Flood forecasting and pumping station warning in urban areas to improve flood-resilience: The project PuwaSTAR

The catchment areas of the Lippeverband (and Emschergenossenschaft, EGLV) in Western Germany are characterized by former coal mining activities. In consequence extensive subsidence areas without drainage have developed. These often densely populated polder areas are dependent on artificial drainage systems such as pumping stations that are crucial for flood protection. If the capacity of the pumping station is exceeded or if the pumping station (partially) fails, the water floods the drainage-free subsidence area. The consequences are life-threatening situations for the population and monetary losses of several billion euros.

The BMBF (Federal Ministry of Education and Research) joint project PuwaSTAR aims to develop a real-time forecasting system for potential flooded areas and their water depth in subsidence areas around pumping stations. Based on artificial intelligence (AI), time-consuming hydraulic simulations are replaced in the event of an incident. In addition to the hydrological forecast the operating status of the pumping station is considered during simulations, and failure scenarios are respected. The AI-model is based on a convolutional neural network (CNN) and designed to generate maximum water depth and inundation areas for the upcoming 24 h using discharge and rainfall data as input data. As part of this project, EGLV's existing flood forecasting system is extended to the forecast of flooded areas including details of the pumping stations status.

Although existing hazard and risk studies provide an overview of the potentially affected flood areas, a dynamic system allows for strategic disaster management. Thus, resulting options of targeted population warnings and initiation of prioritized measures reduce potential damage and protect the population. This enhances flood-resilience. The current operating status of the pumping station as well as a potential failure significantly contributes to the risk of flooding and must be considered likewise.

The real-time prediction based on AI will be demonstrated using the example of the Dorsten-Hammbach pumping station in the Hammbach catchment of the Lippeverband. According to an existing risk study, in the event of a pumping station failure and resulting flooding, the expected damage would amount to around €75 million, about 1,800 people and critical infrastructure would be affected. In collaboration with local authorities and first responders, the opportunities for improved forecasts for practical disaster management are derived. A participatory approach to elaborate and define requirements jointly with the stakeholders is a key aspect of the project. This enables targeted measures in the event of an incident and improves preparedness of the densely populated area around the pumping station.

The results of the project are intended to serve as a basis to transfer the methodology to further pumping stations and other controlling drainage elements, both in the EGLV catchment as well as those of other operators.