Prediction And Stability Test Of Regional Precipitation In China Under Different Downscaling Schemes At 1.5°C And 2°C Global Warming

For the key regions of response of precipitation to the global warming of 1.5? and 2?,the middle and low reaches of the Yangtze River(MLRYR)and the Tibetan Plateau(TP),we first evaluate the performance of the NHMM statistical downscaling scheme(NHMMs),the LMDZ4 dynamical downscaling scheme(LMDZs)and the LMDZ4/CDF-t dynamical-statistical downscaling scheme(LMDZ4/CDF-ts)simulating summer daily precipitation over the MLRYR.The optimal downscaling scheme is used to project future changes in precipitation under the global warming of 1.5? and 2?,and analyze its possible cause.Meanwhile,the difference in daily precipitation over the TP and the Northern Indus-Ganges River Plain(NIGRP)from CORDEX induced by the downscaling,different GCMs,different RCMs and different simulation domains are investigated.Further,the CORDEX/CDF-t dynamical-statistical downscaling scheme(CORDEX/CDF-ts)is established to project more reliable response characteristics of precipitation.Finally,we comprehensively evaluate the capabilities of five downscaling schemes,and examine the stationarity of schemes.Main conclusions are as follows.(1)To establish the best dynamical-statistical downscaling scheme,we first use the four bias correction methods,i.e.Gamma Cumulative Distribution Function(Gam CDF),Quantile-Quantile Adjustment(QQadj),Equidistant CDF Matching(EDCDF)and Transform CDF(CDF-t)to improve the seasonal variation,the probability distribution of daily precipitation,and the spatial distribution pattern over China simulated by RCMs.We put emphasis on the benefits of the bias correction compared to performance of the original RCM simulations.The result shows that CDF-t is the best bias correction method.(2)Compared with LMDZs and LMDZ4/CDF-ts,NHMMs effectively enhances the ability of GCMs in simulating summer daily precipitation over the MLRYR.Under the global warming of 1.5? and 2?,the spectral distribution of regional precipitations(in function of their intensity)shows consistent changes with a decrease of occurrence probability for light rain and an increase for heavy rain.Among variables of interest,PRCPTOT,SDII,R10 mm and P95 all show a gradually increasing trend and geographically an eastward gradient with weak decrease for the west and larger increase for the east.Form the view of the principle of NHMMs,we find the response of precipitation is related with changes in the occurrence frequency of future precipitation patterns and corresponding synoptic atmospheric patterns.(3)RCMs from CORDEX all can reproduce the climatology,seasonal variation and interannual variability of precipitation over the TP and the NIGRP.However,the wet bias in precipitation over the TP for the wet(dry)season decreases(increases)significantly after downscaling.The comparison of precipitation driven by different GCMs shows that MPI-ESM-LR is the best driver,especially for the central TP.However,there is no significant difference in precipitation simulated by different RCMs or simulation regions.Further,CORDEX/CDF-ts projects an increasing trend in precipitation with a decrease of occurrence probability for light rain and an increase for heavy rain.Spatially,PRCPTOT(SDII)will decrease(increase)over most parts of the TP and the NIGRP under 1.5? warming target,with the relative changes keeping above 40% over the southern TP and Pamir Mountains(above 60% over the northern TP,the central TP and Pamir Mountains).The additional half-degree warming makes changes more significant.(4)Five downscaling schemes all have a good consistency for simulating the probability distribution of daily precipitation and the spatial distribution pattern.It's noted that NHMMs(CORDEX/CDF-ts)applied in precipitation over the MLRYR(the TP)performs best.Further,examining the stationarity of schemes show that NHMMs effectively deliver the characteristics of changes in PRCPTOT and SDII to the future for all stations over the MLRYR.However,the capability of NHMMs exhibits a non-linear degradation with time.There are relatively large uncertainties in simulating PRCPTOT(SDII)over the most parts of the MLRYR(the northern and southeastern MLRYR).CORDEX/CDF-ts effectively deliver the characteristics of changes in PRCPTOT(SDII)to the future for the central and northern TP(south edge of TP).However,there are relatively large uncertainties in simulating PRCPTOT(SDII)over some parts of NIGRP,northeastern and central TP(northeastern TP,some parts of TP and northern NIGRP).