There has recently been a strong and very welcome resurgence of interest in biofilters as a powerful and sustainable water treatment technology. Examples include outputs from EBNet’s Water Biofilms WG, and funding awards for the SandSCAPE project and for the EPSRC Network+ Better Water 4 All, as well as ongoing activities in the National Biofilms Innovation Centre (NBIC) and the Environmental Biotechnology Innovation Centre (EBIC).
Developing tools to predict biofilter performance remains a fascinating challenge, however, due both to the complex and dynamic microbial interactions in these systems, and to the sparsity of supporting data.
A recent paper in Scientific Reports by Uzma et al. (2025) describes EnviroPiNet, a novel physics-guided AI framework for predicting biofilter performance by modelling carbon concentration dynamics. The model incorporates a physics-based backbone that allows it to learn the physical properties of complex environments, ensuring predictions are grounded in system behaviour. When benchmarked against conventional methods, EnviroPiNet showed superior performance in identifying variables critical to the evaluation of biofilter performance.
Dr Uzma, the paper’s Lead Author, said “Our work shows how physics-informed machine learning can provide a more robust framework for predicting biofiltration outcomes, making it a vital AI tool for the future of environmental engineering”.
For other recent posts and publications from this very active group at the University of Glasgow, see:
- Microbial ecology and drinking water biofiltration – EBNet
- Microbial ecology in drinking water biofilms – a taster with more to come – EBNet
 A machine learning model guided by physical principles for biofilter performance prediction. Uzma, Cholet, F., Quinn, D., Smith, C.J., You, S. and Sloan, W.T., 2025. Scientific Reports, 15(1), p.34664.Â