Evaluation Of Uncertainty In The Regional Climate Simulation Driving By Different Reanalysis Over China

East Asia is one of the most sensitive and vulnerable areas in the world to the global climate change. The study of climate change in this region is much more difficult than in the other places. With high quality, long time series and high resolution, atmospheric reanalysis have become the most widely used datasets in atmospheric science. However, systematic biases can display negative effect on the quantity of reanalysis datasets. Therefore, an evaluation of the reliability and accuracy of reanalysis datasets is significant for global and regional climate research. Besides, in predicting the regional-scale climate signals, regional climate models (RCMs) have become better choices for their better physical parameterization schemes and higher horizontal resolutions compared to reanalysis. The RCM approach consist of using a fine-resolution grid over a limited-area domain, which requires to be fed at its boundaries by large-scale information usually taken from a general circulation models (GCMs) or reanalysis. The differences in driving data can cause uncertainties on simulated results.Firstly, the high-altitude variables extracted from the six reanalysis datasets(NCEP/NCAR, NCEP/DOE, NCEP-CFSR, JRA-25, ERA-Interim and MERRA) are evaluated with the Integrated Global Radiosonde Archive (IGRA) global sounding observations over China. It is found that the mean values of geopotential height and temperature in each reanalysis dataset are consistent with the observations, but the wind fields, especially the meridional wind, are not. Besides, the reanalysis products do a bad job in revealing the inter-annual variation of meridional wind. The results of empirical orthogonal function (EOF) analysis imply that all reanalysis datasets exhibit better performance in depicting the temporal and spatial distributions of geopotential height and temperature than other variables, especially the wind fields; MERRA performs specific humidity better than other reanalysis products. Generally, NCEP/NCAR, NCEP/DOE and NCEP/CFSR products are not as good as JRA-25, ERA-Interim and MERRA.To investigate the ability of WRF model in simulating the regional climate over East Asia summer monsoon region, long term ensemble simulations with different initial and lateral boundary conditions for summer covering the period of1982-2001are generated. The driving datasets are NCEP/NCAR, NCEP/DOE, NCEP/CFSR, JRA25, ERA40, ERAFN and MERRA respectively. Results show that all experiments are skillful in revealing the spatial distributions of summer mean surface air temperature and precipitation, but the temperature at cold centers are underestimated, especially when using NCEP/NCAR, ERA40and ERAIN as driving datasets. On the contrary, the temperature at warn centers are simulated warmer than observation. The simulated precipitation amounts by RCMs are less than the observations over Northwest and Southwest China, while overestimation appears in the southwest coastal region, especially for the experiments with Exp-CFSR, Exp-ERA40and Exp-ERAIN. The simulations are able to reproduce the inter-annual variations of both temperature and precipitation averaging over the sub-regions, but overestimations are noted for most sub-regions. It is also found that all experiments give a better performance in simulating the spatial patterns of upper geopotential height, temperature, zonal wind and specific humidity than meridional wind. At the middle and high level of tropospheric, the simulation is warmer and dryer than ensemble mean of reanalysis over the regions where high-pressure and anti-cyclonic bias are found, while it’s colder and wetter over the areas with low-pressure and cyclonic bias. It is worth to be noted that the experiments forced by different reanalysis data exhibited remarkable differences in reproducing both surface and upper atmospheric variables, which means that biases in the driving datasets can cause large uncertainties in simulated results. The ensemble of simulations shows better performance than each single experiment result.Finally, the quantitative analysis of model spread among individual members caused by different driving datasets is conducted. For surface air temperature, model spread increases with arise of elevation, and it’s significant over most areas relative to natural variability, especially over high altitude localities. The spatial distribution of model spread for precipitation is strongly related with the spatial distribution of precipitation itself which increases from south to north. Relative to natural variability, the model spread of precipitation is very large over most areas. Furthermore, the model spread of summer mean geopotential height, temperature, horizontal wind and specific humidity attain large values over most regions, and the ratio between model spread and natural variability is also very large, especially for specific humidity, indicating that each ensemble member behaves very differently. This suggests that differences between the forcing reanalysis can cause large uncertainties in simulation results. The analysis shows that the model uncertainties are primarily caused by uncertainties in water vapor influx across the lateral boundaries in the reanalysis. The largest uncertainties in moisture transport are found at west and south lateral boundaries. And in the vertical direction, the largest differences are found below700hPa and middle atmosphere at west and south boundary respectively. Further more, it’s found that the spread between reanalysis datasets is significantly amplified after model calculations, except for specific humidity.