Simulation and evaluation of summer daily precipitation based on nonhomogeneous hidden Markov model over the Yangtze-Huaihe River Basin

The statistical downscaling method of the nonhomogeneous hidden Markov model (NHMM) is introduced in this paper. Based on the observed summer daily precipitation data at 56 meteorological stations in the Yangtze-Huaihe River Basin and the ERA-40 reanalysis data of European Centre for Medium-Range Weather Forecasting (ECMWF) during 1961-2002, the NHMM is established and the simulation capability of NHMM for summer daily precipitation over the East Asian monsoon area in China (represented by the Yangtze-Huaihe River Basin) has been assessed. Results of the NHMM have been compared with that of the BCC-CSM1.1(m). It is found that the NHMM performs well in simulating daily precipitation over the Yangtze-Huaihe River Basin by establishing the relationship between transition probabilities of the precipitation probability distribution states and the synoptic-scale atmospheric predictors. The simulated probability distribution function (PDF) curves are close to observations at individual stations with the Brier score(SB) less than 0.11% and the significance score (Ss) greater than 0.84. Relative errors of summer total precipitation, number of rainy days (≥ 1 mm), number of rainy days with daily precipitation more than 10 mm, simple daily intensity, 95th percentile value of precipitation all are less than 10%, and the first three indices' spatial correlation coefficients between the NHMM outputs and observations are greater than 0.9. The NHMM can also reproduce the interannual variability of the precipitation indices mentioned above. Correlation coefficients of the annual sequences of area-averaged precipitation indices between simulations and observations are within the range of 0.62-0.87. The NHMM is also driven by predictors of BCC-CSM1.1(m) from 1986 to 2005. After downscaling, the SB averagely decreased by 0.57% and Ss averagely increased by 0.23. This result indicates that the probability distribution function curves are much closer to the observations. Absolute values of relative errors for the precipitation indices at most stations decreased from more than 40% to less than 10%, and the spatial correlation coefficients generally increased to 0.8 or more. The NHMM approach can effectively improve the simulation capability of BCC-CSM1.1(m) for summer daily precipitation and has significant "added value" relative to climate models. Thereby it can be applied to the projection of future precipitation changes under the background of climate warming.