Year of Award

2011

Document Type

Thesis - Campus Access Only

Degree Type

Master of Science (MS)

Degree Name

Forestry

Department or School/College

College of Forestry and Conservation

Committee Chair

John Goodburn

Commitee Members

Anna Klene, Carl Seielstad

Keywords

lidar, linear modeling, modeling error, remote sensing, scalability, statistical modeling

Abstract

Prior research has proven the utility of using lidar and field data in a two-stage procedure to predict forest inventory parameters. However, the effects of varying plot size on the prediction errors are not well understood. We investigated the effects of plot size on prediction errors using lidar data for a western Montana forest using five sizes derived from stem-mapped field data and multiple regression modeling techniques. Models were fitted for maximum and mean heights, stand basal area, stem density, and quadratic mean diameter. A validation routine was performed using an independent dataset and models derived from different plot sizes were assessed using goodness of fit, validation root mean squared error (RMSE), mean error, mean absolute error, and the modeling efficiency statistic. Although trends in model quality varied by inventory parameter, there seemed to be some advantages in using plots greater than 300 m^2 in size, as these plots tended to produce models with higher goodness of fit, better modeling efficiency, and lower mean absolute error values. Effects of canopy cover were also examined and showed little effect of varying plot size on low cover plots but potential benefit in using larger plots in high cover areas. The results related to stand structure were contrary to those reported in other studies, but were not surprising based on the complex, multi-story conditions on high cover stands found in the western Montana study area. Finally, the �scalability� of models was explored by fitting a model across the range of training plot sizes and validating it using the middle (300 m^2) size in this study. Larger plots tended to be adaptable as they performed well in predicting forest inventory parameters at different scales.

This record is only available
to users affiliated with
the University of Montana.

Request Access

Share

COinS
 

© Copyright 2011 Jody Paul Bramel