Background & Summary

The cold regions of the Globe are increasingly at risk from the ongoing climate changes. This is leading to rapid environmental dynamics, such as the rapid melting of the polar ice caps, causing sea level rise, changes in ocean currents, and an increase in extreme weather events1,2. Permafrost is solid subsurface material that remains continuously at, or below, 0 °C for at least two consecutive years3. The active layer (that spans between 0.5–2 m on Svalbard) refers to the layer of soil in areas underlain by permafrost that is subject to annual freezing and thawing, and is where most of the subsurface ecosystem processes occur in the permafrost landscape. The increasing trend of warming temperatures since the 1980s has initiated permanent permafrost degradation at a global scale4. The Arctic has warmed four times faster than the rest of the globe since 19792. Irreversible thawing and permanent loss of ice in ice-rich permafrost – known as thermokarst – affects local landforms, ecosystems and, through greenhouse-gas feedback mechanisms, the global climate5,6. This is a common problem in the pan-Arctic area7,8, Antarctica9,10 and Qinghai-Tibet Plateau – the third Pole11,12. As the Arctic warms, methane (the second largest contributor to climate change after carbon dioxide)13,14, carbon dioxide15,16, organic carbon17,18, and nitrogen (nitrous oxide) emissions19,20 are projected to increase in response to permafrost degradation. Warming and thawing permafrost also has the potential to damage vulnerable and unique cultural-heritage sites in the cold regions21,22,23,24, and pose a serious threat to the sustainable development and resilience of Arctic communities25, infrastructure8,11,26,27 and industrial sites28.

Natural hazards at mid to low latitudes are regularly mapped, allowing scientists to (1) analyse their evolution in time and space, (2) build a shared understanding of the genesis of such processes, and ultimately to (3) support susceptibility, hazard and risk assessment protocols29 according to the relevant guidelines30.

This is the case for rainfall-induced landslides31,32,33, co-seismic landslides34, gully erosion35 and floods36. In the present climate context, we advocate for the same consistency to be adopted also for the cold regions of the globe37. There has been observed an increasing trend in the publishing of cryospheric hazard data from Arctic regions, but it is still far from the global inventories from mid to low latitudes38. Some examples of published datasets are from Yukon, Banks Island, Richardson Mountains-Peel Plateau, and Beaufort Delta Canada39,40,41,42, Alaska, USA43, Siberia44,45,46, Svalbard47, and Tibetan Plateau48,49.

With the technological advances and the continuous improvement of automated mapping techniques, efforts to produce automated cryospheric inventories for both TS50,51,52,53 and TEG54 are increasing. However, a consistent accuracy of automatically mapped features requires some degree of adaptation to different data sources (e.g., satellite images) and associated spatial, temporal and spectral characteristics50,55,56. This translates into flexible models capable to transfer the mapping procedures from one area to another53,57.

To assess the available high-resolution (<1 m) aerial images collected by the Norwegian Polar Institute (NPI)58, we chose to visually interpret and manually map both TSs and TEGs, knowing that this option is more time-consuming but often more accurate compared to automated tools59,60,61. However, due to the scarcity of relevant cryospheric hazard inventories to use as validation datasets, training an AI-model for automated mapping would likely have produced erroneous and unreliable results.

Ideally, an active retrogressive thaw slump (RTS) consists of a headwall, a slump floor and a slump lobe, which are colour-differentiated on aerial or satellite imagery. This feature can continue to expand as the headwall scar retreats, making it more visible in the landscape45,59. Conversely, TS are less visible in the landscape and more difficult to map (e.g. Svalbard47,62) because their main characteristic is that they are more stable or have a very smooth headwall that has not yet retrograded (or has retrograded at some point in the past but is no longer active). To date, no automated methods have been used to create the TS inventories. This is because the “footprint” they leave in the landscape is usually not large enough or visible enough in most available remote sensing datasets (e.g. Arctic DEM63) to be able to train a deep or machine learning system.

