Designing an optimal large-scale reintroduction plan for a critically endangered species

Optimization methods are routinely used for landscape-level conservation planning, but still underused in supporting species recovery programs. A possible barrier is the difficulty in representing and optimizing complex multidimensional problems: for example, many species recovery programs require management at the population level, but also allocation of effort and resources across populations and over time. Optimization methods can help, but they must strike a balance: too much realism can be computationally unfeasible, but too much simplification can limit relevance for complex programs, exactly where decision support might be most needed. We show how integer linear programming can be used to solve such a complex problem, combining multiple site-level demographic models with realistic management constraints under different sources of stochasticity and uncertainty. We apply this protocol to reintroduction planning for the critically endangered Montseny brook newt Calotriton arnoldi, optimizing site restoration efforts, captive releases from limited and variable stocks, and short- and long-term monitoring, all across 17 sites over 10 years. For C. arnoldi, the optimal solution was generally to open as many sites as possible, as soon as allowed by budget, and to reinforce sites with additional releases. The number of new populations that could be established was limited not only by the high initial costs of restoring and preparing sites for releases, but also because opening new sites would require subsequent monitoring, eventually adding up to unsustainable costs. Synthesis and applications. Our results suggest releases of Calotriton arnoldi should be dictated first by habitat restoration capacity, then by long-term sustainability. More generally, our study shows how quantitative decision-support methods can improve the value of science for conservation, and help managers find solutions to complex problems. However, deploying those methods requires close collaboration between managers and scientists, to ensure models are realistic, results are relevant, and the whole process is informative.