Several remotely sensed raster data sets and processed versions were used for the identification and mapping of landforms. Lantmäteriet provides a lidar-based digital elevation model (DEM), also known as the national elevation model, with a spatial resolution of 1 m . The DEM has an absolute vertical accuracy of <0.1 m and an absolute horizontal accuracy of <0.3 m, but these can vary depending on point density of the laser scanning, time of scanning, and survey technique . The DEM was processed in ArcGIS Pro 2.9.3 to create a hillshade relief model using an illumination angle with an altitude of 30° and azimuths of 45 and 315°, as these are considered the optimal values for the visualization of hillshade relief models for the purpose of glacial geomorphological mapping . Additional azimuths of 90 and 180°, perpendicular and parallel to the dominant lineation orientation, respectively, were applied to reduce the “azimuth bias” . Additionally, a slope model was derived from the DEM, and contour lines were created with intervals of 10, 20, and 100 m. provides natural-color (RGB 4, 3, 2) and color-infrared (RGB 5, 4, 3) orthophotos with a resolution of 0.5 m acquired in 2018 and 2021 for roughly the western montane region and eastern premontane region, respectively (Fig. c). Although the mapping was primarily DEM-based, in certain cases the aerial imagery enhanced landform detectability. Google Earth Pro enabled visualization of the terrain in 3D, which was mostly used to cross-check mapping based on other imagery. Its imagery is primarily satellite-based and displays varying resolutions.

Vector data sets of previously published studies were used for different purposes. The international database of contains glacially induced faults in northern Fennoscandia (Fig. b), of which many were previously proposed and recently confirmed based on the recent lidar data. The faults in the database were cross-referenced with the lidar-based DEM, but no effort was made to identify new faults. The data set was used to identify cross-cutting relationships between glacial landforms and fault scarps. The deglaciation isochrons reconstructed by were used to evaluate the implications of the direction of mapped landforms and to constrain the chronology (Fig. b). Cosmogenic nuclide 10Be exposure ages of two rock slope failure (RSF) deposits were taken from . The RSF extents were cross-referenced against the lidar-based DEM. The printed geomorphological maps by and were digitized and georeferenced in GIS software using locations on the map with known coordinates for cross-referencing purposes.

Fieldwork was conducted in August 2021. The aim of the fieldwork was to ground-truth landforms. At every site, landform and landscape morphology and sedimentological properties of the landform were assessed. The field sites are concentrated around the southern shore of lake Torneträsk (Fig. c), given the focus on ice-dammed lake traces, such as raised shorelines and perched deltas, and ease of access.

Remote mapping on the scale of a regional sector of a paleo-ice sheet requires a systematic mapping approach, in order to map a large area in a time-efficient yet accurate manner . Relevant for the mapping process itself is the construction of a landform identification table, which describes the morphology, dimensions, identification criteria, and paleoglaciological significance of each glacial landform (Table ). Landforms were mapped at a scale of 1 : 30 000, although the final map is presented at a scale of 1 : 300 000 (Fig. S1 in the Supplement). Depending on the landform, zooming in to a larger scale was required to outline the landforms correctly. Mapping was performed in two iterations.

The application of an inversion model is required for extracting ice sheet properties from mapped glacial geomorphology, such as its thermal regime, subglacial hydrology, or the presence of ice streams. Here, the conceptual framework of is applied to deduce ice sheet evolution through time. The inversion model is composed of a set of assumptions p. 196, a classification system for glacial landform assemblages, and a procedure for managing the landform data and incorporating absolute chronological data. The model thus explains how individual landforms are interpreted in terms of ice sheet properties, which results in ice-sheet-wide glaciologically consistent patterns by aggregation of the individual landforms into swarms. Wet-bed deglaciation swarms include eskers with aligned lineations. These fields of lineations and eskers are formed time-transgressively, parallel to ice flow, and perpendicular to the ice margin. Dry-bed deglaciation swarms typically lack subglacial landforms, due to an absence of sliding when the ice sheet is frozen to its substrate, but include meltwater channels, ice-dammed lake shorelines, and perched deltas. Such meltwater traces are imprinted on a relict surface, which can be non-glacial or glacial, thus demonstrating the subglacial preservation of landforms and landscapes. Ribbed moraine forms when subglacial conditions change from dry- to wet-bed , with its individual ridges oriented perpendicular to ice flow direction. A set of these landforms representing coherent ice flow directions and ice margins can then be outlined to realistically visualize retreat patterns.

