For the GrIS margin and the nunataks in its interior, a multi-band, true-colour mosaic was compiled from Sentinel-2 data using the Sentinel Hub Processing API, a cloud-based platform for satellite data processing (Sinergise Solutions, 2025). An area-of-interest mask was applied to control the spatial extent of the mosaic, encompassing the GrIS margin, nunataks, and the surrounding area. This process generated a pixel-by-pixel, multi-band mosaic using the most recent August 2022 Sentinel-2A data. As part of the API request, the data was filtered to exclude poor-quality data (e.g., cloud-covered areas), ensuring that only valid pixels were included within the specified time frame.

In addition to the Sentinel-2 mosaic, we generated a vector polygon layer indicating the Day of Year (DOY) of image acquisition for each pixel within the mosaic. Approximately 93 % of the area was acquired between DOY 213–243, corresponding to August 2022, while less than 2 % of the area was acquired between DOY 182–212 (July 2022). Fewer than 5 % has no available DOY information. In some areas, the GrIS outline was mapped beyond the predefined area of interest. For these cases, we manually downloaded additional Sentinel-2 scenes from August 2022 via the Copernicus Browser to cover the required regions (ESA, 2022).

Since most of the GrIS terminates on land, the absence of snow in late August makes this the ideal time of the year for acquiring snow-free remote sensing data (Fig. 2). During this period, the ablation zones of the GrIS margin, along with peripheral glaciers and ice caps, are largely snow-free, simplifying the delineation of glacier boundaries between ice and land. For example, in North Greenland, snow cover typically disappears in July and returns in September (Cappelen and Drost Jensen, 2021). This seasonal pattern determines the data acquisition time frame. While marine-terminating glaciers generally reach their maximum retreat between October and early November before beginning their seasonal readvance (Black and Joughin, 2023), these glaciers represent only a small fraction of the GrIS perimeter. Therefore, the need for optimal mapping conditions, achieved with late August snow-free imagery, took precedence over capturing the annual maximum retreat of calving fronts.

The Danish Agency for Climate Data (KDS; Danish: Klimadatastyrelsen) produced a series of Sentinel-2 mosaics (∼ 10 m resolution) for the consecutive years 2019–2022, which are available as Web Map Service (WMS) and can also be downloaded via File Transfer Protocol (https://dataforsyningen.dk/data/4783, last access: 1 July 2025). The 2022 mosaic was compiled from Sentinel-2 data spanning 28 May to 12 September, although in practice, almost all tiles are from the months of June–August, with August tiles making up 68 % of the mosaic. The Sentinel-2 images used in the mosaic are orthorectified using either the GLOBE (1 km) or SRTM (30 m) elevation models (KDS, 2025a).

KDS is also providing a mosaic of high-resolution SPOT6/7 satellite images (∼ 1.6 m resolution) that can be accessed via the same WMS link as the Sentinel-2 images (KDS, 2025a). The SPOT 6/7 satellite images were primarily recorded in 2020 (75 %), supplemented by images from 2016 (25 %) and a few acquisitions from 2013 and 2014. While much of Greenland is covered by the SPOT 6/7 mosaic, there are areas that remain uncovered. The WorldDEM4Ortho (Airbus, 2018) was used for orthorectification.

Åbent Land Grønland (GL50) is a 1:50000 scale topographical map of Greenland, created using 0.5 m high-resolution commercial satellite imagery as the source data (KDS, 2025b). It covers both the GrIS and the glaciers and ice caps (GICs). The SPOT 6/7 images and the 1:50000 scale topographical map of Greenland can be easily accessed via QGreenland v3.0 (Moon et al., 2023). The source data for the entire vector dataset is primarily from the summer months between 2017 and 2021. Specifically, for glacier and snowfield coverage, the acquisition date range spans from July 2013 to October 2020. Currently, 58 % of the glacier outlines have unspecified acquisition dates in the data set. For outlines with assigned source dates, most fall within the period from 2015 to 2020 (41 %). We extracted the surface classes “glacier_s” and “snowicefield_s” from the full dataset, and these were subsequently merged and converted those to vector lines. These lines were then generalized using the Douglas–Peucker (1973) algorithm with a maximum tolerance of 5 m (equivalent to half a pixel). This simplification reduced the number of vertices by approximately 70 %, introduced a negligible maximum error of 5 m and a mean error of 1.3 m, and significantly decreased the amount of manual editing required during subsequent revisions.

