The bioclimatic extent and pattern of the cold edge of the boreal forest: the circumpolar taiga-tundra ecotone

Current configurations of forest structure at the cold edge of the boreal may help understand the future of ecosystem functioning in high northern latitudes. The circumpolar biome boundary at the boreal (taiga) forest and tundra interface is an ecological transition zone (taiga-tundra ecotone; TTE) experiencing changes that affect its forest structure. We accounted for the TTE’s horizontal forest structure with an estimate of its extent and pattern as represented by tree canopy cover (TCC). We quantified TCC patterns with an algorithm that describes its spatial gradient, and summarized landscape patterns of structure to represent heterogeneity, capturing abrupt, diffuse, and uniform forest at mesoscales. We used these landscape patterns to constrain the spatial extent of sparse and open canopy forest, and non-forest (forest-adjacent) edge that defines the TTE extent. The resulting map of the TTE extent is based on forest structure spatial patterns resolved at 30 m, highlights structural variability across landscapes, and helps distinguish tundra from boreal domains. We classified 14 594 landscapes as those associated with the TTE within a circumpolar bioclimatic envelope (11.575 million km2), where 44.83% of the area of these landscapes were forest and non-forest edge, yet 36.43% contributed to the TTE extent. We report the overall extent of the TTE (3.032 million km2) across North America and Greenland (53%), and Eurasia (47%), where 0.697 million km2 is non-forest edge, 0.549 million km2 is sparse forest, and 1.787 million km2 is open canopy forest. Diffuse forest landscapes dominate the TTE (79%), and abrupt landscapes (∼19%) indicate portions of the TTE where sparse forest and non-forest edge are the prevailing structural patterns. This account of the TTE quantifies the area of the cold edge of the boreal forest where previous global estimates show high discrepancies, and can help target monitoring and prediction of circumpolar dynamics.


Introduction
At the northern forest limit, the biome boundary between boreal (taiga) forest and tundra, the taigatundra ecotone (TTE; also referred to as the FTE) is recognized for its patchy gradient of woody vegetation [1,2]. Here, the structure of this vegetation is typically in the form of short trees, tall shrubs or some spatial arrangement of the two across a range of canopy cover, and at coarse scale is associated with the mean July isotherm of 10 • C-12 • C [3][4][5][6].
The general reduction in height, cover, density, and above-ground biomass in this structure across a temperature gradient (both latitudinally and altitudinally) forms a transition zone from forested to tundra landscapes at the cold edge of the boreal forest.
The current structure and arrangement of woody vegetation in this transition zone are linked to a range of biophysical and biogeochemical patterns and processes in and near the high northern latitudes (>60 • N). These include climate [7], fire patterns [8,9], below-ground carbon and permafrost stability [10][11][12][13], seed dispersal [14], stand age and carbon accumulation [15], and albedo [16,17]. Insights into the relative strengths of static (e.g. photoperiod) and shifting (e.g. air temperature) drivers of the TTE provide guidance for how TTE changes continue to be monitored [18]. Model-based predictions of changes suggest continued structural growth and productivity increases with warming temperatures where current woody structure growth in the TTE is controlled primarily by climate [19]. However, identifying where (and at which scales) such primary controls operate may be key to accurately predicting TTE dynamics [20,21]. The variability in primary controls and the likelihood and pattern of structural changes are important characteristics in and near this biome boundary because they modify ecosystem properties that govern how the land surface affects dynamics such as fluxes in trace gases, water, and energy, at seasonal, yearly, and decadal time scales [22,23]. Thus, observations of the various configurations of woody structure with remote sensing may provide a means for inferring the fate of structure itself, the biome boundary that it demarcates, and the biophysical and biogeochemical processes to which this structure is linked.
