Volume 231, Issue 1 p. 94-107
Full papers
Free Access

Seasonal thaw and landscape position determine foliar functional traits and whole-plant water use in tall shrubs on the low arctic tundra

Katherine L. Black

Katherine L. Black

Department of Biology, Wilfrid Laurier University, 75 University Avenue West, Waterloo, ON, N2L 3C5 Canada

Search for more papers by this author
Cory A. Wallace

Cory A. Wallace

Department of Biology, Wilfrid Laurier University, 75 University Avenue West, Waterloo, ON, N2L 3C5 Canada

Search for more papers by this author
Jennifer L. Baltzer

Corresponding Author

Jennifer L. Baltzer

Department of Biology, Wilfrid Laurier University, 75 University Avenue West, Waterloo, ON, N2L 3C5 Canada

Author for correspondence:

Jennifer L. Baltzer

Email:[email protected]

Search for more papers by this author
First published: 28 March 2021
Citations: 5


  • Climate warming is driving tundra shrub expansion with implications for ecosystem function and regional climate. Understanding associations between shrub ecophysiological function, distribution and environment is necessary for predicting consequences of expansion. We evaluated the role of topographic gradients on upland shrub productivity to understand potential constraints on shrub expansion.
  • At a low arctic tundra site near Inuvik, Northwest Territories, Canada, we measured sap flow, stem water potential and productivity-related functional traits in green alder, and environmental predictors (water and nutrient availability and seasonal thaw depth) across a toposequence in alder patches.
  • Seasonal thaw reduced stem sap flow whereas topographic position predicted stem water potential and productivity-related functional traits. Upslope shrubs were more water-limited than those downslope. Shrubs in drainage channels had traits associated with greater productivity than those on the tops of slopes.
  • The effect of thaw depth on sap flow has implications for seasonal water-use patterns and warming impacts on tundra ecohydrology. Topographic variation in functional traits corresponds with observed spatial patterns of tundra shrub expansion along floodplains and concave hillslopes rather than in upland areas. Green alder is expanding rapidly across the low arctic tundra in northwestern North America; thus, anticipating the implications of its expansion is essential for predicting tundra function.


A significant increase in shrub cover has occurred across the tundra biome in recent decades as a consequence of climate warming (e.g. Sturm et al., 2001a; Forbes et al., 2010). Evidence of shrub expansion and increased productivity has been documented through remote sensing (Fraser et al., 2014), fine-scale mapping (Lantz et al., 2013), studies of naturally and experimentally warmed plots (Chapin et al., 1995; Elmendorf et al., 2012), and dendroecology (Myers-Smith et al., 2015). Rates of tundra shrub expansion are projected to accelerate with profound implications for tundra ecosystems, in particular through alterations in land surface characteristics and energy partitioning, which are anticipated to feedback positively to high-latitude warming and shrub productivity (Chapin et al., 2005; Bonfils et al., 2012). However, owing to the heterogeneity of shrub distribution and expansion at regional and local scales, our understanding of the future spatial extent of shrubs on the tundra under continued warming is more limited.

In order to accurately predict the spatial extent and consequences of shrub expansion we need to understand the abiotic factors constraining shrub distribution and productivity across the landscape. Shrub expansion has occurred primarily in areas that accumulate and drain moisture, including drainage channels and concave hillslopes (Tape et al., 2006, 2012; Naito & Cairns, 2011; Ropars & Boudreau, 2012). There also is evidence of stronger growth responses to warming (Elmendorf et al., 2012) and enhanced climate sensitivity (Myers-Smith et al., 2015) of shrubs in wetter sites, and summer precipitation can correlate positively with shrub growth (Wahren et al., 2005; Blok et al., 2011). These lines of evidence suggest that water availability is an important factor limiting shrub expansion. However, disentangling the influence of water availability is difficult as soil moisture can co-vary with other environmental factors that affect shrub growth, such as thaw depth and nutrient availability (e.g. Wahren et al., 2005). It is therefore possible that topographic features, moisture and nutrient availability interact to influence tundra shrub physiological function. Productivity and nutrient cycling rates may be greater in wetter landscape positions on the tundra (e.g. Chapin et al., 1988; Hastings et al., 1989; Curasi et al., 2016). In addition, channels and hillslopes often have deeper snow compared to upslope locations owing to snow redistribution (Essery & Pomeroy, 2004). This can increase nitrogen mineralization through warmer subnivean temperatures and associated snow–shrub feedbacks (Sturm et al., 2001b, 2005; Demarco et al., 2011), and water and nutrient deposition via snowmelt (Bowman, 1992). Despite evidence that resource availability is an important driver of tundra shrub distribution and expansion, the associations between shrub ecophysiology, productivity, topography and abiotic conditions remain understudied.

A valuable approach for understanding the causal mechanisms of plant distribution and productivity is through measurement of plant functional traits, which allows us to understand trade-offs that determine ecological strategies and identify those traits that are responsible for trade-offs (Reich, 2014). A recent study by Bjorkman et al. (2018) characterized circumpolar tundra plant functional traits and responses of these traits to climate. However, intraspecific variation in ecological strategies of tundra shrubs has not been assessed across topographic and/or resource gradients and could provide a mechanistic explanation for patterns of shrub expansion, facilitating predictions regarding the spatial extent of land cover change. In situ measurements of functional traits reflect both genotypic variation and plastic responses to abiotic conditions (Willson et al., 2008; Paquette et al., 2015), and thus are useful for understanding spatial heterogeneity in plant distribution and productivity along resource gradients (McGill et al., 2006; Westoby & Wright, 2006). An integrative approach that ecologists have been using recently to understand relationships between functional traits and plant water transport incorporates the measurement of multiple leaf and stem traits (Worbes et al., 2013; Apgaua et al., 2015). Water transport can be quantified through measurements of stem sap flow which is an indicator of plant water use and transpiration and is strongly coupled to plant productivity (Swanson, 1994). Stem water potential provides a measurement of plant water stress (McCutchan & Shackel, 1992), which is useful in determining whether water may limit productivity under particular conditions. Together, leaf and stem functional traits encompass morphological and physiological features that regulate individual productivity (Pérez-Harguindeguy et al., 2013). However, direct measurements of tundra shrub functional traits and water use are limited, constraining our ability to predict the ecophysiological drivers of tundra shrub expansion or the potential consequences of this for tundra ecohydrology and productivity.

