2021-04-23, 11:15–11:45, Track 1
The goal of the study is to predict dissolved oxygen concentration in a lagoon with XGBoost algorithm, based on a series of explanatory variables (e.g., water temperature, pH value, oxidation-reduction potential, air temperature, salinity). Special focus is given on interpreting the outcomes using Additive exPlanations (SHAP) methodology, aiming to elucidate the environmental windows that cause low levels of dissolved oxygen (anoxic conditions), which may have severe impact on the survival rate of aquatic organisms.
Dissolved oxygen is a key indicator in aquaculture reflecting the change in water quality. In this study, we used the Extreme Gradient Boosting (XGBoost) machine learning algorithm to predict dissolved oxygen in a lagoon (western Greece), considering ten physicochemical and meteorological explanatory variables, that is water temperature, pH value, oxidation-reduction potential, air temperature, salinity, chlorophyll, relative humidity, atmospheric pressure, wind speed and wind direction. XGBoost was trained and then evaluated attaining a root mean square error of 0.66 mg/L in predictions. Results showed that pH, salinity and air temperature were the key factors driving dissolved oxygen variability. Special focus was given on interpreting the outcomes using Additive exPlanations (SHAP) methodology, aiming to elucidate the environmental windows that cause low levels of dissolved oxygen (anoxic conditions), which may have severe impact on the survival rate of aquatic organisms.