Towards understanding multi-model precipitation predictions from CMIP5 based on China hourly merged precipitation analysis data
Highlights
- •
China hourly merged precipitation analysis data used to assess precipitation from CMIP5.
- •
Models tend to overestimate (underestimate) the precipitation in North (South) China.
- •
The performance of projected precipitation highly depended on models and locations.
- •
Model's outputs corrected by quantile mapping algorithm have a better performance in the regions with complex terrain.
Abstract
Large uncertainties still exist in the simulation and projection of precipitation from current climate models. Here, the newly released state-of-the-art China Hourly Merged Precipitation Analysis (CHMPA) data has been used to evaluate the ten models from the fifth phase of the Coupled Models Intercomparison Project (CMIP5). Particularly, the precipitation predictions under the Representative Concentration Pathways (RCP)4.5 and RCP8.5 scenarios in China are assessed for the period from 2008 to 2017. Interestingly, the ensemble mean precipitation under the two emission scenarios does not show systematic differences. Intercomparison analysis of precipitation between multi-model prediction and CHMPA yields a high correlation coefficient (0.85–0.95) on the annual timescale. However, most models tend to overestimate the precipitation in northern China but to underestimate that in southern China, due to the model-simulated monsoon precipitation extending to the north earlier. Relative to UKMO-HadGEM2AO model, other models overestimate precipitation at the southeastern edge of the Tibetan Plateau where the overestimation reaches up to 150%. In terms of the temporal evolution of predicted precipitation, the multi-model ensemble produces relatively small interannual variability except for more summer monsoon precipitation with biases over 0.3 mm/day, which indicates that models are not capable of reproducing the seasonal and meridional propagation of precipitation. Compared with the original model output, the precipitation corrected by quantile mapping algorithm better agrees with the observations for spatial and temporal distributions. The findings have great implications for better utilizing model-predicted precipitation in climate change studies.