Eyes on the Water: Leveraging Citizen Photos and AI for River Health Assessment and Management

River pollution is a global challenge recognized as unacceptable by citizens. Despite increasing awareness and investment in river water quality monitoring worldwide, current monitoring strategies fail to well characterise river health. In particular, the spatial and temporal resolution at which river health is currently monitored is insufficient and falls short to identify e.g., pollution spikes and point pollution sources. At the same time, the rise in citizen engagement in river monitoring, driven by increased awareness and widespread availability of smart phones and other monitoring technologies, has generated opportunities to overcome current monitoring barriers. 

 

In this paper, we share our experience collaborating with the community group Friends of Bradford’s Becks (FoBB, UK) to use citizen-collected photos for AI-based detection of health indicators, leading to enhanced river health management. More specifically, FoBB has collected around 100,000 photos of the streams that flow through and under Bradford. These images offer insights into the health of the becks, including specific pollution issues such as discharging overflows, sewage litter, discolouration, amongst other things, as well as how pollution has changed in time and space. The number of photos makes analysis challenging. In this project, we used AI models for automatic image labelling and prototyped several landing solutions for embedding the labelling model into a tool usable by citizens.

 

The project was initially set up in a Hackathon, funded by Natural England, and aimed to develop solutions using AI models. The landing solutions employed classification and object detection deep learning models to assist citizens by offering automatic detection of river health indicators. This not only reduces the cost of reporting but also improves the quality of reporting with comprehensive labels. Through community engagement, high spatio-temporal resolution data can be collected from citizens to fill the data gaps, including pollution levels, natural habitat conditions, and biodiversity. Additionally, while collecting the data, the deep learning models can be further fine-tuned to better assist citizens and managers in river health assessment and management. In summary, the project presents a holistic approach to river health management, combining the strengths of AI with the insights and engagement of local communities. The success of this approach in Bradford offers a template for similar initiatives globally, marking a step towards more informed and responsive river health management strategies.