Deep learning decodes multi-horizon dynamics and probabilistic risks of harmful algal blooms

Area of Science:

• Environmental Science
• Data Science
• Ecology

Background:

• Harmful algal blooms (HABs) are increasing globally in freshwater systems due to climate change and eutrophication.
• Current early warning systems lack hourly resolution, uncertainty quantification, and mechanistic interpretability, hindering effective intervention.
• Existing models struggle with high-frequency data and dynamic ecological shifts, masking potential risks.

Purpose of the Study:

• To develop an interpretable deep neural network (BloomNet) for accurate, hourly, and uncertainty-aware forecasting of HABs.
• To improve predictive frameworks for proactive water resource management.
• To decode the dynamic ecological drivers influencing HABs at various temporal scales.

Main Methods:

• Developed BloomNet, an interpretable deep neural network for multi-horizon quantile forecasting.
• Integrated future environmental covariates with historical hourly data from a hyper-eutrophic lake.
• Evaluated model performance against LSTM, Transformer, and TCN baselines over a four-year period.

Main Results:

• BloomNet demonstrated exceptional predictive stability across 24-, 48-, and 72-hour horizons (R² up to 0.78).
• The model outperformed baseline deep learning methods, avoiding iterative error accumulation.
• Identified shifting ecological drivers, with water temperature dominant in the short-term and total phosphorus in the medium-term.

Conclusions:

• BloomNet provides a robust, interpretable, and uncertainty-aware early warning system for HABs.
• The model's insights into dynamic driver importance enable precision ecotechnology for water resource management.
• Probabilistic forecasts facilitate a graduated, risk-oriented warning protocol, transforming reactive management into proactive intervention.