The influence of bias correction of global climate models prior to dynamical downscaling using regional climate models over Australia

Karuru Wamahiu

Regional climate models (RCMs) are used to dynamically downscale global climate model (GCMs) which have coarse resolutions of 100 to 250 km to much higher resolutions, typically ranging from 4 to 50 km. A limitation of using GCM outputs as inputs in RCMs is that biases from GCMs get transferred to the RCM and, hence, deteriorate the regional simulations. One solution is to “bias correct” the GCM outputs using re-analysis data-sets as a “surrogate” truth prior to using them in RCM simulations. This approach has been shown to improve regional climate simulations in several regions, but relatively few studies have focused on Australia. This thesis investigates the influence of bias correcting the mean climatological component of GCM outputs against ERA-Interim reanalysis for regional climate simulations over the CORDEX-Australasia domain using the Weather Research and Forecasting (WRF) model. A 4 member GCM ensemble was bias-corrected to investigate the influence of bias correction on historical climate and then on projected changes in climate. For historical climate, results were compared against observationsand showed that over decadal time scales, bias correction removes large systematic precipitation and temperature biases. However, there were instances where bias correction introduced biases where there were none, introduced biases of the opposite sign, or even enhanced existing biases. Overall, the reduction in large systematic biases was greater than either the introduction of new biases, or the enhancement of existing biases. For projected changes in climate, we found that while differences between bias-corrected and noncorrected RCM simulations can be substantial, these differences were generally smaller than the models’ inter-annual variability. Furthermore, results showed that the RCM ensemble solutions tend to converge when the GCMs are bias-corrected against a common re-analysis, which provides an advantage in improving confidence in projections of future climate. Finally, a single GCM was dynamically downscaled to investigate the relative influence of bias correction against other important factors — namely, spectral nudging and model physics—on both historical and projected changes in climate. Results showed that the choice of model parameterisaton schemes had by far the largest influence on model skill and the change in climate. Furthermore, results showed that it is important to first assess the performance of non-corrected GCM-driven simulations against the reference re-analysis driven simulations, as bias correction may not be necessary if the GCMdriven simulation already performs well compared to the reference simulation. We also showed that even if bias correction of GCM lateral boundary conditions and spectral nudging may add value, using a multi-physics approach is more important. Consequently, we recommend that given limited computational and data resources, regional climate modelling groups should prioritize a multi-physics ensemble of the RCM to better account for internal physics-driven variability, and the use of bias-correction or spectral nudging should be secondary.

The influence of bias correction of global climate models prior to dynamical downscaling using regional climate models over Australia

Files and links (1)

Abstract

Details

Metrics