress, and (4) detection of inflection points (i.e., drought thresholds) by piecewise linear regression.

To begin, we performed PCA on four environmental variables: annual average temperature, annual total precipitation, interannual variation of VPD, and tree cover (Mahecha et al., 2017). The first two principal components (PC1 and PC2) were selected, explaining 78.6% of the variance (Figure S1). For each grid cell, we then identified its neighboring grid cells in the two-dimensional feature space composed of PC1 and PC2. If any grid cell is located within a 3 × 3 window whose width accounts for 4% of the total length of the scores of PC1 and PC2, we regarded them as neighboring ones for the targeting grid cell because they shared the similarity in environmental conditions and vegetation covers (Figure S2). To detect the occurrence of drought and vegetation response, we derived the anomaly value of soil moisture and NDVI from the percentile of all samples in the same 3 × 3 window in the PC space.

To identify drought occurrence in Figure 1a, we first obtained the anomaly value for each target grid cell from the 10th percentile of soil moisture samples in the window. Then we identified the drought years when the soil moisture decreased below this anomaly value for the 18-year time slot of the target grid cell (Figure S2). Furthermore, we calculated the coincidence rates (r _i ) between different levels of soil moisture stress (soil moisture anomalies < 1st–50th percentiles) and negative anomalies of vegetation growth (NDVI anomalies <10th percentile) in each window (Rammig et al., 2015). The r _i value was thus calculated as the fraction of the co-occurrence of droughts and vegetation suppression in all drought events (Li et al., 2023). The subscript i represents drought detected by different percentiles, i = 1st, 2nd, …, 50th. r _i is 0 if no responses of vegetation growth to droughts occur while r _i reaches 1 if vegetation responds to all droughts.

Subsequently, we examined the nonlinear relationship between drought stress levels (i) and coincidence rates (r _i ) (Figure S3a). However, the linearity showed better fitness than the nonlinearity from the mathematical perspective in some vegetated areas because the vegetation growth was not largely affected by droughts during 2001–2018. By identifying the inflection point where the slopes of the two distinct segments had the greatest difference, we derived the percentile-based drought threshold, referred to as T _SM (Figure S3b; Toms & Lesperance, 2003). Lower T _SM values correspond to more negative soil moisture anomalies so more rarely occurred drought events featured by higher drought stress are detected, and vice versa.

To verify the robustness of the drought thresholds, we performed a shuffling test (Figure S4). Firstly, the original date of the NDVI series for each pixel was randomly shuffled 500 times. Then we detected drought thresholds based on the original soil moisture series and these 500 surrogate NDVI time series using the same method. Our null hypothesis was that drought thresholds derived from the original NDVI time series showed no difference from those derived from the surrogate time series. We reject the null hypothesis only if the drought threshold from the original NDVI time series is outside the 2.5%–97.5% (corresponding to p < .05) of the distribution for all threshold values from the surrogate NDVI time series. For grid cells with available T _SM values, we projected changes in the probability of T _SM-inferred drought occurring in China during the period 2083–2100, relative to the period 2001–2018, based on model simulation data.

In addition, we examined variation in drought thresholds among different vegetation biomes, including five natural vegetation biomes (needleleaf forest, mixed forest, broadleaf forest, shrubland, and grassland) and one managed vegetation biome (cropland). We applied the Mann–Whitney U test to examine the significance of the difference in drought thresholds among five vegetation biomes (significance level: p < .05). For croplands, we specifically compared drought thresholds across five major agricultural zones that were highly influenced by human activity: North China, the Huang-Huai Area, the Yangtze River Delta, Jianghan Plain, and the Sichuan Basin. Within each agricultural zone, we compared T _SM values across five irrigation rate ranges: 0%–20%, 20%–40%, 40%–60%, 60%–80%, and 80%–100%. It is important to note that only irrigation rate ranges with valid T _SM values are presented for each respective region. This analysis allowed us to assess how irrigation rates impacted the drought thresholds for croplands in different agricultural zones.

3.1 Drought occurrence and relevant vegetation response during 2001–2018

The average drought occurrence during 2001–2018 was 1.8 times across all vegetated areas of China, with varying occurrences across different regions. As depicted in Figure 1a, approximately 47% of the study area, including the Qinghai-Tibet Plateau, Northeast China, and northern Xinjiang, experienced fewer than two drought events during this period. Conversely, around 4% of the regions, such as the central Inner Mongolia Plateau and the western Yunnan-Kweichow Plateau, had a much more frequent occurrence of droughts (>5 times from 2001 to 2018).

