類神經網路（neural network）乃一數學方法，藉由大量的數據資料，模擬人類學習的方式以了解資料的特性，進而取得輸入資料與輸出目標的關係，此網路類似於神經網路，因此稱為類神經網路。這種方式有別於傳統之數值預報模式或統計分析，而且應用於許多科學領域，此方法吾人曾用於颱風降雨估計，結果令人滿意。近幾年豪雨災害嚴重，尤其是山區降雨量正確的估計一直是作業與研究的挑戰，類神經網路的好處是可以透過非線性關係去取得輸入資料（如衛星資料、雷達資料、地形、數值模式預報等等與降雨有關的資料）與輸出值（雨量）之間的相關性。本計畫之主要目的乃使用此方法估計易受災地區之降雨量，擬使用2008年SoWMEX其間之強降雨個案建立神經網路估計降雨之神經網，再利用近年之強降雨個案當作測試，強降水個案的輸入資料包括雷達資料、衛星資料、地形及環境場(包括風場及濕度場)，利用密集雨量站之雨量資料為目標值，以倒傳遞神經網路方法訓練出一組神經網路來估計整個資料涵蓋區域之雨量圖(rain map)，本研究將特別著重在易受災地區的降雨估計。結果亦可用於其他各地之降雨量及集水區之雨量估計。
Quantitative precipitation estimation using the neural network method over the easy flooding area The neural network is mathematic method by linking a large amount of input data and target through a network work. This network is similar to mankind's neural network. This is method is different from the traditional forecast model or statistics method and has been used on many scientific studies. In the past, we have conducted the typhoon rainfall estimation based on this method and we had obtained pretty good results. Besides the conventional data set, such satellite, radar, surface observations, and environmental flow and moist will be all included. The environmental flow and moist are adapted from the 5km x 5km mesoscale model. In this project, the heavy rainfall events collected during 2008 SoWMEX will be used to contract the neural network. And the heavy rainfall events within the recent years will be used as the test. The QPE and QPF product over easy flooding area will be primarily focused.
