Bangladesh has been distressed by freshwater issue for long. Every year the Indian monsoon and melting snow from the Tibetan Plateau has brought considerable amount of water. However it is inconvenient for Bangladesh to capture water resources because of flat terrain. Due to the lack of a complete monitoring network, using satellite image to measure surface water area/volume becomes a practical solution. However, during rainy seasons persistent cloud cover makes optical remote sensing less useful in surface detection. This research focuses on using the knowledge of water occurrence derived from long-term MODIS images, to potentially bypass the cloud cover problem. We utilized the statistics of water occurrence and combined with cloudy images to improve the reorganization of surface water area. We firstly used MODIS 8-day composite data in 2000-2015, and assumed that in the same time period of each year the surface water distribution would be similar and the surface terrain remains unchanged. Thus, a weekly inundation model was constructed as a reference for images with limited surface pixels in cloud openings. We set the lowest chance of water occurrence that shown as water pixel in the image as threshold, and classified the rest of pixels with higher chance as water even if they are under cloud covered. For validation, we applied Sentinel-1A and obtain an overall accuracy as high as 70%. The accuracy can be further improved by using Landsat 30-m imageries with lower temporal resolution.