Brown C FEMC2007

To Optimize or Not to Optimize - A Practitioner’s Perspective

Cam Brown, Forsite Consultants, P.O. Box 2079, #330-42nd Street, Salmon Arm, BC, V1E 4R1, Canada, cbrown@forsite.ca

Forest estate models typically employ one of three solution generation techniques: stepwise simulation, optimization (e.g. linear programming), or heuristics (e.g. simulated annealing). Each of these techniques presents advantages and disadvantages to modeling practitioners. Based on personal experience and discussions to date, this talk provides a BC practitioner’s perspective on the strengths and weakness offered by each approach in the delivery of forest estate modeling services.

Definitions from a harvest scheduling perspective:

Stepwise Simulation (what if?): A solution procedure that find a single solution to a harvest scheduling problem using predefined harvest rules, priorities, management actions, and constraints. The user constructs a scenario and the model is simply a “convenient calculator for a complex/massive problem”.

Optimization (what’s best?): A solution procedure that finds an optimal solution to a harvest scheduling problem given a objective function that measures solution value, a set of management options to chose from, and a set of constraints that ensures an acceptable solution. The user must constructs bounds around solution choices while providing a measure of how good the solution is, and then it is up to the model to explore the solution space and find the optimal answer.

Heuristics (what’s good?): A solution procedure that searches for an optimal solution but can only guarantee a good solution given an objective function that measures solution value, a set of management options to chose from, and a set of constraints or targets that ensures an acceptable solution. This technique is applied in situations where optimization is impossible (problem too large) or impractical (solution times too long). Currently, harvest scheduling problems involving stand level spatial relationships (i.e. adjacency/green-up or patch size) on realistic land base areas cannot be optimized and require heuristics.

If there is a desire to link strategic results to tactical plans, then heuristics allow the most detail to be integrated into the solution procedure and therefore they are likely to provide the most realistic solutions. If you need it all and your willing to wait for it, heuristic approaches are capable of providing it. If you are less interested in stand level spatial relationships but still want treatment options, soft targets, and other features of optimization, then optimization is ideal. When simple, fast, and transparent solutions are necessary for budgetary reasons or to process large land bases in a reasonable amount of time, simulation is likely the best bet.
It is a specific model’s ability to address the problem at hand that should drive choice. Solution technique is only one issue for consideration - the flexibility of the model to represent a problem and its ability to have the solution influenced by key indicators is likely more important than how the solution is generated.