Logistic regression modeling of rock glacier and glacier distribution: Topographic and climatic controls in the semi-arid Andes

Logistic regression is applied to a random sample of surface points in the semi-arid Andes of Santiago and Mendoza (Chile/Argentina, 32° 45′–34° 30′ S, 12 000 km2 above 3000 m a.s.l.) to identify local and regional controls on rock glacier and glacier distribution. The study area is among the areas with the greatest abundance of rock glaciers worldwide. The explanatory variables used for model selection are related to local and regional morphometric and climatic characteristics. The models achieve a very good fit (area under the ROC curve 0.84 for rock glaciers and 0.95 for glaciers) and are able to reproduce the regional distribution pattern. Model results are used to determine the optimal morphographic and climatic niche of rock glaciers and glaciers in the study area, and help explain the presence of low-elevation rock glaciers under favorable topoclimatic conditions.