Process refinement contributed more than parameter optimization to improve the CoLM's performance in simulating the carbon and water fluxes in a grassland

Abstract

The Common Land Model (CoLM) has been widely used to estimate carbon and water fluxes at site or regional scales, but the model is still underperforming in dryland ecosystems. Our research focuses on the joint analysis of both modifying the model process and using parameter optimization techniques to improve the model's performance in a semi-arid grassland ecosystem in Xinjiang, China. The study presents a comparison of the simulated carbon and water fluxes by replacing the root water uptake function (RWUF) of the CoLM and by using particle swarm optimization (PSO) algorithm to optimize the most sensitive parameters. Prior to PSO, the method of Morris one-factor-at-a-time (MOAT) is utilized to screen out parameters that have strong effects on gross primary production (GPP) and latent heat flux (LE) in CoLM. Either modifying the root water uptake process in the CoLM or optimizing model parameters can significantly reduce the biases of the simulated GPP, LE, and water use efficiency (WUE). The coefficient of determination (R2) with the modified RWUF in the CoLM increases from 0.85 to 0.92 for GPP and from 0.76 to 0.81 for LE. Meanwhile, the root mean square error (RMSE) decreases from 3.57 μmol m−2 s−1 to 2.78 μmol m−2 s−1 for GPP and from 50.75 W m−2 to 46.85 W m−2 for LE. Using the PSO approach, the R2 increases to 0.89 and RMSE decreases to 2.92 μmol m−2 s−1 for GPP, while, the R2 increases to 0.79 and RMSE decreases to 46.16 W m−2 for LE. Therefore, modifying the model process contributed more to improve the model simulations than using parameter estimation techniques. Our study recommends that a justified refinement in model structure plays vital role in quantifying the carbon and water fluxes in dryland ecosystems or other ecosystems.