Bayesian calibration and uncertainty analysis of an agroecosystem model under different N management practices

Process-based agroecosystem model provides a powerful tool to evaluate the effects of various field management practices on soil water balance, nitrogen (N) fates, and crop growth for cropland ecosystems. However, parameter calibration of those models is usually time-consuming and uncertain due to the involvement of complex soil and plant growth processes in models. In this study, the Different Evolution Adaptive Metropolis (DREAM) algorithm was successfully used to estimate model parameters and quantify uncertainties of the soil Water Heat Carbon Nitrogen Simulator (WHCNS) model. The calibration efficiency was further improved by identifying the sensitive parameters using global sensitivity analysis. The simulated soil water content, crop N uptake, and yield were in agreement with the measured values, with indices of NRMSE, IA, and NSE ranging from 6.4% to 22.1%, from 0.59 to 0.98, and from 0.43 to 0.92, respectively. The simulation of soil nitrate concentration showed a relatively large NRMSE that varied from 35.9% to 50.2% but was still within the acceptance range. This study suggested that the two-step method of global sensitivity analysis and DREAM algorithm is a robust way to estimate model parameters and quantify uncertainties of agroecosystem models.

Automated summary

The study successfully estimated model parameters and quantified uncertainties of agroecosystem models using the two-step method of global sensitivity analysis and the DREAM algorithm. The simulation results were consistent with measured values, demonstrating the effectiveness of this approach.