Date of Award


Level of Access Assigned by Author

Open-Access Thesis

Degree Name

Master of Science (MS)


Forest Resources


Steven A. Sader

Second Committee Member

Cynthia Loftin

Third Committee Member

Jeremy Wilson


Extensive harvesting practices coupled with major ownership change have led to increasing fragmentation of Maine’s forest, a reduction from larger, contiguous mature forest patches into smaller patches. Using Landsat Thematic Mapper (TM) - based forest cover and change maps (1991-2007), fragmentation metrics, and Principal Components Analysis (PCA), this study determined the extent and configuration of forest fragmentation within three ecoregions and 186 level 5 watersheds throughout the state of Maine. Forests in the Northeastern ecoregion had higher harvest rates and more interspersed patches of undisturbed forest. Forests in the South-Central ecoregion are composed of more, smaller patches than their Northeastern and Western counterparts but had the highest proportion of undisturbed forest at the end of the study period. The cover type PCA indicated that softwood has been the most harvested cover type; mixedwood and hardwood were more prevalent in the residual forest stands. Softwood forests showed a marked decrease in size and area in unharvested forests of the Northeastern and Western ecoregions. The Western ecoregion consisted of small patches of hardwood forest that were closer together, and hardwood forest represented a greater proportion of the landscape. Softwood forest patch shapes were more complex in the South-Central ecoregion. This research provides a numerical assessment of the spatially explicit effects of the 1991-2007 harvesting legacy on the landscape (watershed level) composition of Maine.

With Maine’s northern forest being fragmented and patch size decreasing over time, maintaining a distribution of larger trees may be ecologically valuable. There are no spatially explicit maps for Maine showing the distribution of old growth or large diameter forest and ground data is lacking. Therefore, methods using multiple sources of remotely sensed data, topographic and site index data were combined in a modeling application to predict visible crown diameter (VCD) as a proxy for tree size in recent undisturbed forest (RUF), stands that were not harvested between 1972 and 2007. Change detection maps derived from Landsat TM imagery, raw Landsat TM imagery, two sources of aerial photography, and ancillary data were used as input into a random forests model. Results indicated differences in VCD ranges and importance of predictive variables between softwood, mixedwood, and hardwood forest cover. Recent undisturbed softwood and mixedwood VCD decreased with increasing site elevation and slope. Softwood VCD increased with increasing spectral values of Landsat TM 1, the Normalized Difference Moisture Index and Tasseled Cap wetness, suggesting sensitivity to moisture or shadowing in the canopy. Recent undisturbed hardwood forests were found on the best sites at low elevations. Hardwood VCD responded to spectral variables, especially Tasseled Cap brightness, and the Landsat TM reflected infrared wavebands 4, 5, and 7. This research is repeatable in other regions, provided there is access to historical aerial photography and reliable map information or ground data that could verify the presence of undisturbed forest at earlier dates.