A new multi-model absolute difference-based sensitivity (MMADS) method to screen non-influential processes under process model and parametric uncertainty

Sensitivity analysis under model uncertainty has gained increasing attention for advancing our understanding of complex Earth and Environmental Systems with growing complexity interacting physical, chemical, and biological processes. We develop a multi-model absolute difference-based sensitivity analysis method (MMADS) to screen non-influential system processes and parameters from further investigation. In MMADS, a system process may be represented by multiple alternative process models, and integrating process models of these processes forms multiple system models. The basic ideas of MMADS are to first evaluate the differences of a system model output caused by varying process models and/or parameter values embedded in the process models, and then calculate the mean and variance of the differences for investigating process influence as in the Morris method. MMADS is an extension of the Morris method from a parameter space to a joint parameter-model space to address both process model uncertainty and process model parameter uncertainty. To reduce computational cost of using MMADS, a binning method is developed. MMADS and its numerical implementations are evaluated using two experiments: Sobol’s G*-functions with analytical solutions of the mean and variance of the differences, and groundwater flow modeling with three interacting recharge, geology, and snowmelt processes. Results indicate that , in comparison with the variance-based process sensitivity analysis method, MMADS is computationally efficient and has great potential for in identifying non-influential processes.

[Download paper here]

Recommended citation: Yang, J. and Ye, M. (2022). “A new multi-model absolute difference-based sensitivity (MMADS) method to screen non-influential processes under process model and parametric uncertainty” Journal of Hydrology. 2022 (In presss).