Using neural ordinary differential equations to predict complex ecological dynamics from population density data.

Area of Science:

• Ecology
• Computational Biology
• Machine Learning

Background:

• Ecological systems are complex, often leading to bias and limited predictive power in simple models.
• Neural Ordinary Differential Equations (NODEs) preserve data dynamics, offering a promising approach for time-series analysis.

Purpose of the Study:

• To evaluate the performance of NODEs as a forecasting tool for ecological communities.
• To compare NODEs against traditional and other machine learning forecasting methods.

Main Methods:

• Simulated time series data of competing species in a time-varying environment were used.
• Performance was assessed using point-wise accuracy and interval scores.

Main Results:

• NODEs provided more precise forecasts than Autoregressive Integrated Moving Average (ARIMA) models.
• Untuned NODEs showed similar accuracy to untuned Long Short-Term Memory (LSTM) networks.
• NODEs generally outperformed other methods in interval score evaluation, indicating superior prediction interval accuracy and precision.

Conclusions:

• NODEs demonstrate potential as a powerful forecasting tool for ecological communities.
• NODEs can provide valuable insights into population dynamics, broadening ecological time-series analysis approaches.