Parameter estimation for a global model of terrestrial biogeochemical cycling by an iterative method

We have implemented a geographically distributed ecosystem model TERRA for the carbon, nitrogen, and water dynamics of the global terrestrial biosphere. The ecosystem model in each grid cell has state variables of soil water; vegetation carbon, soil carbon, vegetation nitrogen, soil organic nitrogen, soil inorganic nitrogen, and a variable for allocation. Eight parameters associated with eight carbon or nitrogen fluxes are determined during model calibration at specific sites for each of 17 vegetation types that cover the globe. Calibration is performed by an iterative method that brings calculated fluxes into agreement with observed fluxes over successive iterations. For the 17 vegetation types, calibration required a geometric mean number of 123 iterations for convergence with a geometric S.D. of 2.25. The minimum and maximum iterations required was 52 and 1060 for xeromorphic woodland and tropical evergreen forest, respectively. The parameter controlling gross primary productivity Cmax was found to be correlated with maximum projected leaf area index (plus cover of nonvascular plants) with a correlation coefficient of 0.90 (r2=0.81). Correlation of parameters with the input fluxes and average standing crops corresponding to the iteration equation was high in those cases in which temperature and precipitation were not explicit factors in the iteration equation. The parameters Kfall, Lnc, and Nloss had correlation coefficients of 1.0 each with the appropriate ratio of the matching observed transfer flux and standing crop. In iteration equations with explicit dependence on temperature, soil moisture, and other factors, the correlation coefficients were less than 1.0 with these ratios; the parameters Cmax, Kr, Kd, Nmax, and Nup had correlation coefficients of 0.83, 0.66, 0.60, 0.61, and 0.87, respectively. These correlation coefficients indicate the importance of environmental factors such as temperature and soil moisture in the calibration process and the robustness of the iteration method in determining parameters in systems with time-varying coefficients.