Variability and drivers of grassland sensitivity to drought at different timescales using satellite image time series

Drought is expected to increase in frequency and severity with climate change, leading to more intense impacts on grasslands and their associated ecosystem services. Complementary to ground experiments, remote sensing technologies allow for the study of drought impacts with large spatio-temporal coverage in real-life-conditions. We aimed to quantify the variability of grassland sensitivity to drought using a long-term satellite image time series of 394 temperate permanent grassland plots to identify factors influencing these sensitivities. Accordingly, we assessed the slope of the linear relationship between satellite-based vegetation status, using the standardized anomalies of the vegetation indices (VIs), and drought severity, using a modified version of the Standardized Precipitation Evapotranspiration Index (SPEI), from 1985 to 2019. The process was repeated for 24 VIs and five SPEI timescales. We then conducted a linear model selection procedure, using the grassland sensitivity derived from the most responsive VIs (i.e., VIs for which anomalies indicated a tighter linear relationship with the modified SPEI), to identify which grassland properties influenced sensitivity to drought. A total of 29 properties, grouped into pedoclimate, agricultural management, and vegetation diversity factors, were derived from ground measurements. Overall, we demonstrated that the influence of predictors on grassland sensitivity to drought varied across the drought integration timescales. Our results highlighted the significant mitigating effect of soil water holding capacity on sensitivity to drought for short timescales of fewer than 30 days. The date of first herbage use by farmers was positively related to grassland sensitivity to drought across all timescales. We also demonstrated that higher vegetation diversity significantly reduced sensitivity to drought. However, for the long timescales of drought integration, such influence was mainly redundant with management (i.e., shared partition of variance) suggesting complex cascading effects between agricultural practices and plant community structure that still need to be addressed comprehensively in future studies.