Declining Tradeoff Between Resistance and Resilience of Ecosystems to Drought

Resistance and resilience are widely used to characterize ecosystem drought stability. Tradeoff between resistance and resilience have been reported, but its long‐term trends remain uncertain at global scale. Based on remotely sensed vegetation indices, we assessed the spatiotemporal dynamics of drought resistance and resilience. Result revealed that there was a significant decline in the tradeoff between resistance and resilience, corresponding to a substantial increase in the proportion of areas with high resistance‐high resilience or low resistance‐low resilience. In the South Sahel, South Africa and Central China, the increased precipitation and vegetation coverage contribute to enhanced drought stability constructed by resistance and resilience; while rising temperature, decreased water availability and deforestation lead to declined stability in northeastern North America, South America and the Congo region. Increases in the areas with low resistance‐low resilience resulting from declining tradeoffs warn of increased regional ecosystem vulnerability.


Introduction
Drought is a climate event characterized by a lack of moisture (Wilhite & Glantz, 1985), resulting in reduced photosynthesis (Fu et al., 2022), limited vegetation growth (Gazol et al., 2017;Li et al., 2020) and increased forest mortality (Senf et al., 2020).This greatly harms the structure, functions, and services of ecosystems (He et al., 2018), and threatens global warming mitigation based on natural climate solutions due to declining carbon sequestration of vegetation caused by droughts (Griscom et al., 2017).Under climate change, many regions are projected to experience a decline in terrestrial water storage (Pokhrel et al., 2021), which poses drought threats to terrestrial ecosystems over a larger land area.Therefore, it is necessary to understand how ecosystems respond to drought events, what negative impacts they face, and how they maintain structural and functional stability under climate change.
Drought resistance and resilience describe the two aspects of stability (Hillebrand et al., 2018), resistance refers to the capacity of ecosystems to resist drought and maintain their functions (e.g., vegetation growth and ecosystem productivity) under drought (Isbell et al., 2015;D. Liu et al., 2022), and resilience refers to the capacity of ecosystems to return to normal states (unaffected by disturbances) from the negative anomalies caused by drought (Bottero et al., 2021;Ingrisch & Bahn, 2018).Benefiting from data provided by remote sensing for the detection of drought information and vegetation dynamics (Jiao et al., 2021), previous studies have reported that there are tradeoff patterns between drought resistance and resilience of ecosystems, both at the regional and global scales (Hillebrand et al., 2018;Sturm et al., 2022;Yao, Fu, et al., 2022).The tradeoff pattern between resistance and resilience refers to ecosystem with high resistance-low resilience or low resistance-high resilience, avoiding ecosystems having both low resistance and low resilience, thus ensuring that ecosystems could survive drought.In China dryland ecosystems, Yao, Fu, et al. (2022) and Yao, Liu, et al. (2022) used the leaf area index (LAI) from Global Inventory Monitoring and Modeling System (GIMMS) to assess the drought resistance and resilience of forest and grassland ecosystems, and revealed that forest ecosystems were high resistance-low resilience but grassland ecosystems were low resistance-high resistance.In European ecosystems, Ivits et al. (2016) demonstrated that Mediterranean ecosystems showed the least drought resistance but were resilient based on the fraction of absorbed photosynthetically active radiation (FAPAR) developed using products from GIMMS and Terra Moderate Resolution Imaging Spectroradiometer (MODIS).At the global scale, Chen et al. (2023) found that resistance was negatively correlated with resilience in different biomes around the world, indicating that there was a tradeoff between resistance and resilience.
However, climate change and human activities threaten drought resistance and resilience of ecosystems (Li et al., 2020;D. Liu et al., 2022;Stampfli et al., 2018).Tao et al. (2022) found that both resistance and resilience significantly decreased in tropical rainforests in the Americas and non-significantly decreased in tropical rainforests in Africa based on monthly radar satellite observations with a spatial resolution of 25 km from 1992 to 2018.Resistance and resilience in tropical rainforests in Asia experienced an opposite temporal trend, with resistance decreasing significantly and resilience increasing insignificantly (Tao et al., 2022).Li et al. (2020) found that gymnosperms experienced an obvious increase in drought resilience but a decrease in resistance in 1990-2009compared with 1950-1969. Z. Liu et al. (2023) ) revealed that drought resistance of different biomes in the world showed significant decreasing trends, with evergreen broadleaf forests decreasing most.The temporal changes in resistance and resilience inevitably changed the tradeoff between resistance and resilience, but it is not clear how the tradeoff has changed over time.