A TEG forms when water infiltrates into the frozen soil column warming the surrounding material and causing a loss of cohesion between soil particles. This can occur in or along seasonal frost cracks in the ground, sometimes associated with ice wedge polygons64,65. TEGs rely on both soil creep and fluvial transport of particles and can develop both retrogradely upslope and by widening/deepening of the initial incision66.

The final inventories will be highly relevant for a large suite of scientists from various domains. For biologists, these are useful for habitat assessment, species distribution modelling, conservation planning, and vegetation dynamics67,68. For ecologists, this information supports research related to ecological vulnerability assessment, habitat planning and conservation, and ecological resilience67. For permafrost scientists, these types of inventories constitute the foundation for susceptibility, hazard, and risk assessment69, early warning systems, research, and policy-making70. Hydrologists may use them to evaluate river discharge71, whereas polar archaeology selects and prioritizes sites for investigation upon this type of databases, opting for suitable mitigation and conservation planning, adaptation, as well as long-term monitoring47,62,66,72,73,74. For industry, it is critical for risk assessment and planning, infrastructure development, resource extraction, insurance and finance, and emergency preparedness and response28,75. For statisticians, complete and high-quality cryospheric hazard inventories are the backbone of any prediction model in the cold regions47,76,77,78,79, with the objective of making inferences concerning triggering mechanisms and future evolution80. Ultimately, suitable risk assessment protocols rely on the same information47,76.

Beyond the modelling aspects, inventories also represent a snapshot of the environmental conditions at a given point in time. As a result, any future update of the same, would inevitably provide information on the changes a given landscape has undergone, quantifying increasing/decreasing trends in cryospheric hazard activity.

All of the above are crucial pieces of a much larger puzzle in which the Artic landscape is threatened by global warming, and the only way the geoscientific community can fully understand the landscape’s response is precisely in the form of open data. These considerations apply not only at the level of the Svalbard Archipelago, where the current work has been carried out, but also at the pan-Arctic level. This is also highly relevant in the context of recent studies highlighting a higher temperature sensitivity81 of TS activity in the Arctic compared to the Third Pole82. Therefore, this contribution not only addresses a multidisciplinary need, but also tries to stimulate similar studies elsewhere, as well as follow-up studies in Svalbard.

Methods

The development of TS and TEG inventories for 14 regions of Svalbard followed three main steps: (i) initial mapping of both processes; (ii) validation and quality check of the complete dataset; (iii) processing of geometries, attributes and size characteristics; which were also used before47 to avoid any bias in the data collection.

Data acquisition

To complete and build a comprehensive inventory of the two cryospheric hazards, we analysed the most recent high-resolution ortophotos (0.5 × 0.5 m), obtained between 2008 and 2011 from the Web Map Services (WMS) provided by the Norwegian Polar Institute58 in the ETRS_1989_UTM_Zone_33N reference system. There is no subsequent aerial or satellite imagery clear enough to delineate the two landform types exists over the last decade; the Google Earth and ESRI Wayback imagery are of too low resolution, making them unsuitable for detailed mapping. TSs and TEGs were identified based on their morphology, digitised on-screen as polygons at a maximum zoom level of 1:1.000, and then individually checked in the GIS environment. This process was repeated twice, to minimise any bias in the mapping, first by a geomorphologist (first author) and then quality checked and if needed, adjusted by an Arctic geologist (second author) to check quality and make any necessary adjustments. The aim was to produce the inventories of the highest completeness, quality, and representativeness.

Data Records

The cryospheric inventory of the 14 regions (Andreé Land, Dickson Land, James I Land, Nordenskiöld Land, Bünsow Land, Olav V Land, Sabine Land, Nathorst Land, Heer Land, Wedel Jarlsberg Land, Torell Land, Sørkapp Land, Barentsøya and Edgeøya) (Fig. 1) totalises 8491 polygons, out of which 3679 are TS and 4812 are TEG (Table 1). It should be noted the regions were predefined by naming in the Atlas of Svalbard83,84. The mapped regions are highly representative for the entire ice-free areas of the Svalbard Archipelago. The inventory is available for download at Zenodo85, under the name of SvalCryo, in shapefile (.shp) format, that can be imported into any GIS environment. In detail, the inventory is comprised only of vector-polygons. There are two main shapefiles, named TS_all and TEG_all, which comprises all the features. Within the attribute tables, there are eight columns, which comprise details about each polygon/feature, as follows:

  1. 1.

