The optimal bias correction for daily extreme precipitation indices over the Yangtze-Huaihe River Basin, insight from BCC-CSM1.1-m

Due to the considerable biases of model simulation, bias correction is a widely used method to reduce the model biases. It is necessary to evaluate the performance of multiple bias corrections before selecting the optimal one. In this study, four statistical bias correction methods, i.e., the linear scaling (LS), quantile mapping (QM), distribution mapping (DM), and cumulative distribution function transform (CDFt), are introduced to correct the simulated of daily precipitation in Yangtze-Huaihe River Basin (YRB) in China from BCC-CSM1.1-m model. Furthermore, we evaluate their performance in reproducing the probability distribution of observed summer daily precipitation as well as the precipitation extremes. Results show that BCC-CSM1.1-m exhibits systematic biases in simulating the daily precipitation, especially for the tail of the distribution. The bias corrected simulation can significantly reduce the systematic biases. Among them, QM has the most significant improvement in simulating the probability distributions of daily precipitation with Brier scores (BS) dropping to 0 and while the significance scores (S-score) rise to 1 quickly. For the domain averages and spatial consistency of extreme precipitation, QM, CDFt, and LS have well simulation capabilities in the total precipitation and moderate rainy days. Meanwhile, only QM can reduce the biases of the rainfall days, precipitation intensity, and 95% quantile precipitation simultaneously. For example, the relative errors for 95% quantile precipitation are reduced from -57.8% in raw simulation to -2.76% in bias corrected simulation, and the spatial correlation coefficients are increased from -0.12 to 0.64. It is noted that the continuous dry days, are failed to be well corrected. The DM, based on the Gamma distribution, is less effective in characterizing the probability distribution of model-simulated precipitation, especially for the heavy precipitation. We believe that the QM and CDFt are the optimal methods used for applying the bias corrections in the extreme precipitation in the YRB.

Automated summary

This study introduces four statistical bias correction methods to correct the simulated daily precipitation in the Yangtze-Huaihe River Basin in China. The performance of these methods in reproducing the probability distribution of observed precipitation and extremes is evaluated. Results show that quantile mapping (QM) has the most significant improvement in simulating the probability distributions of daily precipitation, while linear scaling (LS), distribution mapping (DM), and cumulative distribution function transform (CDFt) perform well in total precipitation and moderate rainy days. Only QM can reduce the biases of rainfall days, precipitation intensity, and 95% quantile precipitation simultaneously. However, DM is less effective in characterizing the probability distribution of model-simulated precipitation, especially for heavy precipitation.