Sentinel-1 Spatiotemporal Simulation Using Convolutional LSTM for Flood Mapping

The synthetic aperture radar (SAR) imagery has been widely applied for flooding mapping based on change detection approaches. However, errors in the result are expected since not all land-cover changes flood-induced, and those sensitive to SAR data, such as crop growth or harvest over agricultural lands, clearance of forested areas, and/or modifications urban landscape. This study, therefore, incorporated historical images boost flood-induced during extreme weather events, using Long Short-Term Memory (LSTM) method. Additionally, incorporate spatial signatures detection, we a deep learning-based spatiotemporal simulation framework, Convolutional (ConvLSTM), simulating image Sentinel One intensity time series. will be prepared advance flood then it can used detect areas when post-image is available. Practically, significant divergence between inundated zones, which mapped by applying thresholds Delta (synthetic minus post-image). We trained tested our model three events from Australia, Brazil, Mozambique. generated Flood Proxy Maps were compared against reference data derived Two Planet Labs optical data. To corroborate effectiveness proposed methods, also products two baseline models (closest pre-image mean post-image) LSTM architectures: normal ConvLSTM. Results show that thresholding ConvLSTM yielded highest Cohen’s Kappa coefficients study cases: 0.92 0.78 Mozambique, 0.68 Brazil. Lower values obtained Mozambique case subject topographic effect imagery. These results still confirm benefits terms classification accuracy convolutional operations provide series analysis satellite employing spatially correlated information learning framework.