Optimizing chlorination for water safety and acceptability in emergency water supplies in humanitarian crises using a deep composite neural network

Unprecedented global population displacement in recent years has increased the burden of waterborne illnesses in refugee and internally displaced person (IDP) settlements. Preventing outbreaks of waterborne diseases can be particularly challenging in urban-scale refugee and IDP settlements since recontamination commonly occurs post-distribution period. In this period users manually collect water from public tapstands, transport it to their dwellings where they store and use it over several hours. Unlike contexts where water is piped directly to the home, in urban-scale refugee and IDP settlements effective chlorination in these settlements requires that free residual chlorine (FRC) at tapstands be sufficient to ensure at least 0.2 mg/L of FRC throughout the period of storage and use, while remaining palatable to consumers. Thus, chlorination practice must account for both site-specific dynamics of chlorine decay as well as local attitudes towards chlorinated water taste and odor (T&O). In response to this need, we developed the Safe Water Optimization Tool (SWOT), a “digital water” tool that uses machine learning provide generate evidence-based chlorination decision support that balance over- and under-chlorination risks.We used data collected from the Kyaka II refugee settlement in Uganda to calibrate a tapstand FRC target using the SWOT that maximizes household water safety while minimizing T&O rejection. We evaluated the water safety risk using a deep composite quantile regression neural network (DCQRNN), an artificial intelligence model that predicts the full probabilistic distribution of point-of-consumption FRC concentration using routine monitoring water quality data. We used ordinary least-squares regression (OLS) to predict the percent of the population rejecting chlorinated water as a function of tapstand FRC using forced choice triangle test and flavour rating assessment test focus group data. The final FRC target was selected to balance both risks of unsafe water and T&O rejection.By integrating the predicted risk from both the DCQRNN and OLS models, we determined that an FRC target of 0.7 mg/L in Kyaka II produces the most balanced tradeoff of both risks (38% probability of rejection, 36% probability of unsafe drinking water). The lowest combined probability for both risks was achieved at a tapstand FRC of 1.4 mg/L which would produce only 7% risk of unsafe drinking water but 46% risk of rejection. This integrated risk-based approach allows water system operators to select a target based on their preferred tradeoff of these risks, in consideration of site conditions, especially the safety of alternative sources.This study presents an important digital water solution to ensure safety of water supplies in humanitarian contexts, using the SWOT’s advanced artificial intelligence modelling and analytics to address uncertainty in FRC decay as well as using a data driven approach to quantifying T&O behaviour. This approach yields chlorination guidance that balances risks of both under- and over-chlorination, maximizing access to safe water and improving public health protection. The approach taken in this study can be applied in a range of contexts where water users lack continuous water supply, including in large urban intermittent water supply systems.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.