Multi-decadal degradation and fragmentation of palsas and peat plateaus in coastal Labrador, northeastern Canada

Peatland permafrost landforms, such as palsas and peat plateaus, often represent the most southern lowland permafrost occurrences in the Northern Hemisphere. While peatland permafrost is often found in continental conditions, over a thousand permafrost peatlands were recently identified along the previously understudied coastline of the Labrador Sea in northeastern Canada. The vulnerability of these landscapes to thaw is unknown but is expected to have hydrological and ecological impacts on important caribou habitat, the abundance of culturally relevant berries, and permafrost carbon storage. Using a combination of aerial photography (from 1948, 1985, 1992, 1994, and 2021) and high-resolution satellite imagery (from 2017, 2020, and 2021), we assess multi-decadal areal changes to peatland permafrost landforms at seven peatlands along the Labrador Sea coastline spanning from Red Bay (51.7° N) to north of Hopedale (55.7° N). Analyses reveal declines in permafrost extent of 33%–93% at individual sites, occurring at mean rates of 0.8%–1.5%/a. Permafrost loss was found to occur most rapidly at mixed palsa and peat plateau sites (mean rate of 1.4%/a), followed by palsa sites (mean rate of 1.2%/a) and peat plateau sites (mean rate of 0.9%/a). Patterns of permafrost loss also differed between landform types, with more complete loss of individual landforms at palsa sites and more lateral and internal loss of existing landforms at peat plateau and mixed sites. This widespread degradation of peatland permafrost over the past 28–73 years is attributed to regional warming and peatland greening. Understanding recent change to permafrost peatlands in coastal Labrador is an important step towards predicting future habitat change in northeastern Canada and will inform regional land management in areas dominated by these culturally important landforms.


Introduction
Permafrost peatlands host some of the southernmost lowland periglacial features in the Northern Hemisphere (Sollid andSørbel 1998, Hugelius et al 2020).Peatland permafrost landforms, like palsas (peat mounds with a frozen core of mineral and organic material) and peat plateaus (fields of frozen peat elevated above the surrounding peatland) (Zoltai 1972, Zoltai and Tarnocai 1975, Payette 2004, International Permafrost Association Terminology Working Group 2005), are characterized by thick layers of surface peat that insulate the permafrost.
However, studies have shown that peatland permafrost is sensitive to climate and ecosystem-associated change (Sollid and Sørbel 1998, Zuidhoff and Kolstrup 2000, Luoto et al 2004, Vallée and Payette 2007, Thibault and Payette 2009).Understanding recent changes to these landforms is important for predicting future permafrost thaw, thermokarst potential, and ecosystem change, which together can inform regional infrastructure development and global carbon stores (Hugelius et al 2014).Permafrost peatlands are also frequented for traditional activities, such as berry picking (Karst and Turner 2011, Anderson et al 2018, Norton et al 2021), hunting, and trapping (Way et al 2018), so improved understanding of these environments may inform the development of climate change adaptation strategies for cultural resources and wildlife management.
Peatland permafrost development is reported to follow a cyclic process, involving the formation of ice lenses through frost heave (Seppälä 1986, Zoltai 1993).According to this cycle, peatland permafrost degradation is natural, such that incipient, mature, and degrading or degraded palsas may be found in the same peatland at the same time (Seppälä 1986).However, recent studies throughout the North have identified widespread peatland permafrost degradation, with little to no new permafrost development or aggradation (Sollid and Sørbel 1998, Zuidhoff and Kolstrup 2000, Vallée and Payette 2007, Thibault and Payette 2009).This suggests that regional climatic and/or ecosystem conditions in many parts of the North are no longer favorable for peatland permafrost persistence, let alone permafrost aggradation (Sollid and Sørbel 1998, Zuidhoff and Kolstrup 2000, Luoto et al 2004, Vallée and Payette 2007, Thibault and Payette 2009).
Peatland permafrost is found in both continental and coastal areas of northern Canada but was previously thought to be uncommon in the northeastern region of Labrador.Notwithstanding, a recent study identified large areas of peatland permafrost along the Labrador Sea coastline (Wang et al 2023), including in regions previously mapped as having little to no permafrost (e.g.Fewster et al 2020, Hugelius et al 2020, Olefeldt et al 2021).As these landscapes have only recently been characterized in the literature, there are no previous investigations of historic permafrost change.In this study, we aim to (1) quantify changes in peatland permafrost extent at seven peatlands in coastal Labrador since the mid-20th century, (2) examine and contextualize differences in degradation and fragmentation rates between peatlands, and (3) identify climatic and/or ecosystem-associated drivers of permafrost loss.
We hypothesize that peatland permafrost in coastal Labrador will exhibit a north-south gradient in degradation rates, similar to what has been documented in Fennoscandia (Zuidhoff and Kolstrup 2000, Borge et al 2017, Olvmo et al 2020) and other parts of Canada (e.g.Payette 2004, Vallée and Payette 2007, Carpino et al 2021, Mack et al 2023).We also expect palsas to be less resilient than peat plateaus because palsas are described as taller (Zoltai 1972, Thie 1974, International Permafrost Association Terminology Working Group 2005), leading to greater surrounding snow accumulation (Olvmo et al 2020), and cover less area, making them more susceptible to edge collapse (Thie 1974).This study will contribute to local and regional adaptation strategies related to permafrost thaw and ecosystem change and complements ongoing research on the distribution, thermal properties, and contemporary characteristics of peatland permafrost in coastal Labrador (Beer et al in review, Way et al 2018, Wang et al 2023).

