Author

Rei Hayashi

Date of Award

12-2014

Level of Access

Open-Access Thesis

Degree Name

Master of Science (MS)

Department

Forest Resources

Advisor

Steven A. Sader

Second Committee Member

Aaron Weiskittel

Third Committee Member

Cynthia Loftin

Abstract

The first objective of this study was to evaluate the applicability of using a low density (ca. 1 point m􀀀2) discrete-return LiDAR for predicting maximum tree height, stem density, basal area, quadratic mean diameter, and stem volume using an area-based approach. The research was conducted at the Penobscot Experimental Forest in central Maine, where a range of stand structures and species composition is present and generally representative of northern Maines forests. Using a variety of high dimensional LiDAR metrics, a prediction model was developed using random forest, a nonparametric approach, based on reference data collected in fixed radius circular plots. For comparison, the volume model used two sets of reference data with one being fixed radius circular plots and the other were variable sampling plots. Prediction biases were evaluated with respect to five silvicultural treatments and softwood species composition based on the coefficient of determination (R2), root mean square error, and mean bias as well as residual scatter plots. LiDAR tended to underestimate forest inventory attributes, regardless of silvicultural treatments and species composition. However, the unmanaged units had particularly larger prediction biases, while the prediction biases also tended to be larger when softwood species composition was greatest. The maximum tree height model had the largest R2 (86.9%) followed by the volume model (72.1%), while the stem density had the smallest (R2) (28.7%). Reference data collected in the 0.08-ha fixed radius circular plots resulted in a volume prediction model with a larger R2. While it was difficult to develop models with a large (R2) owing to complexities of Maines forest structures and species composition, low density LiDAR with the area-based approach can be used as a supporting tool in forest management for this region. The second objective of this thesis was to investigate the applicability of low density (ca. 3 pulses m􀀀2) LiDAR data to deploy an individual tree-based approach. Specifically, this study focused on species classifications as well as total height and volume predictions for stem mapped trees. The research was conducted at the Penobscot Experimental Forests in central Maine, where a range of stand structures and species composition is present and generally representative of northern Maine’s forests. First, a random forest technique classified species type and softwood species based on LiDAR metrics. Second, the random forest technique was employed to calibrated individual tree height and volume prediction models. Classification errors for species were evaluated with a confusion matrix, while height and volume prediction biases were evaluated based on the coefficient of determination R2, root mean square error, and mean bias, as well as residual scatter plots with respect to three silvicultural treatments and softwood species composition. Overall, both species type and softwood species classifications had poor classification accuracy, inferring that calibration of LiDAR pulse intensity is necessary. Also, the height and volume models had small R2 values of 0.38 and 0.30, respectively. This limited accuracy of both models likely is caused by low LiDAR pulse density, which prevents an accurate representation of trees in subcanopy positions.

Share