Irregular Spot Detection and Tracking: Automatically Geo-Temporal Tracking of Supra-Glacial Lakes on the Greenland Ice Sheet

• Main Authors: Yu-li Liang, Qin Lv
• Other Authors: The Pennsylvania State University CiteSeerX Archives
• Format: Text
• Language: English
• Subjects: Greenland; glacier; Ice Sheet
• Online Access: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.204.6859; http://www.cs.colorado.edu/department/publications/reports/docs/CU-CS-1078-11.pdf

Supra-glacial lakes (i.e., ponds of melting water on ice sheet) in Greenland have attracted extensive global attention during the recent years. To understand the important role they play in glacier movement, sea level rise, and climate change, scientists need to learn where these lakes are, when they form, and how they change in each melting season and across multiple years. This requires detecting and tracking supra-glacial lakes both spatially and temporally. This problem is challenging due to the diverse qualities of massive amount of remote sensing images, frequent cloud coverage, as well as the diversity and dynamics of the large number of supra-glacial lakes on the Greenland ice sheet. Previous works that use supervised methods to detect supra-glacial lakes in individual cloud-free satellite images are limited in scale, quality, and functionality. With supra-glacial lakes are shown as ”spots ” in images, we propose an effective solution to automatically detect and track time-varying spots from cloudy time-series images. In other words, this framework could be applied to any other problems to automatically detect or track irregular-shape spots from noisy images. Specifically, we propose novel techniques to (1) Select images: select the best-quality image within each time interval; (2) Spot detection: using adaptive thresholding to detect supra-glacial lakes in individual images with diverse quality; and (3) Spot tracking: track lakes across time series of images as lakes appear, change in size, merge or split, and disappear. The proposed solution has been evaluated using 10 years of MODIS data (i.e., 2000 to 2009). The results demonstrate that our proposed solution can automatically detect and track supra-glacial lakes with high efficiency and high accuracy: 96.3 % tracked lakes are Supra-glacial lakes (precision is 0.963 over 1.0), and 99.0 % supra-glacial lakes could be found by our tracking algorithm (recall is 0.990 over 1.0). 1.