Open Collections

Robust optimization for forest resources decision-making under uncertainty

University of British Columbia

There is a general consensus that management decisions concerning forest resources are made in an intrinsically uncertain environment. However, decision-making tools used in forest management assume perfect information, leaving decision-makers to explore the most likely scenarios of uncertainty and determine the most reasonable management alternative. Although techniques that explicitly consider uncertainty exist, they increase the complexity of the models hence precluding their application to large-size problems. This dissertation describes the application of robust optimization concepts that explicitly consider uncertainty in forest management problems while keeping the models computationally tractable. By introducing some simplifying assumptions about uncertainty distributions, i.e. independency and uniformity, this approach allows for including uncertainty in many coefficients of the model. The methodology modifies the constraints for which feasibility is desirable and incorporates uncertainty in the technical coefficients by introducing an additional term. This term is an optimization problem in itself that introduces new constraints into the original model and acts as a buffer that guarantees constraint satisfaction for different uncertainty realizations. By changing the value of a robustness parameter, the trade-off between cost and robustness can be analyzed. The performance of this approach is explored through three structurally different problems: (a) a non-spatial harvest scheduling problem with uncertain volume yields and demands, (b) a multi-objective problem with uncertain preferences, and (c) a spatial harvest scheduling and road building problem with uncertain volume yields. Deterministic and robust formulations of these problems are provided and the performance of their solutions is evaluated under simulated scenarios of uncertain data. In all cases, robust decisions are less sensitive to uncertain data and hence protected from the occurrence of infeasibilities, with a modest reduction in the objective function value. Moreover, deterministic and robust decisions greatly differ, suggesting that traditional solutions may require major corrections to adapt to changing future conditions with a consequent decrease in the quality of the decisions. The effect of the methodology assumptions are discussed and future work is suggested.

Text

2010-04-08

Attribution-NonCommercial-NoDerivs 3.0 Unported

http://hdl.handle.net/2429/23329

Forestry

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/3.0/