To address the abovementioned research gaps, this study explored the long-term (from 1982 to 2016) trends in the tradeoff between resistance and resilience and its drivers.The study was conducted in the following three parts.First, we assessed drought resistance and resilience based on Standardized Precipitation Evapotranspiration Index (SPEI) and remotely sensed vegetation indices (LAI and kernel Normalized Difference Vegetation Index (kNDVI)), and further identified the tradeoff between resistance and resilience according to the calculated negative correlation coefficient between them (Figure S1 in Supporting Information S1).Second, we detected the long-term changes in this tradeoff and analyzed the drivers of these trends by principal component regression.Third, given the application of model outputs in exploring the interactions among climate, hydrology and ecology, we evaluated the performance of the model simulations in characterizing the tradeoff between resistance and resilience.
Vegetation dynamics are assessed using three remotely sensed vegetation indices, including GIMMS LAI3g, Global Mapping LAI (GLOBMAP LAI) product generated by quantitative fusion of Advanced Very High Resolution Radiometer (AVHRR) and MODIS data, and kNDVI.GIMMS LAI3g is from Zhu et al. (2013), which provides LAI observations at a 15-day temporal resolution and a 1/12-degree spatial resolution for global vegetation.We obtained a monthly temporal resolution by averaging the15-day value in the same month (Zhu et al., 2016) and then resampled the monthly data to a 0.5°spatial resolution using the bilinear method to match the spatial resolution of the SPEI.GLOBMAP LAI is from http://www.globalmapping.org/globalLAI/,which provides a global LAI data set at a 15-day temporal resolution and 8 km spatial resolution.NDVI3gv1 is from http://poles.tpdc.ac.cn/en/data/9775f2b4-7370-4e5e-a537-3482c9a83d88/, which provides global NDVI data with a temporal resolution of twice a month and a spatial resolution of 1/12°.We preprocessed the GLOBMAP LAI and GIMMS NDVI3gv1 using the same method for GIMMS LAI3g.The temporal coverage of GIMMS LAI3g, GLOBMAP LAI and GIMMS NDVI3gv1 is 1982-2016, 1982-2016and 1982-2015, respectively.The kNDVI is calculated based on the NDVI from GIMMS 3gv1 based on Equation 1 (Camps-Valls et al., 2021).
Compared with the NDVI, the kNDVI has an obvious advantage in coping with saturation effects in densely vegetated areas (Camps-Valls et al., 2021;Y. Zeng et al., 2022).Regions with monthly NDVI values <0.1 are considered sparse vegetation and are discarded from the final analyses.LAI is widely used measurements for describing plant canopy structure (Yao, Fu, et al., 2022), and kNDVI is close relations to vegetation productivity and density (Yao, Liu, et al., 2022).In the subsequent section, we independently calculated drought resistance and resilience using different vegetation indices to enhance the robustness of results.kNDVI = tanh NDVI 2 ) (1) Based on the SPEI and vegetation data (GIMMS LAI3g, GLOBMAP LAI, and kNDVI), we identified drought events and their corresponding negative vegetation anomalies according to the following steps.Taking LAI as an example, we eliminated the seasonal cycle of vegetation data by subtracting the monthly average values from vegetation data on the time series (Forzieri et al., 2022;S. Zhang et al., 2021) and then removed the long-term trends from the resulting time series by using the "detrend" function in MATLAB (R2020b) (Yao, Fu, et al., 2022).According to the standard deviation (SD) of the detrended vegetation data, we defined the negative vegetation anomaly as the detrended data below 0.5 SD (Figure 1b).Second, we defined drought events as SPEI values below 1 and lasting at least 2 months (Figure 1a).Considering that not all drought events cause negative vegetation anomalies, we conducted the subsequent analysis using the effective drought events (Yao et al., 2023), which were identified based on the cooccurrence of negative vegetation anomalies and drought events (Figure 1).Third, to avoid the impacts of wildfires on negative vegetation anomalies, we removed samples in which there were wildfires during effective drought events based on the burned area data from the Global Fire Emission Database.Based on the vegetation dynamics during effective drought events, we defined the maximum loss as the difference between the minimum detrended LAI and 0.5 SD, defined the resisting time to drought (T1) as the time falling to the minimum detrended vegetation data from the normal state, and defined the recovery time from drought (T2) as the time returning to the normal state from the minimum detrended vegetation data.Earth's Future 10.1029/2024EF004665 YAO ET AL.

