Learning to learn ecosystems from limited data

Abstract

Significance In recent years, machine learning has been successfully applied to complex and nonlinear dynamical systems for improved prediction of the future state, but ecological systems represent a great challenge because of the scarcity of the observational data. This work develops a meta-learning framework with time-delayed feed-forward neural networks to predict the long-term behaviors of ecological systems by leveraging synthetic data from paradigmatic nonlinear and non-ecological dynamical systems for effective machine-learning training. The capability of accurately reconstructing the “dynamical climate” of the system with limited data is demonstrated using three benchmark population models and two real-world ecological datasets. The meta-learning framework can be generalized to other fields where forecasting the dynamics is the goal but the available empirical data is limited. executive-summary A fundamental challenge in developing data-driven approaches to ecological systems for tasks such as state estimation and prediction is the paucity of the observational or measurement data. For example, modern machine-learning techniques such as deep learning or reservoir computing typically require a large quantity of data. Leveraging synthetic data from paradigmatic nonlinear but non-ecological dynamical systems, we develop a meta-learning framework with time-delayed feedforward neural networks to predict the long-term behaviors of ecological systems as characterized by their attractors. We show that the framework is capable of accurately reconstructing the “dynamical climate” of the ecological system with limited data. Three benchmark population models in ecology, namely the Hastings-Powell model, its variant, and the Lotka-Volterra system, are used to demonstrate the performance of the meta-learning based prediction framework. In all cases, enhanced accuracy and robustness have been achieved using five to seven times less training data as compared with the corresponding machine-learning method trained solely from the ecosystem data. In addition, two real-world ecological benchmark datasets: the microbial time-series dataset and global population dynamics database, are tested to demonstrate the applicability of the meta-learning framework to the real world. A number of issues affecting the prediction performance are addressed.