Landscape-level optimization using tabu search and stand density-related forest management prescriptions

Abstract

Spatial and temporal scheduling of forest management activities is becoming increasingly important due to recent developments in environmental regulations, goals and policies. A forest planning model was developed to select activities for stands in a forested area (178,000 ha), from a set of stand-centric optimal prescriptions, to best meet a higher-level landscape objective. The forest-level management problem addressed is complex, if not impossible to solve optimally, with current computing technologies, as integer decision variables are assumed. The higher-level landscape objective is to achieve the highest even-flow of timber harvest volume. Three types of tabu search processes were examined in the model: (1) a process with 1-opt moves only; (2) a process with 1-opt moves and a region-limited 2-opt move process; and (3) a process with 1-opt moves and 10 iterations through a smaller region-limited 2-opt move process. Aspiration criteria and short-term memory were employed within tabu search in an attempt to avoid becoming trapped in local optima. While the 1-opt move process alone created solutions (forest plans) that were adequate, and contained higher average harvest volumes than the other methods, the addition of the 2-opt move processes improved the solutions generated by intensifying the search around local optima. The solutions produced using the 2-opt move processes had less variation in periodic harvest volumes across the planning horizon. While the basic 1-opt tabu search process provides adequate feasible solutions to large, complex forest planning problems, we reinforce the notion suggested, but not proven with previous research, that 2-opt neighborhoods can help improve quality of large scale forest plans generated by tabu search. The contribution of this research is the description of a process which be developed for large forest planning problems (the problem examined is at least 1 order of magnitude greater than previous research, in terms of forested stands modeled), a process that could enable one to produce more efficient forest planning solutions than one could otherwise with standard 1-opt tabu search. In addition, we describe here the use of a set of optimal stand-level prescriptions to choose from when utilizing the 2-opt process, rather than a set of clearcut periods to consider. With this in mind, this research represents a fundamentally new application of operations research techniques to realistic forest planning problems.