In hydrology and flood inundation studies, high temporal and high spatial resolution of surface water extent (SWE) estimates derived from satellite-based remote sensing are crucial for better understanding of natural water variability. However, satellite imagery is often affected by cloud cover, partial satellite coverage, and sensor failures (Pekel et.al., 2016; Ju and Roy, 2008), which can result in gaps in observations of SWE. Therefore, it is necessary to optimize the available satellite information and address the gaps in observations caused by cloud-heavy conditions, partial satellite coverage, and other factors . In this study, we propose to achieve the following objectives: (1) conduct a simple optical-SAR data fusion approach to increase available useful information during the rainy season; and (2) establish a deep learning architecture for effective gap-filling of Landsat-8 and Sentinel-2 SWE observations against any potential causes of missing data. Specifically, the deep learning architecture will leverage the spatio-temporal-pseudo-spectral (STSp ) relationship between observations to fill missing SWE pixels, using a Unet-based convolutional neural networks.
各種検索
サポート
ヘルプ