Drought stress in LPJmL-5: benchmarking and model improvements

Climate change is expected to increase the frequency and severity of Multi-Year Droughts (MYDs), yet their impacts on vegetation remain poorly understood. While satellite records provide valuable insights, they span only recent decades, limiting the number of MYDs available for analysis. Dynamic global vegetation models (DGVMs), such as LPJmL-5 (von Bloh et al., 2018), can help overcome this limitation by simulating vegetation dynamics over longer timescales. However, their ability to capture drought impacts has not yet been systematically evaluated. In this study, we benchmarked LPJmL-5 against MODIS-derived Gross Primary Production (GPP) to assess how well it captures vegetation responses to (multi-year) droughts. We show that LPJmL-5 reproduces GPP reasonably well, but there is a performance decline in parts of the Southern Hemisphere and in regions with croplands. During MYDs, the model captures the main spatial and temporal patterns of GPP decline, yet it tends to overestimate vegetation resilience at drought onset and simulates rapid post-drought recovery, leading to muted overall drought impacts. These biases appear to arise from a simplified representation of vegetation mortality processes in the model. As a result, long-term losses in biomass and shifts in ecosystem structure are often underestimated. To improve this behaviour, we incorporated a drought mortality function into LPJmL-5. This links mortality to water stress and vapour pressure deficit, with vegetation specific parameterization. We calibrated and evaluated its performance across known drought events. With the extended drought mortality representation in LPJmL-5, we refine the process representation, which in turn leads to more realistic vegetation dynamics and ultimately greater confidence in predictions of ecosystem responses under a changing climate.