Incorporation of Vegetation into Mountain Permafrost Distribution Models, Southern Yukon Territory

• Main Author: Kremer, Marian
• Format: Thesis
• Language: English
• Published: University of Ottawa (Canada) 2010
• Subjects: Physical Geography; Environmental Sciences; Remote Sensing; Yukon; permafrost
• Online Access: http://hdl.handle.net/10393/28597; https://doi.org/10.20381/ruor-19348

Three groups of variables (Digital Elevation Model [DEM]-derived variables, fieldwork-derived vegetation variables, and satellite imagery-derived vegetation variables) were combined in Classification and Regression Tree (CART) models to determine the utility of vegetation-based variables for mountain permafrost distribution modelling in the southern half of the Yukon Territory. Four variables were measured in the field: canopy openness, vegetation height, organic mat thickness, and dominant species. Using Landsat TM and ETM+ imagery, three variables were calculated: a Normalized Difference Vegetation Index (NDVI), a vegetation classification, and a canopy closure classification. Individual variables were also examined to determine the one most useful for representing vegetation when modelling permafrost presence or absence. Additionally, models for each of five study areas spread across 5° of latitude were compared to examine the transferability of each variable. The addition of vegetation variables to the CART models created with DEM-derived variables resulted in only a minimal increase in the overall accuracy. Dominant species proved to be the most useful variable, but the relationship between permafrost and each species differed among study areas. Only black spruce (Picea mariana) was consistently classified as permafrost probable and lodgepole pine (Pinus contorta var. latifolia) and trembling aspen (Populus tremuloides) were classified as permafrost improbable over all study areas. These results indicate that models of permafrost distribution across large areas are not likely to be enriched sufficiently by the inclusion of vegetation variables while models covering smaller areas may benefit from the inclusion of vegetation variables. The CART models tended to show a high accuracy in the prediction of areas with no permafrost which could be useful for the purposes of infrastructure development. CART models have not previously been used in permafrost modelling and the high accuracies they produced may indicate their utility for modelling the complex relationships among the variables affecting permafrost.