Evidence for unexpected net permafrost aggradation driven by local hydrology and climatic triggers

Rapid rates of high latitude warming over the past century have led to widespread research on permafrost thaw and its consequences. Studies from lowland plains environments in the discontinuous permafrost zone have highlighted extensive areal loss of permafrost, largely through observations of the collapse of forested permafrost plateaus into wetland features. These low-relief environments tend to have poor drainage, which initiates runaway thaw as increased soil moisture amplifies permafrost degradation. In contrast to lowland plains, the Taiga Shield landscape features a network of lakes, wetlands, soil-filled lowlands, and forests interspersed with bedrock outcrops. With the exposed (or near-surface) bedrock in this landscape, this region may have greater terrain stability under a warming climate than the lowland plains. The hydrological complexity of the Taiga Shield may also contribute to more varied trajectories for permafrost in this landscape. We investigated land cover change and implications for permafrost in an area that typifies the Taiga Shield. We took intensive ground-based measurements of soil organic layer (SOL) thickness and frost table depth to characterize different land cover types. Archival aerial photographs and recent satellite imagery from the area allowed us to assess land cover change between 1972 and 2017. Associations between permafrost, SOL, and land cover allowed us to use land cover as a proxy for change in permafrost extent. Our results suggest that both aggradation and degradation of permafrost has occurred within the Taiga Shield landscape over this 45 year period, but interestingly we found evidence for a net increase in permafrost extent. Permafrost aggradation in this landscape seems to be driven by a combination of local hydrology and climatic triggers that lead to colder, drier soil conditions that are favourable for the development of permafrost. This study highlights the importance of considering diverse and heterogenous landscapes in the study of changing permafrost ecosystems.