Satellite observations of ecological boundaries are often uncertain for a variety of reasons. First, they are difficult to precisely demarcate, often manifesting as a gradient of vegetation patches [24,25]. This spatial uncertainty is a prominent trait of the TTE, and as such there is an inherent scale dependency associated with its detection [1,26,27]. Second, institutional definitions of what constitutes a 'forest' often reflect lower latitude expectations for temperate and tropical forest structure [28]. For example, tree cover >30% satisfies the United Nations Framework Convention on Climate Change's conservative definition of a forest, but excludes important TTE landscapes where forests are important for landscape structure and function. Third, shrubs serve a similar biophysical role as trees in the TTE when they are tall enough to emerge above accumulated snow [29], yet are often not recognized as part of a conventional forested extent. Landsat-derived maps do not reliably distinguish between shrubs and trees, and are likely to include shrubs with sparse forest extents [30,31]. These tree cover criteria for forests, coupled with the uncertainty of identifying boundary forests from satellite data, confound the spatial delineation of this circumpolar biome boundary, potentially affecting accurate accounting of decadal-scale changes in surface albedo, permafrost carbon storage, and vegetation structure change [32][33][34].
To address these issues related to boundary detection, a hierarchical geospatial framework is useful because it links characteristics across scales [35,36]. Within this hierarchical framework, a gradient-based approach can provide a means to quantify patterns associated with gradual and abrupt changes across landscapes [37,38]. This helps to resolve gradients at relevant spatial scales, which is useful because the measurement of gradients can change with scale [39], and inappropriate scales of analysis may not resolve spatial gradients of interest.
As climate changes, there exists an urgency in accounting for forest structure gradients. Studies recognize the importance of forest ecotones when examining drivers of current structure and changes (both underway and forthcoming) [35,[40][41][42][43][44][45][46][47]. Forest structure can be represented vertically (e.g. relative canopy heights) and horizontally (e.g. canopy cover) [48]. The gradient of structure may help understand variations in growth [49][50][51][52] establishment [53], and reproductive potential [54,55]. Ground observations across the TTE suggest these gradients are a result of many factors [43,49,[56][57][58][59]. For example, abrupt gradients (rapid spatial changes) of structure may be decoupled from climate, and more closely linked to site hydrology and soil thermal properties [60,61]. The factors driving these structure gradients are associated with site history of land use, disturbance and rates of regrowth, herbivory, climate, proximity to oceans, and site conditions associated with topography, hydrology, permafrost, snow accumulation, soil, bedrock, wind, and seed availability, germination, and survival [62][63][64]. Pattern provides information on process, and current local-scale TTE structural gradients may be important for understanding the variation in primary factors controlling the TTE.
The identification of these TTE forest structure patterns at site scales across a broad domain in a standardized manner is challenging yet important. This pattern identification is a step towards understanding the overall processes of change, their causes and consequences, and the vulnerability of current structure and landscapes to shifts towards novel configurations of tree cover, density, height, deciduousness, and productivity. Earlier work [31,[65][66][67][68][69][70][71] assessed the extent of the TTE by applying a variety of gradient-and texture-based approaches to delineate structure, often recognizing the need for the consistent circumpolar application of these approaches [40,72]. A common goal of many of these efforts was to identify the extent of the biome boundary while also capturing the variability of its forest structure. Problems with scale, such as the inability to resolve sparse tree cover, have hampered understanding of this variability, and the causes and consequences of ecotone change [73]. These scaling issues also restrict the recognition of overall forest area based on tree cover criteria and satellite uncertainty [28]. Here, we present work that combines broadscale modeled climate, ecological region delineations, mesoscale landscapes, and local-scale forest structure to quantify the extent and assess forest structure patterns of the circumpolar TTE.

Deriving a bioclimatic envelope to identify the taiga-tundra ecotone
To identify the extent and pattern of forest structure in the TTE we compiled a geographic envelope spanning the climatic gradients present across warm tundra and cold boreal forest landscapes. This envelope was used to constrain the analysis of tree canopy cover (TCC) and its gradient to where these forest structure characteristics are likely to be associated with the taiga-tundra biome boundary and not forest-steppe or unmanaged-managed land cover transitions, and to account for broad scale temperature control of the TTE.