In order to address these knowledge gaps, we ask the following: (1) Does shrub water use vary with topographic position and/or associated abiotic conditions?; and (2) How variable are productivity-related functional traits among individual shrubs and is this attributable to topographic position and/or abiotic conditions?

To this end, we measured sap flow, stem water potential, and foliar and whole-plant functional traits of green alder and related this to abiotic conditions across toposequences within upland shrub patches. We found that topographic position predicted variation in productivity-related traits whereas thaw depth was most important for predicting daily cumulative sap volume. Characterizing the drivers of shrub function is critical for understanding the implications of climate warming-induced changes in resource availability thereby supporting mechanistically based predictions of tundra ecosystem responses to ongoing warming.

Materials and Methods

Study species

Green alder (Alnus alnobetula (Ehrhart) K. Koch, formerly Alnus viridis) is a deciduous shrub in the Betulaceae family with a circumpolar distribution. It is expanding rapidly across the low Arctic, particularly in northern Alaska (Tape et al., 2006) and the western Canadian Arctic (Lantz et al., 2013). Green alder is a successful colonizer mainly as a result of its climatic hardiness and partnership with two root symbionts: ectomycorrhizas that are important for water and nutrient uptake (mainly phosphorous) and the formation of root nodules by Frankia bacteria that fix atmospheric nitrogen (N; Mejstrik & Benecke, 1969; Benecke, 1970; Bissonnette et al., 2014). In addition to its N-fixing abilities, green alder is able to grow rapidly after establishment owing to its clonal growth form, which leads to the formation of dense patches of shrubs 1–4 m in height. Green alder has a multistemmed growth form; the mean number of stems per study shrub was 18 ± 2.5 (SE), whereas mean canopy volume (width × width × height) was 16.3 m3 ± 2.1 m3.

We selected this species because (1) it is one of the tallest shrub species undergoing expansion in tundra ecosystems and therefore may have the greatest canopy size (volume) effects (e.g. shading, albedo, litter input, snow capture) and water use; and (2) it may have a different functional response to abiotic limitations compared to other species that have been studied in the context of tundra shrub expansion (e.g. Betula and Salix spp.) owing to its N-fixing abilities.

Study site

Trail Valley Creek (TVC; lat. 68°44′N, long. 133°30′W) is located c. 50 km NNE of Inuvik, Northwest Territories, Canada. The TVC watershed is 57 km2, underlain by continuous permafrost, and characterized by rolling hills, deeply incised river valleys and small lakes (Pomeroy et al., 1997; Marsh et al., 2008). TVC is within the forest-tundra ecotone; sparse stands of spruce (Picea glauca and P. mariana) occur in the southern part of the basin. Graminoids (e.g. Carex spp. and Kobresia spp.), lichens and low-growing woody vegetation (e.g. Vaccinium uliginosum, Rhododendron tomentosum, R. lapponicum, Betula glandulosa) are common upland taxa, whereas common taxa on moist hill slopes, valley bottoms and ephemeral drainage channels include bryophytes (e.g. Sphagnum spp.), acidic/wet-affiliated graminoids (e.g. Eriophorum spp.), green alder and willow (Salix glauca and S. pulchra; Wallace & Baltzer, 2020). Soils are cryosols consisting of an upper peat layer (0.5–50 cm) underlain by silty clay (Quinton & Marsh, 1998). The microtopography of the TVC watershed consists of mineral earth hummocks separated by peat-dominated interhummocks. Mean hummock height in green alder patches is c. 24.8 cm (Wilcox et al., 2019). Green alder at TVC generally grows within the interhummock spaces and rarely on top of hummocks (C. Wallace, pers. obs.). There is evidence of a single fire scar within the basin that supports a distinct, much taller shrub patch and fine-scale disturbance in the form of cryoturbation that supports shrub recruitment (J. Baltzer, pers. obs.). Construction of the Inuvik-Tuktoyaktuk Highway, a gravel road, reached TVC in 2016 resulting in a linear disturbance across the landscape. However, no measurements were taken within c. 1 km of the road after its construction.

Trail Valley Creek has a mean annual air temperature of −8.2°C (1981–2006; Environment and Climate Change Canada (ECCC), 2011), and mean annual rainfall and snowfall of 114.5 mm and 158.6 cm, respectively (1981–2003; ECCC, 2011). The mean air temperature during the study was 11.2°C for 2015 (11 June–22 August; range: 0.2–27.2°C) and 11.7°C for 2016 (6 June–19 August; range: − 2.5–30.1°C; P. Marsh, unpublished). Total rainfall for the same periods was 67.4 and 85.2 mm for 2015 and 2016, respectively (P. Marsh, unpublished).

Patch and individual selection

Green alder patches were selected using high-resolution orthoimagery (mosaiced aerial photographs) collected in 2004 (GNWT Centre for Geomatics, Mackenzie Valley OrthoPhoto Cached Mosaic Service https://www.geomatics.gov.nt.ca/en). Patches were defined as clusters of green alder on slopes surrounded by ‘open’ tundra free of canopy-forming shrubs. Sampled patches spanned from the top of a hill to the bottom of a slope or drainage channel (Fig. 1) and were located on S-to-SE-facing slopes, reflecting the most common slope-aspect for alder patches at TVC. Eight patches were sampled in 2015 and three were more intensively studied in 2016. Mean length along the slope for these eight patches was 116.2 m ± 14.7 m (± SE). Mean elevational change in patches was 13.12 m ± 0.90 m and mean slope was 3.5° ± 0.5°.

Details are in the caption following the image
(a) Map of the lower portion of the Trail Valley Creek (TVC) research basin indicating the locations of sap flow meter-instrumented shrubs for the 2015 (orange circles) and 2016 (blue triangles) growing seasons; yellow star, location of the TVC research station base camp. (b) Aerial image of a green alder patch in TVC (Photo: P. Marsh); topographic positions are labelled. (c) Depiction of the sample locations along this gradient. Orthoimagery used with the permission of the Government of the Northwest Territories (GNWT Centre for Geomatics, Mackenzie Valley OrthoPhoto Cached Mosaic Service https://www.geomatics.gov.nt.ca/en).

