支援向量機模式結合SSM/I 微波頻道反演陸上颱風降水之研究

Abstract

台灣位於西北太平洋颱風之主要路徑上。每當颱風侵襲時，流域上游集水區短且急之強降雨量導致山崩、土石流發生，也在下游低窪地區造成淹水，對人民生命財產造成極大損失。因此，各防洪單位莫不期盼氣象及水利單位能提供有效用之颱風時期定量降雨強度相關資訊，以便及早進行防洪抗災相關作業。本計畫目的在颱風侵台時反演陸上降雨強度。本研究發展一新近支援向量機法進行小時降雨強度反演，資料收集包括微波頻道資料以及雨量資料，其中微波頻道資料為Special Sensor Microwave/Imager (SSM/I)上被動式儀器之亮度溫度，頻道包括19.35、22.234、37.0以及 85.5 GHz；另外，雨量資料來源為自計式雨量站逐時雨量資料。本研究之研究地區為石門水庫集水區。本研究收集12年(1997-2008)侵台之衛星資料以及集水區內 13個雨量站雨量資料。本研究所發展之支援向量機反演模式以及支援向量機混合散射指數模式與單一頻道反演模式和散射指數反演模式進行比較，研究結果顯示支援向量機反演模式與支援向量機混合散射指數模式獲致不錯之結果。Tropical cyclones, also known as typhoons, are among the most devastating events in nature and often attack the western North Pacific region. As soon as a typhoon strikes, the upstream watershed receives voluminous rainfall in a short time. Therefore, it can easily lead to considerable economic loss and casualties. This paper focuses on addressing the rainfall retrieval problem for quantitative precipitation forecast during typhoons. In this study, SSM/I (Special Sensor Microwave/Imager) microwave data and Water Resources Agency (WRA) measurements of Taiwan were used to estimate quantitative precipitation over the watershed of Shihmen Reservoir in northern Taiwan. Various retrievals for the rainfall rate over land are compared by four models. They are the simple regression method, the traditional SIL (scattering index over land) approach, the novel SVM (support vector machine) and the proposed SIL-SVM model. This study collected 70 typhoons affecting the studied watershed over the past 12 years (1997-2008). The channel measurements of SSM/I satellite comprise the terrestrial brightness temperatures at 19.35, 22.23, 37.0 and 85.5 GHz. The results showed the approaches using the novel SVM and SIL-SVM conjunction models have superior products to simple regression and traditional SIL models. This is because the SVM techniques are good at identifying and learning correlated patterns between the input data sets and corresponding target values.