Unexpected relations between temporal resilience indicators and trend breakpoints in a dryland catchment

In this study, we compare two common frameworks to assess ecosystem resilience by looking at temporal changes. One framework is based on the slowing down in the rate of ecosystem recovery from small disturbances, which implies a loss in resilience signalled by higher temporal correlation and variance. The other method applies time-series segmentation to detect breaks in the trend component of the time series, which are interpreted as (positive or negative) shifts in ecosystem functioning. Using remote-sensing vegetation greenness time series for a dryland catchment and a period including a severe drought, we hypothesised that an increased temporal correlation and variance, representing resilience loss, preceded negative drought breakpoints and vice versa. We however found more support for the opposite. Catchment areas responding with the most frequent positive breakpoint to drought (positive reversal) showed higher temporal correlation and variance than areas with the most frequent negative drought breakpoint (interrupted decrease). Further, the lowest temporal correlation and variance were observed in areas with a positive breakpoint in response to drought or without a significant trend in greenness. These results question the robustness of the indicatory potential of temporal early warnings and highlight the need for studies cross-validating resilience indicators.