Within each patch, a transect was established from the top to the bottom of the patch and the nearest healthy green alder stem with a 4–6 cm basal diameter was selected at each of the following topographic positions: shoulder slope, back slope/foot slope, toe slope and, when present, a drainage channel (for brevity, these will be referred to as ‘top’, ‘middle’, ‘bottom’ and ‘channel’, respectively; Fig. 1). The top position was defined as the area at the top of the hill slope where vegetation cover transitioned from shrub-free ‘open’ tundra to a green alder patch. The middle position was located at approximately the half-way point of the total patch length. The bottom position was located at the toe of the hill slope where the gradient was only slightly inclined or nearly level. Drainage channels in this study are low-order, headwater drainage features that are ephemeral in nature and are located just beyond the slope bottom and run perpendicular to the patch slope.

Sap flow

We measured sap flow on 27 and 28 shrubs in 2015 and 2016, respectively (Supporting Information Table S1; Fig. 1). Sap flow was measured continuously from early June to mid-August each year using automated Heat Ratio Method (HRM) sap flow meters (SFM1 Sap Flow Meters, ICT International, Armidale, Australia). Sap flow meters were installed at the base of one stem per selected shrub (hereafter ‘instrumented shrubs’) and measured every 30 and 15 min in 2015 and 2016, respectively. The HRM sap flow meters have three needles inserted into the xylem: a heat pulse needle inserted equidistant between upstream and downstream temperature needles (Burgess et al., 2001). A short heat pulse is used as a tracer in the sap for each measurement (Smith & Allen, 1996; 40 J pulses) and the ratio of the temperature differential between the two temperature needles determines sap velocity (Vh; Burgess et al., 2001, calculated following Marshall (1958):
urn:x-wiley:0028646X:media:nph17375:nph17375-math-0001(Eqn 1)
(k, thermal diffusivity of fresh wood (2.5 × 10−3 cm2 s−1); x, distance between the heat pulse needle and either temperature needle (0.6cm); and v1 and v2, increases in temperature (from initial temperatures; °C) at the two temperature needles). Multiplying by 3600 converts Vh to cm h−1. Volumetric sap flow (cm3 h−1) was derived by multiplying Vh by the cross-sectional area of conducting sapwood (Burgess et al., 2001). To quantify sapwood area, we sampled a stem cross-section at the instrumentation location on each shrub at uninstallation. The sapwood–heartwood boundary of green alders is indistinguishable in untreated wood and was determined by observing differences in stain uptake (over-the-counter betadine) in fresh cross-sections in 2016. Stem and heartwood diameter and bark and sapwood thickness were measured for each cross-section to determine sapwood area. For logistical reasons, we were not able to sample stems from instrumented shrubs in 2015. Instead, sapwood area was estimated for stems sampled during the 2015 growing season using a linear regression of 2016 stem diameter and sapwood thickness measurements (Fig. S1; R2 = 0.51, P < 0.001). ICT HRM meters have two thermistors along each needle (7.5, 22.5 mm); as a consequence of small sapwood thickness (mean = 0.98 cm ± 0.04 cm) of instrumented stems, only data from the outer thermistors were used for sap flow calculations as inner thermistors often were installed in heartwood. Conversion of sap velocities to daily cumulative volumetric sap flows was performed using Sap Flow Tool software (v.1.5.0 beta 1, ICT International). Sapwood area was consistent across topographic positions (ANOVA; 2015: F3,19 = 0.788, P = 0.515; 2016: F3,23 = 0.283, P = 0.837).

Owing to periods of missing data, growing season cumulative sap flow volume could not be calculated in either 2015 or 2016. Instead, we focused on shorter periods of complete data to characterize seasonal variation in daily cumulative sap flow. We divided estimates of daily cumulative sap volume into three periods: early, peak (i.e. seasonal maximum daily cumulative sap volume) and late season. We determined the approximate day of peak physiological activity of each shrub through visual inspection of seasonal sap flow tracings (Fig. S2). To characterize peak shrub water use, we averaged daily cumulative sap volume across a 7-d period encompassing the average day of maximum cumulative sap volume (excluding days with >1 mm of rain; SFpeak). We used a 7-d period to characterize SFpeak because the day of maximum cumulative sap volume varied by a week among individuals. For early- and late-season sap flow we averaged daily cumulative sap volumes for the 5 d after last installation (early season) and before first uninstallation (late season), a time window during which diurnal cumulative sap volume data were complete for all individuals. Seasonal differences in sap flow may be driven partly by variation in day length through the growing season. Daylight hours during the late-season period averaged 17.38 h d−1 (± 0.11 h d−1 (SE)) in 2015 and 19.03 h d−1 (±0.12 h d−1) in 2016, whereas 24-h daylight characterized early and peak periods (Time & Date AS, 2018). We therefore normalized daily cumulative sap volume for each period in a year by the average daylight hours during that period. Table S2 shows the structure and height characteristics of instrumented shrubs by topographic position.

Stem water potential

In order to investigate variation in shrub water status among topographic positions, we measured midday stem water potential (Ψstem; MPa) using a pressure chamber (Model 1000 Pressure Chamber Instruments, PMS Instrument Company, Albany, NY, USA). Midday measurements provide an estimate of maximum water stress (Williams & Araujo, 2002). To avoid destructive sampling of instrumented shrubs, noninstrumented shrubs at the patch closest to camp were selected such that three to four individuals similar in size and apparent health to instrumented shrubs were measured in each topographic position (Table S3). Ψstem was measured according to instrument protocols on three branchlets with four to six healthy leaves for each sampled individual. Ψstem was measured on sunny days with no rainfall in the previous 24 h. Ψstem was measured on 18 June, 26 July and 21 August 2015 to capture seasonal variation. 2016 measurements were limited to 9 August owing to logistical difficulties.

Frost table depth and soil moisture

At each instrumented shrub, we established a 2-m transect perpendicular to the slope with 1 m on either side of the shrub centre. Frost table depth and soil moisture were measured at 20, 100 and 180 cm along this transect. A 2 m graduated steel rod was used to measure the depth to the impermeable frozen layer below the ground surface (i.e. frost table depth). Surface soils at TVC are not underlain by rock (Wilcox et al., 2019); any impermeable layer encountered with the steel rod was assumed to be the frost table. In 2015, soil moisture was measured using a 5-cm pronged probe (Frequency Domain Reflectometry (FDR) Soil Moisture POGO; Stevens Water Monitoring Systems Inc., Portland, OR, USA). In 2016, soil moisture was measured using the same probe when the frost table was < 20 cm and combined with a 20 cm probe (FieldScout Time Domain Reflectometry (TDR) 300 Soil Moisture Meter, Spectrum Technologies Inc., Aurora, CO, USA) when the frost table was > 20 cm to characterize soil moisture conditions throughout the seasonally thawed layer. Probe-specific soil moisture calibrations using TVC soils yielded an accuracy of 0.5 m3 m−3 for either mineral or organic soils (Wrona, 2016). Paired frost table and soil moisture measurements were made on 30 June – 3 July and 19–22 August 2015, and 28–29 June, 13–19 July and 6 August 2016. Soil moisture was measured on days with no rainfall in the previous 24 h.

