Discipline: Ecology Environmental and Earth Sciences
Subcategory: Climate Change
Session: 3
Room: Exhibit Hall
Merrill Storch - Hamilton College
Co-Author(s): Grace Stroh, Wisconsin Lutheran College, Milwaukee, WisconsinEdmund Robbins, Florida Institute of Technology, Melbourne, FloridaRyan White, Florida Institute of Technology, Melbourne, FloridaNezamoddin N. Kachouie, Florida Institute of Technology, Melbourne, Florida
Anthropogenic climate change has caused unprecedented warming of the Earth’s surface since the pre-industrial period. As temperatures rise, having accurate models for future climate scenarios becomes increasingly urgent. One key visible indicator of climate change is the variation of mountain glaciers over time. Glacier retreat also directly impacts communities in both mountainous and coastal regions. Rapid glacier melting threatens the water security of communities that rely on glacier runoff for infrastructural needs. Additionally, glacier runoff contributes to sea level rise which adversely affects coastal communities.Therefore, measuring glacial area variation is crucial to informing climate change adaptation strategies. However, ground-based measurements of glacial area variation are limited and impractical for tracking large areas and quantities of glaciers. Instead, researchers have implemented remote sensing using satellite imagery to measure glacial area recession. Although remote sensing is more efficient at covering large areas than ground-based measurements, it is difficult to discern between snow and ice through satellite images and to overcome cloud and shadow cover. Most notably, rock debris camouflages many mountain glaciers and prevents accurate area measurement. To provide an efficient method of finding glacial areas applicable to a variety of glaciers, a semi-automated method was developed to segment the area of mountain glaciers. The method involves converting false color Landsat images into the L*a*b* color space in order to better differentiate between glacier and mountain. The a* channel threshold was then restricted to only show blue/green glacier pixels in the image. The method was tested on Franz Josef Glacier in New Zealand and Gorner Glacier in Switzerland and the error of the method was quantified by comparing the results from the method with “ground-truth” images from the Global Land Ice Measurements from Space (GLIMS) database. The final segmentations of Gorner and Franz Josef Glaciers had 19.87% error and 21.71% error respectively. While the method was visually successful in segmenting the exposed ice, it could not differentiate the mountain from the debris-covered ice to a high precision. In the future, this project could be extended by quantifying the accuracy with which the L*a*b* segmentation technique finds the area of visible ice. The accuracy could be tested by comparing the segmentations with hand-drawn masks of the visible ice. Then, the areas of debris-covered ice, visible ice, and GLIMS could all be compared at several points in time. If they change similarly over time, the L*a*b* method of segmenting visual ice could be used as a proxy to the full glaciers’ change. Finally, the Landsat images could be merged with Digital Elevation Models to include elevation as a factor and distinguish between debris-covered ice and mountain based on location.
Funder Acknowledgement(s): This research was funded by a grant from NSF REU Program awarded to Dr. Nezamoddin N. Kachouie, Associate Professor in the Department of Mathematical Sciences in the College of Engineering and Science at Florida Institute of Technology in Melbourne, Florida
Faculty Advisor: Dr. Nezamoddin N. Kachouie, nezamoddin@fit.edu
Role: I downloaded images from Landsat, converted them into false color images, and worked on creating the image segmentation method using the L*a*b* color space. I primarily worked on the images of Gorner Glacier in Switzerland. I also compared the final optimal segmentation to the GLIMS database image to quantify the accuracy of the segmentation.