How can the UK improve flood resilience in the Thames Estuary?

The Challenge 

Coastal flooding poses a growing threat to lives, communities and critical infrastructure as sea-level rise and extreme weather place increasing pressure on existing flood defences. In the Thames Estuary, the Environment Agency’s Thames Estuary 2100 (TE2100) Plan highlights the need to improve the operation of existing defences while planning for long-term adaptation. 

However, conventional forecasting models cannot fully capture all the physical processes that influence coastal water levels, limiting forecast accuracy and their usefulness for decision-making. This Sprint aimed to improve coastal flood forecasting by combining established physics-based models with machine learning, while ensuring that the resulting tools could be integrated into existing operational systems and meet the needs of the organisations responsible for managing flood risk. 

The Solutions 

The Sprint developed and validated RTide, a machine-learning framework that augments existing physics-based forecasting models rather than replacing them. Testing across the UK and in the Thames Estuary demonstrated an average improvement of more than 40% in forecasting performance, including for extreme events. 

The research demonstrates that machine learning can learn unresolved coastal dynamics and systematic errors in conventional models, improving forecasts while retaining the robustness of established systems. A streamlined forecasting framework and software wrapper was also developed for operational use, alongside new scientific findings on hybrid AI–physics modelling and a probabilistic framework for storm-surge prediction. 

The Pathway 

This Sprint brought together researchers in oceanography, coastal engineering and machine learning and created a consortium with partners from the Environment Agency, the Met Office and the National Oceanography Centre. Through monthly consortium meetings, workshops and ongoing technical exchange, partners helped to shape the model development, validation and implementation to meet operational needs. 

Although initially focused on the Thames Estuary, the work expanded to validate the approach across the UK, demonstrating its wider applicability. In response to partner input, the team prioritised integrating the forecasting framework into existing systems over developing a separate interactive benchmarking tool. This adaptive, collaborative approach helped translate the research into practical tools that partners could test and implement, while identifying opportunities for further development of probabilistic forecasting. 

What Happened Next? 

The Sprint has taken the forecasting framework from research development to operational testing. Pilot assessments are being undertaken by the Environment Agency, the Met Office and the Thames Barrier teams, supported by streamlined code, implementation guidance and continued knowledge exchange. The Met Office has secured two years of funding to support further testing and integration, providing a foundation for continued operational adoption. 

The research has also generated wider scientific impact, including a manuscript accepted in Nature Communications and further publications in preparation. Partnerships established through the Sprint are supporting discussions on future research and funding. Looking ahead, the team aims to extend the approach to British Overseas Territories and other coastal regions facing significant flood risks, with the potential to improve flood forecasting and inform both operational decisions and longer-term resilience planning under TE2100.