A comparison of numerical methods for optimizing even aged stand management

Abstract

A two-species, whole-stand, deterministic growth model was combined with three optimization methods to derive management regimes for species composition, thinnings, and rotation age, with the objective of maximizing soil expectation value. The methods compared were discrete time – discrete state dynamic programming, direct search using the Hooke and Jeeves algorithm, and random search. Optimum solutions for each of the methods varied considerably, required unequal amounts of computational time, and were not equally stable. Dynamic programming located global optimal solutions but did not determine them accurately, owing to discretized state space. Direct search yielded the largest objective function values with comparable computational effort, although the likelihood of finding a global optimum solution was high only for smaller problems with up to two or three thinnings during the rotation. Random search solutions varied considerably with regard to growing stock level and species composition and did not define a consistent management guideline. In general, direct search and dynamic programming appeared to be superior to random search.