    FID – ID showing the total number of polygons.

  2. 2.

    Shape – Polygon.

  3. 3.

    ID – each polygon from each region has associated an ID for both TS and TEG.

  4. 4.

    Area – expressed in m2.

  5. 5.

    Perimeter – expressed in m.

  6. 6.

    Maximum distance – calculated between two points along the polygon perimeter.

  7. 7.

    Elongation Index – calculated as the maximum distance divided by the square root of the area.

  8. 8.

    Region – the name of the region that the feature belongs to.

Fig. 1
figure 1

Location of the 14 regions (light blue polygons) chosen for this study (base map from Norwegian Polar Institute93).

Full size image
Table 1 Details (numbers) of the two cryospheric hazards for each region.
Full size table

Technical Validation

Compared to the last inventories from Nordenskiöld Land47, which contained 562 TS and 908 TEG, the SvalCryo inventory presented here adds a significant amount of data for the Svalbard Archipelago, reaching a total of 3679 TS and 4812 TEG. Since the data have been produced and validated by two different scientists with different backgrounds, it can be said that they are of high completeness, quality, and representativeness in local, regional and pan-Arctic contexts and scales. To provide a better glimpse into the mapped TS and TEG features, we highlight some of them Fig. 2.

Fig. 2
figure 2

Detail over some of the mapped features (A) Identification of two TS located in Olav V Land, before digitization. (B) The same two TS after digitization (C) Identification of three TEG on Edgeøya, before digitization. (D) The same three TEG after digitization. There are even smaller potential TEGs visible along the snow field in this image, not mapped due to scale compared with decided zoom level at digitising.

Full size image

Completeness and quality of records

The consistency and completeness of an inventory is highly dependent on the quality of the optical imagery as well as the scale at which an area of interest extends and is surveyed86. Any automated techniques used to map TS and TEG from different cartographic materials and aerial photographs require a robust and high quality dataset, usually built up from manual interpretations, to ensure that any AI protocol adequately learns and reproduces relevant features. Some representative features have also been identified and observed in the field (examples in Fig. 2) to verify the remote sensing results. Our inventory includes features, both TS and TEG, that can be identified on screen up to a maximum zoom level of 1:1000. This approach allows unambiguous identification, in balance with the quality of the remote sensing data, while ensuring a consistent spatial accuracy of the final datasets. We did not map any features that we judged ambiguous, or formed by overlapping processes.

Activities to validate the completeness and quality of the inventory were carried out across the entire dataset. Through this approach, the SvalCryo inventory is thoroughly assessed and improved, ensuring a robust and reliable dataset for further analysis and decision-making in the context of climate change and Arctic landscape management. The completeness of an inventory has an impact on the Frequency-Area Distribution (FAD). The FAD of a TS or TEG event quantifies the number of TS and TEGs occurring at different sizes86,87. In Fig. 3, we have plotted the FAD for both processes (TS in red and TEG in blue) for each region individually. Torell Land is not included in the analysis as there were no TS or TEG present.

Fig. 3
figure 3

Plotted FAD for the 13 regions covered by the inventory. In blue we report TS data whereas in red correspond to TEG. A few regions are shown with mL values labelled as NA. This is due to the underlying calculation of the FAD, which involves a fitting a statistical model whose results are dependent on the data size. When the population is too small, the results of the statistical procedure tend to become unstable, and the results are not reported.

Full size image

In addition, we also report the event magnitude (mL), an index that has been scientifically introduced for landslides88 to give a value of the intensity of the present population. In this case, this value can be interpreted as the proportional intensity (number and size) of the TS and TEG as the value increases. This value is obtained by a best-fit line through the TS and TEG respective FADs. In a few regions, we have avoided reporting for one or the other parameter (TS or TEG). The reason for this choice is the limited TS and TEG population in these regions, which inevitably affects the mL estimates. In general, when looking at the respective mL values for TS and TEG (Fig. 3),the former (in blue) is generally associated with higher values, compared to the latter (in red).