Resistance and Resilience of Ecosystems to Drought Events
Drought resilience refers to the capacity of ecosystems to return to a normal state (multi-year average state) (Gazol et al., 2017;Ingrisch & Bahn, 2018).Recovery time, the time it takes for ecosystems to return to a normal state after drought, is an important indicator to assess drought recovery capacity (L.Liu et al., 2019;Schwalm et al., 2017;S. Zhang et al., 2021), but it is not comprehensive to use recovery time alone to characterize resilience considering that the recovery time is affected by vegetation loss (He et al., 2018;Yao et al., 2023).He et al. (2018) reported that greater vegetation damage caused a longer recovery time in the 2003 European drought, and Yao et al. (2023) found that recovery time increased with increasing vegetation losses.Here, we quantified the drought resilience of each grid cell using the ratio of normalized loss (the ratio of maximum loss to multi-year average, representing the proportion of vegetation loss relative to the multi-year average) to recovery time based on Equation 2 (Figure 1b) (Ingrisch & Bahn, 2018); that is, the ecosystem could recover more per unit of recovery time (month), indicating that it is more resilient.For example, for ecosystems suffering from the same loss, a longer recovery time is indicative of ecosystems with low resilience.Drought resistance refers to the capacity of ecosystems to maintain their normal state during drought (Hoover et al., 2021;D. Liu et al., 2022).Greater vegetation losses are indicative of the lower capacity of ecosystem resisting drought (Isbell et al., 2015), but the time it takes for different ecosystems to suffer the same vegetation loss is different.Here, we quantified the drought resistance of each grid cell using the ratio of resisting time to normalized loss based on Equation 3(Figure 1b) (Ingrisch & Bahn, 2018); that is, the ecosystem suffers smaller loss per unit of resisting time, indicating that it is more resistant.For example, for ecosystems with the same resisting time, greater losses suffered by the ecosystem are indicative of low resistance.Based on the same methodology as LAI (Figure 1), we also assessed drought resistance and resilience using kNDVI.
For enhancing the robustness of the results, we supplemented the assessment of resilience (Equation 4) and resistance (Equation 5) according to vegetation loss (Isbell et al., 2015;Li et al., 2020).Where Y n , Y e , and Y e+2 are the normal vegetation state, vegetation state during drought, and vegetation status after 2 months of drought, respectively.
We further calculated the spatial correlation coefficient between resistance and resilience globally.The negative correlation coefficient showed that highly resilient ecosystems tend to have low resistance, which suggested a tradeoff between resistance and resilience.Then, we identified the threshold values of the 10th, 20th, … and 90th percentiles of resistance and resilience, which were used to bin resistance and resilience into 10 bins according to the 0-10th, 10-20th, …, 80-90th, and 90-100th percentiles of resistance or resilience.The mean probability density of each bin of resistance and resilience across all grid cells was used to present the tradeoff distribution of resistance and resilience.

Long-Term Trend in the Tradeoff Between Resistance and Resilience
We applied the nonparametric trend test (Mann-Kendall test) for the spatial correlation coefficient from 1982 to 2016 to evaluate the long-term trends at the 95% significance level.If the trend of correlation coefficient is positive, the tradeoff decreases; if the trend of correlation coefficient is negative, the tradeoff increases.In addition, we analyzed the trend in the probability density of each bin of resistance and resilience to characterize the changes in the tradeoff distribution of resistance and resilience.If the trend of bins with low resistance-high resilience (high resistance-low resilience) is positive, the tradeoff increases; if the trend of bins with low resistance-high resilience (high resistance-low resilience) is negative, the tradeoff decreases.
Ecosystems with high resistance-high resilience are generally considered more stable, and the change in stability could be reflected by the change in the tradeoff mentioned above.For example, the decreases in bins with low resistance-high resilience (high resistance-low resilience) meant the increases in bins with low resistance-low resilience, which indicated reduced stability; or the increases in bins with high resistance-high resilience, which indicated enhanced stability.We identified the spatial pattern of the changed stability according to the following steps.To ensure the comparability of resistance and resilience, we normalized resistance and resilience for each year to a range of 0-1.Point with an resistance value of 0 and an resilience value of 0 was considered the origin of coordinate, such as point a in Figure S2 of the Supporting Information S1.Stability of grid cell i was measured by the deviation of grid cell i from the origin of coordinate, and was calculated based on Equation 6.Specifically, the stability of grid cell b was characterized by the deviation of point b from point a (Figure S2 in Supporting Information S1); and the stability of grid cell c was characterized by the deviation of point c from point a (Figure S2 in Supporting Information S1).For all grids, we calculated the stability and further tested the trend of stability using the Mann-Kendall.
where Stability i , Rt i , and Rs i referred to the stability, resistance and resilience of grid cell i, Rt a and Rs a referred to the resistance and resilience of the origin of coordinate.

