With the intensification of global climate change and the increasing frequency of extreme weather events, flood disasters pose an escalating threat to human society and the ecological environment, severely endangering people's lives and property. Traditional ground-based observation methods have limitations in large-scale flood monitoring and are unable to meet the fast and efficient emergency response needs. This study, based on radar satellite imagery, introduces several methods such as the Dual-Polarization Water Index (SDWI-OSTU), Support Vector Machine (SVM), and Random Forest (RF), and uses the Classification Ternary Combination (CTC) ensemble strategy to identify and evaluate the accuracy of flood inundation areas on typical dates during the flood seasons of May to July 2016 and June to August 2020 in the Chaohu Basin. The monitoring error of flood inundation areas is kept within 10%. Data files are named using the format "Algorithm Name + Date".
The Sentinel-1 satellite, part of the European Space Agency's Copernicus Program (GMES), is an Earth observation satellite system consisting of two satellites: Sentinel-1A and Sentinel-1B. It is equipped with a C-band synthetic aperture radar (SAR) capable of providing continuous imagery (day, night, and in all weather conditions). Sentinel-1 offers four stripmap scanning modes, among which the IW (Interferometric Wide) mode is specifically designed to capture images of land surfaces, featuring VV and VH polarization modes. Therefore, this study utilizes Level-1 Ground Range Detected (GRD) data products in IW mode. Remote sensing imagery during the basin-wide floods of the Chaohu Basin in the flood seasons of 2016 and 2020 was downloaded from the ESA website (https://scihub.copernicus.eu/), with a spatial resolution of 10 meters.
Based on multi-temporal radar imagery data, water body identification simulations were conducted using the dual-polarization water body index method, support vector machines, and random forest methods. The three sets of water body identification results were evaluated using the balanced index calculation unit based on the classification ternary combination to calculate their respective balanced accuracy and integration weights. A weighted combination of the multi-model water body identification results was performed. The integrated water body identification accuracy was assessed using a validation sample set, and the pixel data of the integrated results were converted into vector boundaries to delineate the flood inundation boundary within the basin.
For radar imagery manually labeled sample points, the classification ternary combination method demonstrates higher stability and accuracy over different time periods, with an average accuracy of 0.969 and precision of 0.965. It effectively integrates the strengths of individual models, providing more consistent and accurate classification results for flood inundation mapping. For manually labeled sample points from UAV imagery, the classification ternary combination method achieves both Accuracy and Precision above 0.900, enabling relatively accurate identification of water body areas. Regarding water area identification, the classification ternary combination model provides accurate results in most cases, maintaining flood inundation monitoring errors within 10%.
| # | number | name | type |
| 1 | 2021YFC3000100 | Lower Yangtze River Flood Disaster Integration and Control and Emergency De-risking Technology and Equipment | National key R & D plan |
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.NCDC.NHRI.DB6788.2025
10.12072/ncdc.nhri.db6788.2025