Monitoreo de calidad de agua y predicción de coliformes fecales en playas de Montevideo mediante algoritmos de aprendizaje automático

Abstract

We constructed artificial intelligence (AI) models to predict faecal water quality (CF) to aid management in recreational beaches. Historical data base generated by the Laboratorio de Calidad Ambiental de la Intendencia de Montevideo (IM) was analized and AI models wwere constructed to predict CF excess (CF >2.000). Ten years of monitoring 21 recreational beaches (N=19359, november 2009 to september 2019) presented a wide range of salinity and turbidity variability among beaches. CF showed an asymetric distribution (min=4, median=250, average=1.047 and max=1.280.000) with values exceeding the threshold in all beaches. In situ registered, meteorological and oceanographic variables were used to train AI models. A stratified random forest showed the best performance in the evaluated metrics with an overall accuracy of 86% and 60% of improvement in true positive rates with respect to baseline. High quality data generated by govermental institution together with modeling strategies provided a relevant framework to aid in beach and public health management.