Relative Importance of Drivers on the Long-Term Changes in Stability
We used principal component analysis (PCA) and principal component regression (PCR) to estimate the relative importance of various drivers (climate and land-use variables) on the long-term changes in stability (Abel et al., 2021;Wei et al., 2022).Climate variables included solar radiation (Srad), temperature (Temp), VPD, precipitation (Prec), and SM.Global land-use variables with a spatial resolution of 0.05°× 0.05°included short vegetation (Sv) and tree canopy (Tc) from 1982 to 2016, which was provided by Song et al. (2018).We resampled Sv and Tc to a spatial resolution of 0.5°× 0.5°using the bilinear method to match the spatial resolution of the SPEI.To remove the impact of collinearity among independent (climate and land-use) variables, component loadings and new components with no linear correlation were generated from independent variables based on PCA (Abel et al., 2021;Wei et al., 2022), and then, we performed a multiple linear regression with stability as the dependent variable and new components as independent variables (Seddon et al., 2016).We multiplied the regression slope from the PCR with the respective loadings from the PCA and summed the absolute values of these scores to obtain the relative importance of drivers on stability pixel by pixel.

Outputs of Model Simulations
The "trends and drivers of the regional scale sources and sinks of carbon dioxide" project (TRENDY v.7) models output rich carbon and hydrological data, which could be used to study the interactions among climate, hydrology and ecology (Y.Zhang et al., 2022;Zhao et al., 2022).We used LAI outputs from 6 dynamic global vegetation models (DGVMs) from TRENDY S3 simulations (dynamic CO 2 , land use and climate) to evaluate the performance of recent DGVMs in characterizing the tradeoff between resistance and resilience of ecosystems to drought (Sitch et al., 2008) to compare the differences in results derived from observations and models (Table S1 in Supporting Information S1).According to the same methods as the observations, we calculated the correlation coefficient between resistance and resilience and detected the long-term tradeoff trend using the Mann-Kendall test.

The Tradeoff Between Drought Resistance and Resilience
The drought resistance and resilience of ecosystems were negatively correlated (Figures 2a, 2c, and 2e), and the negative correlation coefficients between resistance and resilience based on GIMMS LAI3g, GLOBMAP LAI and kNDVI were 0.42, 0.45, and 0.38, respectively.Ecosystems with low resistance tended to be more resilient, and ecosystems with low resilience were generally more resistant, indicating a tradeoff between resistance and Earth's Future 10.1029/2024EF004665 resilience.The spatial patterns of resistance and resilience showed that the regions with the highest resistance and the lowest resilience were concentrated in low-latitudes, including the Amazon rainforest, the Congo Basin, and Southeast Asia (Figure 2).The least resistant and most resilient regions were concentrated in mid-latitudes, such as boreal forest regions and northeastern North America.Considering both resistance and resilience, tropical rainforests showed the strongest stability under drought (Figure S3 in Supporting Information S1).
The tradeoff between resistance and resilience is supported by the bimodal distribution of resistance and resilience: low resistance-high resilience and high resistance-low resilience (Figures 2g-2i).The high probability density occurred in the top-left and bottom-right instead of the top-right and bottom-left, showing that the frequency of high resistance-low resilience and low resistance-high resistance is much higher than that of high resistance-high resilience and low resistance-low resistance.Notably, the probability density of the bin with resistance higher than the 90th percentile and with resilience higher than the 90th percentile was 0, meaning that there were no grid cells that were both the most resilient and the most resistant; and the probability density of the bin with resistance lower than the 10th percentile and with resilience lower than the 10th percentile was 0, meaning that there were no grid cells that were both the lowest resistance and resilience.
There was an obvious tradeoff between resistance and resilience among ecosystems of different aridity and vegetation types (Figure 3).In terms of ecosystems of different aridity (Figure S4a in Supporting Information S1), the semi-arid ecosystems had the lowest resistance but the highest resilience.In contrast, humid ecosystems were the most resistant but the least resilient (Figures 3a and 3b).Regarding the ecosystems of different vegetation types (Figure S4b in Supporting Information S1), grasslands were the least resistant but the most resilient among vegetation types (Figures 3c and 3d), and forests showed the strongest drought resistance but the weakest resilience (Figures 3c and 3d).These results indicated the different drought strategies among ecosystems.

Declining Tradeoff Between Resistance and Resilience
Observations consistently showed that the tradeoff between global drought resistance and resilience declined over the past few decades.Trend analysis showed that the significant long-term trends in the negative correlation coefficient were 0.017, 0.021, and 0.009 dec 1 based on GIMMS LAI3g, GLOBMAP LAI, and kNDVI, respectively (Figure 4a).The decline in tradeoff was further supported by the change in the probability density of each bin of resistance and resilience (Figure 4b).Specifically, the probability density of the diagonal from the top-left to bottom-right decreased, while the probability density of the top-right bins and the bottom-left bins increased (Figure 4b).Negative trend values in the probability density of the diagonal from the top-left to bottom-right Positive trend values in the probability density of the top-right bins showed the increases in grid cells with high resistancehigh resilience, meaning that there were areas experiencing an enhancement in stability, including the South Sahel, South Africa and Central China (Figure 4c).Positive trend values in the probability density of the bottom-left bins indicated the increases in grid cells with low resistance-low resilience, meaning that there were areas experiencing a decline in stability, including northeastern North America, the Amazon region and the Congo region (Figure 4c).

