Improved Validation of Earth System Models

Abstract

Earth system models are numerical models that simulate the physical dynamics of the atmosphere and the oceans, as well as other chemical and biological processes. With increases in computational power and model complexity, there is an increased need for performance assessment to better inform decision making. In this thesis, we provide an overview of common statistical methods and develop new approaches for the validation of Earth system models, with a focus on application-specific validation metrics. We apply these methods to the simulation of nearshore ocean temperatures along the Atlantic coast of Nova Scotia, emphasizing techniques for the evaluation of short-term variability and extremes. To assess the effectiveness of these approaches across various types of numerical models, we compare two iterations of a regional FVCOM model, the GLORYS12v1 reanalysis dataset, and a representative CMIP6 model against temperature observations at two nearshore study sites.