The mapping approach, that is, how the landforms are delimited in GIS software, is briefly described for all landforms in Table . Given the focus on ice-dammed lake traces, the mapping approach of raised shorelines and perched deltas and the methodology to identify ice-dammed lake stages are described in more detail below. Raised shorelines are mapped as polylines positioned midway between the toe and inner break of the shoreline. The polylines are then converted to vertices, for which the elevation in meters above sea level was extracted from the DEM by means of the Add Surface Information tool in ArcGIS Pro. Similarly, polylines of identified lake outlet channels are converted to vertices, for which the elevation was extracted for the first vertex, representing the minimum threshold of overflow. Perched deltas are mapped as polygons that demarcate the flat top surface. The minimum elevation along the polygon was extracted with the assumption that it represents the delta front. The elevations of the shorelines, perched deltas, and outlet channels are analyzed against the distance along a reference plane, of which the distance was calculated by means of the Near tool in ArcGIS Pro. The reference plane has an orientation of 325° N (Fig. c), which is oriented perpendicular to the isobases of raised shorelines as inferred from literature . The tilting of the shorelines is described as a gradient in mkm-1, where the elevation difference (in meters) is given over the distance (in kilometers) in the direction of the reference plane. The shorelines corresponding to specific lake stages were isolated, and the outlines of the ice-dammed lakes were drawn by tracing the corresponding contour line of the stage. Interpolation between the contour lines that corresponded to the elevation range of the stage was necessary to account for the tilting. The ice margin that constituted the obstruction damming each lake stage was placed to fit the distribution of ice-dammed lake traces. The volumes of the ice-dammed lakes, approximated by the difference between the ice-dammed lake surface and the present-day lake surface (DEM), were calculated in Python 3.9.18 following the methodologies of . The difference in volume of two consecutive ice-dammed lake stages represents the approximation of the outburst flood volume.

Several ice-dammed lake systems were in operation during deglaciation in basins currently occupied by Torneträsk, Rautasjaure, Vittankijärvi, Nakerijärvi, Vassijaure, and Kaitasjärvi (Fig. ). The identified ice dams are consistent with the distribution of shorelines, perched deltas, and outlet channels; are required to impound the lakes against surrounding topography; and are consistent with the orientation of meltwater channels, lineations, and eskers. Those stages of IDLT for which a plausible outlet could be identified are outlined in Fig.  and described in the following sections.

Because lake Vassijaure (461 m a.s.l.) is located at the far western end of the Torneträsk Basin, ice-dammed lake Vassijaure (Va; Fig. ), with perched delta elevations between 510 and 513 m a.s.l., is included in this subsection. Its spillway, located at 510 m a.s.l., directed meltwater through the valley Norddalen towards Rombaksfjorden in Norway. A set of perched deltas at the head of the fjord are likely related to the incision of the spillway and indicate 95 m of relative sea level lowering due to glacio-isostatic uplift. Ice-dammed lake Vassijaure required an ice dam just west of Sördalen (Fig. ) and was in operation until ice retreat exposed Sördalen, a much lower passage at 412 m a.s.l. into Norway (Fig. a). Multiple channel incisions and a general lack of sediment along the western Sördalen valley side demonstrate the erosive power of the drainage of ice-dammed lake Vassijaure, as the lake level dropped about 50 m.

IDLT stage T2, with raised shoreline elevations between 471 and 495 m a.s.l., required two specific ice margin locations. One ice margin prevented the lake to drain westwards through Sördalen (Figs. a and ). Given the requirement for an ice dam in western Torneträsk Basin abutting the northern shore of Torneträsk and preventing the drainage of both ice-dammed lakes Vassijaure and Torneträsk (T2) through Sördalen (Fig. ), the most plausible configuration yields a piedmont lobe emanating from Abiskodalen. The other ice sheet margin obstructed drainage across lower terrain towards the southeast. This and subsequent damming margins (T3 to T8) outline the retreat of a large coherent ice body towards the south-southwest. A spillway towards Vuoskujärvi, a lake north of the study area (not shown), was controlling the highest elevation of the lake to 500 m a.s.l. Ultimately, T2 drained through Sördalen as the ice lobe retreated from the northern shore, and the Sördalen outlet channel became the spillway of the next lake stage (T3).