However, it is important to note that not all drought events result in vegetation greenness anomalies falling below the 10th percentile threshold (Figure 1b). We found that the average coincidence rate (r _10th) between the occurrence of once-in-a-decade droughts and vegetation response in the vegetated areas of China was .23. This indicates that, on a nationwide scale, vegetation growth has a relatively low likelihood of being severely impacted by once-in-a-decade drought events. Regionally, the lower coincidence rates (r _10th < .1) were identified in 5.6% of the vegetated areas such as the southeastern China, the Great Khingan Mountains, and the Qinghai-Tibet Plateau (Figure 1b). This suggests vegetation is less responsive to droughts. By contrast, higher coincidence rates (r _10th > .4) were observed in the Inner Mongolia Plateau and Huang-Huai Area, and even ~0.6 in Mount Tai, which indicates greater susceptibility of vegetation growth to drought.

The spatial heterogeneity of coincidence rates can be explained by multiple reasons. Specifically, low r _10th values in the southeast are associated with the humid climate, which results in less severe and frequent droughts (Vicente-Serrano et al., 2013; Yang et al., 2016). In the Qinghai-Tibet Plateau where vegetation growth is dominantly limited by low temperatures, low r _10th values were observed because droughts only drive the low plant growth in fewer than one-fourth of the vegetated lands (Zhang et al., 2023). In addition, the low coincidence rates may be associated with the high resistance of vegetation to drought. For example, coniferous forests dominating the Great Khingan Mountains adapt to long-term dry conditions and have good capacities to resist droughts by regulating stomatal openness (Brodribb et al., 2014; Song et al., 2022). Conversely, higher coincidence rates observed in the Inner Mongolia Plateau and the Huang-Huai Area (Figure 1b), are primarily driven by the heightened vulnerability of vegetation growth to drought in these regions (Chen et al., 2022).

Therefore, Figure 1 demonstrates that the occurrence of drought and the response of vegetation do not align perfectly, as the same level of drought stress does not necessarily result in the same magnitude of negative vegetation anomalies. This motivates us to identify the drought thresholds which determine vegetation growth in China, by considering both drought intensity and the performance of vegetation responding to drought.

3.2 Spatial patterns of drought thresholds ( T _SM ) for negative NDVI anomalies in China

Following the applicable framework for detecting drought thresholds developed by Li et al. (2023), we examined the response trajectories of vegetation to increasing drought stress and identified the drought thresholds for the NDVI across China during the period 2001–2018 (Figure 2). Approximately 64.4% of vegetated areas in China exhibited a nonlinear relationship between drought intensity and the coincidence rates of drought-vegetation anomalies, which indicates the existence of drought thresholds in these regions (Figure 2a). However, it is important to note that nonlinearity was not detectable in certain areas, such as the Qinghai-Tibet Plateau or the southeastern regions. We suggest that vegetation growth is not dramatically impacted by droughts during 2001–2018, so coincidence rates remain low and no abrupt shift of vegetation response to drought is detected in these areas (Figure S5). For all vegetated areas where thresholds were detected, the average T _SM value was at the 8.7th percentile (Figure 2b). Furthermore, we observed that T _SM values exceeded the 15th percentile in 13.9% of vegetated areas characterized as ecological-drought-prone regions, whereas T _SM values were smaller than the 5th percentile in a small portion of areas (~6%).

The spatial variability of drought thresholds reveals variations in vegetation responses and underlying mechanisms used to cope with drought stress in different regions. The Yunnan-Kweichow Plateau exhibited similar patterns of drought thresholds ranging from the 6th to the 10th percentiles. The presence of karst landforms in this region leads to severe soil and water loss, resulting in higher drought stress for vegetation due to limited soil water and nutrient availability (Liu et al., 2023; Zhang et al., 2022). Yet, local vegetation has developed adaptive biochemical, physiological, and morphological strategies to cope with water deficits, mitigating the negative impacts of drought (Liu et al., 2021). As a result, the drought thresholds in the Yunnan-Kweichow Plateau are relatively smaller than those of regions such as the Inner Mongolia Plateau and the northwestern regions, where lower magnitudes of drought stress may trigger marked vegetation response (T _SM > 14th percentile). Ecological droughts are more prone to occur in these arid or semi-arid grassland ecosystems due to the frequent and prolonged duration of drought events (Ding et al., 2020; Zhang et al., 2020). In these areas, soil water availability strongly influences vegetation activity, and droughts of even moderate magnitude can lead to a substantial loss of grassland productivity, resulting in a decline in vegetation greenness (Lei et al., 2020; Walther et al., 2019).

