Managing In teracting Species: A Reinforcement Learning Decision Theoretic Approach

Persistence of threatened species relies heavily on the effectiveness of conservation decisions. Yet, conservation strategies may generate positive and/or negative impacts on non-target species through direct (e.g. competition, predation), or indirect (e.g. habitat use) species interactions. Accounting for such interactions rarely occurs in conservation planning due to high biological uncertainty as well as the computational challenge of solving problems of this magnitude. Consequently, the simultaneous implementation of single-species management strategies for species that interact may jeopardize the recovery of one or more of the threatened species. Here we address these obstacles using a simulator and reinforcement learning approach. Reinforcement learning simplifies the representation of complex stochastic processes, and provides an intelligent way of exploring the solution search space. We apply this approach to two threatened species and compare optimal management strategies for ensuring species recovery and coexistence through their ranges.