Introduction
Northern regions are a hotspot for climate warming.Globally, they have experienced some of the most extreme rates of warming in the world over the past century, and have become a focal point due to the thaw-sensitive nature of permafrost terrain (Gibson et al 2021).The boreal forest is one such landscape-approximately 80% of the boreal biome is underlain by perennially frozen ground (Helbig et al 2016).While the northern boreal forest is underlain by continuous permafrost, more southerly areas are in the discontinuous permafrost zone, where discrete patches of frozen ground are interspersed with permafrost-free soils (Smith et al 2010).Areas of patchy, thin permafrost are particularly vulnerable to rapid thaw as they are on the cusp of the 0 • C threshold (e.g.Jorgenson et al 2010).
Discontinuous permafrost terrain is often characterized in the literature by rapid climate-driven thaw, typically leading to dramatic, widespread thermokarst where organic terrain is ice-rich (Kokelj and Jorgenson 2013).In many of these landscapes, thaw begets thaw; changes in surface moisture regime that result from thaw settlement introduce latent heat, requiring more energy to refreeze which accelerates permafrost degradation (e.g.Kokelj et al 2014).Once permafrost begins to degrade, edge effects from increasingly fragmented patches of permafrost are amplified by increased lateral heat flux and advection (Kurylyk et al 2016).This results in a positive feedback, accelerating rates of permafrost loss from the landscape (e.g.Baltzer et al 2014).Ultimately, the runaway thaw in these discontinuous permafrost landscapes may transform into a new state where improved drainage can dry the landscape (Carpino et al 2021, Haynes et al 2021) or alternatively, continue to expand wetland features (Jorgenson et al 2001).However, biophysical feedbacks have been identified as critical to the stability of discontinuous permafrost, and under the right conditions can result in permafrost aggradation (Shur and Jorgenson 2007).Permafrost is estimated to be in this 'ecosystem-driven' state in 29% of the permafrost region in the north (Ran et al 2021), yet evidence of recent aggradation in discontinuous permafrost landscapes has been limited to a few reports of small-scale intermittent aggradation (e.g.Jorgenson et al 2020).
The majority of studies on land cover change and discontinuous permafrost thaw occur in poorlydrained lowland environments (e.g.Jorgenson et al 2001, Quinton et al 2011, Czerniawska and Chlachula 2020), and reports of permafrost loss in the southern discontinuous permafrost zone have ranged from 10% to 50% over recent 50-60 year intervals (Wright et al 2022).In these low-relief landscapes, forested permafrost plateaus are interspersed with permafrost-free wetlands (Vitt et al 1994).However, throughout much of Canada the boreal forest is underlain by Precambrian Shield, a vast area of exposed or near-surface bedrock (Shilts et al 1987; figure 1).The terrain complexity of the Taiga Shield landscape in Northern Canada, with an abundance of bedrock outcrops, will yield greater terrain stability in the face of a warming climate.Assumptions of similar patterns of permafrost change documented in areas of lower relief may not hold true because of the shield terrain driving strong moisture gradients across the landscape.
It is well established that local hydrology plays an important role in the configuration and persistence of permafrost for both lowland plains and shield environments.Studies have shown that higher soil moisture leads to greater vertical thaw within a season (e.g.Guan et al 2010, Williams et al 2013), while improved drainage from the landscape has been shown to stabilize permafrost (Göckede et al 2017).Winter conditions play a dominant role in the annual ground heat flux and formation and persistence of permafrost, and soil moisture is a key determinant of activelayer freezeback and ground heat loss in winter (e.g.Morse et al 2016).In the shield landscape, greater topographic variability produces hydrological complexity leading to spatial and temporal variation in soil moisture conditions across the system (Spence and Woo 2006).Here, 'fill-and-spill' runoff generation can provide ephemeral water sources to lowlying vegetated peatlands (e.g.Spence and Woo 2003); moisture supply can change drastically both seasonally and interannually as contributing area size amplifies (or ameliorates) the effect of variation in precipitation.The hydroclimatic memory of Taiga Shield landscapes can produce multi-year wetting or drying trends (e.g.Spence et al 2011), which may lead to rates of permafrost thaw and land cover change that vary greatly at a local scale in contrast to the general patterns of widespread permafrost loss characteristic of lowland plains environments.
The Taiga Shield is a major ecozone of the boreal biome that is widely affected by permafrost.Southern regions of the Taiga Shield ecozone have exhibited extensive increases in forest cover, in contrast with the southern Taiga Plains where forest extent is increasing only marginally, or in some cases decreasing (Hermosilla et al 2022), motivating questions about the underlying processes driving these changes.In the North Slave region of the Taiga Shield, the ecosystemdriven permafrost is in disequilibrium with climate (Morse et al 2016, Ran et al 2021), emphasizing the importance of the vegetation-permafrost relationship and variation in soil moisture regimes (Morse et al 2015).We use this region as a case study to investigate land cover change in a discontinuous permafrost ecosystem on the shield, and the role of hydrologic conditions in driving change.In this study, we demonstrate differences in soil and permafrost conditions across different land cover types and use these relationships to ask: (1) how has land cover changed over time within the subarctic Taiga Shield landscape?; (2) what are the drivers of land cover change or stability over time?; and ultimately (3) do these changes have implications for permafrost extent?Addressing these questions will contribute to a greater understanding of the change that shield ecosystems underlain by discontinuous permafrost have experienced during a time of rapid warming and inform predictions of future change.

Site description and climate conditions
The Baker Creek watershed is an ∼165 km 2 basin located just north of Yellowknife, NT (figure 1).This area is a typical Taiga Shield landscape underlain by discontinuous permafrost, characterized by a network of streams, lakes, and ponds, exposed bedrock,

