Intensity Forecasting Model Of Tropical Cyclone Over The Northwestern Pacific Based On Partial Least Squares Method

One of the difficulties in current Tropical Cyclone (TC) prediction is the TCintensity forecasting. Presently TC intensity forecasting is mainly based on statisticallinear regression methods and statistic-dynamic methods. The trend of the TCintensity forecasting relies heavily on improving mathematical model and increasingsome new forecasting factors. This paper has improved the current operationalforecasting Partial Least Squares (PLS) scheme, by adjusting the modeling time,screening variable factors and adding new variable factors. In addition, this paper alsogives respectively for TC intensity quasi operational accuracy evaluation method andparameter. The main findings are summarized as follows:1. Taking the prediction results of PLS scheme in June as an example, the newPLS scheme has been compared with Climatology and Persistence scheme (CLIPER),especially in the prediction results of36~60-h. And compared with the PLS, the PLSscheme which contains screening variables factors (Partial Least Squares-FactorScreening, PLS-fs) also increases the forecasting accuracy slightly in the predictionresults of24-h~60-h. And the forecasting accuracy of CLIPER, PLS, and PLS-fs inSeptember are significantly lower than that in June, July and August. The reason forthis phenomenon may be the number of TC increases over the Northwestern Pacific,and the development processes of TC are more complicated in September.2. According to the four main conditions induced TC intensity change----climatology and persistence factors, oceanic and atmospheric thermal effect on TC,environmental flow and TC structure, such factors are added to the PLS scheme as thewhole layer of the vertical wind shear (VWS), potential future intensity change (POT),non symmetry characterization of thermal structure (B), factor of low layer thermalanomaly (VLT), factor of high layer thermal anomaly (VUT) and zonal componentof the TC translation speed6h after t0(CMV) in order to further improve the existingmodel. By using TC data from CMA, the daily NCEP and NOAA re-analysis data for the period of1982-2012, the statistical relationship between TC intensity change andthe above factors is examined. It shows that screening variables factors is necessary inmodeling based on PLS method, but the new variable factors can not obviouslyimprove forecasting accuracy. Therefore, looking for suitable variable factors wouldbe the focus of future research.3. The Wilcoxon rank test method can be used in the meteorology, especially inTC intensity forecasting. Because the sample population distribution based on TCintensity prediction accuracy assessment is often a non-normal distributionIfsignificant results are expected in the operational forecasting of TC intensity, testmethods like t and F are unusable. In order to solve this problem, it has proposed thatWilcoxon rank test method can be used in the meteorological data, and its illustrationof how to use the method in the test of TC intensity forecasting.4. Based on integral deviation (ID), a new TC intensity prediction evaluationparameter has been proposed, which is called Intensity Forecasting Integral Deviation(IFID). IFID is the first use of integral deviation in the test of TC intensity forecasting,and contains two parameter’s information which are the Mean Absolute Error (MAE)and the Trend Rate of Forecasting(TCR). IFID can describe the results of the TCintensity forecasting accurately. The fifth chapter of this paper also has taken the forthtyphoon "Guchol" as an example, and shown a brief description of the calculation ofIFID.