Evaluating key climatic and ecophysiological parameters of worldwide tree mortality with a process-based BGC model and machine learning algorithms

Drought-induced tree mortality has been increasing worldwide under climate change; therefore, forests will become more vulnerable as warming continues. Meanwhile, carbon starvation and hydraulic failure have been proposed as main drought-induced mortality mechanisms, mostly validated through individual tree-level experiments. However, there lack of a unified way to monitor and assess tree mortality across the different biomes and climate regions. In this sense, process-based biogeochemical (BGC) modeling may be an effective tool for simulating and understanding ecophysiological processes for tree mortality at large spatial scales. In this study, a hydraulic vulnerability curve for percentage loss of conductivity (PLC) was added to the BGC-NSCs model, the modified version of the BIOME-BGC with two additional non-structural carbohydrates (NSCs) pools. And then, we simulate the model at the sites around the world where tree mortality were reported. Using sensitivity analysis and machine learning algorithms for hydraulic stress, PLC and NSCs showed a high sensitivity and significance to tree mortality within the modeling framework. The model simulations also reveal the relationship between PLC and NSCs based on mortality stress intensity, plant functional types, and climate conditions, further validated with the results of previous experiment studies at the plot scale. This study proposes a potential to estimate eocphysiological variables at the regional scale using the BGC model, and to use high sensitivity variable, such as PLC and NSCs, as effective diagnostics for hydraulic stress across different biomes and climate regions.