Bridging causality and deep learning for harmful algal bloom prediction.

Area of Science:

• Marine science
• Environmental monitoring
• Machine learning

Background:

• Accurate chlorophyll-a (Chl-a) estimation is crucial for monitoring harmful algal blooms (HABs).
• Existing machine learning (ML) models often lack causal interpretability and robustness.

Purpose of the Study:

• To develop an enhanced causal machine learning framework for HAB prediction.
• To integrate causal discovery, treatment effect estimation, and deep learning.

Main Methods:

• Developed a Causally Informed Neural Network (CINN) incorporating causal graphs and average treatment effects.
• Utilized 31 environmental predictors from MODIS, ERA5, and HYCOM in the Persian Gulf.
• Applied monotonic causal constraints for ecological alignment.

Main Results:

• CINN and MCINN outperformed baseline models (Random Forests, XGBoost, SVM) with R² up to 0.926 (10-17% improvement).
• Reduced Root Mean Square Error (RMSE) by up to 25%.
• Causal validity of drivers like sea surface temperature and nutrient fluxes confirmed.

Conclusions:

• The CINN framework offers an interpretable, data-efficient, and uncertainty-aware solution for HAB prediction.
• Demonstrated potential for operational early-warning systems and policy interventions.
• Addresses challenges in data-scarce, climate-sensitive marine environments.