A common LSTM cell unit.

As a result of climate change and rapid urbanization, urban waterlogging commonly caused by rainstorm, is becoming more frequent and more severe in developing countries. Urban waterlogging sometimes results in significant financial losses as well as human casualties. Accurate waterlogging depth prediction is critical for early warning system and emergency response. However, the existing hydrological models need to obtain more abundant hydrological data, and the model construction is complicated. The waterlogging depth prediction technology based on object detection model are highly dependent on image data. To solve the above problem, we propose a novel approach based on Temporal Convolutional Networks and Long Short-Term Memory networks to predicting urban waterlogging depth with Waterlogging Monitoring Station. The difficulty of data acquisition is small though Waterlogging Monitoring Station and TCN-LSTM model can be used to predict timely waterlogging depth. Waterlogging Monitoring Station is developed which integrates an automatic rain gauge and a water gauge. The rainfall and waterlogging depth can be obtained by periodic sampling at some areas with Waterlogging Monitoring Station. Precise hydrological data such as waterlogging depth and rainfall collected by Waterlogging Monitoring Station are used as training samples. Then training samples are used to train TCN-LSTM model, and finally a model with good prediction effect is obtained. The experimental results show that the difficulty of data acquisition is small, the complexity is low and the proposed TCN-LSTM hybrid model can properly predict the waterlogging depth of the current regional. There is no need for high dependence on image data. Meanwhile, compared with machine learning model and RNN model, TCN-LSTM model has higher prediction accuracy for time series data. Overall, the low-cost method proposed in this study can be used to obtain timely waterlogging warning information, and enhance the possibility of using existing social networks and traffic surveillance video systems to perform opportunistic waterlogging sensing.