Land cover change aerial imagery acquisition and processing
To assess changes in land cover over time, we acquired imagery capturing the Baker Creek watershed from two time points: a 2017 multispectral GeoEye satellite image (DigitalGlobe, Inc, Longmont CO USA; pansharpened to 0.5 m resolution, taken 9 August 2017), and 1972 colour air photographs (National Air Photo Library, Natural Resources Canada, Ottawa ON Canada; scale 20 000; taken 23 July 1972).We cropped the photographs to approximately 2/3 of their original size to remove areas of distortion occurring near the edge of the photograph, and georeferenced them to the GeoEye satellite imagery using ArcMap version 10.6 (Esri, Redlands CA, USA).We selected 31 200 × 200 m areas of interests (AOIs): six of the AOIs overlap with long term observation sites, while the remaining 25 AOIs were randomly placed across the basin in predominantly terrestrial areas (figure 1).
We classified the land cover in each AOI into five cover types by manually tracing feature boundaries in ArcMap version 10.7 (Esri, Redlands CA, USA) in both the 1972 and 2017 imagery.The land cover classes used were dense forest, sparse forest, wetland, rock, and water.Reference photos for each of the forest classes were used to ensure consistency in manual classification (supplementary figure S1).We also modelled 2017 land cover using several DEMderived variables (e.g.elevation, wetness, aspect) and the 1972 vegetated land cover types as predictors (supplementary text S1).This allowed us to determine if parts of the landscape were more susceptible to change, and if transitions between certain land cover types had greater probability of occurring.

Land cover-permafrost associations and permafrost change over time
At the end of the 2019 thaw season (late September/early October), we visited a subset of the AOIs (19/31) to determine basic soil and permafrost characteristics associated with each of the vegetated land cover classes, and to confirm the land cover classifications made using the aerial imagery (supplementary text S2).Within each AOI, we sampled four points in a representative feature of each vegetated class (where present).In total, we took measurements in 16 dense forest, five sparse forest, and 17 wetland features.At each ground sampling point, we measured soil organic layer thickness (SOL; up to 70 cm).SOL was considered 'thin' if less than 40 cm and 'thick' if greater than 40 cm, corresponding with the definition of a peatland (Tarnocai 2009).To determine the presence of permafrost and thaw depth while accounting for small-scale variability in these characteristics, we took three replicate measurements with a frost probe (150 cm graduated steel rod) around each sampling point.Active layer thickness has not been found to exceed 130 cm across different ecotypes in this region (Morse et al 2015) making 150 cm a sufficient probe length for detecting permafrost here.The average depth to refusal was recorded, as well as the substrate hit at depth to refusal-this was classified as frozen, rock, unknown (where the substrate could not be confidently identified), or NA (depth of refusal >150 cm).Further details are provided in supplementary text S2.
We assigned assumed permafrost conditions to land cover types based on the associations found during our ground-based surveys.Our analysis only included dense forest, sparse forest, and wetland cover transitions, as these were the only land cover types in which there are implications for permafrost change.Extent change was inferred if land cover changed from a permafrost to a non-permafrost type, and vice versa.
To support these inferences, we calculated the probability that each land cover transition had implications for permafrost change (supplementary text S3).While there are some limitations to the use of land cover type in historical imagery as a proxy for permafrost conditions (e.g.not being able to confirm permafrost presence/absence prior to the establishment of extensive ground-based monitoring), the use of contemporary permafrost-vegetation cover associations to assume historical conditions has been widely applied (e.g.We then wanted to investigate two important drivers of permafrost change: climate and runoff contributing area.While warming trends are ubiquitous across northern regions and associated with permafrost thaw, 'climatic triggering events' can initiate positive feedbacks that lead to aggradation in some parts of the landscape (Jorgenson et al 2020).We accessed historic climate data records from Yellowknife (Environment Canada Yellowknife A Station, 5 km south of the basin) to assess if there were years (or periods) more favourable for permafrost aggradation-specifically anomalies (relative to the 1981-2010 normals) for Mean Winter Temperature, Total Winter Snow, and Late Season Rain.For years to be considered most conducive to permafrost aggradation, the following criteria must have been met: (i) all three climate variables must be below average conditions, (ii) at least one variable must be <10th percentile, and (iii) at least one variable must be <25th percentile.
We were interested in whether AOIs with large runoff contributing areas would experience extensive permafrost thaw.We calculated net permafrost change between 1972 and 2017 and mean flow accumulation (using the Multiple Flow Direction method, averaged by AOI using Zonal Statistics (ArcMap spatial analyst toolbox)).This gives a metric for areas of the landscape with the largest upstream contributing area using DEM-based flow pathways.We used a nonparametric Spearman's Rank Correlation test to assess the relationship between flow accumulation and permafrost change ('cor.test' in base R (R Core Team 2022)).