Study area
Labrador's Subarctic climate is strongly influenced by the adjacent Labrador Sea, which promotes low air temperatures compared to other regions at similar latitudes (Banfield andJacobs 1998, Way andViau 2015).Mean annual air temperatures (MAAT) across Labrador range from −12 • C at its northernmost tip to 1.5 • C on its southeastern coast near Forteau and L'Anse au Clair (1981-2010 climate normal) (Karger et al 2017(Karger et al , 2021)).MAAT in this region has increased by ∼1.5 • C since the late 19th century, following a pattern of gradual warming up to 1960, cooling from 1960 to 1995, and rapid warming from 1995 to present (Finnis andBell 2015, Way andViau 2015).Labrador exhibits some of the highest precipitation amounts in North America (Hare 1950, Banfield andJacobs 1998), estimated to measure up to 2700 mm a −1 in some locations (Karger et al 2017(Karger et al , 2021)).A slight increase in total annual precipitation (TAP) has been identified for Labrador from 1950 onwards, but the fraction of annual precipitation falling as snow is expected to decline because of ongoing warming (Brown andLemay 2012, Barrette et al 2020).
Surficial materials in coastal Labrador consist mostly of exposed rock or bedrock covered by glacial till or marine and glaciomarine sediments (Fulton 1995).Ecologically, much of the Labrador Sea coastline is classified as the coastal barrens ecoregion (figure 1) (Roberts et al 2006).Wetlands are found throughout Labrador, facilitated by poor drainage conditions over shallow bedrock and hardpans (Smith 2003).The distribution of permafrost in Labrador follows a latitudinal gradient, with continuous permafrost predicted to persist at high latitudes and high elevations in the Torngat Mountains, and discontinuous or isolated patches of permafrost further south (Heginbottom et al 1995).High densities of permafrost peatlands were recently identified in lowland locations along the Labrador Sea coastline (figure 1) (Wang et al 2023), and an analysis of 20 permafrost peatlands describes palsas and peat plateaus in Labrador as being small, fragmented, and irregularly shaped (Beer et al in review).Unlike the peatland permafrost found in some parts of northern Québec (Cyr and Payette 2010, Jean and Payette 2014a, 2014b), peatland permafrost in coastal Labrador is also non-forested and is instead characterized by lichens (e.g.Cladina spp., Cladonia spp., Ocrolechia spp., Cetraria spp.), low or prostrate shrubs (e.g.Empetrum nigrum, Rhododendron groenlandicum, Vaccinium uligunosum), and some herbaceous plants (e.g.Rubus chamaemorus) (Roberts and Robertson 1980).Their non-wooded nature makes them more similar to palsas and peat plateaus found in northern Europe than those found in other parts of North America (Beer et al in review, Dionne 1984, Zuidhoff andKolstrup 2005).
Seven permafrost peatlands, located along a north to south transect from near the communities of Hopedale (55.7 • N) to Red Bay (51.7 • N), were selected for this study (figure 1 and table 1).Study sites were selected from an inventory of likely permafrost peatlands in coastal Labrador (Wang et al 2023) and consist of palsa sites (n = 2), peat plateau sites (n = 3), and mixed palsa and peat plateau sites (n = 2).Sites broadly represent the spatial extent of peatland permafrost in their respective regions (figure 1).

