Analysis

Dr Mónica Rivas Casado explores new ways AI can be integrated into surface water flooding monitoring and management.

In England alone, 5.6 million properties have been reported to be at risk of flooding, with approximately 3.1 million households at risk of flooding due to rivers and the sea.^1,2,3 The impact of surface water flooding is expected to grow over the next few decades, due to the increased intensity and frequency of extreme rainfall events. Therefore, flood management tools that increase preparedness and response to urban surface water flooding are being sought by both national and local authorities alike.⁴ Within this context, both the management of the drainage network and the local microtopographic configuration play a crucial role. Microtopography refers to the small changes in slope and terrain configuration that occur at a fine scale. For example, gullies that are not working to their full efficiency (e.g., due to blockages or damaged infrastructure), as well as localised slopes, have been known to aggravate surface water flood impact to residential properties.⁵

Recent advances in the use of artificial intelligence (AI) in flood management are contributing to a better understanding of microtopography configuration. For example, a recently funded project by Atkins Réalis and EPSRC⁶ is exploring the potential use of AI to identify gullies working below their optimum efficiency level along roads and paved areas. Initial results have generated a range of AI algorithms for the rapid and automated detection of gullies, from high-resolution imagery collected with unmanned aerial vehicles (Figure 1). These algorithms can also be used with LiDAR imagery (Figure 2) without compromising the quality of the results. The imagery collected in the visible spectrum offers unprecedented detail of the microtopographic configuration of the surveyed areas,⁵ enabling the estimation of flow paths and water accumulation in close proximity to households.⁷ The information can then be utilised to advise local authorities and homeowners on the type and location of resilient and resistant measures, including nature-based solutions, that would work best to protect properties.

Several authors have also developed AI architectures to successfully identify flood depth and extent.⁸ Others have explored the used of AI algorithms for the identification of debris from residential properties affected by fluvial flooding (Figure 3).⁹ Homeowners usually start their clearing-up tasks as soon as insurance companies carry out their door-to-door inspections and assess the damage caused by flooding. Refuse, including upholstery, laminated flooring and carpets, amongst other items, accumulate next to affected properties. This debris can be visually identified from UAV aerial imagery when small areas are affected, but a more automated approach is required for wide-area coverage. AI algorithms enable a more rapid and perhaps accurate inspection than visual approaches. This in turn could speed up loss-adjustment inspections of affected properties if adopted by insurance companies. Although the algorithms have only been used at this stage for fluvial (river- or stream-based) flooding, they can easily be adapted to identify debris generated from surface water flooding.

AI approaches have also been developed for flood forecasting,^10,11 using a combination of spatial and time series data (e.g., rainfall data and river flow levels). The Flood Hub,¹² for example, provides spatially explicit flood forecasts up to 7 days in advance using Google’s AI models and global data sources. The input data for the model includes multiple global variables, excluding any reference to streamflow data.^13,14

AI-supported citizen science is also being used for flood forecasting. CrowdWater¹⁵ relies on citizens to collect and upload geo-referenced photos of water bodies to inform a machine learning flood forecasting model for ungauged catchments that are vulnerable to flash floods. AI architectures have also been developed to address any flood-related questions citizens may have. The Environment Agency is working with the “Hello Lamp Post” technology (Figure 4)¹⁶ to develop an AI system that enables citizens to obtain flood-related information. “Hello Lamp Post” uses AI to draw plausible answers to questions placed by citizens using an IoT system habilitated for the purpose. It is a place-based virtual assistant that is accessed via scanning a QR code, or by sending an SMS message to the system. Users are then taken to either a mobile responsive “WebApp” web page, or they interact via a standard SMS interface. Users do not need to download any apps or software to their phones, and the service operates like IoT technology, as it is available in specific locations.

However, it does not need any equipment or sensors installed in those locations. A physical sign is all that is required to inform the users that the location is now interactive. The system is designed specifically to lower barriers to entry for customers and users. All data in each “Hello Lamp Post” implementation is supplied directly by the Environment Agency to ensure the accuracy of the information provided to the user. This information is stored in a purpose-built “Hello Lamp Post” knowledge base, and the AI helps to make sense of the user’s question, identify the correct answer, and return this to the user. With a network of such systems distributed within urban settlements, citizens will have additional mechanisms to quickly address any pressing queries about flooding. The system can also be used to better inform citizens about surface water flood risk within their area, in addition to showcasing the mechanisms homeowners can use to prevent and mitigate impact.

The wide range of AI tools developed for flood science would benefit from a solution that merges them together into a system operating towards a common objective, defined by an increase in urban resilience to surface water flooding. Without such a system in place, opportunities to maximise the benefits of AI in flood science may be missed.

---

1. Surface Water: The Biggest Flood Risk of All (2018). https://www.gov.uk/government/news/surface-water-the-biggest-flood-risk-of-all ↩︎
2. Social Deprivation and the Likelihood of Flooding (2020). https://www.gov.uk/government/publications/social-deprivation-and-the-likelihood-of-flooding ↩︎
3. Flooding in England: A National Assessment of Flood Risk (2009). https://assets.publishing.service.gov.uk/media/5a7ba398ed915d4147621ad6/geho0609bqds-e-e.pdf  ↩︎
4. Improving the UK’s resilience to floods and droughts (2024). https://www.ukri.org/news/improving-the-uks-resilience-to-floods-and-droughts/ ↩︎
5. Accuracy Assessment of Surveying Strategies for the Characterization of Microtopographic Features That Influence Surface Water Flooding (2023). https://www.mdpi.com/2072-4292/15/7/1912 ↩︎
6. Intelligent Dynamic Flood Response and Recovery Strategy (iD-FRe2S) (2021) https://www.cranfield.ac.uk/research-projects/intelligent-dynamic-flood-response-and-recovery-strategy-id-fre2s ↩︎
7. Harnessing long-term gridded rainfall data and microtopographic insights to characterise risk from surface water flooding (2024). https://doi.org/10.1371/journal.pone.0310753 ↩︎
8. A deep learning model for predicting river flood depth and extent (2021) https://www.sciencedirect.com/science/article/pii/S1364815221002280 ↩︎
9. Detection of Flood Damage in Urban Residential Areas Using Object-Oriented UAV Image Analysis Coupled with Tree-Based Classifiers (2021). https://www.mdpi.com/2072-4292/13/19/3913 ↩︎
10. Recent Progress of Artificial Intelligence Application in Flood Forecasting (2024). https://www.frontiersin.org/research-topics/59116/recent-progress-of-artificial-intelligence-application-in-flood-forecasting ↩︎
11. Flood forecasting based on machine learning pattern recognition and dynamic migration of parameters (2023). https://www.sciencedirect.com/science/article/pii/S2214581823000939 ↩︎
12. Flood Forecasting (2024). https://sites.research.google/floodforecasting/ ↩︎
13. Global prediction of extreme floods in ungauged watersheds (2024)  https://www.nature.com/articles/s41586-024-07145-1 ↩︎
14. The Flood Hub (2024) https://thefloodhub.co.uk/ ↩︎
15. CrowdWater: A Citizen Science Revolution for Flood Prediction with Machine Learning (2024). https://blogs.egu.eu/divisions/hs/2024/02/22/crowdwater-a-citizen-science-revolution-for-flood-prediction-with-machine-learning/ ↩︎
16. Environment Agency: Overview (2024). https://www.hlp.city/locations/environment-agency/ ↩︎