New Remote Sensing Methods for Detecting and Quantifying Forest Disturbance and Regeneration in the Eastern United States

Author

Michael Joseph Hughes, University of Tennessee - KnoxvilleFollow

Date of Award

8-2014

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Ecology and Evolutionary Biology

Major Professor

Louis J. Gross

Committee Members

Daniel J. Hayes, Paul R. Armswoth, Carol Harden

Abstract

Forest disturbances, such as wildfires, the southern pine beetle, and the hemlock woolly adelgid, affect millions of hectares of forest in North America with significant implications for forest health and management. This dissertation presents new methods to quantify and monitor disturbance through time in the forests of the eastern United States using remotely sensed imagery from the Landsat family of satellites, detect clouds and cloud-shadow in imagery, generate composite images from the clear-sky regions of multiple images acquired at different times, delineate the extents of disturbance events, identify the years in which they occur, and label those events with an agent and severity. These methods operate at a 30x30 m spatial resolution and a yearly temporal resolution. Overall accuracy for cloud and cloud-shadow detection is 98.7% and is significantly better than a leading method. Overall accuracy for designating a specific space and time as disturbed, stable, or regenerating is 85%, and accuracy for labeling disturbance events with a causal agent ranges from 42% to 90%, depending on agent, with overall accuracy, excluding samples marked as `uncertain', of 81%. Due to the high spatial resolution of the imagery and resulting output, these methods are valuable for managers interested in monitoring specific forested areas. Additionally, these methods enable the discovery and quantification of forest dynamics at larger spatial scales in a way other datasets cannot. Applying these methods over the entire extent of the eastern United States highlands reveals significant differences in disturbance frequency by ecoregion, from less than 1% of forested area per year in the Central Appalachians, to over 5% in the Piedmont. Yearly variations from these means are substantial, with disturbance frequency being twice as high as the mean in some years. Additionally, these analyses reveal that some disturbance agents, such as the southern pine beetle, exhibit periodic dynamics. Finally, although these methods are applied here to the problem of forest disturbance in the eastern United States, the core innovations are easily extended to other locations or even to other applications of landscape change, such as vegetation succession, shifting coastlines, or urbanization.

Recommended Citation

Hughes, Michael Joseph, "New Remote Sensing Methods for Detecting and Quantifying Forest Disturbance and Regeneration in the Eastern United States. " PhD diss., University of Tennessee, 2014.
https://trace.tennessee.edu/utk_graddiss/2831