To help identify this coarse climatic extent, we used the southern limit of the Arctic tundra as defined by the circumpolar arctic vegetation map (CAVM) [74,75] to approximate the northernmost limit of trees. The CAVM treeline served as a rough spatial guide around which we extended a 'cold' domain (generally northward) and a 'warm' domain (generally southward) using the mean temperature of the warmest quarter (June-August) from the World-Clim (Version 2) set of modeled bioclimatic variables [76]. We used a temperature range >7 • C and <14 • C to incorporate landscapes known to support microsite boreal forest refugia in the 'cold' portion of the envelope, and extended the 'warm' portion southward to include most of the Hudson Plains [1]. Within both Nearctic (North America, including Greenland) and Palearctic (Eurasia) realms we used the 'taiga' and 'tundra' portions of the World Wildlife Fund's terrestrial ecoregions [77] as a secondary coarse scale mask to constrain the southern limits of the bioclimatic envelope. This mask primarily reduced the area included in this envelope in the southern Siberian Sayan and Altai Mountains, limited the domain to the coastal ranges of southeastern Alaska, and removed portions of midlatitude Ontario and Quebec. Finally, we applied a satellite-based (Landsat) water mask [78] to reduce the domain to land surface, and excluded all areas south of 50 • N. A satellite-based TCC composite map was assembled within this coarse bioclimatic envelope.

Assembling a circumpolar satellite-based tree canopy cover composite
We assembled a satellite-based TCC composite using data derived from 30 m resolution multi-spectral measurements of vegetation from Landsat to account for horizontal forest structure. We combined two contemporaneous TCC products derived from work described in Montesano et al 2016 [31], Sexton et al 2013 [28], and Hansen et al 2013 [79]. This composite nominally represented TCC for year 2010 within the TTE domain, which spans parts of the two biogeographic realms. In the Nearctic, we used TCC data that was calibrated to represent woody vegetation canopies greater than 2 m in height [30,31]. In the Palearctic, we used TCC from Hansen et al [79], which more clearly represents the woody vegetation gradient in open canopies across Siberia (figures S3, S4, S5 (available online at stacks.iop.org/ERL/15/105019/mmedia)). To temporally harmonize these data, we applied a mask to TCC from Hansen et al to exclude pixels showing forest loss before 2010. We then finalized the circumpolar composite by re-gridding (using nearest neighbor) the Palearctic TCC to match the TCC data in the Nearctic. Landsat-derived maps of vegetation structure often have high pixel-level uncertainty (~±30% TCC), tend to overestimate cover in sparse forests, do not distinguish shrubs from trees, and thus feature noisy representations of all woody vegetation often influenced by the presence of non-woody vegetation at and near the ground level [31]. Yet, the patterns formed by groups of adjacent pixels in these composites introduce important and underutilized forest structure spatial context that mitigates the average uncertainty of any single pixel alone.

Refining tree canopy cover in the bioclimatic envelope
We applied a water occurrence mask to identify tree cover adjacent to sites where standing water has occurred, and reclassified tree cover to 0 where water was mapped with a frequency of ≥1%. This assumed that tree cover is not present in areas inundated at least 1% of the time between March 1984 and October 2015 [80]. This assumption is supported by the fact that the global water estimates are based on Landsat data that are unlikely to identify water underneath forest canopy. Furthermore, infrequently inundated vegetation is often below the 2 m canopy height threshold for trees that formed the basis for determining TCC. Therefore, this step served as a refinement to our tree cover estimates near water, mitigating the tendency for the overestimation of TCC in areas dominated by non-woody vegetation.