Based on the knowledge of T _SM, we identified droughts that trigger the rapid increase in vegetation response using T _SM instead of relying on the empirical drought threshold 10th across vegetated areas in China. These T _SM-inferred droughts are regarded as ecological droughts. We found that on average, the frequency of ecological droughts was 1.2 times during the period 2001–2018 (Figure S6), slightly less frequently than droughts defined by the 10th percentile (1.8 times, as shown in Figure 1a). Note that water-limited regions such as the Hexi Corridor-Hetao Plain and the southern Inner Mongolia Plateau experienced more than three times ecological drought events during the same period, which indicates larger threats of droughts to vegetation in these areas. Whereas ecological droughts were not detected in 63% of the vegetated areas in China, either due to the absence of detected drought thresholds (35.6%, blue areas in Figure 2a) based on observations during 2001–2018 or because the vegetation in those regions did not experience T _SM-inferred droughts (27.4%, gray areas in Figure S6). These findings underscore the importance of identifying drought thresholds that impact vegetation growth for accurately assessing the impacts of drought on ecosystems and informing drought risk management policies in China.

3.3 Divergence of drought thresholds among different vegetation biomes

Our findings indicate significant variation in T _SM values among different vegetation biomes (p < .05 for every two biomes in the Mann–Whitney U test). Woody vegetation biomes, including forests and shrubs, exhibited similar patterns of drought thresholds ranging from the 4th to the 11th percentile, which were relatively smaller compared to T _SM values for grasslands (Figure 3a). Approximately 22% of grassland areas had T _SM values greater than the 15th percentile, which indicates a higher likelihood of ecological droughts occurring. These differences in T _SM values among vegetation biomes suggest that vegetation responds differently to drought stress, influenced by growth conditions and ecophysiological characteristics specific to each biome (Bréda et al., 2006; Farooq et al., 2009; Farooq et al., 2012).

In China, forest ecosystems were characterized by relatively smaller drought thresholds, which indicates less susceptibility to ecological droughts. Different types of forests exhibited similar T _SM values, with needleleaf forests at 6.4 ± 1.1th percentile on average, mixed forests at 5.5 ± 0.8th percentile, and broadleaf forests at 6.1 ± 1.3th percentile. Temperate and subtropical forests in China are mainly distributed in monsoon regions and influenced by seasonal water availability. They have a strong ability to access deep soil water and store water in the root zone, which mitigates the negative impacts of surface soil water deficits on vegetation photosynthesis (McDowell et al., 2019; Stocker et al., 2023; Walther et al., 2019). Therefore, the response of vegetation growth will not shift dramatically except in cases of severe drought. In addition, it is noteworthy that mixed forests exhibited smaller T _SM values than other forest ecosystems. We infer that species diversity enhances the resistance of forests to drought by leveraging compensatory structures and ecosystem functioning for enhanced stability (Fichtner et al., 2020; Gazol & Camarero, 2016; Liu et al., 2022).

However, compared to forests, grasslands exhibited relatively large T _SM values (averaging at the 10.2 ± 4.7th percentile), especially in the northern arid and semi-arid regions. Due to the simple architecture, grasslands are highly dependent on soil water availability (Walther et al., 2019). Once drought occurs, they tend to take the strategies of dormancy during droughts or death but regeneration after drought (Balachowski et al., 2016; Chaves et al., 2003; Zwicke et al., 2015). In addition, the northern grasslands are ecologically vulnerable due to the resource limitations of soil moisture, heat, and nutrients, which results in high risks of drought damage to local ecosystems (Liu et al., 2020). Therefore, we observed the abrupt shifts of grassland's simultaneous response to drought at a milder stress level, corresponding to larger T _SM.

In addition to natural vegetation, we also investigated the drought thresholds for croplands, which are heavily influenced by human activity. Given the positive impacts of agricultural management, in particular irrigation, T _SM values for croplands (averaging at the 6.4 ± 1.1th percentile) tend to be small with little variation compared to T _SM values for grasslands (Figure 3a). Furthermore, we compared T _SM values for croplands among the five major agricultural zones in China, where the croplands account for over 75% of the vegetated area: North China, the Huang-Huai area, the Sichuan Basin, the Yangtze River Delta, and Jianghan Plain. Our findings indicate that T _SM values for croplands differ among these agricultural zones. In general, the smallest average T _SM was for croplands in North China (6.0 ± 0.5th percentile), whereas the largest was for the Huang-Huai Area (7.7 ± 1.1th percentile). The average T _SM for croplands in Jianghan Plain was 5.8 ± 2.2th percentile, equivalent to that in North China but with greater variation compared to the other four regions.

Moreover, we observed varying effects of agricultural management in mitigating the negative impacts of drought on crops across these five agricultural zones, as reflected in the different drought thresholds along the irrigation rate spectrum (Figure 3b). Our results show that irrigation rates in agricultural areas, in particular in the Sichuan Basin (averagely 6.1 ± 0.7th percentile), play a crucial role in shaping T _SM values for crops. In the case of the Sichuan Basin, we observed a decrease in T _SM values as irrigation rates increased, with a decrease of more than 0.1 percentile per 20% increment in irrigation rates. Despite the subtropical monsoon climate of the Sichuan Basin and the fact that its annual average precipitation exceeds 1000 mm, the uneven distribution of precipitation throughout the year results in seasonal soil moisture deficits. As a result, spring droughts can profoundly impact crop growth, as indicated by the higher response of croplands to droughts shown in Figure 1b (Ding et al., 2020; Li et al., 2019). Consequently, agricultural irrigation can mitigate these negative drought impacts and alter T _SM values for crops in the Sichuan Basin. However, we did not observe marked effects of irrigation on T _SM values for crops along the gradient of irrigate rates in the other four agricultural zones.

