Statistical Regression Scheme For Typhoon's Intensity Prediction In Northwest Pacific

Typhoon is a devastating natural disaster.It is very crucial for disaster pre-vention related to typhoon(TY)that timely and accurately predicting the intensity of TY.This study explores the mechanism of intensity change and proposes a statistical regression scheme to forecast typhoons' intensity at 12,24,36,48,60 and 72 hours interval over northwestern Pacific ocean(0-60°N,100-180°E).After weighing the characteristics of the western north Pacific typhoon data issued by various forecast agencies,the best track data published from years 2000 to 2015 by the following agencies is selected for this statistical model:Shanghai Typhoon Institute(STI),China and Joint Typhoon Warning Center(JTWC),United States.Many factors are calculated by the Final Operational Global Analysis(FNL)-issued every six hours by the United States National Center for Environmental Prediction,as well as the weekly mean value of sea surface temperatures(SST)by the National Oceanic and Atmospheric Administration.Besides the conventional factors of climatology and persistence,this study pays special attention to the land effect on the TYs' intensity change by considering a new factor of the ratio of sea water area to land(SL ratio)which describes the underlying surface change and is added into the statistical regression model.TYs'whole life span is divided into categories:Landed,Near Coast and Ocean Basin.Each category is analyzed and modeled separately.Data from 2000 to 2011 are used for model calibration and data from 2012 to 2015 are used for model validation.To ensure predictability of the model,factors are selected using a stepwise regression method with a significance of 0.05.The results from this model are then compared with existing forecasting methods in order to evaluate its accuracy.When comparing each model's factors,we find the following factors are sig-nificant:previous 12-hour intensity change(DVMAX),maximum possible intensity change(POT),and deeply vertical wind shear(SHRD).More than 95%of the pre-vious 12-hour intensity change within lOm/s.The threshold for POT in Near Coast sample(50m/s)is greater than for Ocean Basin sample(40m/s).When SHRD is less than 15(10)m/s,the intensity of the Ocean Basin(Near Coast)sample may increase.When SL ratio<90%,Near Coast sample' intensity is more likely to de-crease.For Landed sample,less than 7%of the typhoons may increase in intensity.Once SL ratio is added,R2 increases by 3-5%,while mean absolute error(MAE)and root mean square error(RMSE)decrease by 2%on average.Landed sample model performs the best,but Near Coast model is the worst.It is maybe resulted from JTWC issues intensity higher that the scheme performs better at STI's than JTWC's.In evaluating the statistical model of landing typhoon intensity,we devel-op a numerical model according to the DeMaria's Decay model.Upon comparing the predictability of the two models,we find that the statistical model performs better than the numerical model.