Calculating abruptness: the spatial gradient of tree canopy cover
We calculated the local spatial gradient in TCC (abruptness) from the Landsat TCC composite across the domain to quantify the uniform, diffuse, or abrupt nature in the change of tree cover across space. The calculation uses the magnitude of TCC ranging from 0 to 100 (magnitude TCC , %) and the spatial rate of change (local gradient, 4-connected neighbors, rescaled from 0-90 • to 0-100 • ) in TCC (spatial_rate_of_change TCC ), and was calculated on a per-pixel basis as such (Equation (1)): Equation (1) provides a normalized spatial rate of change that is not sensitive to the magnitude of tree cover with a unitless range from −1 to 1. Therefore, sparse extents of tree cover with consistent values can have the same measure of uniformity as dense extents of tree cover. The result is a continuous set of values whose endmembers describe a uniform forest (abruptness TCC = −1), where there is no difference in the tree cover value of a pixel and any values of its neighboring pixels and where magnitude TCC > 0) or abrupt forest (abruptness TCC = 1, magnitude TCC = 0), indicating where a non-forested pixel is adjacent to a pixel with magnitude TCC > 0. The mid-point along this continuum describes a gradual gradient used to identify diffuse forest.

Assessing the TTE extent: classifying and quantifying TTE forest structure and pattern across pixels and landscapes
We assessed TTE extent, the extent of non-forest edge, sparse, and open canopy forest that constitutes the TTE, using the bioclimatic envelope and predominant patterns of pixel-level forest structure across landscapes. We built a 30 m scale forest structure pattern classification map to quantify the extent of the TTE and its forest structure patterns with a series of steps to stratify, classify, and combine the magnitude TCC and gradient of tree cover (abruptness TCC ) in the bioclimatic envelope (figure S1). Table 1 shows the forest structure pattern class matrix, described below, that was used to derive the 11 classes describing the pattern of forest structure for each pixel in the bioclimatic envelope.
We stratified magnitude TCC into general zones to describe the sparse portion of the TTE (1-5% tree cover), the open canopy portion (6-30% tree cover), and an Intermediate and Closed canopy extent (>30% tree cover). The 30% tree cover cutoff is a convention [20] that accommodates a variety of forest transitions in the domain, a point of inflection in the accuracy of visual estimates of TCC [81], and approximates the magnitude of pixel-level uncertainty in boreal TCC estimates [31]. Therefore, this limit was inclusive, in that it allowed our analysis to extend further into open canopy boreal forest, which can be important parts of the TTE (e.g. dense patches of microsite forest refugia, riparian bands of forest, etc.). We accounted for the more dense forest structure of the 'Intermediate-Closed' classes to retain knowledge of forest structure within the broader bioclimatic envelope that was likely not associated with the TTE itself.
We stratified the abruptness TCC values to describe six gradient classes, four of which are uniform, diffusegradual, diffuse-rapid, and abrupt. The remaining set of pixels represent the non-forest edge of forest structure where magnitude TCC = 0 and abruptness TCC = 1. This non-forest edge set was divided into 2 classes using the water mask described above, identifying non-forested pixels where water occurs (non-forest edge water ), or does not (non-forest edge land ).
We identified TTE landscapes using the proportions of their forest structure pattern classes within spatial bounds determined by hydrological basins [82]. Landscapes were bounded using level 8 (landscape) hydrobasins, and were nested within level 1 (circumpolar regions) hydrobasins, and biogeographic realms (Nearctic, Palearctic) that intersected the bioclimatic envelope of the TTE. These hydrological basins were a means to introduce basic topographic stratification to the classified pixels. The median size of the landscape hydrobasins used in this study was 388.7 km 2 (figure S2) and describe 'mesoscale' , referenced hereafter.
For each landscape intersecting the bioclimatic envelope, we created a 'landscape pattern class' by grouping its forest structure patterns according to general gradient categories. To do this, we used the pixels with a forest structure pattern class described in table 1 following the schema diagrammed in figure 1. The result was a landscape pattern classification that defined 'TTE landscapes' , whose forest structure patterns were used to derive the 'TTE extent'. We also identified a subset of these landscape hydrobasins that intersected the circumpolar arctic vegetation map (CAVM) treeline to examine the forest structure patterns across a subset of the TTE bioclimatic extent that coincides with a representation of the northernmost extent of trees, a widely used demarcation between the arctic and boreal vegetation domains.

