Data-driven modelling for assessing trophic status in marine ecosystems using machine learning approaches.

Area of Science:

• Marine Ecology
• Environmental Science
• Artificial Intelligence in Environmental Monitoring

Background:

• Existing Trophic Status Index (TSI) models for coastal and transitional waters suffer from multicollinearity, data redundancy, and complex classification.
• Accurate assessment of eutrophication is crucial for managing coastal and transitional water quality.

Purpose of the Study:

• To develop a novel, data-driven tool, the Assessment Trophic Status Index (ATSI) model, for enhanced trophic status assessment in transitional and coastal (TrC) waters.
• To address limitations of existing TSI models by incorporating machine learning (ML) and artificial intelligence (AI).

Main Methods:

• Employed ML techniques, including deep learning, to optimize input data and minimize redundancy for the ATSI model.
• Developed a CHL prediction model using ten algorithms, with XGBoost showing superior performance (RMSE=0.0, MSE=0.0, MAE=0.01).
• Calculated ATSI scores using a novel linear rescaling interpolation function and evaluated model efficiency with R², NSE, and MEF metrics.

Main Results:

• The XGBoost model demonstrated exceptional performance in CHL prediction, indicating high accuracy.
• The ATSI model exhibited heightened sensitivity and efficiency across diverse application domains.
• Significant disparities were observed between the ATSI model and the ATSEBI System in assessing four Irish waterbodies, highlighting ATSI's distinct outcomes.

Conclusions:

• The ATSI model, leveraging ML/AI and a new classification scheme, significantly enhances the accuracy of trophic status assessments in marine ecosystems.
• The ATSI model provides a promising approach for evaluating and monitoring trophic conditions in TrC waters and other waterbodies.
• This research contributes substantially to marine ecosystem management and conservation through improved water quality assessment tools.