臺灣地區近年來所面臨的高衝擊天氣事件問題嚴峻，雨的觀測與預報的精確度更為重要。雲是用來預測天氣的重要資料，也是天氣雲圖在氣象預報中是必定出現的資訊。雲相態及種類、雲頂的溫度、高度及氣壓的變化等資訊都可推估雲的發展及降雨。若要分析雲頂溫度、高度及氣壓的變化，首先必須追蹤雲像元的移動、生成及消散。雲的追蹤可以利用地球同步氣象衛星，提供大範圍且高時間解析度的影像。先前MTSAT-2每30分鐘提供一張影像，遇到訊號干擾則會延長到一小時更新，而較小的對流雲可能在30分鐘內就生成到消失，衛星雲圖很難辨識，移動速度較高的雲也不易追蹤。日本氣象廳發射的向日葵8號（Himawari-8）衛星已於2015年7月正式發送訊號，它不但增加頻道至16個，也提升時間解析度到10分鐘，甚至可對小範圍區域進行2.5分鐘解析度的觀測。再加上日本向日葵九號(Himawari-9)，中國風雲四號(FY-4)、韓國GEO-KOMPSAT-2等，有希望達到半球3分鐘時間解析度。本子計畫將分兩年進行，第一年以粒子群聚法配合類神經網路追蹤雲像元的移動，估計雲濃度變化與雲系統的發展，並預測雲的移動並計算誤差。第二年將整合相關子計畫之參數，利用線性與非線性的多變數回歸函數計算雲特性的變化、溫濕度與降雨的相關性與時間延遲，做為天氣預報的一部分。 ;Extreme weather in Taiwan becomes serious issue recently, and the monitoring and prediction of rainfalls is more important than ever. Clouds present the feature of weather changes, and they are also very important for weather prediction. Cloud phase, cloud type, and the changes of cloud top properties (temperature, height and pressure) all provide information for cloud development and rain fall prediction. To analyze the changes of cloud properties, we must track the movement, appearance and disappearance of clouds in pixel level. The geosynchronous weather satellites orbit with the rotation of the Earth. They can provide semi-sphere image with high temporal resolution. The MTSAT-2 provides an image every 30 minutes, and it might delay to an hour if transmission interference occurs. Some small convection clouds may develop and disappear within 30 minutes, so MTSAT-2 cannot track them. If the clouds move fast, it is also difficult to track them. Japan Meteorological Agency launched Himawari-8 to Earth’s orbit. This satellite began to distribute images in July, 2015. It not only increases the number of channels to 16, but also increases the temporal resolution to 10 minutes. It also has the capability to monitor a small area every 2.5 minutes. With Himawari-9, FY-4 and GEO-KOMPSAT-2, it is possible to monitor hemisphere with 3 minutes resolution. In this 2-year proposal, we will first improve the research of cloud tracking with cloud concentration instead of cloud mask, using particle swarm organization (PSO) and artificial neural network (ANN) to track the cloud pixels and estimate the development of cloud system. We will also calculate the accuracy of short term and long term prediction of cloud movement. In the second year, we will combine the parameters from other related project and find the relationship with rainfalls and time delay by linear and nonlinear multivariable regression function as part of weather prediction.
