氣象衛星影像應用於雨量即時預報系統之研發

Abstract

本研究利用GMS-5同步氣象衛星影像以及中部地區自動記錄雨量站之時雨量紀錄，進行對南投縣陳有蘭溪流域降雨事件時空分配模式之研究，並試圖藉由其中相關性，預測其未來累積一至六小時之總雨量。研究中將時雨量紀錄以空間特性推估為與氣象衛星影像網格相符合之地面資料，並藉此與經處理之氣象衛星影像相配合，利用空間摺合積分模式運算轉換核心函數，由此函數當做狀態變數，利用卡門濾波演算法對累積小時總降雨量進行預測。除降雨量推估及預測模式之建立外，本研究並將此模式撰寫為視窗介面之程式，利用圖形化使用介面之親和性，配合每小時輸入資訊修正預測，可直接由程式輸出累積小時之降雨強度並進行極端雨量之警戒。 經過目前已完成之碧利斯颱風及桃芝颱風降雨事件之不同累積小時試驗演算，程式已可對事件進行趨勢之預測以及表現高強度雨量之警戒性；但相對的，模式推估中對於時間遲延的影響，將造成如桃芝颱風此類短延時高強度降雨事件的錯誤預測；未來可針對更精確的雨量預測進行進一步的研究。

In this study, a feasible system must be built to realize a numerical model that can predict cumulative rainfall of Central Taiwan, by using GMS-5 geostationary meteorological satellite images and hourly rainfall data of some auto-recording rain gauges in Central Taiwan. During the study, rainfall records were used to estimate from dotted data to grid data by Block Kriging Estimation that match the pixels in satellite images, and satellite images were processed to represent the cloud top temperature data (CTT). The kernel function, that means the spatial characteristic of Central Taiwan, was calculated by the grid rainfall data and processed cloud top temperature above using Spatial Convolution Integral. Let the kernel function seemed as the state variable in a time-variant system, Kalman Filtering Algorithm was proposed to forecast this system’s state variable of next time. Then the forecasting can be calculated by Spatial Convolution Integral Technique again to transform into predictions of rainfall. Besides the numerical model development, this study built a practical system that predict rainfall of Central Taiwan by the model. By using the Graphical User Interface (GUI), and updating data hourly, the system will give the predicted data directly. After using two historical typhoon events as trials, the system can predict the trend during whole rainfall events and warn the dangerous districts; but there are still a problem of time-lag makes some error predictions.