Use of prior distributions from aerial photographs in forest inventory

TR Number
Date
1986-06-05
Journal Title
Journal ISSN
Volume Title
Publisher
Virginia Tech
Abstract

Bayesian estimates of gross cubic- foot volume per acre were computed for four stand types (plantation pine, natural pine. hardwood. and mixed wood stands) using aerial photo volume tables as the prior information source. Aerial photographs provided a reliable source of information even though most photographs were nearly five years old.

For a given level of precision within a particular stand, Bayesian methods reduced the required field sample size up to 50% using all or half of the prior information available. Those priors which utilized a regression or a regression/topographic correction in the estimation of photo heights required less field information for the given precision level than those priors which used uncorrected or topographic corrected photo heights.

In order to obtain meaningful gains in sample size reduction corrections to the estimated photo heights should be made. Although the uncorrected prior produced generally less biased estimates. the reduction in sample size was not as large as that observed using other prior types. Greater gains were attributed to the better accuracy of the prior distribution.

Although Bayesian methods are biased, it appeared that these methods tempered severely biased prior distributions. In the hardwood stand for example, the average bias present in the photo volume data amounted to -140%. After combining the prior with the field sample, the greatest average bias was -50%.

Bayesian methods performed better than the traditional estimation methods in terms of precision. In a one to one comparison. the Bayes standard error was consistently less than its non-Bayes counterpart. The one exception to this trend was the regression prior from the hardwood stand. The poor performance of the prior was due to the weak height regression correction equation.

Modal priors utilized were not subject to the extreme input values for prior distribution development as their conservative empirical prior counterparts were. Less overall variation was observed 1n the estimated values. Under the conditions for mode selection set forth in this project, modal priors provided another good source of prior information.

Description
Keywords
Citation
Collections