Today, issues pertaining to anthropogenic and climate forcing are threatening available water resources on a global scale. For remote alpine regions whose primary water resources are seasonally derived from snowmelt and runoff, these issues are particularly threatening. Changes in climate, and global population growth, mandate efforts for further exploration and monitoring of available water resources. This thesis explores the hydrologic significance of rock glaciers via machine-learning methodologies, in the form of a random forest image classifier, to classify rock glaciers in the 21,656 km2 study area extent of the San Juan Mountains, CO. A procedural estimation of volume was developed and run using ArcGIS(R), estimating water and ice volumes contained within all predicted rock glacier landcover in the study area. Water and ice volumes on the order of 0.94 - 1.41 km3 were estimated for a predicted 69.691 km2 rock glacier surface area based on these procedures. In response to the lack of any available validation data in the study area, all observable rock glaciers were mapped manually in ArcGIS(R), resulting in a total of 1,052 observed features. The location of all mapped rock glaciers, and the morphometric data for these locations, served as validation for the semi-automated machine-learning methods used in this thesis. Statistical similarities between manually mapped rock glaciers, with regard to two trial-runs of the random forest image classifier, produced positive statistical results indicating the efficacy of these methods. The present results suggest that machine-learning decision-tree based image classifiers, in addition to the volumetric estimation procedure developed herein, provide a potentially effective means of estimating water and ice volumes present in rock glaciers.
ETD Chair
Zhan, Hongbin Holder of Endowed Dudley J. Hughes '51 Chair in Geology and Geophysics