The PROMICE-2022 ice mask was manually mapped in a custom QGreenland v3.0 QGIS environment (Moon et al., 2023), by tracing and delineating the GrIS margin directly from the August 2022 Sentinel-2 mosaic. In areas where the Sentinel-2 mosaic was not available or not of sufficient quality, we used the Sentinel-2 from 2022 provided by KDS. Rarely, as a third alternative, we consulted the Spot 6/7 images also provided by KDS to delineate the GrIS outline where the available Sentinel-2 images from 2022 rendered unusable. A common practice while delineating the outline was to use the high-resolution Spot 6/7 images to differentiate between snow, ice or debris and subsequently map the exact position of the GrIS margin from the Sentinel-2 mosaic from August 2022. In some cases, we used the GL50 topographic map from KDS as a reference for revision and editing where appropriate. In areas with unfavourable lightning conditions, often due to steep topography, we used a multidirectional hillshade of the Arctic DEM (Porter et al., 2024) with 10 m resolution to guide our PROMICE-2022 ice mask delineation. For the nunatak delineation, we used a combination of manual mapping and editing the nunataks provided in the GL50 dataset. Here, we also used the Sentinel-2 mosaic from 2022 as the primary reference and only occasionally the Sentinel-2 mosaic from 2022 or the SPOT 6/7 images (Fig. 3).

The 10 m resolution of the Sentinel-2 and 1.6 m resolution of the SPOT 6/7 satellite images are suitable for cartographic mapping to scales of 1:20000 and 1:3200, respectively (Tobler, 1987). For the delineation of the GrIS margin and nunataks, we used a variable mapping scale adapted to the level of detail required in different regions. Areas with complex topography or intricate transitions between ice, snow, and sediment were mapped at finer scales than more uniform sections of the ice margin. Throughout the mapping process, we aimed for a minimum mappable feature (MMF) of approximately 1 mm on screen, which corresponds to a mapping scale of about 1:25000. This scale aligns well with the 10 m spatial resolution of the Sentinel-2 imagery used, translating to an MMF of approximately 25 m on the ground (Drusch et al., 2012).

The delineation of the GrIS divides the surface of Greenland into two classes: the PROMICE-2022 ice mask, and everything else. The ice mask includes floating ice tongues and thus was not limited to grounded ice only. A topological definition was applied, consistent with the connectivity concept used in Rastner et al. (2012) and for the GIMP ice mask (Howat et al., 2014). This definition includes all contiguous ice as part of the GrIS, encompassing the ice sheet “proper”, as well as GICs that are connected to it. It also includes debris-covered ice (e.g., lateral and medial moraines), stagnant or dead ice, snowfields, snowdrift glaciers, and supraglacial lakes. Many GICs peripheral to the GrIS are considered connected either through the accumulation area, via a local glacier or ice cap that spans the drainage divide, or where they coalesce with the GrIS in the ablation area. Following connectivity level 1 (CL1) defined by Rastner et al. (2012) we separately delineate GICs that are clearly divided by drainage in the accumulation region and are not connected in the ablation region. In contrast, GICs that coalesce with the GrIS in the ablation region (CL2) are not delineated separately. Additionally, we adapted the glacier catchments provided by Mouginot et al. (2019) to the PROMICE-2022 ice mask.

One notable distinction between the PROMICE-2022 ice mask and the GIMP dataset is the treatment of ice-dammed lakes. While the GIMP mask includes ice-dammed lakes, the PROMICE-2022 ice mask does not. Ice-dammed lakes are excluded because the presence and volume of liquid water in these lakes can fluctuate due to variability in meltwater runoff and periodic drainage events (Dømgaard et al., 2024). The glacier outlines of Rastner et al. (2012) similarly excluded ice-dammed lakes, identifying frozen lakes using hillshaded DEMs and multi-temporal satellite imagery. In our dataset, ice-dammed lakes were identifiable in 10 m Sentinel-2 and 1.6 m SPOT 6/7 true-colour mosaics regardless of whether they were frozen or not. Additionally, lakes in proximity to the ice sheet were detected using vector data from the GL50 topographical dataset (KDS, 2025b).

