Optimizing Parameters of a Terrestrial Ecosystem Model against Eddy Covariance Measurements from Ten FLUXNET Sites using Smoothed Ensemble Kalman Filter

Modeling carbon dynamics need to resolve two challenging issues: (1) adequate quantification of the behavior of model parameters (temporal and spatial changes), and (2) assessment of model output uncertainty. A global network (i.e., FLUXNET) of eddy covariance flux towers measuring carbon, water, and energy fluxes is likely to be the best source of data to address these issues. Because these fluxes are intimately coupled with each other through ecosystem functions, a multiple constraint approach is required to estimate model parameters. In this study we used a Smoothed Ensemble Kalman Filter (SEnKF) method and flux data from ten eddy covariance flux towers in the U.S. to calibrate and validate the General Ensemble biogeochemical Model System (GEMS). The ecosystem types analyzed in this study include cropland, grassland, deciduous forest, evergreen forest, mixed forest, and savanna. We identified that photosynthetic parameters (e.g., the maximum ecosystem-specific net primary production rate) were well-constrained by carbon, water, and energy fluxes, but the parameters controlling respiration and carbon decomposition (e.g., maximum decomposition rates in soil carbon pools) were not robustly determined. The optimized photosynthetic parameters allowed GEMS to better predict the carbon fluxes on the ten flux tower sites. Seasonal variability of photosynthetic parameters in deciduous forests and croplands was evident, but not in evergreen forests and grasslands. The sensitivity of the optimal parameter values to prior parameter values and initial carbon pool sizes was discussed and the posterior uncertainty was also analyzed.