A machine learning model to assess the ecosystem response to water policy measures in the Tagus River Basin (Spain).

Anthropogenic activities are seriously endangering the conservation of biodiversity worldwide, calling for urgent actions to mitigate their impact on ecosystems. We applied machine learning techniques to predict the response of freshwater ecosystems to multiple anthropogenic pressures, with the goal of informing the definition of water policy targets and management measures to recover and protect aquatic biodiversity. Random Forest and Gradient Boosted Regression Trees algorithms were used for the modelling of the biological indices of macroinvertebrates and diatoms in the Tagus river basin (Spain). Among the anthropogenic stressors considered as explanatory variables, the categories of land cover in the upstream catchment area and the nutrient concentrations showed the highest impact on biological communities. The model was then used to predict the biological response to different nutrient concentrations in river water, with the goal of exploring the effect of different regulatory thresholds on the ecosystem status. Specifically, we considered the maximum nutrient concentrations set by the Spanish legislation, as well as by the legislation of other European Union Member States. According to our model, the current nutrient thresholds in Spain ensure values of biological indices consistent with the good ecological status in only about 60% of the total number of water bodies. By applying more restrictive nutrient concentrations, the number of water bodies with biological indices in good status could increase by almost 40%. Moreover, coupling more restrictive nutrient thresholds with measures that improve the riparian habitat yields up to 85% of water bodies with biological indices in good status, thus proving to be a key approach to restore the status of the ecosystem.