In contrast to the GL50 ice and snow delineation, we included debris covered ice as part of the PROMICE-2022 ice mask which is especially notable at large valley and outlet glaciers with prominent medial moraines. We assumed that the basal thermal state of Greenland's outlet glaciers is predominantly thawed and therefore they occupy the full width of the U-shaped valleys that constrain their flow, particularly along the steep fjord systems of eastern Greenland (MacGregor et al., 2022). In cases where Sentinel-2 imagery did not allow an unambiguous distinction between glacier ice, supraglacial debris, and shadow, most commonly in narrow valleys, we supplemented the optical image interpretation with ArcticDEM data. From these we derive hillshade visualizations and surface-roughness metrics used to identify the transition between glacier surfaces (including debris-covered ice) and the topography of adjacent valley walls. This combined approach allowed us to consistently map glacier boundaries in areas where optical data alone were insufficient.

Snow presented a particular challenge in the delineation of the PROMICE-2022 ice mask. The extent of snow cover is highly variable, influenced not only by seasonal changes but also by occasional snowfall shortly before satellite image acquisition. To minimize the impact of snow cover, Sentinel-2 data from as late in August as possible were used for the mosaic. During manual editing and tracing, the KDS Sentinel-2 and SPOT 6/7 mosaics were also utilized to minimize the extent of seasonal and occasional snow cover.

The PROMICE-2022 ice mask and the nunataks within it were divided into several zones, each mapped by different experts following the outlined procedure. To ensure consistency across all of Greenland, each zone was independently reviewed, edited, and refined by a second expert. Quality control included checks for both completeness and correctness, including the verification of conceptual issues such as the connectivity of GICs to the GrIS. Corrections were either made directly by the reviewing expert or returned to the original mapper for revision, followed by a second round of review when necessary.

The absolute accuracy, defined as the measure of how close the vectorized outline is to its true location, can be considered a combination of two main sources of error: the geometric accuracy of the satellite imagery and the operator error associated with outlining the GrIS. Geometric and operator uncertainties were assessed for the PROMICE-2022 ice mask outline, and we assume these uncertainties are comparable for the nunatak mapping. Total uncertainty is expressed as the square root of the sum of the variances of the geometric uncertainty and the operator bias (RSS) (JCGM, 2008). Given the large number and variable spatial extent of the nunataks, we additionally assessed the omission error to quantify and reduce the likelihood of overlooked features during the initial mapping and quality control process.

The mapped and quality-controlled PROMICE-2022 ice mask margin and nunatak features were checked and corrected for dangling nodes, duplicate features, and other invalid geometries. The vector lines representing the margin and each individual nunatak were then dissolved and simplified by distance, using a tolerance of 5 m, equivalent to half a pixel (Douglas and Peucker, 1973). Mean error of the simplification is 1.3 m. Subsequently, the simplified lines were densified, ensuring a maximum spacing of 100 m between vertices. Table 1 summarizes the different file configurations of the PROMICE-2022 ice mass that are being provided. The definitions of the attribute fields associated with the corresponding data products are provided in Table 2. The data are distributed as line and polygon vector features in the GeoPackage format (.gpkg) as well as raster features in the GeoTIFF format (.tif) with coordinates provided in the WGS NSIDC Sea Ice Polar Stereographic North (EPSG:3413) projected coordinate system.

To support users in integrating datasets, we joined the CL1 polygons (file 07) from Rastner et al. (2012) with the drainage basins (file 08) from Mouginot et al. (2019) and added this information as a separate column in the basin-based product within the margins of the PROMICE-2022 ice mask. Each basin is assigned either “ice-mask” or “CL1” based on the largest area of overlap. In file 09, we went a step further and intersected the drainage basins with the smaller CL1 catchments, effectively cutting out the smaller CL1 glaciers from the larger drainage basins. These layers allow users to distinguish CL1 glaciers when working with basin delineations, all within the PROMICE-2022 ice margins. However, differences in the position of drainage basin divides between the CL1 polygons and the basins, as well as variations in the GrIS margin position across the three datasets, may lead to small local inconsistencies within these derived datasets (files 07–09).

