A Machine Learning-based approach for the assessment of water quality variation in the Venice Lagoon

Abstract

Water quality (WQ) is one of the most critical issues in lakes, estuaries, marine and costal water management, affecting not only the socio-economic systems, but also the sustainability of natural processes. As a consequence of the complex interplay between climate and human-induced pressures, changes in marine WQ are observed (e.g. higher turbidity with resulting reduced water clarity, acidification) with cascading effects on the environmental status of natural ecosystems and their capacity to flow services for human wellbeing. To evaluate such effects, a continuous monitoring of WQ parameters is advisable. With advances in space science and the increasing use of computer applications, near real time remotely sensed observations in combination with novel machine learning (ML) methods have become useful tools for monitoring and management of WQ, overcoming limitations posed by traditional in-situ measurements. In the frame of this thesis, a ML-based approach integrating monitoring and satellite data was developed, in order to allow for a multi-scenario analysis and modelling of spatio-temporal dynamics of key WQ parameters (e.g. Chl, turbidity) in the Venice lagoon, also in consideration of changing climate and land-use conditions (e.g. increase of precipitation and temperature, urban sprawl). Resulting output of the application of the methodology at the case study level represents a valuable support for the implementation of relevant EU acquis (e.g. Water Framework and Marine Strategy Framework Directives) in the investigated area, supporting decision makers and managers in the identification of key drivers of deterioration of natural ecosystems and the design of appropriate evidence-based management measures.