__C. Li__, __E. Sinha__, D. E. Horton, N. S. Diffenbaugh and __A. M. Michalak__

### SUMMARY

*Estimates of the impacts of climate change to, for example, water resources and food production, rely on climate model projections. Output from climate models is prone to biases, precluding its direct use in impact studies. Bias correction is routinely applied to circumvent this obstacle. Commonly used bias correction methods fail to correct biases in inter-variable relationships, which are important for assessing impacts. We introduce a joint bias correction methodology (JBC) that explicitly accounts for inter-variable relationships as part of the correction procedure. Application of JBC to precipitation and temperature from CMIP5 simulations shows that JBC not only reduces biases in the mean and variance of precipitation and temperature, but also biases in their correlations.*

### ABSTRACT

Bias correction of meteorological variables from climate model simulations is a routine strategy for circumventing known limitations of state-of-the-art general circulation models. Although the assessment of climate change impacts often depends on the joint variability of multiple variables, commonly used bias correction methodologies treat each variable independently, and do not consider the relationship among variables. Independent bias correction can therefore produce non-physical corrections and may fail to capture important multivariate relationships. Here, we introduce a joint bias correction methodology (JBC) and apply it to precipitation (P) and temperature (T) fields from the CMIP5 model ensemble. This approach is based on a general bivariate distribution of P-T, and can be seen as a multivariate extension of the commonly used univariate quantile mapping method. It proceeds by correcting either P or T first and then correcting the other variable conditional upon the first one, both following the concept of the univariate quantile mapping. JBC is shown to not only reduce biases in the mean and variance of P and T similarly to univariate quantile mapping, but also to correct model-simulated biases in P-T correlation fields. JBC, using methods such as the one presented here, thus represents an important step in impacts-based research as it explicitly accounts for inter-variable relationships as part of the bias correction procedure, thereby improving not only the individual distributions of P and T, but critically, their joint distribution.