The final PROMICE-2022 ice mask vector product is a synchronous polygon layer representing the extent of all contiguous glacierized areas and floating ice of the Greenland Ice Sheet (Fig. 4). Based on this dataset, we calculate the total, glacierized geodesic area of the GrIS to be 1 725 648 (±1061) km², excluding the area occupied by the nunataks. The geodesic perimeter length of the PROMICE-2022 ice mask margin is 53 060 km, with 2386 km marine terminating glacier fronts divided into 841 individual glaciers, and 50 674 km land terminating ice margin. We mapped 19 130 nunataks within the PROMICE-2022 ice mask with a total, projected, area of 17 916 (±1674) km² (Fig. 5). Figures 4a and 5a show the ice sheet outline and nunataks of the PROMICE-2022 ice mask, respectively, while Figs. 4b–g and 5b–g illustrate mapping decisions related to the ice mask and nunatak delineation, overlaid on the August 2022 Sentinel-2 mosaic.

For most scientific and operational applications, we recommend two “master” datasets (Table 3). These two datasets represent the most complete, stable, and widely usable forms of the ice mask: (i) the polygon master dataset 06-PROMICE-2022-IceMask-Nunatak-polygon.gpkg is the best all-purpose polygon dataset representing the clean ice mask with nunataks removed and (ii) the raster master dataset 13-PROMICE-2022-IceMask-raster-150m.gpkg is optimized for modelling workflows that operate on gridded datasets aligned with BedMachine.

The median relative geometric uncertainty of the Sentinel-2 mosaic is 11.9 m (mean: 12.7 m) and the median operator bias is 11.1 m (mean: 17.6 m, Fig. 6a). The combined RSS of geometric uncertainty and operator bias is 19.5 m. Geometric uncertainty and operator bias are spatially consistent across Greenland. The assessment of the completeness of the nunatak mapping showed that the independent mapper identified 1165 nunataks within the selected areas, of which 1141 matched features in the PROMICE-2022 nunatak mapping, and 24 were not present in the original dataset (Fig. 6b). This corresponds to an omission error of approximately 1 % in the PROMICE-2022 nunatak delineation. However, the independent mapper missed to identify 1178 nunataks originally mapped in the PROMICE-2022 dataset, suggesting differences in interpretation or detection threshold between the two mapping efforts. Thus, the total number of nunataks in the selected area is 2343.

Delineating the PROMICE-2022 ice mask margin and nunataks involves uncertainties from image resolution, mapping scale, and operator interpretation. Despite efforts to improve precision, like adjusting scale to local complexity and maintaining a ∼ 25 m MMF, some limitations persist. The use of Sentinel-2 imagery, with a 10 m spatial resolution, constrains the level of detail that can be confidently interpreted, particularly in areas where ice, snow, and sediment boundaries are diffuse or seasonally variable. Although a mapping scale of 1:25000 is theoretically compatible with the resolution of the imagery, the practical limit of delineation is influenced by image quality (e.g., contrast, shadowing) and surface conditions at the time of acquisition. The chosen MMF of ∼ 1 mm on screen is appropriate for most use cases but may not capture very narrow features that are smaller than 25 m in width. The relative consistency of the uncertainties across Greenland helps ensure internal coherence within the dataset but does not eliminate localized misalignments.

Regions with complex terrain or unclear ice boundaries require finer-scale mapping, which improves accuracy but also increases variability in operator judgment. Despite clear guidelines, decisions in shadowed or low-contrast areas, especially near nunataks or persistent snow, can cause minor inconsistencies. We chose August 2022 imagery to minimize seasonal snow cover, but regional year-to-year changes in snow and melt can still obscure the true ice margin. This unquantified temporal uncertainty adds complexity, especially for features near the detection limit.