Methods
Changes in peatland permafrost landform extents were evaluated over time at seven study sites by comparing historical aerial photographs (1948,1985,1992,1994) with recent aerial photographs and satellite imagery (2017,2020,2021).Historical aerial photographs were georeferenced to the best available source of recent imagery.Peatlands were digitized so that changes in permafrost area, permafrost fragmentation, and the proportion of complete versus lateral (i.e.along the perimeter or periphery of the permafrost landform) or internal (i.e. in the center or within the permafrost landform) loss could be calculated (Jones et al 2016).Patterns of change were analyzed in the contexts of climate and ecosystem change.

Imagery sources and pre-processing
Aerial photographs at scales of 1:5000-1:20 000, with corresponding ground sampling distances of 0.06-0.26m, were obtained from the National Air Photograph Library (Natural Resources Canada; Government of Canada) and the Newfoundland and Labrador GIS and Mapping Division (Department of Fisheries, Forestry, and Agriculture; Government of Newfoundland and Labrador) (table 1, appendix 1).Aerial photographs were scanned at the highest available resolution (Newfoundland and Labrador GIS and Mapping Division: 1200 dpi; National Air Photo Library: 2032 dpi).Historical aerial photographs were georeferenced to recent imagery for each site, which was satellite imagery from 2017 to 2021 (0.3-0.5 m resolution) for all sites except RB, which used orthorectified aerial photographs from 2021 (0.3 m resolution) (appendix 1).As permafrost peatlands were located in flat areas, georeferencing was sufficiently accurate for positioning the aerial photographs (Borge et al 2017).Images were georeferenced using 15-60 control points, using the 3rd order polynomial or adjust transformations (appendix 1).Pre-processing and georeferencing were performed using Esri ArcGIS Pro 2.9.5.

Imagery interpretation and mapping error assessment
The extents of permafrost landforms (i.e.palsa and/or peat plateau) and water bodies (i.e.lakes, ponds, rivers) were manually digitized (table 1, appendix 1).Palsa and peat plateau extents were identified from the surrounding peatland based on geomorphological interpretation of landform morphology, vegetation, and hydrology and drainage (Brown 1979).Interpretations of historic permafrost landform extents at all sites were also evaluated by stereoscopic visualization of adjacent overlapping aerial photographs from the National Air Photo Library.Recent permafrost landform extents were verified at HI, MT-PP, and RB based on images collected using handheld digital cameras or a DJI Mini 2 microdrone during field visits in Summer 2016, 2021, and 2023 (Wang et al 2023) (figure 2).Images collected during field visits provided oblique perspectives that helped to inform interpretations of permafrost landform extents at these sites.While field visits to BB1, BB2, MT-P, and SC were not possible due to their more remote locations, the authors did visit similar sites during surveys of other permafrost peatlands for prior work (e.g.Wang et al 2023).
A single mapper delineated the permafrost landforms and water bodies to ensure mapper consistency and comparability between sites.Mapping errors were assessed at MT-P (1992) by comparing total mapped permafrost area to a second estimate derived by an independent mapper (appendix 1).The total difference in area of permafrost landforms was 174 m 2 (RMSE: 28 m 2 , MAE: 18 m 2 ), corresponding to a peatland-scale mapping error of 1.4% (Borge et al 2017, Olvmo et al 2020).Conservatively, we adopted a mapping error of 2% for total mapped permafrost extents at each peatland.Georeferencing errors were also incorporated by generating maximum and minimum buffers around idealized circular representations of the total area of permafrost landforms for each site, using the georeferencing root mean square error (0.3-1.6 m) (appendix 1).This peatlandwide implementation of mapping and georeferencing errors was necessary to account for the small, highly fragmented, and irregularly shaped peatland permafrost landforms in coastal Labrador (Beer et al in review).

