Empirical dynamic programming for model-free ecosystem-based management

Quantitative ecosystem-based management typically relies on hypothetical ecosystem models that are difficult to validate for all but the best-studied systems. Here, we develop a management scheme that is based on predictive models driven by the observed dynamics. We show that near-optimal management policies can be constructed from time-series data by merging empirical dynamic modelling and stochastic dynamic programming. The Empirical Dynamic Programming approach performs well in cases we examined and outperformed a commonly used single-species alternative. We expect model-free ecosystem-based management to be of use wherever ecosystem dynamics are uncertain or observations of the system do not cover all relevant species.