Soil inorganic nutrients

In order to characterize soil inorganic nutrient availability, we installed four Plant Root Simulator (PRS) Probes (Western Ag Innovations, Saskatoon, CA, USA) in early June 2016 at each instrumented shrub 20 cm and 180 cm along the 2-m transect. At each location, cation and anion probes were inserted into the top 10 cm of the soil. Although PRS probes measure availability of a suite of soil ions, we were interested in ammonium (NH4+), nitrate (NO3), phosphorous (P; H2PO4 and HPO4−2) and potassium (K+) as they represent the primary plant macronutrients. PRS probes were removed mid-August (mean burial length = 72 d), washed with deionized water, packaged individually and shipped in a cooler to Western Ag Innovations for analysis.

Functional traits

Leaf nitrogen (N) is considered a proxy for both soil nutrient availability (e.g. Ordoñez et al., 2009) and photosynthetic potential in plants (e.g. Wright et al., 2004) whereas leaf carbon concentration relates to leaf toughness (Enriquez et al., 1993). To characterize these traits, five healthy, fully expanded leaves from each instrumented shrub were harvested in mid-August 2015 and mid-July 2016. 2015 leaf N was analyzed using an elemental analyzer (2400 CHNS Analyzer; PerkinElmer, Guelph, CA, USA). Acetanilide standards yielded an average relative error of 0.003 % %−1. In 2016, samples were analyzed at the University of California Davis Stable Isotope Facility (UCD SIF; Davis, CA, USA) using an elemental analyzer (Elemental Micro Cube, Elementar Analysensysteme GmbH, Langenselbold, Germany) interfaced to a continuous flow isotope ratio mass spectrometer (Sercon 20-20 IRMS, Sercon Ltd., Cheshire, UK) to characterize both leaf elemental concentrations and carbon isotope composition (see next paragraph). Mass values for elemental totals of a peach leaf reference yielded an average relative error of 0.009 μg μg−1 for N and 0.004 μg μg−1 for C.

In order to investigate variation in water use strategies with topography, we measured leaf δ13C and leaf area-to-sapwood area ratio (LA : SA) of instrumented shrubs. Stomatal closure results in greater proportional uptake of the heavier 13C isotope due to reduced discrimination between isotopes during drawdown of 12C in the internal CO2 (Farquhar et al., 1982). The isotopic ratio thus represents water use efficiency (WUE) integrated over the leaf lifetime (Farquhar et al., 1989). The same leaf samples that were sent to UCD SIF for elemental analysis also were analyzed for 13C : 12C ratios. The ratios were expressed as δ13C against a Vienna PeeDee Belemnite standard, where:
urn:x-wiley:0028646X:media:nph17375:nph17375-math-0002(Eqn 2)
δ13C of peach leaf references yielded an average relative error of 0.001 μg μg−1.

Plants often respond to drought by decreasing their LA : SA to increase WUE (Poorter et al., 2009). We therefore determined LA : SA to understand the balance between transpirational surface area and stem water supply to leaves (Waring et al., 1982). Following removal of sap flow meters, all leaves were removed from instrumented stems to estimate total leaf area. Owing to the large number of leaves, fresh leaf scans for all leaves were not feasible. Instead, we developed a shrinkage factor (e.g. Blonder et al., 2012) between fresh and dry leaf area using 400 fresh leaves. Fresh leaves were scanned for leaf area, oven-dried for 48 h at 70°C, and re-scanned to determine dry leaf area and the relationship between fresh and dry leaf area (Fig. S3; R2 = 0.92, P < 0.001). All remaining leaves were oven-dried, scanned in batches, and the above-mentioned relationship used to correct for leaf shrinkage. The fresh area of each batch was summed for each instrumented stem to obtain total leaf area. Leaf area was measured using WinFOLIA (v.2004a; Regent Instruments Inc., Québec, CA, USA). LA : SA was calculated by dividing total fresh leaf area by stem sapwood area.

We measured leaf mass per area (LMA) to characterize leaf thickness and density (Niinemets, 2001). Leaves were weighed and average shrub canopy LMA was calculated by dividing total stem leaf dry mass by total stem fresh leaf area.

Data analysis

All continuous predictor variables were standardized to eliminate scaling issues (subtracted the mean and divided by the standard deviation for each observed value of a variable). Statistical analyses were performed in R v.3.6.2 (R Core Team, 2013).

Q1: Does shrub water use vary with topographic position and/or associated abiotic conditions?

In order to interpret shrub ecophysiological responses to abiotic conditions across a toposequence, it is necessary to characterize variability in abiotic conditions among topographic positions. As such, we used a principal components analysis (PCA; ‘rda’ function in R/vegan (Oksanen et al., 2019)) with 2016 data that included the following abiotic variables: soil moisture (at 5 cm and 20 cm depths), frost table depth, leaf N (proxy for inorganic and organic soil N availability), soil NH4+, soil NO3, soil P and soil K+ (note that the 2015 abiotic variables were limited to soil moisture at 5 cm depth, frost table depth and leaf N). We retained sufficient principal components (PCs) such that they cumulatively explained >80% of the variation in the data. To determine if topography explained multivariate patterns in abiotic conditions, we categorized sampling locations (individual shrubs) by topographic position and overlaid confidence ellipses (95%) onto the PCA. If confidence ellipses showed separation among topographic positions, we used ANOVA to test for topographic differences in PC scores.