In Greenland, the typically oblique solar illumination enhances surface contrast in Sentinel-2 imagery highlighting surface textures, often enabling accurate delineation of land terminating ice. However, mapping remains more challenging for outlet glaciers, small glaciers and generally those covered with debris, where visual contrast is limited, and glacier boundaries are less distinct. In contrast, ice caps are generally easier to delineate due to the absence of debris cover. Yet, the presence of attached snow patches, both seasonal and perennial, introduces considerable uncertainty (Rastner et al., 2012). On the other hand, in steep, narrow valleys, the same oblique sunlight can cast long shadows that obscure glacier margins. The shadows blend with adjacent terrain or cause portions of the ice surface to appear visually similar to surrounding rock or debris-covered slopes. As a result, shadowed areas can lead to underestimation of glacier extent or misidentification of the true ice boundary. This limitation is especially pronounced in high-relief regions, where accurate mapping requires optimal solar conditions, multi-temporal imagery, or complementary datasets such as hillshaded digital elevation models.

The comparison with an independent mapper for the quantification of the omission error highlights both the high completeness of the PROMICE-2022 nunatak mapping and the subjectivity inherent in manual delineation of small ice-free features. The low omission error (∼ 1 %) confirms that very few nunataks were overlooked in our mapping, reinforcing the reliability of the dataset. The large number of nunataks not detected by the independent mapper (n= 1178) suggests a more conservative mapping approach or variability in identifying features near the ice margin or in areas of low contrast. These discrepancies underline the importance of clear mapping protocols and consistent visual criteria, especially when nunataks are small, partially debris-covered, or located in ambiguous topographic settings. Additionally, it emphasises the importance of a subsequent quality assessment following both the initial and independent mapping procedures to ensure confident validation of the final ice margin and nunatak delineation.

The delineation of glacier connectivity and the subdivision of ice complexes into distinct units remain challenging and subject to interpretation. As noted by Weidick et al. (1992) separating local GICs from the contiguous GrIS poses a significant cartographic and glaciological challenge. Despite considerable advances since then, a consistent solution applicable across the entire ice sheet has yet to be universally adopted. In the PROMICE-2022 ice mask, we adapted the classification of ice features from Rastner et al. (2012) to the updated GrIS outline, but we did not reanalyse morphological evidence, including flow patterns, topographic divides, and connectivity in the accumulation and ablation zones.

Generally, temporal changes in glacier geometry over time have affected the connectivity level of several local glaciers and ice caps. As a result, the spatial relationship between the PROMICE-2022 ice mask and earlier inventories such as the RGI is not one-to-one. Several glaciers previously classified as CL2 have detached from the ice sheet and now fulfil the criteria for CL1, whereas other glaciers formerly mapped as CL1 have retreated into the present-day ice-sheet margin and are now part of the PROMICE-2022 ice mask. These processes create regions where the new outlines overlap earlier local-glacier inventories, as well as regions where small CL1 units are absent from datasets that rely on historical connectivity levels. Consequently, combining PROMICE-2022 with the RGI may result in either double-counting or omission of small, dynamic glaciers.

To support users in navigating these inconsistencies, we provide an additional derived product (file 09) in which the drainage basins from Mouginot et al. (2019) are intersected with the smaller CL1 polygons from Rastner et al. (2012). This effectively subdivides the larger basins into basin–CL1 units and preserves the smaller catchments of detached or weakly connected glaciers. File 09 therefore enables users to identify and separately quantify CL1 glaciers within the broader basin framework, reducing the risk of double-counting when combining PROMICE-2022 with the RGI and basin products improving consistency where connectivity levels have changed through time.

Nonetheless, these issues highlight broader challenges in maintaining consistent glacier connectivity information through time. The spatial overlap between ice-sheet inventories and local-glacier inventories, the resolution dependence of ice divides, and the differing modelling requirements for small peripheral ice caps all indicate that a coordinated community effort is needed. We therefore encourage future work on the temporal evolution of connectivity levels, harmonisation across datasets, and the implications for ice sheet and glacier modelling.

