Inter-variable relations in regional climate model outputs

Abstract:

Regional climate models (RCMs) intent to provide physically consistent climate data to the climate change impact research community. However, the effects of parametrisations of unresolved sub-grid processes and systematic biases in the model output requires not only a post-processing in form of bias adjustment but also an analysis of inter-variable relations. Many impact models require several climate variables as input data, which makes it necessary to check if the inter-variable dependence structure is simulated realistically by RCMs.

A common practice is to bias adjust RCM output variables to improve their individual distribution and mean climate characteristics. This can be done by empirical bias adjustment procedures such as quantile mapping. However, applying statistical bias adjustment procedures on individual variables may alter the inter-variable relationships given by the climate model and hence distort the physical consistency.

In our study we examined the inter-variable relations of RCM output variables by using estimates of conditional probability density functions for pairs of variables. Conditional densities obtained from multiple European RCMs were compared with those obtained from observations. We quantified the extent to which these conditional density estimates are distorted by an empirical bias adjustment procedure.

Additionally, the influence of the model physics on the representation of inter-variable relations is analysed for a 24 member perturbed physics ensemble of WRF simulations in the U.S.. Here, multiple observational data sets were used to address the influence of observational uncertainties on the analysis. Finally, the results obtained from the European and U.S. modelling initiatives are compared to provide a common basis on the representation of inter-variable relations in RCM outputs.