Harnessing AI to Decode and Combat Harmful Algal Blooms

A cutting-edge AI model reveals the hidden factors behind harmful algal blooms and provides a roadmap for mitigating their growing impact on ecosystems and public health.

Research: Explainable deep learning identifies patterns and drivers of freshwater harmful algal blooms. Image Credit: Editos foto / Shutterstock

In a significant breakthrough, researchers have developed an advanced, explainable deep learning model to predict and analyze harmful algal blooms (HABs) in freshwater lakes and reservoirs across China. As HABs increasingly threaten water ecosystems and public health, this study offers crucial insights into their underlying drivers and potential mitigation strategies. The research is published in the journal Environmental Science and Ecotechnology.

Harmful algal blooms are complex phenomena influenced by multiple ecological and climatic factors. Traditional models often struggle to predict these blooms or provide interpretable insights accurately. To overcome these challenges, the research team implemented a Long Short-Term Memory (LSTM) neural network enhanced by explainability techniques. The model was trained on data from 102 monitoring sites across China, achieving an average Nash-Sutcliffe efficiency coefficient of 0.48, a significant improvement over conventional machine learning methods.

In addition to water temperature, other factors such as chlorophyll-a, turbidity, and permanganate (CODMn) were also identified as significant drivers, highlighting the multifaceted nature of HAB dynamics. Water temperature emerged as the most influential factor, accounting for an average of 11.7% of the predictive variance. Notably, regions in mid to low latitudes displayed heightened sensitivity to temperature changes, emphasizing the potential impact of climate change on HAB occurrences. The study also revealed that subtropical regions exhibit the strongest sensitivity to water temperature compared to tropical and temperate zones.

"Our explainable deep learning model not only enhances prediction accuracy but also helps policymakers understand the key factors behind harmful algal blooms," said lead author Shengyue Chen. "This approach can inform targeted management strategies for lakes and reservoirs at high risk."

Additionally, the study demonstrated that transfer learning could effectively improve predictions in data-scarce regions by using information from well-monitored areas. However, the researchers identified potential challenges with “negative transfer,” where differences in environmental drivers between source and target regions could reduce model effectiveness. Addressing these challenges requires a careful selection of similar source domains.

This pioneering research highlights the power of combining artificial intelligence with explainability to tackle complex environmental challenges. The study also underscores the urgent need to expand algal monitoring networks, particularly in poorly gauged water bodies, to support broader and more effective HAB management strategies.