Parametric Sensitivity of Vegetation Dynamics in the TRIFFID Model and the Associated Uncertainty in Projected Climate Change Impacts on Western U.S. Forests.

Changing climate conditions impact ecosystem dynamics and have local to global impacts on water and carbon cycles. Many processes in dynamic vegetation models (DVMs) are parameterized, and the unknown/unknowable parameter values introduce uncertainty that has rarely been quantified in projections of forced changes. In this study, we identify processes and parameters that introduce the largest uncertainties in the vegetation state simulated by the DVM Top‐down Representation of Interactive Foliage and Flora Including Dynamics (TRIFFID) coupled to a regional climate model. We adjust parameters simultaneously in an ensemble of equilibrium vegetation simulations and use statistical emulation to explore sensitivities to, and interactions among, parameters. We find that vegetation distribution is most sensitive to parameters related to carbon allocation and competition. Using a suite of statistical emulators, we identify regions of parameter space that reduce the error in modeled forest cover by 31±9%. We then generate large initial atmospheric condition ensembles with 10 improved DVM parameterizations under preindustrial, contemporary, and future climate conditions to assess uncertainty in the forced response due to parameterization. We find that while most parameterizations agree on the direction of future vegetation transitions in the western United States, the magnitude varies considerably: for example, in the northwest coast the expansion of broadleaf trees and corresponding decline of needleleaf trees ranges from 4 to 28% across 10 DVM parameterizations under projected future climate conditions. We demonstrate that model parameterization contributes to uncertainty in vegetation transition and carbon cycle feedback under nonstationary climate conditions, which has important implications for carbon stocks, ecosystem services, and climate feedback. Plain Language Summary: Changing climate conditions can impact the composition of an ecosystem. As plant types better suited for the new climate conditions begin to thrive, they can out‐compete the historical vegetation. This transformation can have impacts on the local water cycle and, if changes are large enough, on the global carbon cycle. Dynamic vegetation models (DVMs) are used to analyze the impacts of climate on vegetation and project future changes. DVMs are used across scales ranging from watersheds, informing management decisions, to the globe as an important component of the global carbon cycle. The mathematical representations of processes in DVMs are imperfect and introduce uncertainty in model projections. In this study we identify processes that introduce the largest uncertainties in the vegetation distribution modeled by a DVM. We then identify settings of model parameters that improve the modeled historical vegetation when compared to observations. While unique model parameterizations simulated the historical vegetation well, each parameterization simulated different future vegetation transitions under future climate conditions. Our study shows that model parameterization introduces uncertainty in the projected future changes in vegetation. Quantifying and appreciating this uncertainty are important when considering management decisions based on projected vegetation changes at the local to global scale. Key Points: Multiple unique parameterizations of a dynamic vegetation model met evaluation criteriaParameterizations simulated varying amounts of vegetation transitions under nonstationary climate contributing to uncertainty in the simulated carbon cycleSimulated vegetation distributions were most sensitive to parameters related to carbon allocation and competition [ABSTRACT FROM AUTHOR]