在過去，人們通常利用水文站以及地下水井監測水資源，然而這些方法不但昂貴、費時且需要許多人力資源進行記錄、維護。因此利用衛星資料來建立水資源監測網路是一個合適的替代方案。孟加拉擁有世界上人口密度第10高的國家，其人口密度是台灣的2倍高。在如此高人口密度的影響下，如何獲取水資源變成孟加拉相當重要的議題。雖然每年雨季時，來自印度洋的季風為孟加拉帶來大量的降雨甚至造成及為嚴重的水災，但因為其平坦的地形以及快速的逕流導致孟加拉無法將這些雨水儲存為可用水資源。因此孟加拉需要一個可以提供大尺度且連續水資源資訊的監測網絡。本研究使用多時期的衛星資料對錫爾赫特平原以及布拉馬普特拉河(雅魯藏布江下游)進行觀測，利用長期的觀測資料量化地表水的體積並結合重力衛星資料預估地下水的變遷。使用的資料包含兩個光學衛星Terra/MODIS 以及Landsat-5/-7/-8 的影像、SRTM之數值高程模型以及重力衛星GRACE的重力異常量資料。為達成量化的目的，首先必須填補光學影像中被雲所遮蔽的部分，我們針對光學衛星影像計算兩個研究區域的改良常態差異化水指數，從中萃取出水面積資訊，並將所有結果相加建立每個星期的淹水機率模型，利用此模形我們可以填補影像中雲的遮蔽。此外由於SRTM之數值高程模型在孟加拉地區的誤差可達數公尺，因此在本研究中我們利用建立出來的淹水機率模型改進SRTM的資料。利用填補後的水面積以及改進後的數值高程模型，我們便可以量化孟加拉的地表水體積。另一方面，在量化地表水體積後，將地表水資訊從GRACE的重力異常量中減去，我們便可以推估地下水的變遷趨勢。本研究方法在兩個研究區所量化的地表水體積曲線中皆可以觀測到於2004及2007年時發生的嚴重洪災之訊號，在地表水體積曲線與GRACE資料的比較中，兩研究區的相關係數皆大於0.9，而方均根誤差約為10公分。而由推估出的地下水變化量中可得知錫爾赫特平原的地下水有下降趨勢(約為0.5公分/年)，而布拉馬普特拉河則呈現下降趨勢(約為1.7公分/年)。此現象符合過去研究所觀測到的水位變化趨勢。;Conventional measures to monitor terrestrial water resources are the deploy of water gauges and in situ well. However, these methods are not only expensive and time-consuming, but also require lots of manpower and infrastructure setups. Therefore, using satellite observations to build a water resource monitoring network becomes an attracting alternative. Ranking as the 10th highest population density in the world, Bangladesh is suffering from multiple freshwater issues. Although the monsoon heading from the Indian Ocean brought lots of rainfall that even induce serious floods every year, it is not practicable for Bangladesh to store surface water due to its flat terrain. Meanwhile, the over-pumping of groundwater has induced extensive land subsidence in many administrative divisions. Therefore, Bangladesh needs a monitoring network that can provide large-scale and continuous data to manage their water resources. This research proposes a method to quantify surface water volume and further estimate the sub-surface water (include soil moisture and groundwater) trend. The study case focuses on Sylhet Plain which has the highest annual precipitation in Bangladesh. We first use the modified normalized difference water index to extract water area from Terra/MODIS MOD09A1 product and Landsat-5/-7/-8 Thematic Mapper (TM)/Enhanced TM plus (ETM+)/Operational Land Imager (OLI) imageries. Then we accumulate a sequence of images to create flood chance model for the recovery of cloud-covered surface. This approach extends the time series of WA with an overall accuracy of 70–80% in rainy season and 40–50% in dry season. This model can be further used to refine Shuttle Radar Topography Mission (SRTM), which has few meters uncertainty. Next, we simulate the flood extent using the modified SRTM and obtain and the overall accuracy of flood extent increases 19% compared to original data. By combining recovered WA and reconstructed DEM, the surface water volume (WV) is quantified and the signals of two extreme flood events in 2004 and 2007 are well observed in the estimated WV curve. The shifting days between estimated WV and GRACE equivalent water heights (EWHs) are 4 days in Sylhet Plain and 15 days in Brahmaputra River. The correlation coefficient and RMS of the EWH difference are 91.7% / 0.10 m in Sylhet plain and 95.48% / 0.12 m in Brahmaputra River. Finally, we subtract surface water from GRACE EWH and the result shows a decreasing trend of sub-surface water at 0.5 cm yr-1 in Sylhet Plain and decreasing trend at 1.7 cm cm yr-1 in Brahmaputra River, which agree with previous studies.
