Rock glacier catalogue and predictive modeling in the Mackenzie Mountains: predicting rock glacier likelihood with a generalized additive model

No Thumbnail Available
Thiessen, Rabecca R.
University of Lethbridge. Faculty of Arts and Science
Journal Title
Journal ISSN
Volume Title
Lethbridge, Alta. : University of Lethbridge, Dept. of Geography and Environment
Rock glaciers are important features of periglacial landscapes and have potentially significant hydrological, ecological, and geological impacts on alpine environments. The purpose of this study is to catalog rock glaciers in three regions of the Mackenzie Mountains, northwest Canada, and to develop a predictive model for rock glacier probability mapping. Identification of rock glaciers follows guidelines set by the International Permafrost Association (IPA) Rock Glacier Action Group, incorporating geomorphological approaches for identification. Of the 530 rock glaciers mapped within three regions of the Mackenzie Mountains, ̴90% were classified as active, with primarily northern orientations. The model utilizes variables such as potential incoming solar radiation (PISR), elevation, slope, aspect, lithology, and topographic position index (TPI) to understand rock glacier distribution. Modeling methods include a Generalized Additive Model (GAM), Random Forest (RF) and Forest-based Classification and Regression (FBCR). The comparison of models revealed that the GAM was the best model, with selected variables. The performance of the GAM was assessed using training (70%) and testing (30%) datasets. The confusion matrix for the training data indicated a total of 321 true negatives (0) and 322 true positives (1), with 50 false negatives and 49 false positives. The training accuracy was calculated to be 0.87, reflecting the proportion of correctly classified instances. An additional test site revealed that the GAM can be generalized to new regions within the Mackenzie Mountains, limiting the time needed for manual identification of rock glaciers in large regions. This research contributes to the existing knowledge on rock glacier formation and persistence and emphasizes the importance of comprehensive inventories for future research and monitoring.
rock glacier , permafrost , Mackenzie Mountains , generalized additive model