Abstract:
This thesis describes a method for designing optimal timber hauling systems
under conditions of uncertainty and risk. This method distinguishes between, and
incorporates, natural uncertainty and parameter uncertainty. Natural uncertainty
encompasses the variation of the trucking activity times (loading and unloading times,
loaded and empty travelling times), and is handled by using probability distributions.
Parameter uncertainty is the inability to accurately estimate the parameters of these
probability distributions. In this study, parameter uncertainty is addressed from a
Bayesian perspective by incorporating, together with some sample data, subjective
estimates from experts. In addition, the decision on the optimum combination of
trucks, loaders and unloaders is based on the decision maker's attitude toward risk.
Design scenarios are analysed by formulating the sequence of hauling activities (i.e.,
loading, travelling loaded, unloading and travelling empty) as a queueing system, and
the equipment combinations are then evaluated by simulation. The inputs to the
simulation model include actions (representing the number of trucks, loaders and
unloaders), probability distributions (of loading time, unloading time, travel loaded
speed and travel empty speed), estimated parameters of these probability distributions,
and constants (such as amount of wood to be hauled, hauling distances, etc). Using a
cost equation, the simulation outputs are converted into cost and payoff tables.
Employing utility functions, which quantify the decision maker's attitude to risk, the
optimum combination of trucks, loaders and unloaders is determined based on
expected payoff and expected utility criteria.
Using basic data supplied by Forest Engineering Research Institute of Canada
(FERIC), a two-parameter gamma distribution is selected to represent the input
variables in the simulation program. Calculations to estimate the parameters of the
gamma distribution using Bayesian methodologies are contrasted with the classical
statistical methods. Other possible applications to illustrate the applicability of
Bayesian decision analysis in forestry are also presented.
The thesis concludes that Bayesian statistics and decision analysis which
combines expert prior knowledge with available sample data provides a better
methodology for designing timber hauling systems than that provided by classical
decision analysis methodologies, especially where woodlands decisions are made with
limited sample data.