Ice-dammed lake stage T3, with raised shoreline elevations between 416 and 462 m a.s.l., presumably had its spillway through Sördalen at 412 m a.s.l. (Fig. a). Because the lowest shorelines in the vicinity, with elevations of 417 m a.s.l., are several meters above the current elevation of the spillway, this potentially reflects deepening of the spillway by ca. 5 m through continuous runoff. A coherent ice margin in the southeast had to plug two valleys, requiring an ice dam with an E–W orientation (Fig. ). An E–W ice margin is also justified from the presence of nearby eskers and lineations with orientations perpendicular to the ice margin. Morphological traces of a drainage event (15 km3, Table ), that lowered the lake level to the next stage, indicate drainage routing to nearby ice-dammed lake Vittankijärvi (Vi, Fig. ). To accommodate this drainage route, the ice margin of IDLT stage T3 was most likely connected to the ice margin of one of the lower stages of ice-dammed lake Vittankijärvi to expose terrain between both lakes (i.e., between Vi1 and Vi4, Fig. ).

The ice front impounding IDLT stage T5, with raised shoreline elevations between 391 and 432 m a.s.l., probably had a rather similar configuration as the ice margin reconstructed for ice-dammed lake stage T3 (Fig. ). The ice sheet retreated around 5 km, although slightly more in the east than in the west. IDLT stage T5 had its outlet channel at 424 m a.s.l., with fluvial deposits downstream of the channel over a distance of 1.5 km (Fig. b) and drained 12 km3 of water (Table ) to ice-dammed lake Vittankijärvi (Fig. ). Drainage routing into the lake explains the absence of traceable GLOF imprints.

IDLT stage T6, with raised shoreline elevations between 372 and 417 m a.s.l., drained through an outlet channel with a width of approximately 750 m at 383 m a.s.l. (see western channel in Fig. c). GLOF traces occur downstream of the outlet channel in the form of streamlined pendant bars on the lee side of bedrock protrusions (Fig. c; northern pathway in Fig. a). Indeed, when the ice dam failed, 19 km3 of lake water (Table ) was routed through Tornedalen.

IDLT stage T7, with raised shoreline elevations between 351 and 382 m a.s.l., required a similar ice margin configuration as for IDLT stage T6 with their outlet channels a mere 1 km apart (Fig. ). The lake level coincided with the elevation of the outlet channel of the previous stage at 383 m a.s.l., now acting as a spillway (western channel in Fig. c). The outlet channel at 365 m a.s.l. (eastern outlet channel in Fig. c) has an internal topography of ca. 5 m. Erosion of at least 5 m appears to have occurred due to the drainage of IDLT stage T7 (2 km3, Table ), thus establishing a new spillway altitude for IDLT stage T8 at 365 m a.s.l. and providing a source of material for downstream GLOF deposits (blue polygons in Fig. a, flat terrain in Fig. d).

The lowest lake level of IDLT, stage T8, with shorelines situated between 347 and 369 m a.s.l., was dammed approximately 18 km south of T7 (Fig. ). The lake level of T8 was probably controlled by the outlet channel of T7 (Figs.  and d). Drainage of ice-dammed lake T8 occurred predominantly through Tornedalen. Two drainage channels paralleling the current river Torneälven occur at 365 and 357 m a.s.l. (Fig. d), which appears to demonstrate that the drainage of 5 km3 occurred in three steps: initial drainage along the ice margin, another drainage after thinning of the ice sheet, and a final drainage as all of Tornedalen opened up (Table ).

