Hot Off The Press — How to Teach AI: Multi-Objective Water Quality Modelling for Emergencies

Post

Published on August 12, 2025

The Humanitarian Water Engineering Lab is happy to announce a new publication entitled “Training for the test: Using multi-objective training to improve ANN ensemble forecasts of household residual chlorine in emergencies” in the journal PLOS Water. The paper co-authored by Michael De Santi, Syed Imran Ali, Jean-François Fesselet, Matthew Arnold, Dawn Taylor, and Usman Khan presents novel multi-objective training approaches for machine learning models of chlorine decay in humanitarian settings.

In emergency settings, residual chlorine, a common disinfectant, provides critical protection against recontamination of household-stored drinking water by pathogens like cholera and hepatitis E, but ensuring sufficient residual chlorine is present can be very challenging because chlorine decays over time. Water system operators need to set chlorination targets that ensure water is protected up to the time when water is consumed. Machine learning (ML) modelling can help operators to set these targets by predicting future decay, but most conventional ML models fail to accurately capture the variability of water quality changes in humanitarian settings, often under-predicting recontamination risks. As part of the Safe Water Optimization Tool project, this research, led by Michael De Santi, presents a new way of teaching ML models to learn this variability, improving the reliability of chlorine decay forecasts, enabling operators to set better water quality targets in humanitarian response settings.

With the emergence of AI and ML in drinking water and humanitarian fields, this paper also highlights the importance of careful design of ML models to ensure that they are fit for purpose, especially in sensitive applications like water treatment.

Find the full article here: "Training for the test: Using multi-objective training to improve ANN ensemble forecasts of household residual chlorine in emergencies"

De Santi, Michael, Ali, Syed Imran, Fesselet, J., Arnold, Matthew, Taylor, D., & Khan, Usman T. (2025). Training for the test: Using multi-objective training to improve ANN ensemble forecasts of household residual chlorine in emergencies. PLOS Water, 4(4), e0000307. https://doi.org/10.1371/journal.pwat.0000307

You may also be interested in...