Estimation of Mountain Glacier Retreat from Landsat Images
Estimation of Mountain Glacier Retreat from Landsat Images
Sunday, 15 February 2015
Exhibit Hall (San Jose Convention Center)
Background: The worldwide retreat of mountain glaciers has been reported in the last decades as an indication of global warming and as a threat to fresh water supplies. However, such a retreat has not yet been properly catalogued. Historical ground records exist for a limited number of glaciers, while estimation methods from satellite imagery involve careful delineation and segmentation of images. Instead, we propose a more direct and automated approach based on tracking of the glacial terminus location over time. Methods: Using spatially registered Landsat images and digital elevation models obtained from Google Earth, a glacial flowline is estimated and image intensity profiles are extracted along the flowline. The movement of the terminus is then captured as the estimated change in location of the main intensity transition in the profiles. This is achieved via spline smoothing of the intensity profiles to estimate first derivatives, minimization of the integrated first derivative over time to track the main intensity transition, and temporal spline smoothing to remove noise. The method requires little human intervention and is designed to be robust to changes in image intensity and disturbances such as clouds, shadows and debris. Estimates are compared to independent ground-based observations for the Nigardsbreen (Norway), Gorner (Switzerland), Rhone (Switzerland) and Franz Josef (New Zealand) glaciers. It is further applied to estimate the retreat of Viedma (Argentina), a glacier with no available ground measurements. Results: The estimates from the proposed method are consistent with the ground measurements to an average accuracy over time of between 8 and 65 m for the four glaciers studied, showing a retreat of 800 m between 2001 and 2011 for Nigardsbreen, 350 m between 1985 and 2005 for Gorner, 180 m between 1984 and 2012 for Rhone, and a retreat and advance of 500 m between 1999 and 2010 for Franz Josef. Further, our method estimates a retreat of 1800 m between 1985 and 2011 for Viedma. Conclusions: The proposed method offers a potential strategy for estimating mountain glacier retreat from Landsat images worldwide without the aid of ground measurements.