Extrapolation of the available data on the elevations of ice-dammed lake traces to determine the western extent of IDLT stage T8 is hampered by invisibility (Fig. a). Indeed, the absence of shoreline traces west of the Pärvie Fault is most likely due to their submergence as a result of differential landscape uplift (creating the illusion of lake tilting relative to the raised shorelines). Ice-dammed lake T8 shorelines would be predicted at elevations at or above approximately 325 m a.s.l. in this sector, which is up to 17 m below the current lake level of Torneträsk. Bathymetric data are therefore required to faithfully outline the perimeter of T8, but the most recent open-source bathymetric map is based on measurements from 1920–1921 . Nevertheless, the available bathymetric information shows that the 20 m bathymetric contour occurs at a relatively short distance inboard from the current shoreline, indicating a rather steep basin. Hence, the current shoreline of Torneträsk is therefore used as the perimeter for the western part of IDLT stage T8 and therefore slightly overestimates its areal extent. Moreover, the current lake surface was used as base surface for the calculations of GLOF volumes, leading to an overestimation of the GLOF volume of stage T8 (Table ).

Drainage of IDLT stage T6 resulted in the largest of the GLOFs with an estimated volume of 19 km3 (Table ). In comparison, a catastrophic GLOF in central Jämtland, instigated by the failure of an ice saddle between the southern and northern domes of the FIS, released an estimated 200 km3 . It appears that the Torneträsk GLOFs terminated in Ancylus Lake, a freshwater lake formed by isolation following isostatic rebound, occupying the current Baltic Sea Basin . Raised shorelines of the highest coastline of Ancylus Lake (Fig. a and b) appear approximately 165 km downstream of the initial drainage location(s) of the GLOFs. Mapping of surficial geology by the Geological Survey of Sweden , carried out from the 1980s to early 2000s based on aerial imagery supported by sparse field observations, shows that vast amounts of glacio-fluvial sediments which form deltaic deposits are found 165 km downstream of the initial drainage location(s) along both Tornedalen and Kalixdalen (Fig. d). The most parsimonious explanation for these glacio-fluvial deposits is that they originated from the debris-laden GLOFs entering Ancylus Lake at the highest coastline at an elevation of 170 m a.s.l., where discharge velocities decreased, sediments were deposited, and a deltaic landform was created.

Shrinkage of FIS across the study area (Fig. c) yielded a complicated pattern, with southeastward retreat in the far western corner (isolating ice-dammed lake Vassijaure), initial southeast–southward retreat of the FIS across the Torneträsk Basin, and eventually curving towards southwestern retreat in the southern part of the study area (Fig. ). This overall pattern aligns with former reconstructions and is reflected by the orientation of the small-scale lineations, eskers, and subglacial meltwater channels in the premontane region (Figs. c and S1). Disentangling glacial geomorphological traces in the mountains, determining their succession, construction of swarms, and assigning absolute ages (even determining if landforms are deglacial or not), is challenging. For example, it can remain unclear whether moraines were formed by advancing valley glaciers situated up-valley of the moraine or by outlet glaciers from a thinning ice sheet . This problem is exacerbated by the knowledge that landforms in these mountains are not exclusively from the last deglaciation, including moraines . Nonetheless, the retreat in the premontane region is clear, and moraines, lateral meltwater channels, and relict shorelines in the montane region require the former existence of ice lobes (Figs. c and S1).

There is a clear distinction between retreat in the premontane and montane regions in terms of the configuration of the ice sheet. In the premontane region, ice sheet retreat is orderly, and the margin maintains its shape and outlines a coherent ice body (Figs. c and ) as ice flow directions transition from northwest through north to northeast, while the ice margin pivots around the higher topography. In the montane region, meanwhile, the ice sheet disintegrates into several ice lobes (Fig. ). Hence, a strong control of topography on ice retreat patterns and rates is evident, as other studies have demonstrated for the FIS , the British–Irish Ice Sheet , and the Cordilleran Ice Sheet . Bed topography becomes increasingly dominant as the ice thins ; hence topographic control is especially significant during ice expansion and final deglaciation. Because we cannot determine the precise disintegration of the ice sheet in the mountains, save for a few margin locations such as the up-valley ice dam required for ice-dammed lake Rautasjaure (Fig. e and f), we have simply characterized this area with uplands becoming ice-free before adjacent valleys e.g., with “nunataks” (Fig. c and d) and “question marks” (Fig. e–g). Dynamically, montane ice separated from premontane ice, leading to the damming of lakes in all major easterly valleys, including farther south .

Ice-dammed lakes influence ice sheet mass balance and ice dynamics, while they amplify glacier mass loss and velocity . On the ice sheet scale, it is well known that the FIS retreated faster in the (south)east than west due to a calving margin in the Baltic Basin . Higher ice losses of the FIS due to the ice-dammed lakes in the Torneträsk region are consistent with the amplifying effect that ice-marginal lakes have on retreat rates e.g.,. Thus, the presence of the ice-dammed lakes led in part to the pivoting motion of ice retreat in this region.