Data analysis of physiographic and climate and ecosystem-associated characteristics and change
Changes to permafrost landform extents were examined by comparing absolute and relative permafrost area losses and by considering the proportion of lateral or internal loss from existing landforms versus complete loss of individual landforms.Permafrost fragmentation was assessed using FRAGSTATS (McGarigal and Marks 1995), a program that assesses spatial patterns in landscape structure and that has been previously used to study peatland permafrost landforms (Beer et al in review, Mamet et al 2017).Permafrost polygon vectors were converted to rasters (pixel resolution: 0.5 m) and processed through FRAGSTATS to quantify the area, perimeter, proportion of the peatland, and mean patch fractional dimension (MPFD) of the permafrost landforms, which is a measure of shape complexity.The MPFD, combined with the total area of permafrost landforms (area of interest) and the total area of the peatland (area of influence), was then used to calculate a patch fragmentation index (PFI) for the permafrost landforms, following (Rivas et al 2022): ) .
Temperature and precipitation metrics were used to contextualize change in peatland permafrost landform extents over time.ERA-5 daily temperature and precipitation records for 1950-2022 (Hersbach et al 2020) were downloaded from the KNMI Climate Explorer (https://climexp.knmi.nl/).Daily temperature records were used to calculate MAAT, mean annual freezing degree days (FDDs), and mean annual thawing degree days (TDDs).Daily precipitation records were aggregated to TAP and total winter precipitation (TWP).TWP was calculated from December to March to capture the month with the maximum snowpack, which is usually March in Labrador (Environment and Climate Change Canada 2023).Theil-sen slope estimates of change in MAAT, TAP, and TWP from 1985 to 2021 were calculated for each study area in R.
Changes in ecosystem characteristics were investigated through trend analyses of stacks of multispectral Landsat satellite imagery (30 m resolution) from annual summer scenes from 1985 to 2021.Images were transformed into tasseled cap indices following Beer (2021) using Google Earth Engine (Gorelick et al 2017).Mean annual brightness, wetness, and greenness values were calculated for each study site for 1985-2021.Theil-sen slope estimates of change in mean annual brightness, wetness, and greenness from 1985 to 2021 were calculated for each study site in R (Frappier et al 2023).

Changes in peatland permafrost area from as early as 1948 up to 2021
Peatland permafrost area decreased at all study sites over time (figure 3 and table 2) by 33 (RB) to 93% (MT-P) (figure 3).Rates of peatland permafrost degradation varied from 154 to 1886 m 2 a −1 or 0.8%-1.5%/a(figure 4 and table 3).At MT-P, MT-PP, and SC, accelerated degradation was observed in the most recent period (1992-2017-2021), compared to the earlier period  (table 3).
The total rate of change of peatland permafrost area over time was associated with the initial area of peatland permafrost landforms (y = 0.0151 × −133.5;R 2 = 0.96; p-value = 0.0001), where sites with greater initial permafrost extents generally experienced more annual permafrost loss (figure 4(h)).However, rates of change of peatland permafrost area over time were also found to vary by landform type.For example, the smaller MT-P palsa site experienced faster degradation and greater loss than the nearby larger MT-PP peat plateau site (table 3).Analysis of the residuals of the linear fit showed that the palsa sites and mixed sites (BB1, BB2, HI, MT-P) had positive residuals (faster degradation), while two of the three peat plateau sites (MT-PP, SC) had negative residuals (slower degradation).RB, which contains a heavily degraded and dissected peat plateau that may resemble palsas, had a positive residual.The fastest degradation rates were reported for the mixed sites, BB1 and BB2 (mean rate of 1.4%/a), followed by medium rates for the palsa sites, HI and MT-P (mean rate of 1.2%/a), and the slowest rates for the peat plateau sites, MT-PP, SC, and RB (mean rate of 0.9%/a) (table 3).

