Automated Detection of Marine Glacier Calving Fronts Using the 2-D Wavelet Transform Modulus Maxima Segmentation Method

Abstract

Changes in the calving front position of marineterminating glaciers strongly influence the mass balance of glaciers, ice caps, and ice sheets. At present, quantification of frontal position change primarily relies on time-consuming and subjective manual mapping techniques, limiting our ability to understand changes to glacier calving fronts. Here we describe a newly developed automated method of mapping glacier calving fronts in satellite imagery using observations from a representative sample of Greenland's peripheral marine-terminating glaciers. Our method is adapted from the 2-D wavelet transform modulus maxima (WTMM) segmentation method, which has been used previously for image segmentation in biomedical and other applied science fields. The gradient-based method places edge detection lines along regions with the greatest intensity gradient in the image, such as the contrast between glacier ice and water or glacier ice and sea ice. The lines corresponding to the calving front are identified using thresholds for length, average gradient value, and orientation that minimize the misfit with respect to a manual validation data set. We demonstrate that the method is capable of mapping glacier calving fronts over a wide range of image conditions (light to intermediate cloud cover, dim or bright, mélange presence, etc.). With these time series, we are able to resolve subseasonal to multiyear temporal patterns as well as regional patterns in glacier frontal position change.

Synchronous Retreat of Southeast Greenland's Peripheral Glaciers

“…We use an automated terminus position delineation method adapted from a gradient‐based image segmentation technique used previously in the fields of biomedicine, solar physics, etc., (Khalil et al., 2007; McAteer et al., 2010), to determine changes in the peripheral glacier terminus positions. The adapted method (Liu et al., 2021) applies a continuous wavelet transform to delineate contours in the image brightness gradient at various spatial scales. Thresholds on the brightness contour line properties are used to select the final terminus delineation.…”

“…The thresholds were calculated using an optimization algorithm that minimized the difference between the automated delineations and a manual validation data set (512 manual delineations). When applied to Landsat 8 panchromatic satellite images over the peripheral glaciers, the method produces delineations that are accurate within one pixel of manual delineations in a variety of image conditions (Liu et al., 2021). The method efficiently generates time series of relative glacier terminus positions at unprecedentedly high temporal resolution, which varies on the order of weekly to monthly, depending on the overlap in Landsat satellite tracks and the image conditions (i.e., cloud cover, shadows, etc.,) for each glacier.…”

“…Here, we use a novel automated method (Liu et al., 2021) for delineating glacier terminus positions to investigate patterns in glacier terminus change for the Greenland peripheral marine glaciers. We focus on southeast Greenland, where the majority (59%) of these glaciers are located and where their GrIS outlet counterparts have been observed to respond sensitively and synchronously to changes in glacier stress balance initiated at their marine margins (Enderlin et al., 2014; Howat et al., 2008; Pritchard et al., 2009).…”

“…We focus on southeast Greenland, where the majority (59%) of these glaciers are located and where their GrIS outlet counterparts have been observed to respond sensitively and synchronously to changes in glacier stress balance initiated at their marine margins (Enderlin et al, 2014;Howat et al, 2008;Pritchard et al, 2009). With sub-seasonal resolution time series of terminus position generated from the automated terminus delineation method (Liu et al, 2021), we show that 56 of 135 peripheral marine glaciers in SE Greenland experienced a synchronous retreat anomaly in 2016.…”

Elimination of Image Saturation Effects on Multifractal Statistics Using the 2D WTMM Method

“…The 2D Wavelet Transform Modulus Maxima (WTMM) method has been adapted and applied in three different forms: a multiscale segmentation method ( Kestener et al, 2001 ; Khalil et al, 2007 ; Roland et al, 2009 ; Grant et al, 2010 ; Kestener et al, 2010 ; McAteer et al, 2010 ; Batchelder et al, 2014 ; Marin et al, 2018 ; Liu J. et al, 2021 ), a multiscale anisotropy method ( Khalil et al, 2006 , 2009 ; Tilbury et al, 2021 ), and a multifractal formalism ( Arneodo et al, 2000 ; Decoster et al, 2000 ; Roux et al, 2000 ; Arneodo et al, 2003 ; Khalil et al, 2006 ). The 2D WTMM multifractal method is a multiscale formalism perfectly suited for the analysis of self-affine rough surfaces such as mammograms by identifying density fluctuations and spatial correlations within these surfaces ( Batchelder et al, 2014 ; Plourde et al, 2016 ; Marin et al, 2017 ; Gerasimova-Chechkina et al, 2021 ).…”