Results
We found a strong association between land cover type and permafrost presence (figure 2).Wetlands were generally characterized as permafrost-free with thick SOL.SOL was almost always thick in sparse forests, which were typically underlain by permafrost.While dense forests could be found with variable belowground conditions, when SOL was thick permafrost was likely to be found.The probability of finding permafrost in each land cover type is described in supplementary table S1.These cover types showed interesting patterns in change over time; we found an increase in dense forest cover between 1976 and 2017 across the 31 AOIs studied, with net decreases in both sparse forest and wetland cover while rock and water remained stable (figure 3, supplementary figure S2).Transitions of sparse forest to and from wetland cover were significantly probable, and there was a high likelihood of sparse forests becoming dense forest through infilling (supplementary text S1, figure S3, table S4).2017 sparse forest cover points were generally found on gentler slopes, and sparse forest points that remained unchanged between 1972 and 2017 were most likely in gentler, more north-facing slopes (supplementary figure S4).Wetlands were found in areas with gentler slopes and at lower elevations (supplementary figure S4).
The distinct trends in SOL thickness and substrate at depth of refusal among the different vegetated land cover classes (figure 2) allowed us to use landcover change to infer shifts in permafrost extent over time (supplementary table S2 summarizes the probability of permafrost change for transitions between land cover types, and Supplementary Text S3 outlines uncertainty and assumptions made in this process).Figure 4 describes the outcomes for permafrost under each possible transition between vegetated land cover types.Transitions from wetland to both sparse and dense forest suggest areas of permafrost gain between 1972 and 2017, while transitions to wetland from both forest cover types represent loss of permafrost (figure 4).Ground-truthing at sites that transitioned from wetland to sparse forest confirmed the presence of permafrost in these forest features when permafrost was unlikely to have been present in the wetland state observed in 1972 imagery given the absence of permafrost in any of the wetland features sampled in 2019.Permafrost has likely persisted where cover has shifted from dense to sparse forest, sparse to dense forest, or sparse forests have remained unchanged (figure 4).The uncertainty and assumptions made in our probability of permafrost change outcomes were accounted for under different scenarios in supplementary table S3, which further supported the inferred permafrost changes in figure S4.From the land cover transitions we documented, there was a net areal permafrost gain of 4.3% between 1972 and 2017 (figure 5).Transitions between vegetated cover types and water were also considered, but we could only assume that shifts between sparse forest and water would have implications for permafrost change.These transitions accounted for <0.01% of the area studied, and therefore did not contribute to overall trends in permafrost change.
We found that contributing area (mean flow accumulation) was an important predictor of permafrost dynamics.At the AOI scale, we found a strong correlation between flow accumulation and permafrost extent change; permafrost loss was greatest in AOIs with high flow accumulation (figure 6; p < 0.001, Spearman's rho = −0.644).We also identified periods in the climate records that were more conducive to permafrost aggradation.Several years of anomalous cold temperatures, low snowfall, and dry late summer months occurred either during or leading up to our land cover change analysis period  which could have contributed to land cover changes and permafrost aggradation : 1962, 1964-67, 1979, 1982-83, and 1994 (supplementary figure S5).

Discussion
In contrast to most of the literature on permafrost change, our study suggests net permafrost gain across the landscape over a 45 year period.Our field data collection established strong linkages between landcover type and permafrost presence, and analyses of historical imagery indicate a net increase in landcover types that were strongly associated with permafrost.This trend appears to be driven by periods of ideal climatic conditions in tandem with shifts in local hydrology.These conditions favourable for permafrost aggradation have likely offset local thaw leading to net permafrost gain within this Taiga Shield ecosystem.

