Research On Sub-seasonal Scale Prediction Of Sudden Drought In Chin

Flash drought is a type of drought characterized by rapid onset and strong intensity.In recent years,flash droughts have occurred frequently in China,and showed an increasing trend,posing a serious threat to the food,ecology and water resources security in China.Unlike traditional drought,flash drought usually occurred at the subseasonal timescale,its sudden onset and high intensity presented new challenges to the drought monitoring and forecast.Based on the hindcast dataset from the European Center for Medium Range Weather Prediction(ECMWF)and the National Center for Environmental Prediction(NCEP)systems,this study evaluated their predictive skills for flash droughts during the growing season in China.The impact of extending samples and ensemble members,as well as the different methods for constructing cumulative distribution function(CDF),on the predictive skill were also analyzed.Finally,the study further performed a group of hindcast experiments based on a high-resolution land surface hydrological model and meteorological ensemble forcings generated by the ensemble streamflow prediction(ESP)method,and explored the impact of land surface initialization on the flash drought forecast.The main conclusions are as follows:(1)We found that the forecast skill of the ECMWF model for flash droughts in China was higher than that of the NCEP model,and the ensemble averaging of the two models can further improve the forecast skill.Based on the ERA5 Reanalysis soil moisture dataset,this study evaluated the skills of the ECMWF and NCEP models for predicting flash droughts at the sub-seasonal scale.Results show that the flash drought frequency were underestimated by both ECMWF and NCEP forecast models,with average values of 5%and 19% at the first lead week.The hit rates averaged over China for ECMWF and NCEP models were 0.22 and 0.16 respectively,and the values could reach 0.29 and 0.18 over South China.In terms of probability forecast,the brier skill score(BSS)of the ECMWF model were also higher than that of NCEP model,especially over Eastern China.This may be related to the higher predictive skill of ECMWF for temperature and precipitation.The equitable threat score(ETS)based on the ensemble mean of the two models was higher by 8% and 40% than ECMWF and NCEP models respectively.Meanwhile,the BSS was also higher,indicating that the multi-model ensemble method could effectively improve the predictive skill for flash droughts.(2)We found that extending the hindcast samples and ensemble members can improve the drought forecast skills at the longer lead times,but the impact from different cumulative distribution functions(CDF)based on different ensemble members was not significant.Results show that extending the hindcast samples can improve the forecast skills of the ECMWF model.However,extending samples and ensemble members could not improve the forecast skills of NCEP model at the lead-1 week,but can improve the skills at the 2-3 lead weeks.In addition,this study compared the impact of different CDFs constructed from individual ensemble member and the unified CDF constructed from all ensembles on the prediction results,and found that the difference between the two was not significant.Therefore,more long-term hindcast dataset is needed to objectively evaluate the forecast skills of the forecast models,which can provide a basis for correcting the real-time forecasts.(3)Based on the combined surface and subsurface process(CSSPv2)model that could well simulate soil moisture,this study performed a group of ESP hindcast experiments,and revealed the influence of memory of initial land surface condition on the predictive skills for flash drought at sub-seasonal timescale.Results show that the flash drought frequency was underestimated by 19%,50%,and 65% at 1-3 lead weeks,respectively.It can be seen that at the lead-1 week,the predictive skill of ESP hindcast was similar to the NCEP model but was slightly lower than that of the ECMWF model.This indicates the importance of land surface initialization on the flash drought forecast.While at the longer leads,the impact of climate models on flash drought forecast is more significant.