The bioclimatic envelope that contains the taiga-tundra ecotone
The bioclimatic envelope in which the TTE exists is a large spatial margin (11.575 million km 2 ) that encapsulates portions of warm tundra and cold boreal domains (figure 2). The envelope extends across part or all of six broad circumpolar regions, and includes sites at the northernmost ranges of the boreal forest where recent studies have documented a variety of climate-induced changes to forest structure [5,59,[83][84][85][86][87][88][89][90][91].
3.2. The extent and pattern of the taiga-tundra ecotone 3.2.1. The extent and pattern of forest structure in the bioclimatic envelope Across North America (which hereafter includes Greenland) and Eurasia this bioclimatic envelope has Table 1. The forest structure pattern classification matrix showing 11 pixel-level classes formed from the combination of the gradient and magnitude of tree canopy cover. Bold text identifies forest structure pattern classes. Gradient classes are stratifications of the range of abruptness TCC and magnitude classes are a stratification of magnitude TCC . The two pattern classes for which tcc = 0 represent land adjacent to forest and indicates whether water does (non-forest edgewet), or does not (non-forest edge dry ), occur.    figure 5(a) and (b) presents a landscape pattern classification of landscapes intersecting the bioclimatic envelope. Figure 5(a) shows the area of the forest structure patterns (non-forest edge, sparse, and open canopy forest) that determine the landscape patterns and figure 5(b) maps these landscapes. The landscape pattern classification imposed a mesoscale constraint on the total extent of non-forest edge, and sparse and open canopy forest that we associate with the TTE. In the bioclimatic envelope, a total of 26 566 landscapes were classified. Landscapes associated with the TTE Figure 2. Examples of the forest structure pattern classification across the TTE. The bioclimatic envelope (center, dark grey) and five regional inset maps show forest structure pattern classes of tree canopy cover. The inset maps show classified forest pixels atop the dark grey non-forest area of the bioclimatic envelope, and highlight five regional extents across the TTE within the envelope. Areas of detail for two of the regional inset maps show where classified forest structure is overlain atop high-resolution spaceborne imagery (HRSI) from Worldview-2 at~2 m resolution. In these areas of detail, transparent non-forest areas reveal this HRSI, which is displayed in false-color where vegetation appears red, water appears dark, and non-vegetated areas appear white. The bottom row shows the unobscured corresponding HRSI extent with detailed forest patterns. (TTE landscapes, 14 594) cover 8.323 million km 2 , where 44.83% of the area of these landscapes are associated with forest ≤30% tree cover and edge. However, only a portion of these TTE landscapes were non-forest edge, and sparse and open canopy forest that contribute to the extent of the TTE, as discussed below.  The proportion of each landscape that has a forest structure pattern. This proportion explains the fraction of each landscape's pixels (within the bioclimatic envelope) that is classified as forest or non-forest edge, relative to the total area of each landscape (within the bioclimatic envelope). The CAVM treeline is marked in white for reference. Additionally, table 2 reports prominent differences between the TTE extent across the two circumpolar biogeographic realms. First, the Nearctic (North America, Greenland) accounts for the majority of the TTE extent (52.63%). This is in part due to the greater proportion of Intermediate and Closed canopies accounting for forest area in bioclimatic envelope in the Palearctic. Second, in North America, the TTE extent occupies primarily the southern portion of its bioclimatic envelope, where the Brooks Range and Canadian Shield are associated with the northernmost limits of its extent. In contrast, the TTE extent in Eurasia approaches the northern limits of the bioclimatic envelope, particularly across Siberia. Here, the coupled permafrost-larch forest system accounts for forest structure that achieves the northernmost limits of forest growth. Third, the TTE extent across TTE abrupt (edge) landscapes (figure 4) is 0.014 million km 2 in North America, 0.033 million km 2 in Eurasia, and accounts for 1.57% of all TTE extent. Yet, these landscapes are distributed conspicuously across the northern edge of the bioclimatic envelope in Siberia, and to a lesser extent the North American subarctic. A prominent subset of this extent is the forest-adjacent (but unforested) nonforest edge dry . This portion of the TTE, which helped  The corresponding map of landscape patterns of forest structure across the circumpolar bioclimatic envelope overlain with CAVM treeline as reference. This classification of landscapes based on their prevailing forest structure is a mesoscale constraint on the area of forest structure used to determine the extent of the taiga-tundra ecotone.