In order to test for differences in Ψstem among topographic positions at different points in the growing season, we used a linear mixed effects (LME) model with an interaction between topographic position and measurement period (‘lme’ function in R/nlme; Pinheiro et al., 2017). We used pre-planned contrasts to test for differences in Ψstem between topographic positions (top vs middle, bottom, channel; middle vs bottom, channel; and top vs bottom) and measurement periods (mid- vs early-season; mid- vs late-season). The planned comparisons approach avoids making every possible combination of comparisons and instead focuses on a few scientifically sound comparisons that reflect the study design and associated predictions (Crawley, 2012). Making a more limited number of comparisons increases the statistical power of each comparison (family-wise error is not likely to be a concern) and avoids ‘fishing’ for statistically significant results which is often an implicit factor in post hoc comparisons (Saville & Rowarth, 2008). Our a priori topographic contrasts were based on an expected gradient in growing conditions from the top of the slope to the channel, with the expectation that the top of the slope would be the most water-limited, and the bottom of the slope would be the optimal growing location along this gradient (ample moisture without the potential waterlogging effects that the channel position may impose). Comparisons therefore were made between (1) what we expected would be the most stressful position for a shrub from a water availability perspective (top of slope) and all downslope locations combined, (2) the second most resource-limited position for a shrub (middle of slope) and all downslope locations, and (3) the most resource-limited position for a shrub (top of slope) and the least resource-limited position (bottom of slope). Pre-planned contrasts between measurement periods (early, peak, late) used peak season as the reference against which early and late were compared. We used ‘individual’ as a random effect to account for repeated measurements through the growing season. To compare shrub water status between years, we compared late-season Ψstem using an ANOVA with topographic position, study year and their interaction as predictors. We used pre-planned contrasts to test for differences in mean late-season Ψstem among topographic positions as outlined above. Model assumptions of homoscedasticity and normality were verified visually.

We tested the relative importance of water availability, nutrient availability, and topographic position on mean daily cumulative sap volume (SFdaily) averaged over 7-d periods that corresponded with the timing of paired frost table depth and soil moisture measurements as indicated above. We used 7-d periods as it took a week to perform abiotic measurements for all shrubs. These 7-d periods were categorized into three time periods for each study year: early-, mid- and late-season. Using SFdaily as a response variable allowed us to determine which abiotic variables were important drivers of shrub water use. SFdaily was modelled through candidate sets of LME models (one set of candidate models per growing season; Tables S4, S5). Candidate models represented the individual effects of water availability, nutrient availability, topography and their combined effects on SFdaily. We calculated AICc values (adjusted Akaike’s information criterion for small sample sizes; Burnham & Anderson, 1998) for each candidate model and ranked them using R/AICcmodavg (Mazerolle, 2017). The function ‘modavg’ was used to model-average the estimates of parameters in candidate models. The parameter estimate for a single variable can vary depending on the model in question; model averaging allows the variable estimates within each model to be averaged by giving equal weight to each model (regardless of the fact that AICc values show that some models are stronger than others). The model averaging technique therefore averages the fits for a number of models with many terms instead of picking a single best model based on potentially inflated estimates. The result is a model with the strongest predictive capability (Mazerolle, 2017).

We tested for collinearity between continuous predictor variables using a Pearson’s correlation matrix and excluded collinear variables that exceeded a correlation coefficient of r = 0.3 (Tables S6, S7), a conservative threshold that acknowledges the variability inherent in observational ecological data. We tested for collinearity between continuous predictors and topographic position through visual inspection of boxplots. Because no obvious trends were present, topographic position was included as a categorical predictor in models to represent topography-induced variation in SFdaily through variables not directly measured (e.g. topographically driven differences in snowmelt or soil characteristics). We used pre-planned contrasts to test for differences in SFdaily among topographic positions (as above). Study years were analyzed separately owing to different levels of nesting and replication between years. The random effect for both years was ‘individual’ nested within ‘patch’ to account for our spatially nested sampling design (‘patch’) and repeated measures through the growing season (‘individual’). For both years, SFdaily was not normally distributed and model residuals showed unequal variances. To meet model assumptions, SFdaily was square-root transformed for 2015 and log-transformed for 2016.

Q2: How variable are productivity-related functional traits among individual shrubs and is this attributable to topographic position and/or abiotic conditions?

In order to investigate how green alder functional traits relate to resource use, we used a PCA following methods described for Q1. This PCA included the following functional trait variables collected in 2016: leaf N, leaf C, LMA, LA : SA, δ13C and SFpeak. Similar to statistical methods described for Q1, we categorized data points by topographic position to determine if topography explained variability in resource use strategies. We used ANOVA to determine if PC scores significantly differed among topographic positions.


Q1: Does shrub water use vary with topographic position and/or associated abiotic conditions?

Variability in abiotic conditions among topographic positions

Data reduction using PCA resulted in five axes explaining variation in abiotic conditions along the topographic gradient in shrub patches in 2016. These five PCs cumulatively explained 85.4% of the variation in abiotic variables investigated (Table S8). PCs 1 and 2 together represented 46.7% of variation in abiotic conditions (Fig. 2). Based on the magnitude and directionality of the abiotic variable loadings for these two PCs, PC1 mainly represented measures of surface soil moisture and potassium availability whereas PC2 represented frost table depth, phosphate availability, leaf N and soil moisture at 5 cm. Overlaying topographic position on the PCA biplot demonstrated that variation in abiotic conditions represented by the first two PCs did not change as predicted along a toposequence. Notably, the top and channel positions overlapped considerably. However, abiotic conditions experienced by shrubs in middle positions did separate from those of shrubs in top and channel positions. Subsequent ANOVA confirmed that the values of both axes differed significantly with topographic position (Fig. 2; PC1: F3 = 5.66, P = 0.004; PC2: F3 = 3.02, P = 0.049).

Details are in the caption following the image
Principal components analysis biplot from an analysis used to characterize variability in abiotic conditions among topographic positions. Abiotic variables include frost table depth (FTD), soil moisture at 5 cm depth (SM5), soil moisture at 20 cm depth (SM20), soil nitrate supply rate (NO3), soil ammonium supply rate (NH4), soil dihydrogen phosphate and hydrogen phosphate supply rate (P), soil potassium supply rate (K) and leaf nitrogen concentration (Leaf N).

Differences in Ψstem among topographic positions at different points in the growing season

In 2015, Ψstem significantly differed across the toposequence such that downslope locations had less negative Ψstem values than upslope locations (Table S9). Ψstem also was significantly less negative early in the 2015 season relative to mid-season and more negative in late season relative to mid-season (Table S9). There was no significant interaction between topographic position and measurement period in 2015 (Fig. 3a; Tables S9, S10). The lowest Ψstem experienced by a shrub (maximum water deficit) during 2015 was −2.4 MPa and the highest value (minimum water deficit) was −0.28 MPa.

Details are in the caption following the image
(a) Mean midday stem water potential (Ψstem) of green alders (± SE) measured early, mid- and late season in 2015. (b) Mean (± SE) late-season Ψstem in 2015 and 2016. See Supporting Information Tables S9S11 for complete results of pre-planned contrasts).

