高時空解析度衛星降雨推估在台灣地區的特性與誤差分析

本研究將評估GSMaP (Global Satellite Mapping of Precipitation)各個版本衛星資料反演台灣不同季節與日循環的降雨表現能力，並根據不同的降雨類型(颱風降雨、午後對流)探討GSMaP衛星資料的差異。經分析後發現季節下的空間掌握分布能力以GSMaP MVK V5在濕季(梅雨季、夏季、秋季)表現最佳，但是時常有低估的情況發生。另外GSMaP RNL與GSMaP MVK V6普遍存在高估的情形，所有版本中以GSMaP MVK V7的降雨量估計較接近地面觀測資料。日循環的部分，衛星資料與地面觀測資料在乾季的降雨量差異皆不大，但是對於時序變化表現不佳，濕季則是每個版本皆有掌握到一日內最大降雨時段發生在午後，其中以GSMaP MVK V7對於梅雨季與夏季的時序變化表現最優越。GSMaP RNL與GSMaP MVK V6在濕季的清晨時段有相當嚴重的高估情形，尤其以夏季最為明顯，經分析後發現GSMaP RNL與GSMaP MVK V6清晨時段高估主要來自於某些颱風的降雨。第二部分以探討降雨類型為主，分為颱風降雨(Tropical Cyclones Precipitation, TCP)與午後對流(Convection Afternoon Precipitaion, CAP)兩種降雨類型，因衛星資料年份不同，所以分成兩組比較(第一組: GSMaP RNL與GSMaP MVK V5，第二組: GSMaP MVK V6與GSMaP MVK V7)。在颱風降雨(TCP)中，以GSMaP MVK V7的表現最佳，不但在空間分布有一定的掌握外，在所有GSMaP版本中針對降雨量估計存在著最小的誤差。午後對流(CAP)中，即使各版本對於平坦或沿海區域的掌握較好，但是時有低估的狀況發生，且對於每小時日循環降雨貢獻掌握存在一定的極限。已知在不同季節下，不同GSMaP版本的表現皆不相同外，不同降雨類型的表現也不盡相同，但是從季節、降雨類型的表現中可以發現GSMaP MVK V7的降雨量估計為所有版本中最穩定的。衛星資料的降雨估計仍存在一定的極限，因此在未來需再針對其他天氣型態或降雨類型做更多的評估，以利使用者在使用GSMaP資料評估台灣時有一定的參考依據。

關鍵字

季節；日循環；颱風降雨；午後對流

並列摘要

This study will evaluate rainfall performance of different seasons, diurnal cycles and rainfall types over Taiwan of each GSMaP version in first part. In the second part, we will discuss the performance of each GSMaP version in different rainfall types such as Tropical Cyclones Precipitation (TCP)、Convection Afternoon Precipitaion (CAP). After analysis, although we found that GSMaP MVK V5 had the best spatial distribution in wet season (MJ、JA、SO) of all GSMaP versions but it always underestimated observation data (Gridded Rain Gauge Data). In wet season, GSMaP RNL and GSMaP MVK V6 are always overestimated, GSMaP MVK V7 had the best rainfall estimation in wet season. In diurnal cycle part, each GSMaP version had low bias in dry season, and they can captured maximum rainfall occurs in afternoon of the day during wet season. Especially, GSMaP MVK V7 had the best timing change in MJ and JA. Among them, GSMaP RNL and GSMaP MVK V6 are severe overestimated over southwestern Taiwan in the morning. However, it has been known that overestimation is mainly caused by some TCP events. In the second part, we mainly discuss CAP and TCP these two rainfall types. Due to GSMaP data had different period in each version. We split two teams : (1) GSMaP RNL and GSMaP MVK V5. (2) GSMaP MVK V6 and GSMaP MVK V7. In TCP part, GSMaP MVK V7 had the best spatial distribution and rainfall estimation in all versions. In CAP part, although all GSMaP version were underestimation, they can captured rainfall in flat and coast area. We know that GSMaP data were poor at perform timing changes in CAP hourly contribution. For all we know, GSMaP had different perform in each version, but we can know GSMaP MVK V7 had most stable rainfall estimation in seasonal and rainfall types. It illustrated that satellite products can’t perfect to capture rainfall truly, so we need to assessment other rainfall types in the future. It can provide GSMaP user for a reference.