General trends in permafrost change
The increase in forest cover and associated permafrost aggradation that we observed contrasts with much of the literature on permafrost change.Reports of extensive thaw are widespread; however, these studies may be limited in representing the heterogeneity of landscapes in which permafrost can be found.Discontinuous permafrost in lowland landscapes has been intensively studied across the circumpolar north, in the Taiga Plains ecozone in the southern Northwest Territories (e.g.Baltzer et al Figure 2. Soil organic layer (SOL) thickness above each of the four substrates identified while frost probing-frozen soil, rock, unknown (where the substrate could not be confidently identified), and NA (depth of refusal was greater than the 150 cm frost probe).SOL depth above each substrate is shown for the three vegetated land cover classes to visualize the associations between SOL depth (thick vs. thin), land cover class, and presence of permafrost.Square points show the mean, while circles show measured values (note there is some overlap of points).Dashed line at 40 cm indicates the threshold for 'thick' SOL.Number of samples (n) in each category is given above each box plot.a gap in studies of thaw-driven land cover change in Taiga Shield landscapes and proposed that thawdriven change may be more limited in this ecozone.Our study suggests that while thermokarst is observed within the Taiga Shield landscape, for example lithalsa collapse (figure 6(b)), evidence of aggradation across our sites outweighed degradation (figure 5).Given the focus on permafrost-driven land cover change in lowland discontinuous permafrost systems in the  2. PF changes with probability (P) >0.8 were considered high certainty.PF changes labelled 'likely' are by far the most probable, but with moderate certainty where P is between 0.5 and 0.8.All probabilities are reported in supplementary table S2. ( 1) Wetlands have thick organic soils (SOL) and no PF detected, and the majority of dense forests with thick SOL have permafrost present, therefore a transition from wetland to dense forest suggests permafrost gain while dense forest to wetland indicates permafrost loss (probability of PF change (loss or gain; P ∆PF = 0.89); (2) Sparse forests generally had permafrost, so a transition between permafrost-free wetlands to sparse forest cover suggests permafrost gain, while sparse forest to wetland transitions are evidence of permafrost loss (P ∆PF = 0.96); (3) Almost all sparse forests have thick organic soil underlain by permafrost, and almost all dense forests with deep organic soil have permafrost present.Organic soils accumulate a few centimetres over the course of a century, so a transition between sparse forest cover and dense forest cover has most likely occurred in areas of thick organic soil with persistent permafrost (probability of PF stable P = 0.72); (4) Areas where wetland cover did not change indicates the persistence of permafrost-free terrain (probability of no PF stable P = 0.91); (5) Where sparse forest cover did not change, the permafrost has persisted (probability of PF stable P = 0.89); (6) Due to the variability in SOL thickness and substrate within the dense forest cover class, we cannot make conclusions about the implications for permafrost in areas where dense forests have not transitioned to another vegetated cover class.Tree graphics from Natural Resources Canada, Canadian Forest Service.literature, our results highlight the need to consider landscape heterogeneity across the north, and the differences landscape and climate variability may have on the spatial distribution of water that affects permafrost resilience.

