Urban Waterlogging Prediction Based on Feature Extraction and Transfer

In waterlogging prediction, all or part of the real-time waterlogging data may be missing due to sensor failure, too sparse sampling interval setting, or sensor sensitivity problems, resulting in the failure of waterlogging prediction. In this study, we propose an urban waterlogging depth prediction method based on the transfer of waterlogging point feature extraction. The method quantifies the relationship between rainfall and waterlogging depth by extracting and constructing rainfall features at waterlogging points. Using only current or future rainfall data as model input to achieve future waterlogging depth prediction, it can effectively overcome the limitations of sparse distribution of monitoring stations and insufficient current real-time waterlogging data and can achieve more accurate medium-term waterlogging prediction and transfer prediction of water level at potential waterlogging-prone points.