Global vegetation resilience linked to water availability and variability

Abstract

Quantifying the resilience of vegetated ecosystems is key to constraining both present-day and future global impacts of anthropogenic climate change. Here we apply both empirical and theoretical resilience metrics to remotely-sensed vegetation data in order to examine the role of water availability and variability in controlling vegetation resilience at the global scale. We find a concise global relationship where vegetation resilience is greater in regions with higher water availability. We also reveal that resilience is lower in regions with more pronounced inter-annual precipitation variability, but find less concise relationships between vegetation resilience and intra-annual precipitation variability. Our results thus imply that the resilience of vegetation responds differently to water deficits at varying time scales. In view of projected increases in precipitation variability, our findings highlight the risk of ecosystem degradation under ongoing climate change.

Fig. 1. Spatial distribution of data used in this study.

Long-term median A vegetation optical depth (VOD, 1987–2017,) and B normalized difference vegetation index (GIMMS3g NDVI, 1981–2015,). C IGBP Land-cover classes, masked for anthropogenic influence (Methods). D Global aridity index, adapted from WorldCLIM. Note that higher values correspond to drier conditions. E Walsh-Lawler Seasonality Index and F normalized inter-annual precipitation variability (Methods) based on ERA5 data (monthly, 1981–2021). See Supplementary Fig. S1 for a similar map of MODIS NDVI.

Fig. 2. Comparison of aridity and intra- and inter-annual precipitation variability in their relative importance for vegetation resilience at the global scale.

A, B Vegetation optical depth (VOD), C, D GIMMS3g normalized difference vegetation index (NDVI), and E, F MODIS NDVI. Aridity compared to intra-annual (left column) and inter-annual (right column) precipitation variability. Hexbins colored by recovery rate computed from AC1 (minimum five points per bin). Values of the recovery rate λ closer to zero imply lower resilience. Transition from water surplus (aridity <1) to deficit marked with dashed vertical line; there is a sharp increase in resilience as water availability increases. Higher inter-annual precipitation variability (right column) consistently leads to lower resilience; intra-annual precipitation variability, i.e., seasonality, has a more varied impact. See Supplementary Fig. S3 for a direct comparison of intra- and inter-annual precipitation variability.

Fig. 3. Vegetation resilience as a function of aridity at the global scale, separated by land-cover type.

Vegetation resilience λ estimated empirically (A, B) and via the AC1 (C, D) for vegetation optical depth (VOD, left column) and MODIS NDVI (right column). Binned medians shown as solid dots (Kendall-Tau (KT) p <0.05) and transparent arrows (KT p>0.05), with 25–75th percentiles of each bin shown as connected vertical lines capped with hatches. Land covers with less than 1000 points or less than 10 bins of at least 50 members are omitted. E KT coefficients (aridity vs AC1-derived λ, panels C, D) for each land-cover type. Significant (p < 0.05) KTs shown as a black triangle (KT of median binned data, cf. C, D), insignificant relationships (p > 0.05) shown as a black circle. Additional box-plot of 1000 randomly sampled surrogates (box edges: 25–75th percentiles, black line: median) shown with red for MODIS NDVI, orange for AVHRR NDVI, and blue for VOD. KT of medians consistently higher than box plots due to random sampling (see Methods). Both VOD and NDVI exhibit lower resilience—i.e., λ closer to zero, see Methods—with lower water availability across the majority of land-cover types. Equivalent figure for mean annual precipitation (MAP) shown as Supplementary Fig. S4, and for mean annual soil moisture shown as Supplementary Fig. S5. Figure for aridity showing all three instruments and metrics as Supplementary Fig. S6.

Fig. 4. Vegetation resilience as a function of precipitation seasonality in terms of the Walsh-Lawler seasonality index (Methods), separated by land-cover type.

Vegetation resilience λ estimated empirically (A, B) and via the AC1 (C, D) for vegetation optical depth (VOD, left column) and MODIS NDVI (right column). Binned medians shown as solid dots (Kendall-Tau (KT) p < 0.05) and transparent arrows (KT p>0.05), with 25–75th percentiles of each bin shown as connected vertical lines capped with hatches. Land covers with less than 1000 points or less than 10 bins of at least 50 members are omitted. E KT coefficients (aridity vs AC1-derived λ, panels C, D) for each land-cover type. Significant (p < 0.05) KTs shown as a black triangle (KT of median binned data, cf. C, D), insignificant relationships (p > 0.05) shown as a black circle. Additional box-plot of 1000 randomly sampled surrogates (box edges: 25–75th percentiles, black line: median) shown with red for MODIS NDVI, orange for AVHRR NDVI, and blue for VOD. KT of medians consistently higher than box plots due to random sampling (see Methods). Both VOD and NDVI exhibit lower resilience—i.e. λ closer to zero, see Methods—with lower water availability across the majority of land-cover types. While for all three considered vegetation datasets empirical recovery rates generally decrease with more concentrated precipitation, the relationship between Walsh-Lawler seasonality and recovery rates is less steep than for aridity (Fig. 3). Figure showing all three instruments and metrics as Supplementary Fig. S7.

Fig. 5. Vegetation resilience as a function of normalized inter-annual precipitation variability (Methods), separated by land-cover type.

Vegetation resilience λ estimated empirically (A, B) and via the AC1 (C, D) for vegetation optical depth (VOD, left column) and MODIS NDVI (right column). Binned medians shown as solid dots (Kendall-Tau (KT) p < 0.05) and transparent arrows (KT p > 0.05), with 25–75th percentiles of each bin shown as connected vertical lines capped with hatches. Land covers with less than 1000 points or less than 10 bins of at least 50 members are omitted. E KT coefficients (aridity vs AC1-derived λ, panels C, D) for each land-cover type. Significant (p < 0.05) KTs shown as a black triangle (KT of median binned data, cf. C, D), insignificant relationships (p>0.05) shown as a black circle. Additional box-plot of 1000 randomly sampled surrogates (box edges: 25–75th percentiles, black line: median) shown with red for MODIS NDVI, orange for AVHRR NDVI, and blue for VOD. KT of medians consistently higher than box plots due to random sampling (see Methods). For both VOD and NDVI we infer lower resilience for higher relative inter-annual precipitation variability. Equivalent figure for normalized inter-annual soil moisture variability shown as Supplementary Fig. S8. Figure showing all three instruments and metrics as Supplementary Fig. S9.