A novel multi-strategy hydrological feature extraction (MHFE) method to improve urban waterlogging risk prediction, a case study of Fuzhou City in China

Abstract

Reliable hydrological data ensure the precision of the urban waterlogging simulation. To reduce the simulation error caused by insufficient basic data, a multi-strategy method (MHFE) for extracting hydrological features is proposed, which includes land use/land cover (LULC) extraction (LE) and digital elevation model (DEM) reconstruction (DR). First, the high-resolution remote image, satellite DEM, precipitation, flood points and depth, and planned LULC were collected. Second, the buildings, roads, and other areas of the satellite image were segmented using the U-Net model, and the LULC data with drainage features were extracted by combining the segmentation result with the planned LULC and drainage data. Then, the terrain features of the road were enhanced to construct high-precision DEM based on the fusion of multi-source data, such as elevation points, LULC, and satellite DEM. Finally, the waterlogging model was implemented under different return periods of rainfalls and typhoon rainfall to obtain the waterlogging distribution and water depth. The simulation results were compared with historical waterlogging event data and water depth observations. The results indicated that the proposed method significantly improved the accuracy of the simulation. In terms of identifying the waterlogging points, the average F1 score increased by 0.36, 0.20, and 0.07 compared to the raw model and the single LE and DR methods, respectively. In terms of water depth simulation, the average Nash-Sutcliffe efficiency (NSE) was increased from −0.24 to 0.86, with DR and LE contributing to 79.1 % and 20.9 %, respectively. The principal contribution and novelty of this paper is to explore the generic method that enhance the hydrological data, and the findings of this study improved the performance of urban waterlogging simulation.