Bias correction of climate model output for impact models

Abstract

Bias correction is an unavoidable, and often undesired, step in climate projection and climate impact modeling. Here we briefly explain the mathematical and physical framework behind bias correction and propose a working definition for bias in climate simulations. We then present the simplest, univariate, methods for bias correction along with their underlying assumptions, limitations, and a brief historical background. In this context, we will discuss in particular the applicability to extreme events, and among others, issues concerning appropriate observational datasets, cross-validation, the stationarity assumption, and the limitations of the univariate approach. We follow this section by introducing the framework behind simple multivariate bias correction methods, continue with a presentation of some of the most prominent state-of-the-art multivariate bias correction methods used today, and discuss some of the limitations and trade-offs involved when adjusting multiple variables. Lastly, we describe and discuss the application of observation-based constraints to model ensembles as a potential (multivariate) complement to traditional statistical bias correction. The chapter concludes with a discussion of the most promising lines of research and development, the looming challenges, and a few best-practices recommendations for bias correcting impact simulations.