Evaluation And Revision Of The Forecast Of High Temperature And Heat Waves In The Yangtze River Basin Based On The Subseasonal-seasonal (S2S) Forecasting Model

Based on the reforecast data(1999-2010)of three operational models [the China Meteorological Administration(CMA),the National Centers for Environmental Prediction of the U.S.(NCEP)and the European Centre for Medium-Range Weather Forecasts(ECMWF)]that participated in the Subseasonal to Seasonal Prediction(S2S)project,we systematically evaluate the performances of the models' surface air temperature(SAT)in summer over China,with a focus on revealed the major sources of subseasonal prediction skill for heatwaves over the Yangtze River basin(YRB).Based on case and statistical analyses,the relative influences of 10–30-day and 30–90-day circulation prediction biases and the contributions of land surface conditions to heatwave prediction skill are discussed.Finally,the statistical post-processing models over YRB are constructed by employing linear regression(LR),systematic bias correction(SBC),random forest regression(RF)and cumulative distribution function transform(CDFt),and their applicability are explored in each model.Systematic biases of models vary with model configuration,region and forecast lead time.The amplitude and variability of SAT in Southeast China(including YRB)are consistently underestimated.Useful skill of intraseasonal SAT revealed by temporal correlation coefficients(TCCs)and mean square skill scores(MSSSs)over China are within a13–15/10-day(CMA),17–20/15-day(NCEP)and 19–22/19-day(ECMWF)lead time.And its pattern correlation coefficients(PCCs)drop rapidly exceed a 15-day lead time and with significant inter-annual variability.The prediction skills of all models for warm events are better than that for extreme warm events,heatwave events and average events averaged over China.The more skilled the event kind,the faster the skill drops with the lead time increase.Differently,for the YRB,both the NCEP and ECMWF models perform better in predicting average events than extreme warm and heatwave events.The three models show limited prediction skills in terms of the fraction of correct predictions for heatwave days in summer;the Heidke Skill Score drops quickly after a 5-day forecast lead and falls down close to zero beyond the lead time of 15 days.The superior skill of the ECMWF model in predicting the intensity and duration of the YRB heatwave is attributable to its fidelity in capturing the phase evolution and amplitude of high-pressure anomalies associated with the intraseasonal oscillation and the dryness of soil moisture induced by less precipitation via the land–atmosphere coupling.The effects of 10–30-day and30–90-day circulation prediction skills on heatwave predictions are comparable at shorter forecast leads(10 days),while the biases in 30–90-day circulation amplitude prediction show close connection with the degradation of heatwave prediction skill at longer forecast leads(>15–20 days).The biases of intraseasonal circulation anomalies further affect precipitation anomalies and thus land conditions,causing difficulty in capturing extremely hot days and their persistence in the S2 S models.The calibration effectively enhance the prediction skill while the degree of improvement is limited by models' predictability.After bias-correction,the SAT skill of CMA and NCEP model are optimized significantly,and the prediction of ECMWF model is also improved and with the highest forecasting performances.LR method has excellent and stable effect in reducing forecast deviation,and RF and SBC are also quite effective.CDFt method has more advantages in improving the correct hit rate of the model to different climate probability events.For YRB heatwave events,the SBC method effectively reduced the prediction biases of CMA and NCEP models,while the LR and RF methods played a better effect in ECMWF model.The CDFt method can consistently improve the correct hit ratio of the model to the regional heat wave days.