Comparison of population-based algorithms for optimizing thinnings and rotation using a process-based growth model

Abstract

Stand management optimization has long been computationally demanding as increasingly detailed growth and yield models have been developed. Process-based growth models are useful tools for predicting forest dynamics. However, the difficulty of classic optimization algorithms limited its applications in forest planning. This study assessed alternative approaches to optimizing thinning regimes and rotation length using a process-based growth model. We considered (1) population-based algorithms proposed for stand management optimization, including differential evolution (DE), particle swarm optimization (PSO), evolution strategy (ES), and (2) derivative-free search algorithms, including the Nelder–Mead method (NM) and Osyczka’s direct and random search algorithm (DRS). We incorporated population-based algorithms into the simulation-optimization system OptiFor in which the process-based model PipeQual was the simulator. The results showed that DE was the most reliable algorithm among those tested. Meanwhile, DRS was also an effective algorithm for sparse stands with fewer decision variables. PSO resulted in some higher objective function values, however, the computational time of PSO was the longest. In general, of the population-based algorithms, DE is superior to the competing ones. The effectiveness of DE for stand management optimization is promising and manifested.