Improvement of the Statistical Typhoon Intensity Prediction Model by Using the SHIPS Developmental Data

• Main Authors: Shuo-Yan Lin, 林碩彥
• Other Authors: Hsiao-Chung Tsai
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/74617940071730341777

碩士 === 淡江大學 === 水資源及環境工程學系碩士班 === 104 === The purpose of this study is to improve the typhoon intensity forecast skill. A statistical five-day typhoon intensity prediction model called WAIPs is developed by adapting the Weighted Analog Intensity Predict model (WAIP; Tsai and Elsberry, 2014), and the environmental factors (e.g., vertical wind shear, maximum potential intensity, sea surface temperature, ocean heat content, etc.) obtained from the Statistical Hurricane Intensity Prediction Scheme (SHIPS) developmental data. The improvement of the forecast skill over the original WAIP model is investigated if the SHIPS predictors are included. In addition, a model that only uses the SHIPS predictors (named SHIPSa), and the SHIPSa without using the ocean heat content predictors (named SHIPSb) are also investigated. In this study, the SHIPS data from 2000 to 2007 are used as the training samples, and the predictors are selected by using the stepwise regression method. The analysis results show that the vertical wind shear should be used from 12 to 120 hours, but the importance decreases after 72 hours. Also, the ocean heat content is used in all forecast periods as revealed by the stepwise regression. The SHIPS data from 2008 to 2012 are used as the independent testing samples. As compared to the original WAIP model, the forecast skill is improved at every forecast period if the SHIPS predictors are included. For example, the R2 values are increased by 19-39% from 60 to 120 h. The WAIPs can also outperform the SHIPSa and SHIPSb by 16-30% and 30-48%, respectively. The RMSE of the WAIPs is also smaller than that of the WAIP, SHIPSa, and SHIPSb by about 11%, 12.5%, and 15%, respectively. The cases that undergo RI (rapid intensification) are also investigated. It is shown that the skills are improved during the 72-120 h forecast periods. For example, the R2 improvement of the WAIPs over the WAIP, SHIPSa, and SHIPSb are 0.21, 0.14, and 0.07, respectively. As for the MAE, the WAIPs is 3-7 kt smaller than the WAIP. The RMSE of the WAIPS is also 1-4 kt and 2-11 kt smaller than that of the SHIPSa and the SHIPSb. The forecast skill is also evaluated according to the geographical distributions. At 24 h, the MAE and RMSE over the area ranging from 150-160o E and 10-20o N can be reduced by 5.8 kt and 5.5 kt, respectively. At 72 h, the MAE and RMSE over the area ranging from 140-150o E and 10-20o N can be reduced by 10 kt and 10.5 kt. For the Taiwan area, the MAE and the RMSE can be reduced by about 2.9 kt and 3.4 kt, respectively. At 120 h, the forecast improvement over the area near Japan is significant, and the MAE and RMSE can be reduced by about 3.3 kt and 4.5 kt. The MAE and RMSE near Taiwan area can also be reduced by about 2.2 kt and 4.4 kt, respectively. The improvement of the RI cases is quite significant, especially the area near Taiwan. The MAE and RMSE over the area ranging from 150-160o E and 10-20o N are reduced by 8.2 kt and 7.5 kt, respectively. At 72 h, the area ranging from 120-130o E and 30-40o N has the most significant improvement. The MAE and the RMSE can be reduced by 14.4 kt and 13.6 kt. At 120 h, the most significant area is the region near Taiwan. The MAE and the RMSE can be reduced by 8 kt and 10 kt. Also, the MAE and RMSE for the area over the west of the Philippines are reduced by 7.2 kt and 7.8 kt, respectively.