Conditions for permafrost aggradation
Permafrost resilience is driven by complex interactions of climate, vegetation, and geology (Shur and Jorgenson 2007, Jorgenson et al 2010).Dry organic soils before freeze-up, cold winters, and shallow snowpack are conducive to the aggradation or maintenance of permafrost (e.g.Morse et al 2016, Jorgenson et al 2020).However, permafrost aggradation is not ubiquitous across the organic soil-filled valleys that characterize the Taiga Shield landscape in our study, despite all having experienced comparable climatic conditions.We provide evidence that variation in hydrological conditions can make certain parts of the topographically complex shield landscape more responsive to climatic conditions that promote ground cooling, as we found that the variation in permafrost aggradation inferred through land cover change associated with permafrost aggradation is inversely related to flow accumulation patterns (figure 6).
Two examples of how local flow accumulation may work in conjunction with climate to enhance or diminish permafrost resilience are shown in figure 7.Under cool, dry climate conditions, forest or wetland features with a smaller contributing area will have more pronounced soil drying than features accumulating the runoff from a large contributing area (figure 7(a)).These smaller catchment areas should be more responsive to hydrological shifts; during periods of low precipitation, the drying of organic soils and associated decrease in latent heat will promote more rapid ground freezing, conducive to aggradation or maintenance of permafrost (Jorgenson et al 2020).The formation of permafrost plateaus in wetlands then provide a solid foundation with improved drainage for trees to establish upon.In warmer, wet climate conditions, features with larger contributing areas will either sustain wet soil conditions or experience greater increases in soil wetness because of the greater runoff and associated advective heat transfer, which is conducive to permafrost degradation (Phillips et al 2011;figure 7(b)).Lowlying areas with larger upstream contributing areas are more likely to sustain saturated soils, making them (a) Wetland in a small contributing area: in a warm wet year, land cover and frost conditions would remain unchanged.In a cool dry year, thermal offsets between air and soil temperature could increase and initiate permafrost formation and feedbacks leading to persistence of permafrost and forest establishment.(b) Forest on permafrost (PF) in a large contributing area: a warm wet year could lead to permafrost degradation through advection of latent heat from high drainage off the bedrock, while permafrost and forest cover would be stable in cool, dry conditions.Tree graphics from Natural Resources Canada, Canadian Forest Service.
unfavourable for permafrost aggradation regardless of whether air temperatures are favourable.The 'filland-spill' characteristics of this hydrological system can lead to ephemeral connectivity along a drainage channel network when storage capacity of upstream catchment segments is not exceeded (Spence et al 2010).Anomalous climatic conditions can trigger multiyear changes in ephemeral hydrological connections that further amplify the effects of dry climate and contributing area on soil moisture (Spence and Rausch 2005); this plays an important role in when and where conditions favour formation or maintenance of permafrost, and should be investigated more broadly in shield permafrost ecosystems.
Aggradation has been reported locally in Central Alaska by Jorgenson et al (2020), where cold, low snowfall triggering events supported by ecological feedbacks led to permafrost formation in a discontinuous permafrost landscape over a similar timescale to our study; however, in more recent site imagery this has degraded.Interestingly, this aggradation occurred in a fen system, which contrasts with our findings of permafrost aggradation in areas of low flow accumulation.Climatically, MAT near this site (Fairbanks, Alaska, 2.4 • C; www.akclimate.org) and at the site where much of the work on permafrost in the Taiga Plains has occurred (Fort Simpson, NT, −2.8 • C; www.climate.weather.gc.ca) are both warmer than our Taiga Shield site (Yellowknife, NT, −4.3 • C; www.climate.weather.gc.ca).The colder temperatures in conjunction with the shield hydrological regime make exceeding thresholds for permafrost aggradation across parts of the landscape more likely than in the well-studied lowland environments of Central Alaska and the Taiga Plains.Further studies in other heterogenous shield landscapes will help establish how widespread aggradation processes are in shield landscapes, and pinpoint climate and hydrological thresholds conducive to permafrost formation.

Vegetation and permafrost relationships
Permafrost aggradation, driven by climatic trigger years and local hydrological conditions, is associated with an increase in forest cover in the landscape (figure 3).This trend contrasts with widespread reports of thaw-driven boreal forest loss at regional and site scales which are largely attributed to thermokarst (e.g.Thie 1974, Jorgenson et al 2001, Payette et al 2004, Carpino et al 2018).Our model suggests a high likelihood of sparse forests infilling to dense forests and significant probabilities of sparse forest establishment in wetlands as well as sparse forest transitioning to wetland (supplementary figure S3), although there was a greater extent of forest establishment than loss (figure 2).Sparse forests are strongly associated with areas that are generally flat or have gentle north-facing slopes (supplementary figure S4)-areas shown to best support permafrost in some discontinuous permafrost landscapes due to the reduced incoming solar radiation (Viereck et al 1986).The strong associations we found between land cover types, soil characteristics, and permafrost (figure 2) make us confident in the use of land cover transitions to infer permafrost change (figure 5) in this Taiga Shield landscape.
In addition to acting as proxies for permafrost change, the ecosystem properties of treed land cover classes can play an important role in permafrost resilience (Shur and Jorgenson 2007).Climatic trigger years may initiate aggradation in areas where flow accumulation is low by cutting off ephemeral hydrological connections for a multi-year period.As taller vegetation such as trees establish during this time, shading also serves to cool the soil (Jorgenson et al 2020).This 'ecosystem-driven' permafrost (Shur and Jorgenson 2007) encompasses 29% of permafrost area in the circumpolar north (Ran et al 2021).However, vegetation-permafrost relationships are variable within the broad discontinuous permafrost zone-while forest establishment in the Taiga Shield is associated with permafrost aggradation (figure 5), afforestation of bogs in lowland landscapes is occurring without frost development (Carpino et al 2021, Haynes et al 2021).This highlights the importance of the hydrological complexity and differences in MAT in driving the aggradation trends we observed.

