Monitoring surface water dynamics is essential in understanding the effects of natural phenomenon and anthropogenic processes in water resources. Thus, satellite-based remote sensing has been utilized to visualize natural water variability. However, satellite images experience serious missing pixel information caused by cloud cover, partial satellite coverage and satellite sensor failures. This leads to limited surface water extent observations, especially during the rainy season – from as long as one month to a full season. Therefore, this research aims to pioneer data gap-filling of missing surface water pixel information from Landsat-8 and Sentinel-2 satellite imageries. A novel STSp-Unet framework is proposed to study the nonlinear relationship between the spatial, temporal and pseudo-spectral information domains. Sentinel-1 SAR data are integrated in the surface water extent inputs to solve the gap in optical satellite data availability during the rainy season. Effect of surface and subsurface soil moisture as preliminary pseudo-spectral information is also investigated. Performance analysis of the approach reflects strong data recovery even at high percentage of simulated pixel contaminations. Inclusion of SAR data contributes significantly in the gap-filling results, while soil moisture inputs have little to no effect. Actual data-gap filling implementation over the region of interest shows notable improvement in the mean revisit interval of surface water extent observations up to three-folds (from 6.2 days to 1.8 days). Utilization of the proposed framework and combination of optical and SAR data sources offers the capability to provide both high spatial and temporal resolution SWE data. This promotes opportunities for the extended application of satellite remote sensing in land surface hydrological modelling and flood inundation mapping, especially in poorly gauged to ungauged regions.
各種検索
サポート
ヘルプ