Using Machine-Learning Algorithms for Eutrophication Modeling: Case Study of Mar Menor Lagoon (Spain).

Area of Science:

• Environmental Science
• Eutrophication Studies
• Machine Learning Applications

Background:

• The Mar Menor, a hypersaline lagoon in Spain, faces severe eutrophication crises impacting its ecological value.
• Anthropogenic pressures have led to abrupt water quality degradation, causing significant public concern.

Purpose of the Study:

• To investigate machine learning (ML) methods for predicting chlorophyll-a (Chl-a) levels.
• To enhance the management of the Mar Menor's complex hydro-ecosystem through accurate Chl-a modeling.

Main Methods:

• Evaluation of Multilayer Neural Networks (MLNNs) and Support Vector Regressions (SVRs) for Chl-a prediction.
• Utilized water quality parameters as input data, with feature selection to identify optimal combinations.
• Employed wrapper feature selection methods to simplify models and improve efficiency.

Main Results:

• Both MLNNs and SVRs demonstrated satisfactory performance in predicting Chl-a concentrations.
• Support Vector Regression (SVR) models exhibited superior performance during the validation phase compared to MLNNs.
• The best-fit models achieved a cross-validated coefficient of determination (R²_CV) of up to 0.7.

Conclusions:

• Machine learning algorithms, particularly SVR, are effective tools for predicting Chl-a in eutrophic coastal lagoons.
• Feature selection enhances the accuracy and efficiency of ML models for water quality management.
• These findings support the use of ML for proactive environmental management of sensitive ecosystems like the Mar Menor.