A Unifying Framework for Detecting Outliers and Change Points from Time Series
IEEE Transactions on Knowledge and Data Engineering
An online kernel change detection algorithm
IEEE Transactions on Signal Processing - Part II
| Hi-index | 0.10 |
This paper addresses the quantification of glacier retreat through remote sensing. Specifically, we use multi-spectral Landsat satellite images for the estimation of glacier termini locations. Different frequency bands-including visual, infrared, thermal, and processed bands-are examined with respect to their utility in identifying the location of glacier termini and the associated standard error across several scenes. The methodology is to extract an intensity profile along the glacier path from the spatially registered Landsat imagery such that the complexity of the problem is reduced from 2D (image intensity) to 1D (glacier profile intensity). Local polynomial regression is then used to smooth the 1D glacier intensity profile, where the underlying function is assumed to be corrupted with correlated noise. The glacier terminus is then detected by locating an inflection point in the smoothed glacier profile, where a constrained bandwidth selection method is introduced to ensure a single inflection point along the glacier path. Using our method with thermal band B62 and a standard processed band called normalized difference snow index (NDSI) often permits for separating ice from soil but does not lead to a consistent identification of termini location, relative to ground based observations. We therefore introduce a new processed band that combines B62 and NDSI, termed normalized difference thermal snow index (NDTSI). Applying our method along with NDTSI to multiple frames from the Franz Josef, Gorner, Rhone, and Nigardsbreen glaciers indicates an ability to accurately and robustly identify the position of glacier termini, though confirmation of skill awaits application to a larger population of observations.