Paper Published in Journal of Geophysical Research: Atmospheres

Title: Factorial Sensitivity Analysis of Physical Schemes and Their Interactions in RegCM

Journal: Journal of Geophysical Research: Atmospheres

DOI: https://doi.org/10.1029/2020JD032501

Abstract: It is well known that the choices of physical schemes in Regional Climate Models (RCMs) can cause considerable uncertainties in future climate projections. In this study, a factorial sensitivity analysis method has been proposed to screen out statistically significant schemes and interactions, which assists in selecting the optimized physical scheme combination from a long‐term perspective with affordable computational costs. The Regional Climate Model (RegCM) is used as an example to illustrate how the approach works. In detail, all schemes are fully tested through 120 experimental runs based on a factorial design; the contributions and statistical significance (P value) of individual schemes and their interactions to temperature, precipitation, wind speed, and wind direction are then quantified. The performance of the proposed approach is then demonstrated through a case study of Canada. The results indicate that there exist considerable spatial and temporal simulated variations associated with different scheme combinations. It is also suggested that individual physical schemes have dominant influences on simulated variations, but some effects explained by their interactions are statistically significant and thus cannot be neglected. In particular, the planetary boundary layer (PBL) scheme, moisture scheme, and land surface model are found to be the dominant factors affecting the uncertainties of temperature, precipitation, and wind speed in future climate projections over Canada, respectively. Furthermore, the potential relationships between the vegetation cover conditions and the sensitivity of physical schemes are explored. The proposed approach is an attempt to analyze the sensitivity influenced by not only individual physical schemes and also their multilevel interactions.