The variability of forest structure patterns across TTE landscapes
The variability of forest structure patterns of the TTE extent are shown in figure 6. For all 14 594 TTE landscapes they summarize the final pixel-level results, reveal differences in the forest structure patterns used to classify landscapes across each circumpolar region, and highlight the variability in forest structure representation within and between North American and Eurasian landscapes. The difference in non-forest edge dry proportions between the two circumpolar realms is the most prominent source of variability in TTE pattern. In Eurasia, 62% of TTE landscapes are 25% non-forest edge dry , in contrast to 36% in North America. For the subset of TTE landscapes that intersect the CAVM treeline, this figure shows that there is greater consistency in forest structure pattern across this portion of the circumpolar TTE than across the broader, more variable, full TTE extent. These 1061 landscapes encapsulate a portion of the TTE that is associated with particularly sparse forest structure near the northern limit of the TTE extent in the bioclimatic envelope. Of this subset of CAVM landscapes in the TTE, 818 (77%) are either TTE abrupt (edge) or TTE abrupt. The landscapes where non-forest edge dry dominate the withinlandscape pattern suggest where forest structure is most discontinuous, and where they be most uncertain.

A new account of a global biome boundary in the high northern latitudes
These results present a new account of the extent and pattern of the circumpolar TTE circa 2010, a global scale land cover transition between the boreal and tundra domains in the high northern latitudes. The TTE is contained within a broad scale bioclimatic envelope, and is constrained by mesoscale prevalence of the patterns in the magnitude and gradient of forest structure that are resolved at 30 m. It forms a broad geographic band in which forest structure patterns indicate a transition from primarily diffuse to abrupt forest gradients across the boreal-tundra biome boundary. The preponderance of landscapes with abrupt patterns of forest structure indicates the limits of woody structure, which likely include extensive dense shrublands.
These results, at a practical level, enable TTE studies to account for forest structure pattern and its spatial variation. These patterns update existing TTE spatial information that is represented with linear features (e.g. treelines) [74,75] and coarse tree cover patches [67]. The pixel-level patterns and the mesoscale landscape summaries quantify forest features in the TTE at local, regional, and circumpolar scales. This TTE representation is important because it reveals forest gradients in all directions, useful for identifying detailed east-west gradients linked to local disturbance patterns, and soil and topographic conditions, and complexities in vegetation greening [92,93]. These gradients may in part explain the causes of current vegetation patterns and help predict vulnerability to change, ultimately helping assess a variety of forest structure and tundra vegetation changes in the TTE, beyond just the northward advance of woody vegetation.
This gradient-based representation reveals and quantifies a broad biome boundary with large extents of diffuse forest structure landscapes, and identifies landscapes where abrupt forest gradients represent areas of frequent disruption of forested extents. This map shows that while these 'abrupt' landscapes do not have extensive forest structure (<20% of the overall TTE extent), the non-forested edges with sparsely forested sites may indicate priority monitoring and prediction sites to target with high resolution remote sensing [94]. These abrupt and edge subsets of the TTE will provide the spatial bounds for closely examining detailed changes in vegetation structure near the limits of vertical tree growth forms for all of its circumpolar sub-domains. Furthermore, this gradient-based TTE extent may provide useful prior information on seed availability to individualbased models that incorporate seed dispersal routines to predict TTE changes [95].
These results highlight the variability of TTE forest structure to elevate its importance for studying TTE change. This variability, quantified and illustrated in multiple ways in this study (particularly in figures 4, 5(b) and 6), is useful for identifying structurally similar TTE landscapes. Whether spatially adjacent or disparate, similar landscapes can provide a starting point for understanding the fate of different parts of the TTE, and its variability will be important prior information for tracking ongoing, and predicting upcoming changes. Divergence in the change trajectories of structurally similar TTE landscapes may illustrate differences in how forest structure responds, ultimately, to warming.

