Urban flood prediction in real-time from weather radar and rainfall data using artificial neural networks

This paper describes the application of Artificial Neural Networks (ANNs) as Data Driven Models (DDMs) to predict urban flooding in real-time based on weather radar and/or raingauge rainfall data. A 123manhole combined sewer sub-network from Keighley, West Yorkshire, UK is used to demonstrate the methodology. An ANN is configured for prediction of flooding at manholes based on rainfall input. In the absence of actual flood data, the 3DNet / SIPSON simulator, which uses a conventional hydrodynamic approach to predict flooding surcharge levels in sewer networks, is employed to provide the target data for training the ANN. The ANN model, once trained, acts as a rapid surrogate for the hydrodynamic simulator. Artificial rainfall profiles derived from observed data provide the input. Both flood-level analogue and flood-severity classification schemes are implemented. We also investigate the use of an ANN for nowcasting of rainfall based on the relationship between radar data and recorded rainfall history. This allows the two ANNs to be cascaded to predict flooding in real-time based on weather radar.