Expectations for future conditions
Studies that have modelled change in permafrost extent in the Taiga Shield suggest a reduction in permafrost extent from 72% in the 1950s to 52% in the 2000s and predict that only 12.5% areal extent of permafrost would remain by the 2050s (Zhang et al 2014).However, the models by Zhang et al (2014) used 20 m resolution imagery from a single timepoint, and permafrost extent was hindcast and forecast without accounting for feedbacks from vegetation changes or consistently cooler/drier climatic periods.Our ground-truthed data suggest that the remotely sensed permafrost extent may have been underestimated, thus impacting forecasted change.
With climate change projections suggesting increases in temperatures, especially winter minimums, and small increases in precipitation that are highest in the winter months (Price et al 2011), the cold temperature and low snowfall triggering events to initiate formation of (or sustain) ecosystem-driven permafrost may become less common.Direct impacts of more rainfall could also lead to thaw (Douglas et al 2020).These changes could lead to a shift away from the net permafrost aggradation trend that we inferred for the period between 1972 and 2017.Though aggradation under warmer conditions may become less common, these projected drier conditions in the Taiga Shield may enhance the circumstances for the persistence of existing permafrost-particularly in topographically complex landscapes where there are stronger gradients across vegetation, soils, and changing connectivity within the hydrological network that all play a crucial role in permafrost resilience.This region is likely on a trajectory away from its current 'ecosystem-driven' permafrost type, under which permafrost can form in particular landscape conditions, toward a more 'ecosystem protected' state where aggradation is not possible (Shur and Jorgenson 2007, Ran et al 2021).

Conclusions
Our study has found compelling data supporting net permafrost aggradation in a Taiga Shield basin.This is evidenced by afforestation patterns across the landscape-strong associations between land cover type, and soil conditions (permafrost, SOL thickness) allowed us to use remote sensing images to quantify change over a 45 year period.This serves as a case study for discontinuous permafrost change in shield ecosystems, where climatic and hydrologic conditions can align to create an environment conducive to permafrost aggradation-contrary to findings from most studies of permafrost change.Heterogeneity of the shield landscape and multiyear changes in the connectivity of hydrological systems lead to finescale aggradation, which may not be captured by coarser analyses.Additionally, studies on permafrost change are largely focused on systems with dramatic thermokarst, potentially biasing the literature toward thaw-driven landscapes.This emphasizes the need for a broader representation of the diversity and heterogeneity of permafrost systems in studies of ecosystem change in the north.
This research was performed on land in the asserted territory of the Akaitcho Dene First Nations.This area is traditionally used by other Indigenous groups including the Tłı̨chǫ First Nation and NWT Métis.J B and C S proposed the research.All authors developed the methodology and contributed to the development of the manuscript.C S and A S collected data, and A S analyzed it.Thank you to Jason Paul, Niels Weiss, and Anna Coles for assistance with data collection in the field.This work was funded by Global Water Futures (GWF), the GWF Project Northern Water Futures, and Environment and Climate Change Canada.We are grateful for the support provided through the Government of the Northwest Territories-Wilfrid Laurier University Partnership.We also appreciate the helpful comments of three anonymous reviewers on this manuscript.

Figure 1 .
Figure 1.Map of the Baker Creek Watershed.Purple boxes indicate each of the 31 areas of interest (AOIs).Inset map shows the location of the Baker Creek Watershed, with brown shading representing the extent of the boreal forest on the Canadian Shield.The dotted outline shows the extensive discontinuous permafrost zone.