Glacial isostatic adjustment is a response of the Earth to a redistribution in ice load . A combination of ambient tectonic and glacial isostatic adjustment-induced stresses due to ice sheet thinning and retreat during deglaciation can cause faulting, a reactivation of pre-existing faults, and earthquakes . The reactivated faults are referred to as glacially induced faults. The Pärvie Fault, 155 km long, is the longest glacially induced fault in Sweden . The fault is composed of more than 200, generally 5–10 m high, predominantly west-facing, fault scarps .

The NNE-trending Pärvie Fault cross-cuts Torneträsk, and we have shown that it offsets IDLT shoreline stages T3–T6 (and presumably T1 and T2) but not T7 and (by inference) T8. Because we can now show, for the first time, that the FIS margin was positioned in between the ice-dammed lake stages T6 and T7 reconstructed ice margins when the Pärvie Fault ruptured (Figs. , e and f), it will now be possible to combine Pärvie Fault analyses from lidar e.g., with glacial configurations. At the very least, the implication is that the fault ruptured within 40 km from the ice margin. In fact, if the fault ruptured along its entire length at this time, then as much as 95 km of the fault trace was ice-covered at that point.

Along the length of the Pärvie Fault within the study area, cross-cutting relationships between the glacial geomorphology and fault scarp traces can be studied relative to an inferred 9.9–10 ka cal BP reactivation age of the Pärvie Fault. If all of the Pärvie Fault ruptured at once, such as is typically considered when calculating the amount of energy released, cross-cutting should post-date deglaciation north of the inferred ice sheet margin at the time of rupture (between T6 and T7, Fig. e and f) and pre-date deglaciation south of this ice sheet margin, as originally suggested by . However, such a systematic relationship does not exist (Fig. a). In fact, our documentation yields examples of pre-deglaciation faulting north of the ice margin (Fig. b), postglacial faulting south of the ice margin (Figs. c and c), a few cross-cutting relationships indicating the fault ruptured multiple times (Fig. d and e), and sites where the relationship remains unclear (Fig. a).

A complicating factor is that the region in northern Sweden is known for palimpsest landscapes where traces of older glaciation may be preserved despite subsequent ice sheet overriding . Indeed, illustrate that some glacial landforms cross-cut by the Pärvie Fault (farther south) are not necessarily formed during the last deglaciation, and, if so, that the expression of the fault scarp itself could have been preserved beneath ice sheets. However, glacial landforms cut by the Pärvie Fault within the study area align with reconstructed deglaciation directions, and so faulting here was demonstrably postglacial.

A recurring hypothesis is that the Pärvie Fault was created (or reactivated) through a single seismic event . If true, the formation of the Pärvie Fault would require a thrust earthquake with a moment magnitude of Mw≈8.0 (e.g.,). Currently, such an approach appears unrealistic given the mounting evidence for different types of cross-cutting relationships (Fig. ), reinforcing alternative interpretations that the Pärvie Fault ruptured multiple times. Supporting this interpretation, found that a fluvial terrace truncated by the Merasjärvi Fault scarp, another glacially induced fault southeast of the study area (67.520833° N, 21.941944° E), displayed multiple ruptures.

describe a situation which resembles that of Torneträsk where a section of the Pärvie Fault north of lake Langas cross-cuts an esker (67.430859° N, 18.711962° E) and a series of poorly preserved raised shorelines (67.424151° N, 18.701256° E), except for the lowest shoreline, which is not displaced by the fault . This setting also appears to satisfy rupture immediately after local ice margin retreat from the area . Therefore, this segment of the Pärvie Fault must have ruptured later than the rupture in the Torneträsk Basin between IDLT stages T6 and T7 because lake Langas would have been ice-covered at that time.

