The Forecast Verification And Improvements Of Northwestern Pacific Tropical Cyclones By NAEFS Ensemble Prediction Systems

The Northwest Pacific Ocean is the basin where tropical cyclones appear most frenquently,and it is of great importance to improve the forecast skill of numerical forecast model because of the strong destruction and broad influence a tropical cyclone may cause.This thesis aims to investigate the forecast skill of tropical cyclones by operational ensemble prediction systems(EPSs)from the North American Ensemble Forecast System(NAEFS),which includes the EPSs at the United States National Centers for Environmental Prediction(NCEP)and Canadian Meteorological Centre(CMC).Based on the storm track data of the NCEP global ensemble forecast system(GEFS),CMC global ensemble forecast system(CENS)and the best track data produced by China Meteorological Administration(CMA),the track,pressure and maximum sustained wind(MSW)forcast of tropical cyclone in the northwest pacific ocean from 2012 to 2014 are analyzed.A weighted motion vector consensus of ensenmble mean track prediction is also designed,verified and then compared between the EPSs of NCEP and CMC.In addition,a promotion method of calculating the strike porbability is proposed and verified.The analysis show that the different index of tropical cyclone intensity leads to different result.To GEFS,the high resolution model of global forecast system(GFS)is the best and GEFS are similar to control forecast when pressure is chosen as the index of intensity,while the ensemble is the best and the control is the worse when MSW of 2 minutes is instead.Referring to CENS,ensemble mean is the best in initial,but then its error increases rapidly and became the worse,and it's the same in pressure and MSW of 2 minutes.To the track forecast,GEFS has a tendency of introduce more along track error than cross track error into the prediction.The analysis of spread vs error indecates that GEFS shows a satisfing performance in short time range forecast of track but a slight underdispersive in long time rang,while the intensity forecast result is always extream underdispersive,no matter pressure or MSW is chosen.The probable reason may be that the mechanism of tropical cyclone developing and strengthen is still unclear,and model phyisical frame is still imperfect.The longer of initial location correct time it use,the better motion vector consensus method performed in ensemble mean track forecast,but the effect is limited in a short time range after corrected.All of the four weighted scheme show significant abilities in correcting track forecast error,and the modified binomial coefficient and variance reciprocal are better,especially.Refered to a time range larger than 5 days,variance reciprocal is the best in CENS,while modified binomial coefficient is the best in GEFS.When weighted motion vector consensus is applied,a rapidly increasing track forecast error apprears in the next time step,then fall down to normal range soon.In addition,the different values of initial location correct time make no diffrence to the weighted scheme performance.Above two phenomena approve that a real-time-adjusted weighted schenme has a remarkable and steadily correction to forecast error.A promoted method to calculate strike probability has a positive effect on improving reliability and the Relative operating characteristic(ROC)skill of ensemble forecast.The promotion of strike probability reliability is mainly in 50%?70%in GEFS,whose ROC area(ROCA)increases from 0.65 to 0.68.Refering to CENS,the promotion is mainly on 50%-80%,with a ROCA increasing from 0.6 to 0.64.In general,the promoted method performs better in CENS.GEFS and CENS show almost no reliability difference in low probability interval,neither before nor after the promotion,and both of them show an underestimation of strike probability in low probability interval,while a better reliability in middal probability interval and an over estimation in high probability interval.CENS has a better reliability in high probability interval than GEFS both before and after the promotion,but GEFS is better in ROC curve in the probability interval less than 80%,which is in accord with the throughout better ROC A of GEFS.The promoted method performs better in improving the probability interval where landfall point falls in.In CENS,the promotion appears when the landfall point falls in low,middle and high probability interval.While in GEFS,only when the landfall point falls in middle and high probability interval,the promotion is significant.