Modeling freshwater plankton community dynamics with static and dynamic interactions using graph convolution embedded...

Area of Science:

• Environmental Science
• Computational Biology
• Ecology

Background:

• Freshwater plankton dynamics are linked to water quality, prompting the development of predictive models.
• Existing models often overlook the crucial role of biotic and abiotic interactions in plankton communities.

Purpose of the Study:

• To investigate the importance of interaction terms in predicting plankton community dynamics.
• To apply graph convolution embedded long short-term memory networks (GC-LSTM) for enhanced plankton forecasting.

Main Methods:

• Developed GC-LSTM models using temporal graph series of plankton genera and environmental drivers from reservoir and river ecosystems.
• Compared GC-LSTM performance against LSTM and GCN models at various lead times.
• Utilized GNNExplainer to interpret the importance of nodes and edges in model predictions.

Main Results:

• GC-LSTM models significantly outperformed traditional LSTM models in predicting plankton community dynamics, showing higher accuracy.
• Performance degradation was observed in all models at longer lead times, but GC-LSTM maintained superior performance.
• GNNExplainer provided interpretable insights into key plankton genera and interactions influencing community dynamics.

Conclusions:

• The proposed GC-LSTM approach effectively forecasts plankton community dynamics by integrating interaction terms.
• Graph signals representing interactions are critical for accurate plankton community predictions.
• This methodology enhances our understanding of freshwater ecosystem dynamics and supports water quality management strategies.