Transferable machine learning for rock glacier mapping
This Master's project was designed for Brage Norman Auestad Nilsen who started the Master's program in Earth sciences, UiB, fall 2024. The Master's project is given by the research group Quaternary geology & paleo climate.
Main content
Project description
In recent years, climate change has led to rising global temperatures, causing most glaciers to retreat. This poses challenges for areas with low precipitation, where glacifluvial meltwater is a crucial source of drinking water. Rock glaciers, another important part of the cryosphere,
respond more slowly to climate change and may provide an alternative source of water in such regions. Rock glaciers are a mixture of ice and debris, typically with a distinctive morphology. This includes a lobate shape, often resembling a small glacier, and they are commonly found in high-latitude and mountainous terrains. Rock glaciers generally feature ridges, furrows, and a steep front at the angle of repose. Rock glaciers are permafrost landforms and are the amongst the only indicators visible on the surface of permafrost ground conditions. As such, creating rock glacier inventories can be a first step in mapping permafrost distribution.
Mapping rock glaciers has however proven challenging, as it typically relies on highresolution imagery and manual interpretation. One key difficulty lies in the fact that rock glaciers share spectral similarities with their surrounding materials, making them hard to distinguish using methods that work well for conventional glaciers. As such the majority of rock glacier inventories have been produced manually, which is both time consuming, but also could introduce potentially subjective biases. These factors underscore the need for more efficient methods of mapping rock glaciers.
This master's project will focus on evaluating which machine learning techniques are best suited for transferable rock glacier mapping. The project will involve the use of satellite imagery, digital elevation models, and deep learning methods to determine the most effective approach for various environments. The study will examine areas in Svalbard and Austria, each with different climate and topographic conditions. Svalbard represents a permafrostdominated, high-latitude region with low elevations and an Arctic climate. Austria, by contrast, features lower latitudes, higher elevations, and increased precipitation. The goal of the project is to find the method that will work聽best for different regions, with respect to the different environments the areas are in.
Proposed course plan during the master's degree (60 ECTS)
First semester:
- GEOV336 Field and Laboratory Couse in Quaternary Geology (10 ECTS.)
- GEOV324 Polar Paleoclimate (5 ECTS.)
- GEOV217 Geohazards (10 ECTS.)
- GEOV300 Scientific writing and communication in Earth Science (5 ECTS.)
Second semester
- GEOV325A Glaciology (10 ECTS.)
- GEOV316 Practical Skills in Remote Sensing and Spatial analysis (10 ECTS.)
- GEOV302 Data analysis in earth science (10 ECTS.)
Prerequisites
GIS or Python experience is an advantage. GEOV205 and INF100.
Field-, lab- and analysis work
Computer based study