3.4 Future risk of T _SM -inferred drought in China

Based on the satellite-derived T _SM values in Figure 2b, we identified ecological droughts. Then we utilized outputs of surface soil moisture from 13 ESMs to project the occurrence of ecological droughts as well as the possible change of ecological drought risk in China during 2083–2100 relative to the past periods 2001–2018 under two SSPs (SSP2-4.5 and SSP5-8.5). Figure S7 illustrates that, on average, there will be approximately 1.5 times ecological droughts in China during the period 2083–2100 under both SSP2-4.5 and SSP5-8.5. Certain regions, such as the southern Inner Mongolia Plateau and the Hexi Corridor-Hetao Plain, which were already prone to T _SM-inferred droughts in the historical periods (2001–2018, as shown in Figure S8), will continue to be hotspots for ecological droughts in the future (i.e., they are projected to experience around three times ecological droughts during the period 2083–2100, as shown in Figure S7).

In comparisons of the occurrence of T _SM-inferred droughts during the period 2001–2018 with future projections, ESMs indicate that the risk that vegetation experiences ecological droughts will change across China (Figure S9). As shown in Figure 4a, of the areas for which satellite T _SM values are known (from Figure 2b), approximately 46.6% show an average increase of 1.8% in the risk of experiencing ecological droughts under SSP2-4.5. Conversely, in another 44.0% of these areas, the future risk of ecological droughts is projected to decrease. Under SSP5-8.5, a greater proportion of areas (48.3% of areas with known satellite T _SM values) will face a higher risk (averagely 2% increase) of experiencing ecological droughts. Therefore, our findings indicate that if the current high intensity of greenhouse gas emissions is not mitigated, the risk for drought-induced suppression of vegetation growth will likely increase in more regions of China. This highlights the importance of taking effective actions to constrain greenhouse gas emissions and manage the risks.

Moreover, our findings highlight marked changes in the future risk for ecological droughts in several regions, including the Inner Mongolia Plateau, Central Plain, Shandong Peninsula, the Hexi Corridor-Hetao Plain, and the northern and southern Yunnan-Kweichow Plateau (Figure 4b). Under continued global warming, there will be an increase in the risk for ecological droughts of 1.1%–3.9% in the Central Plain, Shandong Peninsula, the Hexi Corridor-Hetao Plain, and the northern Yunnan-Kweichow Plateau. Note that the Hexi Corridor-Hetao Plain, where severe desertification is occurring and vegetation growth faces a high risk for ecological droughts (2.7 times for 2001–2018; Figure S8), is projected to have an increase of 3.9% for the ecological drought risk under SSP5-8.5. By contrast, the Inner Mongolia Plateau shows marked decreases in the risk for ecological droughts, with a 2.4% decrease under SSP2-4.5 and a 3.4% decrease under SSP5-8.5. Although ecological droughts will still occur frequently in the Inner Mongolia Plateau (around three times during the period 2083–2100; Figure S7), the future projections suggest a relative alleviation of ecological drought stress compared to historical periods during 2001–2018 (Figure 4b; Figure S9). In contrast to the northern Yunnan-Kweichow Plateau, which will experience an increasing drought risk, the southern Yunnan-Kweichow Plateau is expected to have a decreasing risk for ecological droughts, with a 1.5% decrease under SSP2-4.5 and a 0.8% decrease under SSP5-8.5.

We also projected that the risk for ecological droughts would decrease more (−3.4%) in the Inner Mongolia Plateau and increase less (+1.1%) in the northern Yunnan-Kweichow Plateau under SSP5-8.5, compared to that under SSP2-4.5, which reveals that these two regions are less likely to have the exposure risk of ecological droughts under a high emissions scenario compared to a mild emissions scenario. This suggests the potential alleviation of drought stress, which may be attributed to the increasing water use efficiency resulting from higher atmospheric CO₂ concentrations (Keenan et al., 2013).

The divergent changes in the risk of the occurrence of ecological drought across different regions indicate that there is a need to take targeted measures in order to cope with future climate challenges. In regions that are experiencing increasing risk for ecological droughts, such as the Hexi Corridor-Hetao Plain, more effective measures should be taken to reduce ecosystem degradation. Despite the marked decrease in the risk for ecological droughts, ecological droughts are still frequent in the Inner Mongolia Plateau under future climates; thus, drought-induced suppression of vegetation growth cannot be ignored in this region.