Late-season Ψstem significantly differed between years, with shrubs having significantly more negative Ψstem in 2015 than 2016 (Table S11). Green alders on the tops of slopes had significantly more negative late-season Ψstem than downslope locations in 2015 but not in 2016 (Fig. 3b; Table S11). Late-season Ψstem ranged from −2.4 to −0.95 MPa in 2015 and −1.2 to −0.63 MPa in 2016.

Relative importance of water availability, nutrient availability, and topographic position on mean daily cumulative sap volume

Peak daily cumulative sap volumes occurred between 22–28 June 2015 and 9–15 July 2016. Early-season daily cumulative sap volume was calculated for 13–17 June 2015 and 9–13 June 2016. Late-season daily cumulative sap volume was calculated for 15–19 August 2015 and 11–15 August 2016. Variations in daily cumulative sap volume among these periods and among topographic positions for each year are shown in Fig. S4.

Variation in SFdaily in both study years was best characterized by candidate models representing frost table depth (Tables 1, 2). Frost table depth explained 39.2% and 3.5% of the variability in SFdaily in 2015 and 2016, respectively. It is important to note here that the estimates for the random variance in 2015 were very close to 0 for the frost table model, suggesting issues with model singularity. It is possible therefore that the model fit discrepancy between years may be due in part to incorrect estimation of the random term due to a small number of levels (n = 7).

Table 1. Rankings for candidate linear mixed effects models predicting 2015 green alder daily cumulative sap volume (SFdaily).
Model K AICc Δi LogLik wi R 2 m R 2 c
M2 Frost table depth 5 171.69 0.00 −79.85 0.96 0.392 0.392
M5 All ‘nutrient’ variables 8 178.38 6.69 −78.52 0.03 0.438 0.440
M0 Null 4 186.52 14.83 −88.61 0.00 0.00 0.00
M1 Soil moisture 5 187.32 15.63 −87.66 0.00 0.053 0.053
M6 Global model 12 190.93 19.24 −76.68 0.00 0.502 0.521
M3 Topography 7 194.51 22.82 −88.26 0.00 0.020 0.020
M4 Topography × Soil moisture 11 203.91 32.22 −85.45 0.00 0.165 0.165
  • Model rankings were based on Akaike information criterion (AICc) and associated statistics (Burnham & Anderson, 1998) including the difference between each model and the minimum AICc model (Δi), number of parameters per model (K), log-likelihood values (LogLik), the Akaike weights (wi), variation in SFdaily explained by fixed effects (pseudo-marginal R2; R2m), and total variation SFdaily explained by each model (pseudo-conditional R2; R2c; fixed and random effects combined). The random effect was ‘individual’ nested within ‘patch’.
Table 2. Akaike information criterion (AICc)-based rankings for candidate linear mixed effects models predicting 2016 green alder daily cumulative sap flow (SFdaily).
Model K AICc Δi LogLik w i R 2 m R 2 c
M3 Frost table depth 5 63.52 0.00 −26.15 0.84 0.035 0.830
M0 Null 4 68.03 4.51 −29.61 0.09 0.000 0.773
M1 Soil moisture 5 69.10 5.59 −28.94 0.05 0.015 0.798
M4 Topography 7 71.67 8.15 −27.64 0.01 0.132 0.800
M7 All ‘nutrient’ variables 11 73.67 10.15 −22.77 0.01 0.228 0.831
M2 Soil inorganic nutrients 7 74.55 11.03 −29.08 0.00 0.044 0.777
M6 All ‘water’ variables 12 75.24 11.72 −21.91 0.00 0.170 0.877
M5 Topography × Soil moisture 11 79.40 15.88 −25.63 0.00 0.149 0.842
M8 Full model 15 84.35 20.83 −21.02 0.00 0.228 0.862
  • See Table 1 for full description.

Across the growing season, the relationship between frost table depth and SFdaily was negative in both 2015 and 2016 (Fig. 4). This negative relationship was statistically significant for both years as model-averaged confidence intervals for frost table depth did not overlap zero (Tables S12, S13).

Details are in the caption following the image
Relationship between frost table depth and green alder sap flow (SFdaily) during 2015 (a) and 2016 (b). The dashed lines represent the overall relationship between frost table depth and SFdaily (across early, mid- and late season). The shading around each line indicates 95% confidence intervals for the relationship. Y-axis units are presented in cm3 h−1 instead of cm3 d−1 as a result of normalization by the average amount of daylight hours for each period in the growing season.

Q2: How variable are functional traits among individual shrubs and is this variability attributable to shrub location along topographic and/or resource gradients?

Data reduction using PCA resulted in four axes explaining variation in functional traits. These four PCs explained 89.4% of the variation in functional traits investigated (Table S14). Together, PC1 and 2 represented over 60% of the functional trait variation and the trait loadings suggest that these PCs could be considered separate functional trait spectra (Fig. 5). PC1 represented a gradient in productivity and resource use: leaf N and LA : SA were negatively related to LMA and leaf δ13C. PC2 captured a gradient in whole-plant water use and leaf carbon allocation: LA : SA and SFpeak were negatively related to leaf C and leaf δ13C.

Details are in the caption following the image
Principal components analysis biplot from an analysis used to characterize green alder functional traits among topographic positions. Functional traits include percentage leaf nitrogen (Leaf N), percentage leaf carbon (Leaf C), leaf mass per area (LMA), mean daily cumulative sap volume during peak physiological activity (SFpeak), leaf water use efficiency (δ13C) and leaf area-to-sapwood area ratio (LA : SA).

Overlaying topographic position onto the PCA biplot demonstrated that topographic position was related to resource-use strategies (Fig. 5). Functional traits relating to productivity and resource-use (PC1) of shrubs in channels clearly separated from shrubs at the tops of patches. ANOVA further confirmed that PC1 scores significantly differed between shrubs at the tops of patches and shrubs in channels (top vs channel; F3,19 = 5.54, P = 0.005); channel shrubs had higher leaf N and LA : SA and lower LMA and leaf δ13C (meaning a lower WUE). To facilitate comparison to other studies, abiotic conditions and functional trait data are summarized by topographic position in Table S15.


Here we provide the first empirical data on tall tundra shrub sap flow. Using a range of predictors of daily cumulative sap volume, we found that frost table depth was a central predictor, with implications for understanding seasonal patterns of water use and anticipating the effects of warming-induced permafrost thaw on tundra ecohydrology. Though not central in determining shrub SFdaily, topographic position drove variability in shrub functional traits pertaining to resource use and productivity. Given the rates of shrub proliferation on the low arctic tundra, information on tall shrub water use and drivers of shrub productivity are critical for predictive models of ecosystem function.