Multiple ruptures of the Pärvie Fault appear to explain the observed cross-cutting relationships best (Fig. ). The shorelines of IDLT stages T3–T6, which are cut by the Pärvie Fault, are followed by IDLT stages T7 and T8, which are not offset, hence providing a geomorphological marker for the timing of this rupture of the Pärvie Fault (Figs.  and ). Locations demonstrably underneath the ice sheet during the IDLT stages 6–7 Pärvie Fault rupture also exhibit cross-cutting relationships, indicating at least some fault scarp segments ruptured subaerially (Figs. e and f and c). The most inclusive explanation is therefore that the fault ruptured several times within the study area: once shortly after the formation of IDLT stage T6 shorelines and north of the ice margin at T7 and once or several times subsequently, as evidenced by postglacial cross-cutting relationships south of IDLT stages T7 and T8 (Figs. f and g and ). Clearly, this latter faulting south of Torneträsk did not reactivate fault traces further north, allowing the shorelines of IDLT stage T7 to escape offsetting. The variety of cross-cutting relationships clearly illustrates the complexity and spatial reach of the reactivation of faults in response to glacial isostatic adjustment.

Ice retreat in the Torneträsk region coincided with the formation and drainage of ice-dammed lakes and (multiple) ruptures of the Pärvie Fault. Geomorphological relationships provide insights into the relative timing of these events, but a lack of absolute ages limits our ability to precisely constrain their temporal evolution. Here, we attempt to bridge that gap by integrating existing knowledge regarding the timing of ice sheet retreat with our reconstruction of the ice-dammed lakes and faulting.

Ice retreated from the coastal zone in Norrbotten between 10.2–9.9 ka cal BP at the same time as Ancylus Lake transgressed the coastal zone, peaking at 10 ka cal BP . There are no cross-cutting relationships between the GLOF landforms and the Ancylus Lake shorelines along the flood path. However, as pointed out, the deposition zone of GLOF sediments corresponds with the location of the locally highest coastline. The age of the GLOFs must therefore overlap with the age of Ancylus Lake (10.5–9.5 ka cal BP; ) and, through its association with the highest coastline, can therefore perhaps be narrowed down to having occurred close to 10 ka cal BP. According to the reconstruction by , Tornedalen opened up approximately 9.9 ka cal BP, which is consistent with this age estimation for the GLOFs.

The shoreline gradients of Torneträsk have a tilting direction towards the northwest (325°) and decrease from 0.7 mkm-1 for the oldest ice-dammed lake stages to 0.5 mkm-1 for the youngest ice-dammed lake stages. The gradients clearly reflect the glacio-isostatic uplift pattern following the deglaciation of Fennoscandia . The estimated gradients of Torneträsk are (i) slightly higher than the gradients of 0.4–0.5 mkm-1 that calculated for ice-dammed lakes Akkajaure and Sitojaure, located approximately 120 km farther south between the Kebnekaise and Sarek mountains, and (ii) bracketed by the gradients of 0.3–0.9 mkm-1 of the shorelines of ice-dammed lakes in central Jämtland . These are favorable comparisons because ice-dammed lakes Akkajaure and Sitojaure also formed in response to the final deglaciation of the Fennoscandian Ice Sheet. Their timing is closest to the youngest GLOF of Torneträsk, and so are their shoreline gradients. Additionally, lake evolution in central Jämtland spans 10.5–9.2 ka cal BP , which brackets ice retreat from the Torneträsk Basin, rendering it reasonable that the gradients of IDLT fall within the range of values from central Jämtland. We are, however, hesitant to put too much confidence in the derived gradients and comparisons between the regions. The calculations depend on the direction of the tilt along which they were calculated, the precision of the mapping of shorelines, and on post-depositional faulting.

In the reconstruction of , the retreating ice margin swept across the study area in a time span of 500 years (Fig. b). The ice-marginal positions that dammed the successive ice-dammed lake stages of Torneträsk fall approximately in between their 10.1 and >9.9 ka cal BP isochrons (Fig. b), which would suggest the ice-dammed lake system of Torneträsk existed for a total duration of <200 years. Shorelines are estimated to require at least a few decades to develop . Waves play a key role in the development of shorelines , although in periglacial environments lake ice potentially influences shoreline development as well . Regardless, the formation of shorelines is largely dependent on the duration of exposure to shoreline-building processes, which requires a stable lake level for a certain period. Drainage of an ice-dammed lake (67.100232° N, 16.403404° E) in front of Sállajiegna, a glacier on the Swedish–Norwegian border , in 2013, revealed that part of a prominent shoreline had been built since 2009 (within 5 years) judging from aerial imagery. This is observational evidence that a shoreline can form in mere years, not necessarily in decades. A total duration of the ice-dammed lake system of Torneträsk of 200 years is therefore certainly plausible. Even the shorelines of the 22 stages of the ice-dammed lake system of Rautasjaure would have been able to develop within the overall time window.