Morphology of peatland permafrost loss from as early as 1948 up to 2021
Permafrost landforms at all sites became more fragmented over time, with three sites going from medium (0.4-0.6) to high (0.6-0.8) fragmentation (BB1, MT-PP, SC), two sites going from high to higher fragmentation (BB2, RB), and two sites going from high to very high (>0.8)fragmentation (HI, MT-P) (Rivas et al 2022).The MT-P and HI palsa sites, which exhibited the highest recent PFIs (>0.9) (table 2 and figure 5), experienced a greater proportion of complete loss (table 3).By contrast, the MT-PP, SC, and RB peat plateau sites and the BB1 and BB2 mixed sites demonstrated lower contemporary PFIs (0.68-0.80) and experienced mostly lateral or internal loss (figure 5 and tables 2, 3).MAATs (1950MAATs ( -2021) ) at the five study areas ranged from −3.0 • C (Big Bay, 55.7 • N) to +0.6 • C (Red Bay, 51.7 • N) (figure 6, appendix 4).All five study areas experienced similar climatic trends including gradual warming (up to 1960), cooling , and rapid warming (1995 to present), as described by Way and Viau (2015) (appendix 4).Trends in mean annual TDD and FDD mimicked this pattern, with a decrease in TDD and an increase in FDD from 1960 to 1994, followed by an increase in TDD and a decrease in FDD from 1995 to present (appendix 4).All sites experienced statistically significant increases in MAAT in the most recent study period from 1985-1994 to 2017-2021 (figure 7(a)).HI and RB experienced statistically significant increases in TAP, but no other trends in TAP or TWP were observed (figures 7(b) and (c)).Tasseled cap trend analysis from 1985 to 2021 revealed heterogeneous patterns of ecosystem change.No statistically significant trends in brightness or wetness were observed (figures 7(d) and (f)), but a significant increase in greenness was observed at six sites from 1985 to 2021 (figure 7(e)).Verdonen et al 2023), and many permafrost peatlands in coastal Labrador exhibit geomorphological characteristics (i.e.small, fragmented, irregularly shaped) that suggest high vulnerability to permafrost thaw (Beer et al in review).Thaw of peatland permafrost at the seven study sites in coastal Labrador occurred at relative rates of 0.8%-1.5%/a.These rates of permafrost loss are within the range of rates reported for similar studies in northwestern North America (Alaska, Yukon Territory, Northwest Territories, and northern Prairie provinces: 0.2%-1.Through time, peatland permafrost degradation rates are expected to increase as permafrost landforms become more fragmented (Olvmo et al 2020).This is reflected by higher thaw rates for palsas, which are by their nature more fragmented than peat plateaus (figure 5).As hypothesized, peat plateaus were more resilient than palsas (table 3), and this is consistent with comparisons between landform types from other regions, including western Canada (Thie 1974, Beilman and Robinson 2003), Alaska   In the Northwest Territories, peatland permafrost landforms found near the southern limit of the discontinuous permafrost zone were found to be more sensitive to climate, while peatland permafrost landforms further north were found to be sensitive to both climate and ecosystem changes, such as changes in local hydrology (Beilman and Robinson 2003). .By contrast, in coastal Labrador, we found that the highest relative rates of permafrost loss did not occur at the southernmost sites, but this may be attributed to the distribution of palsa versus peat plateau sites that were studied.

Drivers and ecosystem implications of peatland permafrost degradation
Peatland permafrost development typically requires colder climate conditions and thinner snow depths than peatland permafrost persistence (Sollid andSørbel 1998, Coultish andLewkowicz 2003).For example, palsas and peat plateaus found in southern Labrador (Brown 1979, Dionne 1984), as well as in other locations in northeastern Canada (Arlen-Pouliot and Bhiry 2005, Bhiry et al 2007, Cyr and Payette 2010), are thought to have formed under the cooler conditions of the Little Ice Age.Under the context of a warming climate, peatland permafrost persistence is attributed to the insulative nature of the surface peat, but regional climatic and local ecosystem changes can both impact the resilience of the permafrost as well (Borge et al 2017).
Regional climate conditions act as the primary control on palsa and peat plateau distribution (Borge et al 2017).Degradation of peatland permafrost landforms in coastal Labrador coincides with regional warming from 1880 to present (Way and Viau 2015).Despite a period of regional cooling from 1960 to 1995, the MT-P palsas and the MT-PP and SC peat plateaus degraded at rates of 0.8%-1.4%/afrom 1948 to 1992 (table 3).In other regions, continued degradation during periods of regional climatic cooling has been attributed to increased winter precipitation, leading to thicker snow packs and warmer ground temperatures (Zuidhoff and Kolstrup 2000).From 1992 to 2017-2020, rates of palsa and peat plateau degradation at MT-P, MT-PP, and SC almost doubled relative to the previous period, to 1.4%-2.9%/a(table 3).This period of accelerated degradation coincides with rapid regional warming from 1995 onwards (figure 7(a)) (Way and Viau 2015), with higher MAAT, higher TDD, and lower FDD.Long-term trends in peatland permafrost degradation in coastal Labrador thus reflect regional climate trends, aligning well with previous studies that have used peatland permafrost degradation as indicators of regional warming climate conditions (Sollid and Sørbel 1998).
Peatland permafrost sensitivity is also linked to local conditions, including peat thickness, water content, groundwater flow, and vegetation (Borge et al 2017).As peatland permafrost thaws, the habitat is expected to transition from raised, dry, lichen-rich landforms to wet, graminoid-dominated ecosystems (Camill et al 2001, Anderson et al 2018).These wetter environments are less favorable for culturally important plants and wildlife, so thaw of peatland permafrost is expected to have impacts on local biodiversity and longstanding traditional and cultural practices.Few to no trends in TAP, TWP, or summer wetness were observed at the seven study sites from 1985 to present (figure 7), despite peatland permafrost degradation and thermokarst pond development at all sites over time (appendix 3).This may be attributed to rapid terrestrialization of thermokarst ponds by sedges and mosses (Laberge and Payette 1995, Robinson and Moore 2000, Turetsky et al 2000, Zuidhoff and Kolstrup 2000, Payette 2004, Arlen-Pouliot and Bhiry 2005).Terrestrialization is most effective in shallow ponds with greater light availability (Temmink et al 2021).This process may be supported by the presence of near-surface sediments or bedrock that would limit the depth of the pond (Anderson et al 2003, Smith 2003), efficient water drainage through increased hydrological connectivity, and initially shorter permafrost landforms with lower volumes of excess ice (Beer et al in review).Greening trends detected for six of the seven permafrost peatlands reflect significant changes in vegetation (figure 7), which may be occurring through increased shrub growth both on and around permafrost landforms and through increased graminoid growth in locations where permafrost has thawed (Camill et al 2001, Anderson et al 2018).The presence of shrubs, like R. groenlandicum, Betula nana, and Vaccinium spp., is often considered to be an indicator of stable or degrading permafrost landforms versus aggrading permafrost landforms (Zuidhoff and Kolstrup 2005).Rapid establishment and growth of shrubs has been documented since the 1980 s and 1990 s in other parts of Labrador in response to warmer summer (Davis et al 2020) and winter conditions (Larking et al 2021).However, at the permafrost peatlands, it is not clear whether this increase in shrubs is attributed to climate warming or permafrost thaw, or a combination of both.Permafrost thaw itself may also be driven by the encroachment of tall, woody shrubs, which may negatively impact permafrost integrity through snow-trapping processes (Sturm et al 2001, Komarov andSturm 2023).