Figure 3 .
Figure 3.Total areal extent of each land cover class in 1972 and 2017 across 31 200 × 200 meter areas of interest (AOIs).

Figure 4 .
Figure 4. Schematic of permafrost (PF) change conditions inferred by land cover transitions and ground data from figure2.PF changes with probability (P) >0.8 were considered high certainty.PF changes labelled 'likely' are by far the most probable, but with moderate certainty where P is between 0.5 and 0.8.All probabilities are reported in supplementary tableS2.(1) Wetlands have thick organic soils (SOL) and no PF detected, and the majority of dense forests with thick SOL have permafrost present, therefore a transition from wetland to dense forest suggests permafrost gain while dense forest to wetland indicates permafrost loss (probability of PF change (loss or gain; P ∆PF = 0.89); (2) Sparse forests generally had permafrost, so a transition between permafrost-free wetlands to sparse forest cover suggests permafrost gain, while sparse forest to wetland transitions are evidence of permafrost loss (P ∆PF = 0.96); (3) Almost all sparse forests have thick organic soil underlain by permafrost, and almost all dense forests with deep organic soil have permafrost present.Organic soils accumulate a few centimetres over the course of a century, so a transition between sparse forest cover and dense forest cover has most likely occurred in areas of thick organic soil with persistent permafrost (probability of PF stable P = 0.72); (4) Areas where wetland cover did not change indicates the persistence of permafrost-free terrain (probability of no PF stable P = 0.91); (5) Where sparse forest cover did not change, the permafrost has persisted (probability of PF stable P = 0.89); (6) Due to the variability in SOL thickness and substrate within the dense forest cover class, we cannot make conclusions about the implications for permafrost in areas where dense forests have not transitioned to another vegetated cover class.Tree graphics from Natural Resources Canada, Canadian Forest Service.

Figure 5 .
Figure 5. (a) Land cover feature transition categories between 1972 and 2017, where W = wetland, D = dense forest, and S = sparse forest (e.g.WS is a transition from wetland in 1972 to Sparse forest in 2017).Area shown is the % of total vegetated land cover in the 31 areas of interest (AOIs).Not shown are the DD category (43.1% of total area) which has uncertain implications for permafrost change, WW transition category (20.6%) in which permafrost was not likely present over the 1972-2017 period, and land cover transitions that likely represent areas of stable permafrost (permafrost that persisted between 1972 and 2017) which include SD (11.3%),DS (2.2%), and SS (6.0%).(b) Total permafrost (PF) gain, loss, and net change for the vegetated land cover in the 31 AOIs between 1972 and 2017.

Figure 6 .
Figure 6.Relationship between mean flow accumulation (log-transformed weighted sum of cells in upstream contributing area) and net permafrost change within each of the 200 × 200 m areas of interest (AOIs).Point 'a' is an example of an AOI with low flow/permafrost aggradation (as seen by the tree establishment in a previously treeless wetland), and point 'b' shows an example of permafrost thaw in a high flow area.The collapse of a lithalsa (ice-rich permafrost mound) is indicated by the red arrows.Permafrost change is determined from transitions between vegetated land cover types described in figure 4. Positive values of net permafrost change represent an increase in permafrost extent between 1972 and 2017 imagery, while negative values indicate permafrost loss.Results of Spearman's Rank Correlation test are shown.

Figure 7 .
Figure 7. Two examples of how contributing area (flow accumulation) could affect permafrost and land cover changes.Arrow thickness shows how contributing area can enhance or diminish runoff into the vegetated feature under different climate conditions.(a) Wetland in a small contributing area: in a warm wet year, land cover and frost conditions would remain unchanged.In a cool dry year, thermal offsets between air and soil temperature could increase and initiate permafrost formation and feedbacks leading to persistence of permafrost and forest establishment.(b) Forest on permafrost (PF) in a large contributing area: a warm wet year could lead to permafrost degradation through advection of latent heat from high drainage off the bedrock, while permafrost and forest cover would be stable in cool, dry conditions.Tree graphics from Natural Resources Canada, Canadian Forest Service.