Uncertainty and sensitivity analysis of process-based tree mortality models
One of the largest uncertainties in Earth System Models (ESMs) lies in our understanding of abrupt drought-induced disturbances to terrestrial ecosystems and their feedbacks to future climate change. The primary culprit underlying this uncertainty is our limited capability to quantify different mechanisms that directly leads to drought-caused tree mortality. Most current ESMs use statistical relationships between mortality and growth rate or growth efficiency; however, these models do not account for potential acclimation and may fail under future climate conditions. Therefore, there is an increasing interest to develop processes-based models that simulate key components of tree mortality. One key limitation of such process-based models is that the number of parameters increases to account for complex processes, which can potentially lead to large uncertainties in model predictions. For purposes of improving modeling at large scale and reducing cost of measurement at field sites, it is critical to identify those key parameters that dominate uncertainty in mortality prediction. We applied a comprehensive global sensitivity to explore the importance of more than 30 parameters that determine the mortality based on the assumption of carbon starvation, hydraulic failure, and their interdependency, under the modeling framework of the Ecosystem Demography model.
Our results showed that the maximum photosynthetic rate, the minimum stomatal conductance under drought, critical carbon storage that initiates mortality, the reference mortality rate under stress as determined jointly by plant defense and insect attacks, and the minimum leaf water potential at which net photosynthesis reaches zero are all important parameters that substantially influence tree mortality predictions. Our results suggest that further study that potentially account for belowground process and the interaction between carbon and water limitation could substantially improve mortality prediction.