Dataset for \An auto-adaptive thresholding method for urban flood mapping using Sentinel-1 images\

As satellite remote sensing provides enormous spatial-temporal data for flood modelling, the potential of Sentinel-1 images in urban environments is yet to be exploited. This paper presents an auto-adaptive thresholding method for improved urban flood mapping using Sentinel-1 images. Specifically, an initial water layer is expanded to auto-adaptively construct masks with enhanced bimodal distribution; the expanded mask is used as the input to map urban flood inundation; and an error analysis supported by multi-source datasets is presented. A case study is devised for Linhai City under the Super Typhoon Lekima in August 2019. The results show that the auto-adaptive thresholding method enhances the accuracy of urban flood mapping compared to the original method, particularly for the polarization combinations with better performance. These improvements stem from the enhanced bimodal distribution and the derived thresholds that effectively balance the retention of flooded objects against the exclusion of non-flooded objects. Among various polarization combinations, the VV polarization outperforms the VH polarization in suppressing building-related noise, while the combinations of addition, multiplication and squared addition achieve superior performance in urban flood mapping with overall accuracy exceeding 0.90. The proposed method demonstrates particular effectiveness in mitigating the interference sources common in urban environments. Notably, for 81.27% to 100.00% of the commission errors that can be classified, the contribution of roads and buildings is substantially diminished, while the relative importance of mountain shadows remains below 5.00%. These findings provide valuable insights for exploiting automated thresholding approaches in urban flood mapping algorithms using Sentinel-1 images.