Limitations
This study includes seven sites, but the north to south distribution of mixed sites, palsa sites, and peat plateau sites made it difficult to contrast north-south degradation rates.Inclusion of additional southern palsa sites or northern peat plateau sites would have improved comparisons and interpretations of differences in permafrost loss rates due to geography.
Unfortunately, the distribution of peatland permafrost landform types (Wang et al 2023) and the availability of historical aerial photographs at suitable scales and coverages limited the scope of our analysis.
All seven study sites still contained permafrost landforms in the most recent period of study (2017)(2018)(2019)(2020)(2021), but there are far fewer contemporary permafrost peatlands in southern Labrador than in central Labrador (Wang et al 2023).Limited aerial photograph availability over southern Labrador may have led to survivorship bias, with the selection of especially resilient sites compared to others that may have already degraded (Beer et al in review).Ponding suggestive of historic thermokarst processes is evident on some contemporary satellite imagery in southern Labrador.For example, some peatlands near the communities of L'Anse au Clair and Forteau visited by Wang et al (2023) (appendix 2) may have previously contained permafrost landforms, but site visits to these selected peatlands revealed only seasonal frost in Summer 2021.Expanded analyses may reveal other areas of peatland permafrost that have disappeared.

Conclusion
Analysis of aerial photographs and satellite imagery indicates rapid degradation of peatland permafrost in coastal Labrador over study periods of 28-73 years, from as early as 1948 up to 2021, at rates of 0.8%-1.5%/a.Peatland permafrost degradation rates almost doubled from a historic to a more recent study period, consistent with rapid regional warming since the 1990 s (Way and Viau 2015).Sites containing palsas degraded faster than sites containing peat plateaus, and the processes of degradation differed between palsa sites, peat plateau sites, and mixed sites, with greater complete loss of individual landforms occurring at palsa sites, and more lateral and internal loss occurring at peat plateau and mixed sites.As in other regions, permafrost loss is associated with a warming climate, but it is also linked to ecosystem changes, like increased greening.This study finds that peat plateau sites, despite being found further south and in locations with higher MAATs, were more resilient to change than palsa sites.This distinction between palsa and peat plateau resilience will help to inform adaptation strategies for local and regional activities that will be affected by thaw-associated ecosystem change.

