Tailoring a decomposition method to a large forest management scheduling problem in northern Ontario

A forest planning problem with multiple market locations and multiple products is formulated for a 3,166,000 ha forested region in northern Ontario. The resulting linear programming model is extremely large, with millions of decision variables and tens of thousands of constraints. A modelling method proposed by Hoganson and Rose (1984) was extended and applied successfully for eight model formulations. Optimal, near-feasible solutions were consistently produced in 200 to 300 iterations. Results showed definite, interpretable, patterns in the values of key dual variables that correspond to harvest level constraints in the primal LP and that tie the forest planning problem together.