極端降雨事件一直是台灣重要的天然災害之一，當這些劇烈降雨事件發生在都會區時，對於民生經濟的影響更為顯著。這樣的降雨事件本身具備了跨尺度的特性，也導致在傳統定量降雨預報上有非常大的不確定與限制。本研究將使用全新的降雨預報概念，將過去針對台灣極端降雨與長期天氣系統分析的經驗，結合目前最新的人工智慧(AI)分析工具-機器學習/深度學習，發展特定天氣型態之極端降雨事件預警系統。我們將使用過去建立相對應之台灣大氣事件資料庫，發展客觀化天氣系統判讀模組，用以克服過去對於天氣診斷工具的諸多限制並延長可分析之資料時間。我們也將利用這些天氣系統的資訊與過去的極端降雨事件的相關性分析結果，以機器學習技術為核心由大台北都會區為範例，建立適合台灣都會區極端降雨事件的人工智慧預警系統。同時也針對長期的天氣事件變化進行分析，並利用所開發之診斷工具探討在不同氣候變遷情境下的天氣與極端降雨事件的變化特徵。
The extreme rainfall event is one of the major nature disasters in Taiwan and has high impact to the livelihood and economy. This kind of weather event has complicated multi-scale interactions and also cases some limitation and error of the Quantitative Precipitation Forecasts (QPFs). In this study, we will use a new approach of QPFs. We will base on the previous experience of weather events related extreme rainfall studies and use the Artificial Intelligence (AI) analysis methods to develop an objective weather classification method. We will also use those objective weather events data, real-time observations and atmospheric numerical forecasting model outputs to develop an artificial intelligence based extreme rainfall warning system in Taiwan Metropolitan Area. We will also analysis the long-term variation of weather events. The weather classification tool and extreme rainfall diagnose method can be used to estimate the change of extreme rainfall within different climate change scenarios.