Data availability statement
The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers.The data that support the findings of this

Figure 1 .
Figure 1.Palsa, peat plateau, and mixed palsa and peat plateau study sites and selected communities in coastal Labrador superimposed on a generalized representation of Labrador's ecoregions as barrens, forest, string bog, and tundra (Government of Newfoundland and Labrador 2020).Inset map showing the distribution of permafrost in Canada (Heginbottom et al 1995) and Labrador's position in Canada.

Figure 2 .
Figure 2. Images from field visits to (a) the HI site near the community of Rigolet in Summer 2021, (b) the RB site near the community of Red Bay in Summer 2021, and the MT-PP site near the community of Cartwright in (c) Summer 2016 and (d) Summer 2023.

Figure 3 .
Figure 3. Digitized permafrost, water, and other (peatland) extents from georeferenced aerial photographs and satellite images for the MT-P: Main Tickle-Palsa study site for (a), (b) 1948, (c), (d) 1992, and (e), (f) 2020 and for the MT-PP: Main Tickle-Peat Plateau study site for (g), (h) 1948, (i), (j) 1992, and (k), (l) 2020.Tracings for other sites are available in appendix 3. Aerial photos reproduced with permission from Her Majesty the Queen in Right of Newfoundland & Labrador.Satellite images reproduced with permission from Esri World Imagery layer.

Figure 4 .
Figure 4. Change in area of palsas (square), peat plateaus (triangle), or mixed palsas and peat plateaus (circle) at (a) BB1: Big Bay 1, (b) BB2: Big Bay 2, (c) HI: Henrietta Island, (d) MT-P: Main Tickle-Palsa, (e) MT-PP: Main Tickle-Peat Plateau, (f) SC: Sandhill Cove, and (g) RB: Red Bay study sites from as early as 1948 up to 2021 and comparison of (h) annual absolute loss rate (m 2 yr −1 ) and (i) annual relative loss rate (%/a) as a function of initial permafrost area (m 2 ) for all study sites.A linear model of the annual absolute loss rate as a function of initial permafrost area (y = 0.015 11 × −133.5;R 2 = 0.96; p-value = 0.0001) is presented in panel (h) as a thick black line.Thin black lines on panels (a-g) represent cumulative error of mapping and georeferencing error while the vertical dotted lines indicate years with available aerial photographs or satellite imagery.

Figure 5 .
Figure 5. (a) Patch fragmentation indices for peatland permafrost landforms at two palsa sites (square), three peat plateau sites (triangle), and two mixed palsa and peat plateau sites (circle) in coastal Labrador from as early as 1948 up to 2021.Vertical dotted lines indicate years with available aerial photographs or satellite imagery.Gray horizontal lines indicate thresholds for medium (0.4-0.6), high (0.6-0.8), and very high fragmentation (>0.8) (Rivas et al 2022).Example of increased fragmentation over time at the SC peat plateau site from (b) 1948 to (c) 2017.

Figure 6 .
Figure 6.Mean annual air temperature and total annual and winter precipitation at (a) BB1: Big Bay 1 and BB2: Big Bay 2, (b) HI: Henrietta Island, (c) MT-P: Main Tickle-Palsa and MT-PP: Main Tickle-Peat Plateau, (d) SC: Sandhill Cove, and (e) RB: Red Bay study sites from 1950 to 2022.Mean annual air temperature and total annual and winter precipitation were calculated from the daily ERA5 reanalysis product.Vertical dotted lines indicate years with available aerial photographs or satellite imagery.Annual temperature and annual and winter precipitation records were aggregated from daily ERA-5 temperature and precipitation records (Hersbach et al 2020), downloaded from the KNMI Climate Explorer (https://climexp.knmi.nl/).
Similar north-south gradients in permafrost sensitivity have been identified in Fennoscandia and other parts of Canada (e.g.Zuidhoff and Kolstrup 2000, Payette 2004, Vallée and Payette 2007, Borge et al 2017, Olvmo et al 2020, Carpino et al 2021, Mack et al 2023)

Table 1 .
Summary of study sites and their nearest communities, coordinates, years of available imagery, and the classification of peatland permafrost landforms found at each site.

Table 2 .
Area, mapping and georeferencing errors, and patch fragmentation index for seven permafrost peatlands in coastal Labrador from as early as 1948 up to 2021.

Table 3 .
Absolute and relative loss, annual absolute and relative loss rates, and proportion of loss as complete or lateral and internal loss for seven permafrost peatlands in coastal Labrador from as early as 1948 up to 2021.