The brief period during which the ice-dammed lakes formed and drained makes it challenging to determine the role of faulting. The largest drainage, following IDLT stage T6, coincides with the fault rupture offsetting shorelines of IDLT stages T6 and older. These two events, fault rupture and drainage, appear to have happened simultaneously (i.e. 9.9–10 ka cal BP). For example, the rupture might have triggered an instability of the ice dam, causing the ice dam to fail and the lake surface elevation to lower to a new spillway level (or the other way around; the record release of water from the lake might have triggered the fault reactivation). Although there is no indisputable evidence for correlation of these two events, the implied influence of glacially induced faulting on the stability of ice dams, and hence on the ice retreat dynamics, should not be discarded.

Dating of the Pärvie Fault could provide a minimum age for IDLT stage T6 and a maximum age for stage T7. Unfortunately, no absolute ages are available for the Pärvie Fault. However, our mapping might provide future avenues to dating the timing of fault rupture by means of dating-related deposits. The drainage of IDLT stage T6 is the largest in terms of volume (Table ) and the first to have resulted in the deposition of recognizable GLOF deposits (Fig. a and d). Macrofossils incorporated in basal sediments of the Torneträsk GLOF deposits at the Ancylus Lake highest shoreline could potentially yield an age for the drainage of T6. Alternatively, radiocarbon dates on macrofossils straddling the boundary between glacio-lacustrine sediments from the ice-dammed lake phase and non-glacial sediments (after drainage), in small lakes located between IDLT stages T6 and T7, could offer an additional approach to dating the T6 GLOF and, by association and within dating uncertainty, activity on the Pärvie Fault e.g.,. Finally, because the outlet of IDLT stage T6 (western channel in Fig. c) also was the spillway channel of IDLT stage T7, bedrock in this channel would be first fully exposed to cosmic rays after failure of the IDLT stage T7 ice dam only a decade or decades after the rupture of the Pärvie Fault. Cosmogenic in situ 14C dating of outlet channel bedrock is likely the most direct methodology to determine the age of faulting (within uncertainty). The promise of this technique to deliver accurate deglaciation ages in Sweden was demonstrated by .

Similarly, secondary deposits related to the fault rupture itself could be dated. The formation of landslides in northern Fennoscandia has been associated to earthquakes caused by post-glacial faulting . There are two large RSF deposits in the study area (RSF-K and RSF-B, Fig. c). If they were triggered by ruptures along the Pärvie Fault, their ages could be used to date fault activity. A boulder from a RSF deposit in Kärkevagge (RSF-K, Fig. c; ) has been dated to 9.5 ± 1.1 ka using cosmogenic nuclide 10Be , and a RSF scar bedrock sample (S-02-05) and boulder tongue sample (S-02-06) in Bessešvággi (RSF-B, Fig. c) returned 10Be apparent exposure ages of 8.9 ± 0.8 ka and 10.9 ± 0.8 ka, respectively. All three samples are expressed using the LSDn scaling method with external uncertainties (cf. supplementary data set from ). Clearly, these results alone are inconclusive in dating fault activity. For example, although one may argue that deep-seated (>3 m depth) bedrock in a scar should provide the most reliable age, considering that a boulder is more likely to have inherited previous exposure, we do not know from a mere two samples whether its younger age is indeed due to boulder inheritance or whether slope erosion after the main failure could have resulted in a younger apparent age of the bedrock scar. On aggregate, however, the timing of RSF activity from all three samples, within uncertainty, overlaps with our inferred timing of the Pärvie Fault of ca. 9.9–10 ka cal BP.

