Heuristic solution techniques for a spatial harvest scheduling problem involving wildlife habitat and timber income

Graduation date: 2000Three heuristic techniques: simulated annealing (SA), tabu search (TS), and\ud tabu search with strategic oscillation (TSSO), were used to schedule silvicultural\ud activities designed to accelerate development of older forest structure at both stand\ud and landscape scales over a 2450 acre forest located in northwestern Oregon.\ud Goals for the forest over a 100-year planning horizon included reaching at least 500\ud acres of older forest structure with at least one contiguous 200-acre (or larger)\ud block as soon as possible. The configuration and location, but not the amounts, of\ud the older forest structure acres and the contiguous block were then free to move\ud about the forest through time while best meeting the goal of producing a high,\ud steady revenue flow over the entire planning horizon subject to restrictions on\ud maximum clearcut patch size.\ud The heuristic techniques were able to provide feasible tactical schedules\ud fulfilling the strategic goals over the entire horizon in ways which traditional forest\ud planning tools cannot. Of the three techniques examined, TSSO produced\ud schedules with the best, most consistent objective function values. SA yielded a\ud wider range of values which were always slightly worse but required only a\ud fraction of the computing time. Straightforward TS produced relatively poor\ud objective function values, most likely because of its inability to search the\ud infeasible regions of the diverse solution space. Estimation of the globally optimal\ud objective function value using Weibull distributions suggested that all TSSO\ud solutions were within 1.8% of the optimum, the best being within .03%, while all\ud SA solutions were within 7.6%, the best being within 1 .7%. However, 95%\ud confidence intervals of the Weibull location parameter estimates for the SA and\ud TSSO distributions did not overlap, despite the fact that both distributions of results\ud failed to be rejected as fitting a Weibull distribution. This disparity again suggests\ud that statistical inference by itself of global optima for heuristic results may be an\ud inadequate means of assessing how "good" a heuristic is