Calibrating multi-constraint ensemble ecosystem models using genetic algorithms and Approximate Bayesian Computation: A case study of rewilding at the Knepp Estate, UK

This paper presents a new ensemble ecosystem model (EEM) which predicts the impacts of species reintroductions and optimises potential future management interventions at the Knepp Estate rewilding project, UK. Compared to other EEMs, Knepp has a relatively high level of data availability that can be used to constrain the model, including time-series abundance data and expert knowledge. This could improve the realism of outputs and enable more nuanced and context-specific management intervention recommendations. Calibrating EEMs can be challenging, however, and as the number of constraints increases, so does the complexity of the model fitting process. We use a new Genetic Algorithm-Approximate Bayesian Computation (GA-ABC) approach wherein GA outputs are used to inform the prior distributions for ABC. To reduce the parameter search space, we fixed twelve parameters - the consumer self-interaction strengths αi,iand negative growth rates – based on theoretical assumptions. While the GA-ABC method proved effective at efficiently searching the parameter space and optimising multiple constraints, it was computationally intensive and struggled to identify a broad range of outputs. Ultimately, this led to an ensemble of models with similar trajectories. Several potential ways to address this are discussed. Our results reinforce the findings of previous studies that the EEM methodology has potential for guiding conservation management and decision-making. Outputs suggest that reintroducing large herbivores was key to maintaining a diverse grassland-scrubland-woodland ecosystem, and optimisation experiments informed species characteristics and stocking densities needed to achieve specific goals. Ultimately, refining the EEM methodology to improve calibration and facilitate the integration of additional data will enhance its utility for ecosystem management, helping to achieve more effective and informed outcomes.