Targeting priority sites for circumpolar monitoring and prediction using the TTE extent
The TTE extent provides a mesoscale lens that can be used to target priority sites in greater detail. In the TTE, where spaceborne estimates of forest structure have shown high discrepancies [28] this study's proportional representation of forest structure patterns supports targeting of structurally uncertain landscapes in a standardized and spatially consistent manner, critical across such an extensive and heterogeneous domain [73,75], and facilitates comprehensive monitoring with stratified, high-resolution remote Figure 6. The variability of forest structure patterns within all TTE landscapes provides a comparison of these patterns between landscapes, and suggests where within-landscape forest structure may be most uncertain. These proportional bar plots show the fraction of forest structure pattern classes for all TTE landscapes (top) and for the subset of those TTE landscapes that intersect the reference CAVM treeline (bottom) for all circumpolar regions. Each landscape is represented with a single vertical bar, divided into 10 sections that correspond to the relative proportions represented by the within-landscape forest structure pattern classes, and arranged from left to right according to the longitude of the centroid of the landscape. sensing or field-based sampling of circumpolar vegetation at various scales. This support of multi-scale monitoring may aid thorough explorations into the variation in vegetation structure [94], change [96,97], flora-fauna interactions [98], climate feedbacks [17], and divergent or mixed structural and functional changes in vegetation [99,100]. We retained the pixel-level classifications to assist with high resolution targeting, and to be able to refine landscape classification.
This TTE extent can also help target high resolution predictions for exploring their variability in light of current structure patterns. Spatially-explicit individual-based forest gap models [101][102][103][104] are suited for studying important forest structure changes at the cold boreal edge because, aside from those resulting from rapid disturbances (e.g. fire), woody structure changes at these latitudes often occur at magnitude and rates that challenge spectral-based change detection methods from current spaceborne time-series [105,106]. Such models can integrate factors of change across all scales to model how the local expressions of change vary across broad extents. Yet, because these models run at high resolutions (individual tree level), they may be most effective when they are deployed with a stratified random sampling approach to build sets of predictions at various micro-sites that represent the range of ecologic conditions across the TTE. This TTE extent provides a practical means for identifying these micro-sites for high resolution modeling of likely variation in multidecadal forest structure responses across the cold edge of the boreal.
Circumpolar monitoring with this TTE extent need not be limited to forest structure. For example, these maps may be used to: (a) identify a variety of warm tundra domains and update the CAVM tundra extent, (b) provide a means for examining the interaction of permafrost and forests [107], (c) update uncertainties in canopy-snow interactions for snowalbedo feedback [108] and (d) can be coupled with paleo-environmental biomarkers [109] to build ecological identities of sub-districts within the arctic and sub-arctic. Such identities could be powerful prior input to predictions of biome boundary changes. Fundamentally, this TTE extent bounds a fairly indeterminate transitional environment to assist future studies in knowing where to look for changes across a vast domain.

Depicting the circumpolar TTE: an update to previous work, and sources of uncertainty
We updated previous work on the depiction of the circumpolar TTE by adopting a conceptual framework derived from studies that suggest forest structure patterns are relevant for examining temporal changes in the TTE [39,45,49,[110][111][112]. We note four primary guidelines of this framework as applied to forests along the cold edge of the circumpolar boreal: (a) forest structure is a fundamental feature used to define the ecotone, (b) a coupled areal-and gradient-based delineation of forest structure within the domain (instead of linear demarcation) acknowledges the indeterminate nature of the ecological transition zone itself and allows for flexible interpretation of it, (c) Mesoscale landscapes (e.g. hydrobasins) are part of a spatial hierarchy that serves as the basis for quantifying the spatial variability in the local patterns of forest structure, and representing differences across the domain.
(d) Environmental factors at broad scales help constrain the domain to a bioclimatic envelope.
These guidelines provide a useful way of partially mitigating the effects of the pixel-level uncertainty of spaceborne forest structure estimates on TTE extent estimates.
