资源环境科技发展态势分析平台

A new satellite-based precipitation data set, Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Climate Data Record (PERSIANN-CDR), with long-term time series dating back to 1983, can be one valuable data set for climate studies. This study investigates the feasibility of using PERSIANN-CDR as a reference data set for climate studies. Sixteen Coupled Model Intercomparison Projection Phase 5 (CMIP5) models are evaluated over the Xiang River basin, southern China, by comparing their performance on precipitation projection and streamflow simulation, particularly on extreme precipitation and streamflow events. The results show PERSIANN-CDR is a valuable data set for climate studies, even on extreme precipitation events. The precipitation estimates and their extreme events from CMIP5 models are improved significantly compared with rain gauge observations after bias correction by the PERSIANN-CDR precipitation estimates. Given streamflows simulated with raw and bias-corrected precipitation estimates from 16 CMIP5 models, 10 out of 16 are improved after bias correction. The impact of bias correction on extreme events for streamflow simulations are unstable, with 8 out of 16 models can be clearly claimed they are improved after the bias correction. Concerning the performance of raw CMIP5 models on precipitation, IPSL-CM5A-MR excels the other CMIP5 models, while MRI-CGCM3 outperforms on extreme events with its better performance on six extreme precipitation metrics. Case studies also show that raw CCSM4, CESM1-CAM5, and MRI-CGCM3 outperform other models on streamflow simulation, while MIROC5-ESM-CHEM, MIROC5-ESM, and IPSL-CM5A-MR behave better than the other models after bias correction.