Assessing the biological quality of freshwater bodies with machine learning technique

The deterioration of superficial water quality is a relevant issue worldwide and most European rivers do not achieve the qualitative standards required by the Water Framework Directive (WFD). Furthermore, the ecological status is defined referring to surveyed data, which is available only along main watercourses and often appears erratic in time and space. Given the goals of the WFD, a short-cut methodology to perform the assessment of water pressures on rivers starting from easily accessible data is proposed. The methodology relies on machine learning techniques and implements a procedure to: (1) identify river segment exposed to pollution spills with a raster-based numerical model; (2) introduce and estimate the spatial allocation of a Biochemical Quality Index (BQI) for each exposed river segment. The study proposes a predictive tool to assess the water quality status using a machine learning algorithm trained starting from easily available input data, such as climatic and hydrological variables, anthropic pressures, water management techniques. In this prospective, the BQI is used as a reliable proxy variable to represent the anthropogenic pressures that impacts on superficial water bodies. Results show that the BQI is well reflected in the monitoring values of COD, used as proxy variable for the quality status of watercourses. We argue that the methodology can represent a solid tool for decision-making processes and predictive studies in areas with no, or poor, monitoring data.