Drivers of the Long-Term Changes in Stability Constructed by Resistance and Resilience
The relative importance of drivers on the long-term changes in stability constructed by resistance and resilience varied regionally.In northeastern North America, temperature and VPD were the two most important factors for the changes in stability (Figure 5a), and the long-term trends showed that hotter temperatures and higher VPD were unfavorable to regional stability.In South America, vegetation coverage had the highest importance (Figure 5a), and the transition from trees to short vegetation led to a decline in stability (Figure 5b).In the South Sahel, precipitation was the most important driver of increasing stability, followed by the tree canopy (Figure 5a).This suggested that the increased precipitation over the past few decades contributed to vegetation growth, enhancing the stability of regional ecosystems to drought (Figure 5b).In the Congo region, increased VPD and reduced SM exposed regional ecosystems to a compound land-atmosphere dry environment, which made ecosystems vulnerable to droughts.In South Africa and Central China, vegetation coverage was the most important driver (Figure 5a), and the increase in vegetation coverage was beneficial for enhancing stability.

Performance of Models in Characterizing the Tradeoff Between Resistance and Resilience
Resistance and resilience based on LAI from DGVMs from the TRENDY project consistently showed a tradeoff, which was not different from the observed results that the mid-latitudes were low resistance-high resilience and   the low-latitudes were high resistance-low resilience high resistance-low resilience (Figure S5 in Supporting Information S1).The distribution of resistance and resilience was concentrated on the diagonal from the top-left to bottom-right (Figures 6a-6f).The probability density of the top-left bin (high resistance-low resilience) was up to 8%, and the mean probability density of the bins on the diagonal from the top-left to bottom-right was more than 2%, which was much higher than the probability of the top-right and bottom-left bins.However, the temporal trends of the tradeoff relationship between resistance and resilience based on the model simulations were not consistent with the observations.Among the 6 DGVMs from the TRENDY project, the results based on ISBA-CTRIP, LPJ-GUESS, and VISIT showed an increasing trend of the tradeoff (Figures 6g and 6h), which was contrary to the observations.Only the SDGVM reflected the significant decreasing trend of the tradeoff (Figure 6h).This result suggested that DGVMs could characterize the tradeoff between resistance and resilience, but most models failed to capture the temporal dynamics of this tradeoff.

Discussion
Droughts challenge the stability of ecosystem structure and function, and the development of remote sensing provides data for understanding the response of ecosystems to droughts at the global scale.This study assessed the drought resistance and resilience using remotely sensed vegetation data sets, clarified that the tradeoff between resistance and resilience existed globally and found a temporal decline in the tradeoff.The tradeoff between resistance and resilience is the result of long-term selection of vegetation by droughts (Chen et al., 2023;Miller & Chesson, 2009), ensuring that ecosystems are not in a vulnerable pattern of low resistance-high resistance when exposed to droughts.Being high resistance and resilience at the same time minimizes the damage caused by drought, but each of them consumes resource costs (Chen et al., 2023).As a result, the development of one often comes at the expense of the other (Russo et al., 2020), thus evolving into tradeoff.However, this tradeoff has declined over the past few decades (Figure 4; Figure S6 in Supporting Information S1), suggesting that a vulnerable pattern of low resistance-low resilience became possible and warning that there were regions suffering from decreased drought stability.Although the models could identify the tradeoff between resistance and resilience, it is difficult to reveal the declined tradeoff.This may be due to the large uncertainty of the response of ecosystems to key climate change processes in models, such as increasing temperature and changes in soil water (Sitch et al., 2008), which largely causes the long-term trend of the simulated LAI to mismatch the observed LAI (Murray-Tortarolo et al., 2013) and makes it difficult for models to capture the patterns of changes reported in observations (Zhao et al., 2022).

