Hypoxia extreme events in a changing climate: Machine learning methods and deterministic simulations for future scenarios development in the Venice Lagoon

Abstract

Climate change pressures include the dissolved oxygen decline that in lagoon ecosystems can lead to hypoxia, i.e. low dissolved oxygen concentrations, which have consequences to ecosystem functioning including biogeochemical cycling from mild to severe disruption. The study investigates the potential of machine learning (ML) and deterministic models to predict future hypoxia events. Employing ML models, e.g. Random Forest and AdaBoost, past hypoxia events (2008–2019) in the Venice Lagoon were classified with an F1 score of around 0.83, based on water quality, meteorological, and spatio-temporal factors. Future scenarios (2050, 2100) were estimated by integrating hydrodynamic-biogeochemical and climate projections. Results suggest hypoxia events will increase from 3.5 % to 8.8 % by 2100, particularly in landward lagoon areas. Summer prediction foresee a rise from 118 events to 265 by 2100, with a longer hypoxia-prone season. This model is a valuable tool for developing hypoxia scenarios, aiding in identifying restoration hotspots for climate-threatened lagoons.