Multi-Scale Assessment of Prediction Uncertainty in Coupled Reactive Transport Models Conducted at the Florida State University

Abstract

This report summarizes the research activities in the Florida State University for quantifying parametric and model uncertainty in groundwater reactive transport modeling. Mathematical and computational research was conducted to investigate the following five questions: (1) How does uncertainty behave and affect groundwater reactive transport models? (2) What cause the uncertainty in groundwater reactive transport modeling? (3) How to quantify parametric uncertainty of groundwater reactive transport modeling? (4) How to quantify model uncertainty of groundwater reactive transport modeling? and (5) How to reduce predictive uncertainty by collecting data of maximum value of information or data-worth? The questions were addressed using Interdisciplinary methods, including computational statistics, Bayesian uncertainty analysis, and groundwater modeling. Both synthetic and real-world data were used to evaluate and demonstrate the developed methods. The research results revealed special challenges to uncertainty quantification for groundwater reactive transport models. For example, competitive reactions and substitution effects of reactions also cause parametric uncertainty. Model uncertainty is more important than parametric uncertainty, and model averaging methods are a vital tool to improve model predictions. Bayesian methods are more accurate than regression methods for uncertainty quantification. However, when Bayesian uncertainty analysis is computationally impractical, uncertainty analysis using regression methods still provides insights into uncertainty analysis. The research results of this study are useful to science-informed decision-making and uncertainty reduction by collecting data of more value of information.