Robustness Analysis of the Tradeoff
We re-explored the tradeoff between resistance and resilience and its long-term trend using SPEI and GIMMS LAI3g with a spatial resolution of 0.25°for the robustness of the tradeoff at different spatial resolutions.Results showed that the high probability density occurred in the top-left and bottom-right (Figure S7a in Supporting Information S1), indicating that the frequency of high resistance-low resilience and low resistance-high resilience was much higher than that of high resistance-high resilience and low resistance-low resilience.The positive trend in the correlation coefficient between resistance and resilience suggested a declining tradeoff (Figure S7b in Supporting Information S1).In addition, resistance and resilience quantified using vegetation loss were tradeoff, and this tradeoff showed significant declines (Figure S8 in Supporting Information S1).This result suggested that the spatial resolution, as well as the quantification of resistance and resilience, did not affect the tradeoff between resistance and resilience and its declining trend.
Given that the drought legacy effects are crucial for the vegetation growth and its dynamics (Kannenberg et al., 2019(Kannenberg et al., , 2020)), we supplemented the impact of the legacy effects on drought resistance and resilience.Specifically, ecosystems affected by drought legacy effects were defined as those in which vegetation was still in negative anomalies caused by the previous drought when the current drought occurred.We then quantified the resistance (Rt legacy ) and resilience (Rs legacy ) of ecosystems affected by drought legacy effects to the current drought.Results showed that the difference between resistance and Rt legacy and the difference between resilience and Rs legacy was positive (Figures S9a and S9b in Supporting Information S1), indicating that legacy effects reduced ecosystem resistance and resilience.Notably, the spatial pattern of Rt legacy and Rs legacy revealed that ecosystems with low Rt legacy tended to be more resilient, for example, southwestern United States, eastern and southern South America; ecosystems with low Rs legacy were generally more resistant, for example, the Amazon rainforest region (Figures S9c and S9d in Supporting Information S1).In summary, although legacy effects reduced the resistance and resilience of ecosystems, they did not affect the tradeoff between them.

The Tradeoff Between Resistance and Resilience Among Different Ecosystems
The tradeoff between resistance and resilience among ecosystems reflected the different strategies of ecosystems to cope with droughts (Hoover et al., 2021).For ecosystems with different aridity, dryland ecosystems are more resilient than humid regions but less resistant to drought (Figures 3a and 3b), which was consistent with previous evidence.D. Liu et al. (2022) found that drought resistance increased with increasing water availability based on remotely sensed data, and Gazol et al. (2017) noted that forest ecosystems in humid regions were more resistant to droughts based on tree-ring data.Ecosystems in drylands are affected by long-term moisture deficits and barren soils and are hardly resistant to droughts (Lian et al., 2021;D. Liu et al., 2022;X. Zeng et al., 2022); in turn, as a result of long-term evolution, they rely more on rapid recovery after drought to achieve the restoration of structure and function.The underlying mechanism could be explained by the fact that the main limitation of biological activities in dryland ecosystems is moisture availability (de Bashan et al., 2022;Lian et al., 2021), and precipitation after drought in drylands increases the content of water, soil nutrients, and microbial and plant activities (Barnard et al., 2020;de Bashan et al., 2022).Therefore, ecosystems can quickly return to their normal state once rewetted (Nguyen et al., 2017;S. Zhang et al., 2021).Compared with drylands, humid regions had higher water availability, more complex structures and higher biodiversity, and resisted drought by drawing soil moisture at different depths and the heterogeneous responses to drought among individuals (D. Liu et al., 2022;Z. Xu et al., 2021).
For different vegetation types, forests were more resistant than grasslands (Figures 3c and 3d), which was in line with previous studies (Yao, Fu, et al., 2022;S. Zhang et al., 2021).This could be explained by the fact that forest ecosystems tend to have higher biodiversity, greater species richness and deeper rooting depth (Figure S10 in Supporting Information S1) (Isbell et al., 2015;D. Liu et al., 2022;Stuart-Haëntjens et al., 2018).The interannual variation in productivity among tree species buffers the stress-related productivity declines, thereby mitigating the negative impacts of droughts and increasing resistance (Schnabel et al., 2021;H. Xu et al., 2019).The high resilience of grasslands is mainly because grasslands are predominantly inhabited by annual or biennial herbaceous plants, whose rapid renewal and growth are conducive to drought recovery (Stampfli et al., 2018;Stuart-Haëntjens et al., 2018).

