Multi-Sensor Deep Learning for Glacier Mapping

TL;DR

This paper surveys how multi-sensor deep learning enables scalable glacier mapping outside the ice sheets, focusing on glacier extent delineation and calving-front detection. It synthesizes data modalities (optical, SAR, and DEM) and DL architectures (e.g., GlacierNet/DeepLabv3+ and transformer-based approaches) that fuse multiple sources to handle debris cover, snow, and ocean-frontier challenges. The review highlights substantial progress in extent mapping, rock-glacier mapping, and calving-front detection, along with practical challenges in temporal change analysis, data quality, and reproducibility. It advocates for large benchmark datasets, time-series inputs, uncertainty quantification, and physics-informed DL to advance glacier mass-balance modelling and predictive evolution under climate change.

Abstract

The more than 200,000 glaciers outside the ice sheets play a crucial role in our society by influencing sea-level rise, water resource management, natural hazards, biodiversity, and tourism. However, only a fraction of these glaciers benefit from consistent and detailed in-situ observations that allow for assessing their status and changes over time. This limitation can, in part, be overcome by relying on satellite-based Earth Observation techniques. Satellite-based glacier mapping applications have historically mainly relied on manual and semi-automatic detection methods, while recently, a fast and notable transition to deep learning techniques has started. This chapter reviews how combining multi-sensor remote sensing data and deep learning allows us to better delineate (i.e. map) glaciers and detect their temporal changes. We explain how relying on deep learning multi-sensor frameworks to map glaciers benefits from the extensive availability of regional and global glacier inventories. We also analyse the rationale behind glacier mapping, the benefits of deep learning methodologies, and the inherent challenges in integrating multi-sensor earth observation data with deep learning algorithms. While our review aims to provide a broad overview of glacier mapping efforts, we highlight a few setups where deep learning multi-sensor remote sensing applications have a considerable potential added value. This includes applications for debris-covered and rock glaciers that are visually difficult to distinguish from surroundings and for calving glaciers that are in contact with the ocean. These specific cases are illustrated through a series of visual imageries, highlighting some significant advantages and challenges when detecting glacier changes, including dealing with seasonal snow cover, changing debris coverage, and distinguishing glacier fronts from the surrounding sea ice.

Figures (8)

• Figure 1: Band-ratio method. Example of the Band-ratio method being applied to Vadret da Misaun, a glacier in Switzerland (46.42$^{\circ}$ N, 9.89$^{\circ}$ E). The upper panel uses the RGB bands from the Sentinel-2 acquisition on 26/08/2015 (courtesy of EU Copernicus program). We then apply the classical band-ratio method (central panel) using the Red and SWIR bands, with a threshold of 2.8, to delineate glacier areas (lower panel). This threshold was manually chosen to extract shadowed pixels. When comparing to the glacier & debris outlines from the linsbauer_new_2021, we notice that the clean-ice and snow are accurately extracted, but this approach does not capture the debris-covered parts.
• Figure 2: Spatial resolution comparison. The effect of spatial resolution on visual products is here illustrated for Rottalgletscher, a glacier in Switzerland (46.52$^{\circ}$ N, 7.95$^{\circ}$ E). We use the RGB bands from the Landsat-8 acquisition on 22/08/2018 (image courtesy of the U.S. Geological Survey), from Sentinel-2 on 20/08/2018 (courtesy of EU Copernicus program) and the aerial image from swisstopo_swissimage_2024 (flight year = 2018). As resolution increases, an increasing level of detail can be observed, particularly pronounced for the debris-covered and shadowed glacier parts. The right panels provide a zoomed-in view of the glacier tongue, illustrating, among others, how the crevasses, a distinct feature of glaciers, become visible with increasing spatial resolution.
• Figure 3: Calving fronts - optical and . Two calving fronts that belong to the Kangiata Nunaata Sermia and Akullersuup Sermia glaciers, in Greenland (64.33$^{\circ}$ N, -49.65$^{\circ}$ E), are observed using optical data (upper panel, RGB bands from the Sentinel-2 acquisition on 24/07/2023), and data (lower panel, false-color composite extracted from the Level-1 Ground Range Detected of the Sentinel-1 acquisition on 06/07/2023. Courtesy of EU Copernicus program.
• Figure 4: at two different spatial resolutions. The figure displays the tongue of the Rottalgletscher glacier in Switzerland (46.52$^{\circ}$ N, 7.94$^{\circ}$ E) (see also \ref{['fig:res-comparison']}). The are extracted from the swissALTI$^{\text{3D}}$swisstopo_swissalti3d_2024 and the Copernicus GLO-30 , and are here displayed with a superimposed shaded relief. While both capture the shape of the valley the glacier flows in, only the with a sub-meter resolution captures smaller scale features such as crevasses.
• Figure 5: Differencing of . Here, we illustrate the role of resolution in differencing for an entirely debris-covered glacier, Glatscher da Sut Fuina, in Switzerland (46.535$^{\circ}$ N, 9.473$^{\circ}$ E). While both differences allow identifying the location of the glacier, the version (central panel), based on two swissALTI$^{\text{3D}}$swisstopo_swissalti3d_2024, is significantly more accurate. Note that for this differencing, the two were not co-registered before differentiation, which would allow for some artefacts to be removed. The differencing map based on the lower-resolution hugonnet_accelerated_2021 (lower panel) is only able to roughly capture the location of the glacier, with some potential outliers (e.g., the significantly positive pixels). This figure illustrates that when using differencing to detect glaciers, the role of the spatial resolution becomes important for relatively small glaciers.