Frost table depth as a central predictor of daily cumulative sap volume in green alder

Overall, there was a negative relationship between frost table depth and SFdaily in both study years (Fig. 4), implicating thaw depth as an important determinant of green alder water use in the upland tundra landscape of our study site. The negative relationship between frost table depth and SFdaily is somewhat unexpected within the broader context of tundra shrub expansion. Many studies have found that tall shrubs tend to occur in locations with deeper active layers often associated with disturbance, such as floodplains (Tape et al., 2012), water tracks (Curasi et al., 2016), thaw slumps (Lantz et al., 2009), burned tundra sites (Lantz et al., 2010) and abandoned drilling mud sumps (Johnstone & Kokelj, 2008). However, in permafrost regions the water table is perched on the impermeable frost table (Quinton & Baltzer, 2013), and as the frost table and water table deepen seasonally, plant access to soil moisture decreases (Patankar et al., 2015). Patankar et al. (2015) likewise found that seasonal thaw decreased black spruce sap flow in a subarctic boreal peatland in the Northwest Territories. Their results indicated 50–60% reductions in sap flow with seasonal active layer development, which they attributed to reduced soil moisture availability due to the paired deepening of the frost and water tables and shallow rooting of black spruce. Although we lack green alder rooting depths at TVC, the overall negative response of green alder mean daily cumulative sap volume (SFdaily) to thaw depth likely is a consequence of seasonally drier surface soils in which green alders are rooted. The 2016 growing season received 26% more rain than the 2015 growing season resulting in shrub stem water potentials (Ψstem) in August 2016 that were significantly greater than those in August 2015 (Fig. 3). The relationship between frost table depth and SFdaily was weaker in 2016, further supporting the idea that thaw-induced water stress was reduced in 2016 due to greater growing season precipitation (Fig. 4). However, our model selection and model averaging did not indicate soil moisture as a strong predictor of SFdaily for either 2015 or 2016, potentially because proximity of water table to the soil surface is more relevant to shrub productivity than soil moisture in the top 20 cm of the soil profile.

Despite the overall negative relationships between frost table depth and SFdaily, there was a positive trend between frost table depth and green alder SFdaily early in both growing seasons (Fig. 4). This early-season trend is likely the result of warmer surface soil temperatures with early ground thaw. Warmer soil temperatures could enhance root function, soil microbial activity, shrub water use and overall shrub function (e.g. Keuper et al., 2012a). Thus, warmer soil temperatures combined with a frost table depth that still permits sufficient surface water availability could explain the positive relationship between frost table depth and early-season green alder SFdaily. Because this early-season relationship switches to a neutral relationship with deeper frost tables towards the middle (2016) and end of 2015 and 2016 growing seasons, there may be a threshold at which active layer thickening no longer benefits shrub productivity, resulting in the overall negative relationship through the growing season.

It could be argued that SFdaily decreased with active layer thickening simply because end-of-season senescence coincides with the timing of maximum thaw. However, our water potential data support the idea that seasonal thaw reduces water availability. Green alders had significantly more negative Ψstem values mid-season compared to early season during 2015. Whether this change translates to drought stress in green alder is not clear, but similar water potentials have been shown to induce drought stress in other forest-tundra ecotonal species (Oberbauer & Miller, 1982; Dang et al., 1997). Regardless, the decrease in Ψstem with seasonal progression implicates decreased surface water availability (Fig. 3a). Furthermore, we corrected SFdaily for differences in day length that occur with seasonal progression. Thus, the negative influence of thaw depth on SFdaily was likely important during the period in which green alders had the highest water demand, not just during senescence.

Variability and drivers of productivity-related functional traits in green alder

Using the conceptual framework of the global ‘fast-slow’ plant economics spectrum (Reich, 2014) which expands on the leaf economics spectrum (Wright et al., 2004), we provide evidence of variation in ecological strategies in green alder that reflect a gradient in fast to slow water and nutrient use that corresponds with topographic position. PC1 explained 33.1% of the variation in functional traits and corresponded with traits relating to resource use: photosynthetic potential (leaf nitrogen, N), leaf structural investment (leaf mass per area (LMA)), leaf water use efficiency (WUE) (δ13C) and whole-plant WUE (leaf area-to-sapwood area ratio (LA : SA)). In channels, green alder had higher leaf N, lower LMA and less conservative water use, whereas at the tops of slopes the opposite was true. These differences in productivity may provide a partial explanation for observed trends in green alder expansion in channels at Trail Valley Creek (TVC) (C. Wallace, unpublished) and other tundra locations (e.g. Naito & Cairns, 2011) and allow general predictions to be made about the variability in resource use and productivity among the topographic positions within patches. For instance, high LMA leaves are part of a resource conservative strategy (e.g. Mooney & Gulmon, 1982; Parkhurst, 1994; Terashima et al., 2001). Because channel shrubs had comparatively cheap and thin leaves with higher N concentrations, they likely also have higher photosynthetic and growth rates compared to shrubs at the tops of slopes. Channel shrubs also may return more N to the soil pool via higher quality leaf litter, which may increase available soil N in this position, potentially resulting in a positive feedback to shrub productivity in channels or increased nutrient export from the system (e.g. Compton et al., 2003). Tundra shrubs play an important role in trapping snow (Sturm et al., 2001), and channels often have deeper snow compared to upslope locations as a consequence of snow redistribution by wind (Essery & Pomeroy, 2004). Deeper snow in channels likely increases thermal insulation of the soil surface (e.g. Sturm et al., 2001) which can result in increased plant available N through effects on the summer soil microenvironment (Demarco et al., 2011). Shrubs at the tops of slopes invested more in conservative water use and associated leaf construction costs compared to channel shrubs. This resulted in more rigid leaves that are less susceptible to wilting with smaller transpirational surface areas that reduce water requirements under drier conditions (i.e. increased LMA and δ13C and decreased LA : SA; Poorter et al., 2009). Our Ψstem data suggest that the tops of slopes are more water-limited, and so differences in functional traits pertaining to water use indicate that shrubs are responding to local moisture conditions. Ultimately, these relationships suggest that channels are higher resource environments compared to the tops of patch slopes and provide a mechanistic explanation for observed distributional patterns of shrub expansion along floodplains rather than upland areas. However, we acknowledge other key limitations of shrub expansion, notably the arrival of seeds at sites, which has been shown to be related to factors other than topography (C. Wallace, unpublished).

