Four decades of declining stability in global dryland ecosystems despite widespread greening

Dryland ecosystems play a key role in regulating both the trend and interannual variability (IAV) of the terrestrial carbon sink and provide key ecosystem services to over 2 billion people. Recent work has highlighted the sensitivity of dryland ecosystems to changing climate, yet uncertainties from satellite data, reliance on land-surface models, and analytical techniques have led to mixed conclusions. Here, we use a unique set of long-term satellite-derived leaf area index (LAI) datasets that include novel machine learning algorithms to remove artifacts that have hindered greening trend analyses in the past. From 1982 to 2020, we find a persistent increase in LAI over 54.7% of drylands, but surprisingly this trend is accompanied by an increase in its interannual variability ( ). Increasing LAI_cv is found over 81.8% of drylands, indicating a decline in stability despite long term greening trends. This increasing variability is driven by a divergence between increasing annual growing-season maximum LAI and simultaneously a declining minimum. The observed rise in  also correlates well with an increasing vegetation sensitivity over time to rainfall and to shifts in intra- and interannual rainfall variability. While current dynamic global vegetation models (DGVMs) can reproduce long-term greening, they fail to capture increases in , instead simulating declining variability and convergent LAI trends. Given their disproportionate role in driving the interannual growth rate of atmospheric CO₂, declining stability of global dryland systems is indicative of a potential transition to alternative states that will further impact the carbon cycle and critical ecosystem services that drylands provide.