This can be intuitively justified by the genetically much larger planimetric area of a standard TS as compared to a TEG, since a TS is by nature a laterally extending surface of thawed sediment which are displaced laterally, whereas a TEG is formed along a thin fluvial centreline by mainly grain-by-grain displacement and subsequent transport. In terms of what the FAD parameter conveys at the scale of the whole Archipelago, it shows that TS are much larger in the region 13 (Edgeøya), compared to the others, while TEGs are much more developed in region 8 (Nordenskiöld Land) as compared to the others.

We do not go deeper into the mechanism in this paper, but one could speculate on the respective influence of geology and local climatic parameters in influencing the different processes. TS are naturally very sensitive to the internal ice content of the permafrost. This, in turn, is related to the porosity and morphology/connectivity of the pore spaces that form the conduits for the movement of soil-water in the deposits, and hence the potential rate of water movement to freezing fronts89. Pore spaces and pore morphology are in turn closely related to the grain size and grain size distribution in the soil, which is physically related to the bedrock provenance (minerology). It is therefore reasonable to assume that variations in TS FAD are, at least in part, related to parent bedrock types in different areas of Svalbard, as well as to thaw season temperatures. TEGs are logically related to similar mechanisms concerning soil ice-content, but their areal development and growth rate are also closely related to the removal of solid grains by surface water flow to allow further exposure and degradation of soil-ice in the gully sides. Surface water flow is related to both the regional precipitation pattern around Svalbard and to the topography, as much of the winter precipitation falls irregularly across the landscape as wind-driven snow. The variation in TEG FAD may therefore be less easily deciphered as being related to geology and more dependent on other, independent factors.

Overall representativeness and coverage of records

The inventory covers a very large part of the ice-free area of the Svalbard Archipelago. This means areas where both processes can initiate and further develop. In the current climate context and for further understanding of the triggering mechanisms and development patterns of both TS and TEG, this is undoubtedly a step forward, and in the context of TS and TEG risk assessment and multi-hazard susceptibility, a high quality, representative inventory across the majority of the Svalbard Archipelago is a fundamental achievement. This is also of great importance in the context of Svalbard’s geology, which is known to be represented by rock formations from virtually all geological eras90. Due to its structure, the SvalCryo inventory offers users the possibility to carry out research and different types of analysis from local to regional scales in an easier and more consistent way. Examples include, but are not limited to: analysis of the controlling environmental factors (which can be derived from the ArcticDEM database63) of the two processes, their size parameters, analysis of the possible impacts of TS and TEG on Arctic infrastructure and cultural heritage sites, their distribution in selected areas, such as geologically homogeneous areas or in relation to administrative boundaries. Therefore, they were grouped according to the administrative regions of Svalbard. This is because each region has its specific climatic factors (precipitation, temperature) that influence both processes91,92. Below are some examples where we have plotted the parameters Maximum Distance (Fig. 4) and Elongation Index (Fig. 5) to give an example of what can be achieved with further processing.

Fig. 4
figure 4

Distribution of Maximum Distance across the 13 regions. TS are shown in blue, TEG are reported in red.

Full size image
Fig. 5
figure 5

Elongation Index for the 13 regions.

Full size image

The Maximum Distance is essentially the longest segment that can be drawn between any pair of points distributed along the TS or TEG polygons. It is clear from Fig. 4 is that the TEG are generally underdeveloped, with their longest expression of approximately 400 m being reached in region 5 (Barentsøya). Another general indication of TEG underdevelopment can be seen in the heavy tailed and positively skewed distributions typical of each of the regions considered, with the bulk of the distribution centred within a few tens of meters. In terms of TS maximum distance distributions, these are also heavily tailed and positively skewed, but the range of the bulk is much wider. Interestingly, the TEGs can be quite long, with a maximum overall distance of about 700 m being reached in three regions (4 – Dickson Land, 5 – Barentsøya, and 13 – Edgeøya). The long TEGs in some regions may be the result of one or several combined factors, such as the geographical extent of sedimentologically homogeneous deposits, e.g. large coastal plains of isostatically uplifted fine-grained marine sediments, together with favourable precipitation and climatic conditions for a long thawing season with regular precipitation, providing the excess water needed for fluvial sediment transport out of the gullies. The opposite may be the explanation for regions with generally smaller TEGs; laterally confined favourable deposits in the valley bottom, and/or a higher degree of internally varying sediment types, e.g. with spatially varying grain-size compositions from the central valleys towards the till- or talus-covered slopes. High spatial variability of unconsolidated sediment types would hinder TEG development, even if climatic factors are favourable.

