Quantifying Contributions of Uncertainties in Physical Parameterization Schemes and Model Parameters to Overall Errors in Noah-MP Dynamic Vegetation Modeling

Abstract

Quantifying contributions of errors in model structure and parameters to biases in a land surface model (LSM) is critical for model improvement. This paper investigated the uncertainties in parameterizations and parameters in the Noah with multiparameterization (Noah-MP) LSM with dynamic vegetation using eddy flux data. First, we conducted full factorial experiments of eight land subprocesses, followed by sensitivity analysis (SA) to identify the subprocesses for which possible parameterizations made significant difference. Then, based on the full factorial experiments and SA results, we selected the statistically optimal parameterizations combination and the most biased parameterizations combination. Lastly, we calibrated the parameters in two selected parameterizations combinations. The results showed that five subprocesses—surface exchange coefficient, soil moisture β threshold, radiation transfer, runoff and groundwater, and surface resistance to evaporation—had significant influence on the model performances, and the interactions were generally low but contributed up to 80% of the variation in the performance at some sites. In the optimization period, following the criterion by Moriasi et al. (2007, https://doi.org/10.13031/2013.23153), parameter optimization improved the performance of both parameterizations combinations at most sites to be satisfactory, and the superiority between two parameterizations combinations was preserved; in the validation period, adjusting the parameterizations was more robust than parameter optimization in improving LSMs. Finally, we found that uncertainty in soil parameters was much higher than that in vegetation parameters because the optimal soil parameters were significantly different among sites with the same soil types and recommended that spatially calibrating the soil parameters was a major issue for Noah-MP dynamic vegetation modeling.