Lack of topographic variation in abiotic conditions

Although most green alder functional traits varied with topography, it is not clear whether topographic differences in abiotic conditions were the drivers of this. Bjorkman et al. (2018) demonstrated that plant community height, leaf N and specific leaf area across the tundra biome were significantly higher at wetter sites, highlighting the importance of site moisture on tundra plant ecological strategies. Even though we found differences in shrub water status and leaf functional traits among topographic positions, we were not able to detect systematic differences in surface soil moisture, frost table depth or surface nutrient availability between top and channel positions (Fig. 2). This was unexpected as it was quite evident in the field that channels were wetter than upslope locations, for example. Soil moisture at greater depths than what we measured (5–20 cm) may differ among topographic positions, especially because the water table perches on the frost table in permafrost landscapes (Quinton & Baltzer, 2013), which, at our site, varied in depth from 13 to 94 cm. It also is possible that variation in abiotic conditions across microtopographic positions masked variation at the level of hillslope topography. For example, differences in soil chemistry and water content may dramatically differ between hummocks and interhummocks at TVC owing to differences in permafrost configuration and organic matter content (Wilcox et al., 2019) and organic layer thickness. On average, the bulk density of hummocks is an order of magnitude greater than adjacent organic interhummocks which are highly porous, and hillslope runoff is channelled preferentially through inter-hummocks (Quinton et al., 2000).

Ammonium (NH4+) and nitrate (NO3) supply rates also did not differ between the tops of slopes and channels, yet channel shrubs had significantly greater leaf N than shrubs at the tops of slopes, suggesting greater N availability and/or uptake in channels. Because green alder forms a symbiotic relationship with N-fixing Frankia, it is possible that differences in foliar N are driven by differences in atmospheric N-fixation. However, a paired study performed at TVC by Rabley (2017) demonstrated that mean nodule count and average nodule biomass per shrub did not significantly differ between the tops of slopes and channels, implying that N-fixation rates may be similar if nodules have equivalent function across topographic positions. We hypothesize that differences in leaf N thus may be due to greater organic N availability in channels rather than differences in inorganic N availability or symbiotic N-fixation rates. Arctic plants including both sedges and woody shrubs have been shown to use organic N sources (mainly amino acids; Chapin et al., 1993; Kielland, 1994) because inorganic forms are limited as a result of low litter inputs and slow decomposition and N-cycling rates (Haag, 1974; Schimel & Bennett, 2004). Because several studies argue that tundra shrub productivity is nutrient- rather than water-limited (Chapin et al., 1995; Keuper et al., 2012b), measures of dissolved organic N availability and uptake across the toposequence of shrub patches would be a valuable next step in determining how organic N availability influences shrub productivity.


Recognizing thaw depth as an important driver of green alder water use provides insight into how tundra shrubs may respond to future environmental changes. For instance, if high-latitude climate warming continues as projected, permafrost thaw is expected to accelerate (Lawrence et al., 2008). Our results suggest that this may reduce water availability, decreasing productivity of these tall shrubs, unless changes in rooting depth occur. Thus, predictions of future permafrost conditions may be useful in forecasting changes in shrub productivity and distribution but studies of plasticity in belowground biomass allocation will be important for evaluating this. Characterizing variation in rooting depth of tundra shrubs and additional belowground environmental variables also may provide further insight into relationships between water and nutrient availability and shrub productivity. Soil organic N availability and proximity of water table to roots and/or the soil surface are likely highly relevant to shrub productivity. These variables may help explain the topography-induced differences in productivity-related functional traits we found here along with observed trends of shrub expansion on the tundra in general.

This study produced the first dataset on tall tundra shrub sap flow, but further steps are required to upscale these estimates so modellers can better evaluate the ecohydrological impacts of tundra shrub expansion. The estimates of daily cumulative sap volume and sapwood area provided in this study can be used to calculate transpiration on a shrub patch basis by incorporating shrub patch area, the number of shrubs in a patch and the number of stems per shrub. Although our data for these variables are too coarse to undertake a rigorous upscaling of transpiration, a simple upscaling using site-level estimates of the above-mentioned variables indicates that patch-level transpiration could be as high as 2 mm d−1 (data not shown). This estimate is likely conservative as it includes only green alder transpiration (i.e. other shrub and ground vegetation species are not included). For comparison, ecosystem evapotranspiration (ET) measured for this site over open tundra ranges from 1.5 to 1.82 mm d−1 (M. Hurkurck, P. Marsh, & O. Sonnentag, unpublished data). If ground surface ET in green alder patches is comparable to values from the open tundra, then encroachment by tall alder shrubs could conceivably double the ET at this site. Studies aimed at a rigorous quantification of shrub patch ET will support an understanding of how much additional atmospheric water flux shrub patch transpiration may be adding to total tundra ET, which will be essential in quantifying the implications for tundra ecohydrology and patterns of regional warming in the face of shrub patch expansion.


Funding was provided by ArcticNet, the Natural Sciences and Engineering Research Council of Canada (Changing Cold Regions Network), Canadian Northern Studies Trust, Wilfrid Laurier University, the Polar Continental Shelf Program, Polar Knowledge Canada, Canada Foundation for Innovation, Ontario Ministry of Research and Innovation, and the Northern Scientific Training Program. J. Rabley, T. Giguere and E. Way-Nee supported data collection. We gratefully acknowledge logistical support provided by P. Marsh through access to TVC. N. Tran provided technical support for sap flow data processing. N. Day provided statistical advice. P. Marsh and K. Stevens provided input regarding study design. A. Berg provided soil moisture equipment and E. Wrona developed soil moisture calibrations. C. Pappas and O. Sonnentag reviewed early manuscript versions. Four anonymous reviewers provided thoughtful and constructive feedback on a previous version of this manuscript. Logistical support was provided by the Government of the Northwest Territories – Laurier Partnership. This research was carried out in the Inuvialuit Settlement Region under Aurora Research Institute Scientific Research License #15609.

    Author contributions

    KLB and JLB conceived the ideas and designed the methods and analyses; KLB collected the data with support from CAW; KLB and CAW analyzed the data; and KLB and JLB wrote the manuscript. All authors contributed critically to the drafts and gave final approval for submission.

    Data availability

    All data are available online via Dataverse: https://doi.org/10.5683/SP2/5O5BRA