Most of the small-scale visual differences between the PROMICE-2022 ice mask and earlier GrIS outlines, namely GL50, GIMP, BedMachine, the RGI, and the PROMICE-1980s ice mask, can be attributed to a combination of factors: variations in mapping approaches (e.g., GL50), differences in dataset resolution (e.g., BedMachine), contrasting ice delineation techniques (e.g., GIMP), and temporal changes in the GrIS itself (e.g., PROMICE-1980s ice mask). As a result, direct, quantitative comparisons of area and perimeter length are not meaningful.

Compared to the GL50 outline, the most recent high-resolution dataset of the GrIS, we applied a different mapping approach by including debris-covered ice as part of the ice sheet. A visual inspection, however, highlights two main areas included in the GL50 outline that we did not consider connected to the PROMICE-2022 ice mask based on our connectivity criteria. One is the local GIC on the Petermann Peninsula in North Greenland, which was previously connected to the GrIS via Faith Glacier and the floating tongue of Petermann Glacier. In the Sentinel-2 imagery from August 2022, no such connection was observed. However, due to the high interannual variability of marine-terminating glacier tongues, future iterations of the PROMICE ice mask might again include this GIC as part of the GrIS.

The second area is Nuna Dronning Louise in Northeast Greenland, which was previously mapped as a nunatak. In the 2022 imagery, we observed a continuous rift between the termini of the converging glaciers Storstrømmen (flowing from north to south) and L. Bistrup Bræ (flowing from south to north), indicating that Nuna Dronning Louise is no longer a nunatak but instead lies outside the current GrIS margin. It should be noted that our mapping approach delineates the seaward edge of the floating ice tongues of these two marine-terminating glaciers, rather than the grounding line (Andersen et al., 2025). In contrast, the local GICs on Suess Land and Lyell Land in East Greenland are connected to the GrIS in the PROMICE-2022 ice mask, but not in the GL50 outline. However, most of this area qualifies as CL1, given that the drainage divides are clearly separated and the GICs do not merge with the GrIS in the ablation zone, as they do not contribute to the GrIS's mass balance.

This mapping of the PROMICE-2022 ice mask is an important step towards a more comprehensive understanding of Greenland's cryospheric dynamics and its contribution to global sea-level rise. The PROMICE-2022 ice mask provides a high-quality, expert-based, and scientifically validated GrIS outline to scientific communities, including those working in ice sheet modelling, glacier dynamics, hydrology, remote sensing, and machine learning. The dataset has been thoroughly validated and expands the availability of training data for automated mapping techniques. The consistently high amount of time required to manually map and validate ice sheet outlines contrasts with the growing availability of high-resolution satellite data. Furthermore, this accurate and high-resolution outline provides local, regional, and international stakeholders with a base dataset that policy and legislative frameworks addressing ice sheet changes, glacier retreat, and associated impacts on coastal and global systems can be built upon.

Future challenges in ice sheet mapping lie in developing methods to reliably automate this process. The present dataset serves as a resource to train and refine future algorithms. Especially when combined with freely available DEMs and remote sensing data (e.g., Sentinel-2), automated methods can leverage elevation data and optical imagery to improve mapping accuracy. Additionally, this dataset includes a statistical estimation of mapping errors, providing critical insights into the reliability and completeness of the mapped ice sheet outline.

Delineating the margin of the GrIS involves a degree of interpretation. While we have applied rigorous quality control throughout the mapping process, we acknowledge that some inaccuracies such as localized misalignments may remain. We therefore welcome feedback from the scientific glaciology community to help improve the PROMICE-2022 Ice Mask dataset. Any issues, suggestions, or corrections can be submitted via the project's GitHub repository (https://github.com/GEUS-Glaciology-and-Climate/PROMICE-ice-mask, last access: 8 July 2025).

With the expected acceleration of ice sheet mass loss due to climate change, a detailed and regularly updated outline of the GrIS is more important than ever. In Greenland, a combination of warming temperatures, increasing melt rates, and dynamic ice flow changes will continue to reshape the GrIS margins. Understanding these changes and their implications is critical for effective climate adaptation strategies. To support this, the PROMICE ice mask will be regularly updated with future inventory years, following the methodology and data sources presented here. We anticipate that upcoming versions will also include glaciers and ice caps peripheral to the GrIS.