Precipitation Spatiotemporal Pattern And Long-range Forecasting Technology In The Three Gorges Reservoir Basin

The spatiotemporal pattern of precipitation and precipitation forecasting are two hot issues for the international community.It is crucial to understand spatiotemporal changes in precipitation and forecast precipitation precisely,to prevent and mitigate natural disasters and guarantee economic society development.However,because of the high heterogeneity of precipitation,particularly extreme precipitation,at regional scale spatiotemporal changes in extreme precipitation are still poorly known.In addition,due to the chaos and stochastic characteristics of climate and weather systems,the impact factors and uncertainty of precipitation forecasting are increasing with longer lead times,thus resulting in much difficulty in making medium-and long-range precipitation forecasting.Unfortunately,there are very few international or national authorities that can release precipitation forecasting products with lead times longer than one month,whilst limitations still exist in methodologies of long-range precipitation forecasting of which the accuracies are not high engouth for practice.Therefore,it is vital to expore new longrange precipitation forecasting method to ensure higher forecasting accuracy.To overcome the shortages described above,in this thesis,the spatiotemporal changes in extreme precipitation together with new technologies for monthly precipitation forecasting during the flood season over the Three Gorges Reservoir basin(TGRB),are studied with the aim of providing technical support for scheduling of flood control and water storage of the Three Gorges.The main contents and conclusions are summarized below:(1)A concept of Event-based Extreme Precipitation(EEP)is proposed to consider the preceding and succeeding precipitation of daily extremes.The time distribution patterns of EEP and the related spatiotemporal characteristics of EEP over TGRB are investigated.Results indicate that unimodal EEP with late peak dominates TGRB,and that the average precipitation amounts of unimodal and bimodal EEPs over the source region of Yangtze River and the midstream are smaller than those in the other regions.The average durations/concentration ratios of unimodal and bimodal EEPs are shorter/higher over the downstream of Xiangjiaba than over the upstream.Moreover,the frequency and precipitation amount of unimodal EEP over TGRB both show significant upward trends,while those of the other types of EEP exhibit insignificant trends.(2)The time distribution patterns of extreme daily precipitation and the related spatiotemporal changes over the Xiangjiaba-Three Gorges region(XTGR)are analyzed.In addition,unimodal daily extreme precipitation with late peak is dominant over XTGR.The frequency and amount of unimodal daily extreme precipitation with early peak and bidmodal daily extreme precipitation with early and middle peaks have increased during 2003-2016,while those of unimodal daily extreme precipitation with late peak and bidmodal daily extreme precipitation with middle and late peaks show opposite trends.Moreover,the mean level of daily extreme precipitation is generally lower than 100 mm,but in some places over the upper region of the XTGR the level can be higher than 100 mm.In the lower region of XTGR,unimodal daily extreme precipitation with early peak have become the dominant precipitation pattern since 2010,replacing unimodal daily extreme precipitation with late peak.(3)The simple linear regression,multi-linear regression,random forest,support vector machine models show different performances in the hindcast of monthly precipitation during May and October over TGRB for the historical period of 1961-2017.The simple linear regression and multi-linear regression models are adaptable for forecasting Augest and September precipitation,respectively,whilst the random forest model is adaptable for forecasting June and October precipitation and the support vector machine model is adaptable for forecasting May,July and September precipitation.However,although these four models can forecast May and June precipitation satisfactorily,their performances in forecasting September and October precipitation are still poor.(4)A concept of Time Varing Multipole Sea Surface Temperature(TVMSST)is proposed.TVMSST introduces 13 fluctuation modes of preceding sea surface temperature,along with the key sea surface temperature(SST)pole and the time varing SST pole.The TVMSST concept includes three kinds of parameters,i.e.the combination coefficients of key SST pole and time varing SST pole,and the contribution degrees of key SST pole and time varing SST pole,in which the combination coefficient of time varing SST pole and the contribution degrees of key SST pole and time varing SST pole are time variant.Depending on the choice of fluctuation mode of preceding SST,the TVMSST index can be used for deterministic forecasting or ensemble forecasting.Afterwards,the regression model based on TVMSST is developed to hincast monthly precipitation during flood seasons for the period 1961-2017 over the TGRB,and the results demonstrate the model capability of simulating monthly precipitation.For lead times of 1-3 months,both the accuracies of deterministic and ensemble hindcasts decrease as lead time gets longer.However,its accuracy is higher than those of the simple linear regression,multi-linear regression,random forest,support vector machine models.In particular,the TVMSST-based model shows distinct superiority in hindcasting June and September precipitation.(5)A multiple regression model is developed with consideration to large-scale atmospheric circulation factors.Further,the multiple regression model is combined with the TVMSST-based simple regression model to forecast monthly precipitation,where a predictor forecasting opinion is introduced to determine whether to use the multiple regression model or the TVMSST-based simple regression model.More specifically,when the absolute value of the predictor forecasting opinion is equal to or larger than half of the number of total predictors,the multiple regression model is employed,otherwise the TVMSST-based simple regression model.The hindcast results demonstrate that the combined model is able to simulate monthly precipitation during flood seasons for 1961-2017 over the TGRB,whilst the forecasting of monthly precipitation during the flood season in 2018 demonstrates the adaptability and capability of the combined model.In the hincasting,the model accuracy decreases along with longer lead times.However,both the accuracies of deterministic and ensemble hindcasting are higher than those of TVMSST-based simple regression model,particularly for the hindcasting of July and Augest precipitation.