Mapping Outburst Floods Using a Collaborative Learning Method Based on Temporally Dense Optical and SAR Data: A Case Study with the Baige Landslide Dam on the Jinsha River, Tibet

Outburst floods resulting from giant landslide dams can cause devastating damage to hundreds or thousands of kilometres a river. Accurate and timely delineation flood inundated areas is essential for disaster assessment mitigation. There have been significant advances in mapping using remote sensing images recent years, but little attention has devoted outburst mapping. The short-duration nature these events observation constraints cloud cover significantly challenged This study used the Baige dam on Jinsha River 3 November 2018 as an example propose new method that combines optical Sentinel-2, synthetic aperture radar (SAR) Sentinel-1 Digital Elevation Model (DEM). First, cloud-free region, comparison four spectral indexes calculated time series Sentinel-2 indicated normalized difference vegetation index (NDVI) with threshold 0.15 provided best separation flooded area. Subsequently, cloud-covered analysis dual-polarization RGB false color composites backscattering coefficient differences SAR data were found apparent response ground roughness’s changes caused by flood. We carried out range prediction model based random forest algorithm. Training samples consisted 13 feature vectors obtained Hue-Saturation-Value space, differences/ratio, DEM data, label set prepared images. Finally, field investigation confusion matrix tested accuracy end-of-flood map. overall Kappa 92.3%, 0.89 respectively. full extent was successfully within five days its occurrence. multi-source merging framework massive sample preparation proposed this paper, provide practical demonstration similar machine learning applications sensing.