Climate Changes and Land-Use Change Drive the Enhancements in Stability Constructed by Resistance and Resilience
The decline in the tradeoff between drought resistance and resilience indicated a decrease in areas with the previous tradeoff pattern of low resistance-high resilience or high resistance-low resilience but an increase in areas with high resistance-high resilience or low resistance-low resilience (Figures 4a,4b,and 7), indicating a change in ecosystem strategies to cope with droughts (Forzieri et al., 2022;Hoover et al., 2021;Li et al., 2020).Benefitting from climate changes and land-use change, the South Sahel, South Africa, and Central China experienced enhancements in stability (Figures 4c and 5).The main drivers for the increased stability in the South Sahel were increased precipitation and tree canopy (a transition from short vegetation to tree canopy) (Figure 5; Figure S11b in Supporting Information S1).On the one hand, the high precipitation anomaly contributed to regional greening and declined drought intensity (Herrmann et al., 2005;Hickler et al., 2005;Song et al., 2018), on the other hand, the reforestation planned for the Great Green Wall increased tree coverage (Mirzabaev et al., 2022;Sacande et al., 2021).In South Africa, increased stability was mainly attributed to the positive trend values in short vegetation (Figure 5; Figure S11b in Supporting Information S1).Previous studies have noted that woody plants have replaced grasslands and bare ground in South Africa (Mani et al., 2021;Mogashoa et al., 2021;Song et al., 2018).Although shrub invasions have reduced grassland species (Mani et al., 2021;Mogashoa et al., 2021), shrubs have shown stronger tolerance to drought than grasslands (Sandoval et al., 2019;Winkler et al., 2019).The improvement of the tree canopy and short vegetation in Central China enhanced the stability of regional ecosystems, mainly due to the transformation of bare or arable land to shrubland or forests (Piao et al., 2020;Wu et al., 2020), making ecosystems more stable to drought (Isbell et al., 2015;Schnabel et al., 2021).

Climate Changes and Land-Use Change Drive the Declines in Stability Constructed by Resistance and Resilience
In northeastern North America, South America, and Congo regions, there was a decline in stability (Figure 4c).A previous study confirmed that South American rainforests experienced significant declines in resistance and resilience, and the African rainforest experienced a significant decline in resistance and an insignificant decrease in resilience (Tao et al., 2022).Reduced tree canopy, enhanced VPD, decreased SM and increased temperature were the main driving factors for the declines in stability (Figure 5; Figure S11b in Supporting Information S1).
Previous studies reported consistent findings that there were obvious declines in tree canopy density/vegetation activities in Amazon rainforests and Congo regions due to deforestation driven by commodity or agriculture (Curtis et al., 2018;Song et al., 2018).Saatchi et al. (2021) noted that a marked increase in VPD and an obvious decline in water balance led to the vulnerability of tropical rainforests.Climate change has also adversely affected northeastern North America, where VPD was negatively correlated with vegetation growth (Li & Xiao, 2020;L. Liu et al., 2020), with increased temperatures and VPD reducing the environmental suitability of drought recovery (Figure 5) (Yao et al., 2023).

Limitations and Perspectives
The assessment of resistance and resilience is largely influenced by the data used.First, to match the temporal resolution of the SPEI, we analyzed the tradeoff between resistance and resilience on the monthly scale.However, key time nodes for ecosystems to respond to droughts may occur on smaller timescales, such as daily or weekly.We expect that with the development of remote sensing technology, high temporal resolution data will improve our understanding of the dynamic processes of ecosystems in response to droughts.Second, the carbon and hydrological data outputs by models are widely used to study the interactions among climate, hydrology and ecology.We found that although the models revealed the tradeoff between resistance and resilience, they could not capture the declining trend of the tradeoff.We recommend strengthening research on the response mechanisms of ecosystems to extreme events and incorporating them into the optimization of models to better simulate and predict the carbon-water cycles of ecosystems.Third, the intrinsic physiological and ecological mechanism of the tradeoff between resistance and resilience is not yet clear, and we recommend the comprehensive use of flux sites, remotely sensed and model data to conduct multi-scale researches at sample, regional and global scales to deeply disclose the internal mechanisms of the tradeoff and further explore how to prevent the declining trend of the tradeoff.

Conclusion
Based on three sets of remotely sensed vegetation indices, we assessed the drought resistance and resilience of terrestrial ecosystems, clarified the tradeoff between resistance and resilience globally, analyzed the long-term trends of this tradeoff and its drivers, and finally evaluated the performance of model simulations in characterizing the tradeoff between resistance and resilience.The results showed that there was an obvious tradeoff pattern with low resistance-high resilience or high resistance-low resilience across global ecosystems.For ecosystems with different aridity, dryland ecosystems are less resistant to drought than humid ecosystems but have higher resilience; for ecosystems with different vegetation types, grasslands are the least resistant but the most resilient, and forests show the strongest resistance but the weakest resilience.The tradeoff between resistance and resilience has declined significantly over the past few decades, while the proportion of grid cells with high resistancehigh resilience (low resistance-low resilience) has increased.Driver analysis shows that climate change (warming and drying) and land-use change (deforestation for agriculture or commerce) drive a decline in stability constructed by resistance and resilience, including northeastern North America, South America and the Congo region.In contrast, increased precipitation and vegetation coverage contribute to increases in stability, including the South Sahel, South Africa and Central China.Simulations from TRENDY could identify the tradeoff but could not characterize the temporal dynamics of this tradeoff.These results indicate a mismatch between model simulations and observations in capturing the responses of ecosystems to droughts, emphasizing the importance of developing dynamic global vegetation models that consider the response of ecosystems to key climate change processes, such as global warming and drying.

