Describir: Forecasting point-of-consumption chlorine residual in refugee settlements using ensembles of artificial neural networks