Another parameter we have chosen to present is the Elongation Index (Fig. 5). This metric is obtained by combining the two metrics presented above (planimetric area and maximum distance). The formula links the two parameters by dividing the Maximum Distance by the square root of the planimetric area. The interpretation is therefore that, from a physical point of view, the Elongation Index is unitless, and its interpretation assumes that larger values would indicate polygons whose length is much greater than their transformed planimetric extent. In other words, values around one correspond to round to very round polygons, while values progressively larger indicate an elongation along a single axis. For TS, this value is less obvious to interpret, as the elongation index is not related to any lateral or retrogressive development of the TS. This information requires a more detailed study of each feature, which we did not focus on the first inventory. However, as we know from our mapping experience that the TS occurrences far outnumber the RTS in the Svalbard inventory, we may theorise that the more elongated TS are related to development along lateral erosional scarps in the terrain – usually associated with incising glaciofluvial rives. Thus, this index may capture a combination of favourable conditions for TS (discussed in the section above) with the presence of incised rivers in different regions. For TEG, the Elongation Index is logically most related to the same factors as the maximum distance index, since the TEG process itself works narrowly backwards into low angle deposits with homogeneous sediment properties, we expect favourable conditions in terms of both size and elongation to be the same.

SvalCryo inventory limitations

Some technical limitations must be mentioned due to the approach used to develop the SvalCryo inventory. First of all, the main limitation of the inventory is its temporal coverage. The polygons represent the reality on the ground as it was in 2009–2011. From another point of view, this is a strength for future change analysis. Considering the large area of the Archipelago, manual mapping at 1:1000 scale was quite a time-consuming process. However, we did our best to overcome the potential negative effects on homogeneity of quality, taking into account the double validation process by two independent scientists. The limitation of ground truthing should be mentioned. There were limited opportunities to be in the field during the summer season to validate our inventory, due to both scheduling and Arctic field challenges and high costs. However, as shown in Fig. 6, some limited ground truthing was carried out as well as drawing on previous experience from other Svalbard mapping projects by operator from NGU/UNIS. Another limitation is that TS and TEG were not mapped beyond the 1:1000 zoom level.

Fig. 6
figure 6

Ground truthing of some mapped features. (A) UAV photo of active thaw slumps on the right side of Linnéelva river, close to Russekeila in 2020. (B) Freshly initiated thaw slump on the left side of Linnéelva river in 2020. (C, D) Arrows indicate active thermo-erosion gullies in the proximity of Barentsburg in 2021. (E) UAV photo of a freshly initiated thermo-erosion gully on the left side of Linnéelva river (in red is highlighted the gully head, in yellow is highlighted the deposition area) in 2020. (F) The same thermo-erosion gully in 2022.

Full size image

Usage Notes

The SvalCryo inventory is available in the repository85. It is intended to be open and available to all permafrost researchers, users and authorities. The entire digitisation process has been carried out in ArcGIS v10.6.1, but all items have been produced in a format that can be imported into any GIS environment. We did our best to map all the visible features representing a TS and a TEG at a 1:1000 zoom level. There is therefore a possibility that smaller features visible beyond this zoom level may not be mapped. Future versions of the inventory will be extended to other ice-free areas, with the aim of covering the entire Archipelago. It is also hoped that more recent ortophotos will become available, allowing the repeated mapping of both processes in order to analyse their spatio-temporal evolution.