Figure 1 .
Figure 1.Examples of SPEI and LAI for 36 consecutive months are randomly selected to illustrate drought definition and vegetation dynamics.(a) SPEI on a time series, an SPEI value of 0 indicates that there is no water surplus or deficit, and an SPEI value of less than 0 indicates a water deficit.We use 1 as the drought threshold.(b) Detrended LAI on a time series, a detrended LAI value of 0 indicates the multi-year average, and we use 0.5 SD as the threshold of negative vegetation anomalies.Normalized loss refers to the difference between the minimum detrended LAI and 0.5 SD, T1 refers to the time falling to the minimum detrended VIs from the normal state, and T2 refers to the time recovering to the normal state from the minimum detrended LAI.Orange shading in (a) indicates a drought event.

Figure 2 .
Figure 2. Resistance and resilience of ecosystems to droughts.(a, c, and e) Represent the spatial pattern of resistance based on GIMMS LAI3g, GLOBMAP LAI and kNDVI, respectively.(b, d, and f) Represent the spatial pattern of resilience based on GIMMS LAI3g, GLOBMAP LAI, and kNDVI, respectively.The value of each pixel represents the average of resistance and resilience calculated for multiple drought events in (a)-(f).(g)-(i) represent the probability density of each bin of resistance and resilience across all grid cells based on GIMMS LAI3g, GLOBMAP LAI, and kNDVI, respectively.Regions with sparse vegetation or no droughts are masked with white.The CC RR in (a), (c), and (e) represents the correlation coefficient between resistance and resilience.

Figure 3 .
Figure 3. Resistance and resilience of ecosystems with different aridity and vegetation types based on GIMMS LAI3g.(a and b) Represent resistance and resilience in four different aridity zones, respectively.(c and d) Represent resistance and resilience of four different vegetation types, respectively.Asterisks indicate statistically significant differences (**P < 0.05; ***P < 0.001).

Figure 4 .
Figure 4. Declining tradeoff between resistance and resilience.(a) Represents the long-term changes in the correlation coefficient between resistance and resilience.(b) Represents the trend in the probability density of each bin of resistance (Rt) and resilience (Rs) based on GIMMS LAI3g.(c) Represents the spatial pattern of the changes in stability based on GIMMS LAI3g.** indicates a significant difference at the P < 0.05 level.

Figure 5 .
Figure 5. Drivers of the changes in stability constructed by resistance and resilience based on GIMMS LAI3g.(a) Represents the relative importance of driving factors of stability changes in the six regions shown in Figure 4, including (northeastern North America (NE NA), South America, South Sahel, Congo region, South Africa and Central China).(b) Represents the long-term trends of the two most important drivers.Srad, Temp, VPD, Prec, SM, Sv, and Tc represent shortwave radiation, temperature, precipitation, soil moisture, short vegetation and tree canopy, respectively.Asterisks indicate statistically significant differences (**P < 0.05; ***P < 0.001).

Figure 6 .
Figure 6.Tradeoff between resistance (Rt) and resilience (Rs) from model simulations.(a-f) Represent the probability density of each bin of resistance and resilience based on SBA-CTRIP, LPJ-GUESS, LPX-Bern, ORCHIDEE, SDGVM and VISIT, respectively.(g and h) Represent the long-term changes in the correlation coefficient between resistance and resilience.** indicates a significant difference at the P < 0.05 level, and * indicates a significant difference at the P < 0.1 level.

Figure 7 .
Figure 7.The process of vegetation resisting to drought and recovering from drought.Ecosystems suffer from drought at t1, reach the greatest losses at t3, and begin to recover thereafter.Ecosystems in (a) and (b) recover at t6, ecosystem in (c) recovers at t7, and ecosystem in (d) recovers at t5.The loss of vegetation in (a) is the same as that in (d), and the loss of vegetation in (b) is the same as that in (c).(a and b) Display tradeoff patterns of high resistance-low resilience or low resistance-high resilience.Obviously, the ecosystem in (a) suffers a smaller loss than that in (b) during the same resisting time, showing a higher resistance.The ecosystem in (b) restores more loss in the same recovery time, showing higher resilience.(c and d) Display an example of a declining tradeoff, showing a decrease in areas with the tradeoff patterns of high resistance-low resilience or low resistance-high resilience but an increase in areas with a synergistic pattern of low resistance-low resilience or high resistance-high resilience.The ecosystem in (c) shows a pattern of low resistance-low resilience due to deforestation (Def), drying (Dry) and warming (War), while the ecosystem in (d) shows a pattern of high resistance-high resilience benefiting from increased regional precipitation (Pre) and vegetation coverage (Veg).