The TTE is, by nature, an uncertain zone that eludes precise demarcation. Ecotones, conceptually, have indeterminate extents, thus the area results we present may change with a different set of rules and methods for the remote sensing of forest structure. These changes may arise from uncertainty in the sparse and open canopy forests of the TTE extent also arises from a mean pixel-level uncertainty of ±30% TCC from Landsat in boreal forests arising from overestimates of some sparse cover, discontinuities from data acquired from Landsat-7 after its scan-line corrector failed in 2003, and short seasonal windows for acquiring data in the TTE in northern Siberia [31].
For these reasons, the map's non-forest edge results may provide some measure of landscape-level uncertainty of the final estimates. For example, northeastern Siberia features clustering of landscapes classified as TTE abrupt (edge) beyond known limits of tree cover. In these landscapes, in particular, the ratio of non-forest edge (dry) to actual forested area may suggest where noise in the estimate exceeds the signal of forest structure. Additional remote sensing data on surface topography and vegetation height, e.g. from HRSI or ICESat-2, will be an important complement to the magnitude and gradient of TCC for reducing large relative errors in forest structure across TTE extents at fine scales.
A detailed comparison with previous TTE delineations was not the focus of this study. Given the inherent difficulty in defining geographic extents with indeterminate boundaries, comparisons are fraught with inconsistent definitions of features of interest. We note that the TTE extent from Ranson et al [66] covers 1.9 million km 2 (below 70 • N) while this study estimates 2.335 million km 2 of forested TTE extent, and an additional 0.697 million km 2 of non-forest edge. This study's~27% increase in the forested component of the TTE extent may be due to the consideration of more area (particularly in Eurasia), and the ability to resolve smaller and sparser forested areas.
This study of a broad-scale circumpolar domain and its heterogeneous landscapes benefited from a computational platform that facilitated a multi-scale analysis, whereby pixel-level estimates were transferred up to landscapes to mitigate some forest structure uncertainty [113]. Using this platform (Google Earth Engine), we accessed locally-scaled data (30 m pixels) and analyzed a forest gradient across ecological domains, incorporating biogeographic context from the proportional representation of forest structure that emerged across landscapes. This platform helped address the need for a standardized protocol centered on the concept that the heterogeneity of forest structure across the TTE is based on estimates derived at spatial scales consistent with changes being examined [2,71,73].

Conclusions
These methods and results provide a new account of the extent and pattern of the circumpolar TTE that is contained within a broad scale bioclimatic envelope and constrained by mesoscale patterns in the magnitude and gradient of forest structure. The map of the TTE extent uses the spatial gradient of forest structure to quantify its variability across landscapes, to help distinguish tundra and boreal domains. We classify 14 594 landscapes as those associated with the TTE, covering 8.323 million km 2 within a circumpolar bioclimatic envelope (11.575 million km 2 ), where 44.83% of that area of these landscapes are forest and non-forest edge pixels, yet 36.43% contribute to the TTE extent. The overall extent of the TTE is 3.032 million km 2 across North America and Greenland (53%) and Eurasia (47%), where 0.697 million km 2 is non-forest edge, and 2.335 million km 2 is forested (0.549 million km 2 is sparse forest, and 1.787 million km 2 is open canopy forest). Diffuse forest landscapes dominate the TTE accounting for 79% of its extent, and abrupt landscapes (~19%) indicate portions of the TTE extent where sparse and non-forest edge prevails. This multi-scale account of the TTE quantifies the area of the cold edge of the boreal forest where previous global estimates often show high discrepancies, and serves as a basis for closely examining ongoing changes in the heterogeneous circumpolar biome boundary.

Acknowledgments
This work was funded primarily by NASA grants NNH13ZDA001N-CARBON, NNH16ZDA001N-CARBON and NNH18ZDA001N-TE.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors. The forest structure pattern classification and the bioclimatic envelope are available upon request as a Google Earth Engine asset, hydrobasins are available from www.hydrosheds.org. MAXAR data were provided by NASA's Commercial Archive Data for NASA investigators (cad4nasa.gsfc.nasa.gov) under the National Geospatial-Intelligence Agency's Nex-tView license agreement.