Caitlin Rogers, the FloodAI innovation lead from Isle Utilities, shares the resilience and innovation aspirations of FloodAI.
In Northumberland, there are six catchments that do not have access to the regular Environment Agency flood warnings. This means communities have no warning to protect their property, or even their safety. This is because when it rains, the water levels in these small to mid-sized catchments rise so quickly that hydraulic models cannot predict the flooding with enough time for communities to act on the information. The FloodAI project is attempting to alleviate this problem, by providing a means of flood warning to these “rapid response” catchments. This is being done through the development and testing of an artificial intelligence (AI) model to better predict flash flooding.
FloodAI is funded by Defra as part of the £200 million Flood and Coastal Innovation Programmes, which is managed by the Environment Agency. The programmes will drive innovation in flood and coastal resilience, and adaptation to a changing climate.
Flood resilience
At the heart of FloodAI are the six communities of Hepscott, Ovingham, Riding Mill, Stocksfield, Haltwhistle and Acomb in Northumberland, as shown in the maps below.
One of the main benefits the project is trying to achieve is the creation of a more resilient, community-based flood response across these six communities, through the pilot of the new enhanced flood warning service. In practice, this means that FloodAI will give flood risk planners, responders, and community groups better information to improve their preparedness for flooding. These before and after photos of the Whittle Burn, Ovingham (taken in 2019) show the very local effects of this flooding.
In the words of the Northumberland flood wardens, flood resilience is about “good honest communication with at risk areas” and “knowing when and where flooding is likely, enabling a quick and appropriate response to mitigate the worst effects when flooding happens”.1
Ultimately, it is only the residents and responders who know what information they require when and in what format. In order to capitalise on this important, locally-situated knowledge, Arup are co-creating the flood warning messages with the local stakeholders.
Transparent artificial intelligence
The innovation enabling this enhanced resilience is a significant change from the current hydraulic model-based forecasting.
| Traditional vs AI Forecasting |
|---|
| Professor Wai Lok Woo, Professor of Machine Learning at Northumbria University and the FloodAI artificial intelligence developer, explains the difference between the business-as-usual of hydraulic model forecasting and FloodAI’s innovative forecast modelling. “Traditional hydraulic model forecasting involves using established physical laws and empirical relationships to simulate the behaviour of water flow in natural and man-made systems. These models typically use a combination of mathematical equations, such as the Saint-Venant equations for open channel flow, to predict water levels, flow rates, and other hydraulic parameters based on input data like rainfall, soil moisture, and topography. The accuracy of these models relies heavily on the precision of the input data and the assumptions made during model development. In the FloodAI project, our AI-based forecasting models offer several advantages over traditional hydraulic models. These AI models, often utilising machine learning techniques, can automatically learn complex patterns and relationships from large datasets without the need for explicit physical equations. This allows them to handle more diverse and dynamic inputs, such as real-time sensor data and climate projections. Additionally, interpretability techniques enable stakeholders to understand and trust the model’s predictions by providing insights into how different factors influence the forecast. As a result, AI-based models can provide more accurate, adaptable, and understandable predictions, especially in environments where conditions change rapidly and traditional models struggle to keep up.” |
Rebecca Croft, an FCERM Officer at Northumberland County Council and FloodAI project lead considers this departure from business-as-usual: “The use of artificial intelligence in flood risk management is an emerging innovation that is still in the early stages of use and research. This project will provide valuable information for future use. The AI models will inform human decisions”. Whilst it is true that there are AI models currently being used for flood prediction, the most well-known probably being Google’s Flood Hub, these are typically used for larger rivers and not for the smaller, rapid response catchments that FloodAI is tackling.
The discussion about trust in AI is important in society as a whole: therefore, considerations on this subject are being factored into the FloodAI project. Northumbria University are developing the AI models to be ‘transparent’, which means that the model results will come with explanations of how they were derived, as opposed to many AI models, some of which operate as a black box with no insight for the end-user into how the outputs were determined.
Responding to data challenges
Artificial intelligence models require large datasets to be accurately trained. In the context of FloodAI, this means gathering and transmitting significant amounts of data, such as soil moisture and rainfall, from the six catchments. These data will be connected and transmitted using data loggers and a wireless sensor network being developed by Northumbria University, and it is novel in the way the loggers balance power consumption with data collection and transmission requirements. To supplement this real-world data, Arup are generating ‘synthetic data’ to train the AI.
Innovation adoption
FloodAI is being developed with features to make it attractive to both future adopters (i.e. those who will own and operate the system) and end-users (i.e. those who will receive the flood predictions or warnings). These include transparent AI, predictions with sufficient warning and information for local stakeholders to act, and a wireless sensor network with optimised power consumption. To facilitate uptake of the model, the project is determining the costs and benefits of the flood warning service, with a view to better understanding the skills and capacities required to deliver the service, and developing an operational and technological framework that can be shared, repeated or scaled nationally for other communities facing similar flooding risks.
If you are interested in learning more about FloodAI, or if you are a potential adopter or end-user, you can receive regular project updates here or contact caitlin.rogers@isleutilities.com directly.
Acknowledgements
The FloodAI project is delivered in partnership with Northumberland County Council (FloodAI project management), Environment Agency (FloodAI advice), Northumbria University (FloodAI wireless sensor network and artificial model development), Arup (FloodAI service co-creation, service platform and synthetic data development, operational framework development) and Isle Utilities (FloodAI innovation governance).
References
- Opening rural communities to a new flood warning system (2024) https://engageenvironmentagency.uk.engagementhq.com/floodai ↩︎
