Integrating of Bayesian model averaging and formal likelihood function to enhance groundwater process modeling in arid environments

Abstract

Predictive uncertainty has influenced by traditional assumptions about the residual error. This study attempts to perform an uncertainty analysis of ensemble groundwater modeling through Bayesian Model Averaging- BMA in conditions that these assumptions are violated. This study hired a framework accompanied by BMA to generate an anticipative inference of numerical groundwater contents with non-stationary, dependent, and non‐Gaussian errors. Groundwater levels were numerically simulated using three different methods for an arid aquifer in Iran. Subsequently, the BMA approach generated an improved estimate of groundwater levels by incorporating various likelihood contexts (i.e., formal and informal) to address assumptions related to residual errors. Results showed that the formal likelihood function deals with residual assumptions well, primarily for stationary and normality. Additionally, the results of the uncertainty analysis revealed that the formal function-based BMA outperforms the informal function-based BMA. Furthermore, the final predictions generated by the formal function-based BMA are comparable to the outputs of the Mesh free method in terms of RMSE.