The absence of a larger group of landslides in the vicinity of the Pärvie Fault challenges the potential earthquake-induced origin. It is predominantly the scattering of a group of landslides across a discrete area, in close proximity to a fault, and their synchronous age rendering it likely that they were triggered by an earthquake e.g.,. The spatial distribution of the two RSF deposits and the corresponding ages are therefore not conclusive regarding whether they were triggered by an earthquake or by other triggers, such as glacier debuttressing after deglaciation. However, the absence of a larger group of landslides could hint towards the nature of the Pärvie Fault rupture. It is in stark contrast to the large groups of earthquake-induced landslides nearby glacially induced faults in northern Finland e.g.,. The presence of fault scarps but absence of landslides could support the occurrence of earthquakes underneath the retreating ice sheet. The cross-cut shorelines of IDLT indicate that the fault scarps locally ruptured at a close distance to the retreating ice margin. Although there is mounting evidence that the Pärvie Fault was not the result of a single rupture, it cannot be ruled out that there was a partial subglacial rupture. suggest morphological signs of subglacial rupture could be anastomosing networks of eskers (Fig. b) and subglacial crevasse fillings, which all seem to be present in the Torneträsk area .

Previously published work on glacial geomorphology in the Torneträsk Basin is broadly consistent with the results of this study but lacks the detail allowed by the use of the lidar-based DEM. The most detailed geomorphological maps of the Torneträsk region were produced by . His landform interpretation is based on aerial photographs and extensive field verification and resulted in a comprehensive geomorphological map presented at 1 : 250 000. Our mapping (Fig. S1) is consistent with his mapping but adds considerable detail in terms of the number of raised shorelines (resulting in more ice-dammed lake stages) and the number of channels in flights of lateral meltwater channels. Additionally, whereas we map different types of meltwater channels, only categorizes glaciofluvial channels by size. For example, some large glaciofluvial channels correspond to outlet channels of ice-dammed lakes in this study. A critical difference between our maps is the number of lineations; our mapping includes significantly more lineations in both the premontane and the montane regions. The last glacial geomorphological map covering the Torneträsk region was produced from aerial photographs by at 1 : 1 250 000. Unlike , this map includes large- and small-scale lineations, ribbed moraine, De Geer moraines, and Veiki moraine. Our landform distributions of those features are consistent with the map but provide more detail, as individual lineations are outlined rather than a representative for a larger area. Thus, our mapping based on high-resolution lidar data, as expected, adds more detail in terms of landform count but is consistent with previously mapped landform distributions. The critical implication of added detail in our mapping resides in a more detailed reconstruction of the ice-dammed lakes but does not alter general inferences on ice retreat from ice flow directional indicators.

A lidar-based reconstruction of the ice-dammed lakes in the Torneträsk Basin resulted in eight stages, of which the two highest stages (T1 and T2) and the lowest stage (T8) were the least clear in their expression. Not surprisingly, therefore, identified at least five ice-dammed lake stages, which overlap in elevation with the stages T3–T7. It has been debated whether the shorelines reflect the presence of open lakes or ice-marginal lakes . The presence of shorelines on either side of the lake, with consistent elevations along the lake for the different ice-dammed lake stages, strongly supports the notion of open lake systems for IDLT stages T3 to T8, while an absence of shorelines on southern valley slopes for IDLT stages T1 and T2 supports the existence of smaller ice-marginal lakes at that time. The notion of Sördalen and its canyon as an outlet for ice-dammed lakes in Torneträsk finds strong support in the literature . additionally suspected two potential outlets at the southeastern end of IDLT, but as the shorelines could not be traced to these proposed outlets, subglacial drainage is mentioned as an alternative. Neither nor mapped shorelines farther south than Jiekajärvi (not shown), while the southernmost shoreline in this study occurs 25 km farther south (than Jiekajärvi) at Alanen Kallovaara along Torneälven (Fig. d). Hence, this reconstruction shows that the lowest level of IDLT was more extensive than previously thought. Furthermore, an additional seven outlet channels could be identified and connected to the IDLT stages.

The literature on outburst floods in this region of northern Sweden is lacking , while regions farther to the south have received more research attention e.g.,. The geomorphic traces of the GLOFs of IDLT stages T6–T8, which terminated in Ancylus Lake, have not been described before. Although the Pärvie Fault has been the subject of much investigation , the cross-cutting relationship between the Pärvie Fault and the oldest raised shorelines in the Torneträsk Basin has only become evident thanks to a regional analysis of shoreline gradients facilitated by recently released lidar data (Fig. ).