UC San Diego

To make useful predictions about the behavior of a system, a model of the system is built. Complex models, such as environmental or electrochemical models, include numerous inputs that influence the model outputs. The true values of these inputs are often unknown and must be estimated using empirical data from experiments. The accuracy of these estimates improves as the uncertainty associated with the inputs decreases. While careful measurements can reduce input uncertainty, rigorous measurement of all inputs in complex models can be prohibitively expensive due to high experimental or computational costs, making it essential to prioritize which inputs to measure precisely. Therefore, it is crucial to identify the inputs that have the greatest influence on the model outputs. Sensitivity analysis provides a framework for distinguishing important inputs by quantifying or qualifying the effects of inputs on model outputs. This allows researchers to focus resources on improving the accuracy of the most influential inputs, leading to more reliable model predictions.

This dissertation applied global sensitivity analysis (GSA) to two complex models: a groundwater flow model and a lithium-ion battery model. Two well-known GSA methods were employed: the Morris method and the Sobol' method. The Morris method assesses input influence by computing the sample mean and standard deviation of elementary effects of each input. The Sobol' method calculates first-order sensitivity indices and total sensitivity indices to quantify input importance.

For the groundwater flow model, the GSA results indicated that recharge flux was the primary driver of variations in the net flux of seawater intrusion. In the lithium-ion battery model, the thickness of the positive electrode had the greatest impact on battery life.