Initial condition uncertainty exerts a large and persistent influence on model simulations of ecosystem carbon dynamics in California

Abstract Uncertainties in terrestrial ecosystem models limit their predictive power. Efforts to reduce projection error have rarely focused on constraining uncertainty in the initial state of the ecosystem, however, despite evidence that matching model initial conditions to real-world observations reduces overall model bias. Here we use an integrated land change and carbon gain-loss model to evaluate the influence of initial condition uncertainty on simulations of California wildland ecosystems during the years 1985–2020. We generated 36 initial conditions scenarios by varying the source data used to initialize state variables and then ran simulations based on each of these scenarios under a constant set of historical conditions. We found that discrepancies in initial forest extent and initial forest age among scenarios generated wide uncertainty ranges in model estimates of terrestrial ecosystem carbon stocks and flux rates at the outset of the simulation period, but differences in initial forest composition had no impact. Over time, forest age became more homogeneous across model scenarios leading to exponential rates of decline in the uncertainty ranges of live biomass and dead wood carbon but little to no impact on uncertainties in litter and soil organic carbon. Uncertainties in individual carbon flux rates were consistent with uncertainties in their source pools. In contrast, model estimates of ecosystem carbon balance demonstrated a shift in system behavior not apparent in trends for individual carbon stocks and fluxes. Specifically, estimates of ecosystem carbon balance converged across scenarios for the first 20 years of the simulation period but then began to diverge at an accelerating rate, possibly due to weakened resilience to the increased frequency and severity of climate-driven disturbances. Our results demonstrate that uncertainty in the initial state of the system can have large and persistent impacts on the predictability of ecosystem carbon dynamics, and that ongoing shifts in external forcing by climate and climate-driven disturbances can exacerbate these impacts.