The persistence of future forests depends on the success of tree seedlings which are experiencing increasing physiological stress from changing climate and air pollution. Although the moss layer can serve as an important substrate for tree seedlings, its potential for reducing environmental stress and enhancing the establishment of seedlings remains poorly understood. We tested if the moss layer decreased environmental stress and increased the abundance of balsam fir seedlings dominant in high-elevation forests of the northeastern United States that are sensitive to changing climate and mercury deposition. We surveyed balsam fir seedling density by substrate (moss, litter, other) on 120 quadrats (1×1 m) in two contrasting canopy environments (in gaps and under canopies), measured seedling stress, and quantified mercury content in seedlings and substrates. We observed that, in both canopy environments, tree seedlings established on moss exhibited (i) increased density, (ii) decreased physiological stress, and (iii) higher potential to recruit into larger size class, compared to seedlings established in litter. Regardless of canopy environment, seedling foliar mercury levels did not correspond to substrate mercury despite large differences in substrate mercury concentrations (relative to moss, litter concentrations were ~4-times greater and soil concentrations were ~6-times greater), likely reflecting the dominance of foliar over root uptake of mercury. Since the moss layer appeared to mitigate seedling drought stress and to increase seedling establishment and recruitment compared to other substrates, these microsite effects should be considered in models predicting forest regeneration and dynamics under increased drought stress associated with ongoing climate warming.

Study area

This study was conducted in the high-elevation fir-dominated forests on Whiteface Mountain (44.22° N and 73.54° W) in the northeastern United States. Details of the study area were previously described in Berdugo and Dovciak (2019) and elsewhere (e.g., Battles and others 2003; Wason and others 2017b, 2021). Briefly, Whiteface Mountain is an isolated massif with a summit at 1,485 m above sea level (asl) and well-developed elevational climatic and vegetation zones whose ecology has been extensively studied over the past half century (e.g., Sprugel 1978; Battles and others 2003; Aleksic and others 2009; Wason and others 2017b). The vegetation of the area belongs to the Adirondack-New England Highlands ecosystem province (Bailey 2014) and is representative of regional forest communities with northern hardwood forests occurring below ~800 m asl, spruce-fir forests between ~800 m and ~1,300 m asl, and alpine communities above ~1,350 m asl (Battles and others 2003). Balsam fir (Abies balsamea (L.) Mill.) is a dominant tree species between ~1,100 and ~1,250 m asl where recurring linear openings in the forest canopy, known as fir waves, facilitate fir regeneration (Sprugel 1976; Silvertown and Dodd 1999). Soils in the area are mainly Histosols derived from the accumulation of organic material due to slow decomposition and slow weathering of anorthosite (Witty and Arnold, 1970; Sprugel, 1976).

The character of the moss layer in high-elevation fir forests of Whiteface Mountain is relatively unresponsive to the microclimatic and other environmental changes caused by gap dynamics (Berdugo and Dovciak, 2019). The moss layer here typically covers 57.5 + 2.2 % of the forest floor surface, is ~2.5 cm thick, and tends to contain ca. 5 species per square meter both in gaps and under closed canopies (Berdugo and Dovciak, 2019). Notwithstanding, two aspects of the moss layer differed between canopy environments: Dicranum fuscescens Turner, Hypnum imponens Hedw., and Tetraphis pellucida Hedw., are more abundant under closed canopies than in gaps, making these moss species indicator species for closed canopies in the area; and functional dispersion (derived from growth form and the Ellenberg indicator values for soil fertility and acidity) is higher under closed canopies than in gaps (Berdugo and Dovciak, 2019). While the moss layer in the study area includes up to 25 bryophyte species (Table S2.2 in Berdugo and Dovciak, 2019), those identified as seedling substrates in the current study corresponded to the liverwort Lophozia ventricosa (Dicks.) Dumort. and four mosses, D. fuscescens, H. imponens, Pleurozium schreberi (Brid.) Mitt., and Polytrichastrum alpinum (Hedw) G. L. Sm. In all our observations, mats of H. imponens held sparse fertile shoots of Pohlia nutants (Hedw.) Lindb.

