The capability of CMIP6 models on seasonal precipitation extremes over Central Asia

The accurate extreme precipitation simulations of the GCMs during the historical period are more likely to offer reliable forecasts in the future. In this study, we evaluated the abilities of 22 GCMs of CMIP6 in simulating extreme precipitation over Central Asia in the early spring, summer, and late winter, when extreme precipitation can be very threatening. Based on seven extreme precipitation indices, robust skill score techniques such as Taylor diagrams and DISO (Distance between Indices of Simulation and Observation) are employed to rank the overall model performance. The results show that all models have difficulties in simulating the overall spatial patterns of all extreme precipitation in early spring and late winter. Differently, in summer, most of them can fairly well capture the spatial pattern of PRCPTOT and the CDD. Most models show consistent bias of extreme precipitation indices in early spring and late winter, with overall overestimated CDD and CWD and under-estimated SDII, Rx5day, R95PTOT, and R10mm. However, there is high inter-model variability in the regional bias of summer, and the areas of inconsistency that emerge always contain most of Xinjiang or its surrounding mountain regions, possibly due to a lack of topographic correction in these areas with very complex topography. The model ensemble simulations generally outperform the individual models in terms of CC (Correlation coef-ficient) and DISO, while the improvement is limited regarding RMS (Centered mean square error) and RSD (Relative standard deviations). This study may serve as a reference to further investigate the sources of persistent systematic biases in current GCMs over Central Asia, which is expected to assist policymakers in adopting strategies reflecting the state of scientific understanding of the likelihood.

Automated summary

The study evaluated the abilities of 22 GCMs of CMIP6 in simulating extreme precipitation over Central Asia and found that models have difficulties in capturing overall spatial patterns, with better performance in summer but still room for improvements in consistent bias and spatial distribution.