Bath Engineering & Design Studentships: Machine learning for early warning of debris screen blockage in urban rivers

This project is one of a number in competition for a studentship from the Faculty of Engineering and Design at the University of Bath. If successful, this studentship is expected to commence 04 October 2027.

An alternative start date may be possible if agreed with your intended supervisors and the Doctoral College.

This PhD project will develop new methods for forecasting debris and trash screen blockages in urban rivers, enabling a transition from reactive maintenance to proactive flood risk management. Across the UK, thousands of culverts are protected by debris screens that prevent large objects from entering hydraulic structures. However, these screens can become blocked by vegetation, litter and other debris transported from upstream catchments. Such blockages reduce conveyance capacity, increase flood risk, and often require urgent maintenance interventions that expose operational staff to health and safety risks during high-flow conditions. Current management relies largely on routine inspections or responses after blockage has occurred, creating a need for predictive tools that can identify elevated risk in advance. Although recent advances have demonstrated that machine learning can detect blocked screens from CCTV images (Smith et al., 2025), there is no framework for forecasting future blockage events. The project is founded on the hypothesis that debris blockage is governed by identifiable hydro-meteorological controls and can therefore be predicted before it occurs. In particular, the interactions between rainfall characteristics, antecedent conditions, catchment properties and debris mobilisation remain poorly understood, limiting the ability to forecast blockage risk. The research will address three core challenges. First, it will develop image-based machine learning models capable of identifying blockage conditions and transferring knowledge between sites with limited historical data. Second, it will investigate how weather observations, rainfall forecasts and catchment descriptors can be combined with image-derived information to predict the probability of future blockage events. Particular emphasis will be placed on understanding the role of antecedent dry periods and rainfall intensity in controlling debris accumulation and transport. Third, the project will couple probabilistic blockage forecasts with hydraulic flood models to quantify the implications of screen blockages for downstream flood risk. Through real-world case studies, the project will analyse large CCTV image archives, meteorological datasets and hydraulic models to create a new scientific framework for debris-blockage forecasting. The resulting methods will enable risk-based maintenance planning, reduce unnecessary inspections, improve operational safety, and enhance the resilience of urban drainage infrastructure under future climate extremes.