The climate on Whiteface Mountain has been characterized in several studies (Table 1) and it can be considered a high-elevation form of a modified continental climate (McNab and others 2007) experiencing the warming climate trend typical of the northeastern United States (Huntington and others 2009; Wason and others 2017a). The climate records specific for the fir wave zone (1,100–1,250 m asl) are rather sporadic compared to those at the summit (1,485 m asl) and the shoulder (~600 m asl) of the mountain where climate stations are located (Schwab and others 2015). The weather on Whiteface Mountain during the study period (in September 2015, see Study design and field data collection) was mild, but the month was considerably drier in 2015 (48.0 mm of precipitation) than the long-term average (93.6 mm, mean September precipitation for 1985–2019 period) (National Atmospheric Deposition Program 2020).

Study design and field data collection

 To assess the relevance of the moss layer for forest regeneration, we selected a total of 40 study plots using a stratified sampling design with two strata (forest gaps, closed canopy) that represented the two typical canopy environments in these high-elevation forests (cf. fir waves; Sprugel 1976). We selected 20 fir waves and placed one plot in each canopy gap and another plot under the forest canopy nearby (on average within 50.2 ± 3.9 m, mean ± SE, from the closest gap). Each plot consisted of three 1-m² survey quadrats placed along a transect in the gap center and parallel to the gap length; minimum spacing between survey quadrats was 5.6 m. Transect lengths varied between ca. 14 m and 25 m long, depending on the gap length while transects under forest canopy were consistently 25 m long. The total number of surveyed quadrats was 120 (3 quadrats × 40 plots).

Seedling bank surveys: We characterized tree seedling abundance and composition on all quadrats in 2014 (May to September) by counting the number of tree seedlings (≥5 and ≤ 25 cm tall) by species, height class, and substrate. Based on specific aging methods developed to describe the seedling bank structure of balsam fir (Parent and others 2003), this height range likely captures ~98% of balsam fir seedling bank, composed of seedlings recruited in, at least, the previous 10 years. Balsam fir also dominated the seedling bank as it was found in all plots and represented ~95 % of all surveyed seedlings. The remaining 5% of tree seedlings was split among other four species (Picea rubens, Betula papyrifera var. cordifolia, B. alleghaniensis, and Populus tremuloides) that were present on less than half of the plots. Seedlings were divided into two height classes: small (≥5 and ≤ 10 cm tall) and large (>10 and ≤ 25 cm tall). Seedling substrate, characterized as moss, litter, and other (decaying logs, bare soil, peat, and rocks), was sampled to a depth ≤ 5 cm to collect the substrate under the moss layer (with thickness of 2.5 ± 0.3 cm, mean ± SE; Berdugo and Dovciak 2019) while accommodating the generally shallow soil depth (< 10 cm deep; Gerson and others 2017). The two most common seedling substrates were moss and litter (Figure S1).

Microsite characteristics: In addition to seedling substrate (see above), other environmental characteristics potentially affecting tree seedlings were also measured on all quadrats. Canopy cover (%) was measured with a concave spherical densiometer (Forestry Supplies Inc., Jackson, Mississippi, USA) at 50 cm above the ground. Despite lower accuracy under closed canopies, densiometer measurements of canopy cover (or openness) were shown to provide satisfactory contrast between closed forest canopies and open canopies such as those in forest gaps (Russavage and others. 2021). We also measured tree stem (≥2 m tall) density and height by species, density of standing dead trees, tree diameter at the breast height (DBH) by species, and the density of balsam fir saplings (individuals >25 cm but < 2 m tall). We estimated the cover (%) of understory vegetation, coarse woody debris (CWD; the largest diameter >5 cm), and fine woody debris (FWD; the largest diameter < 5 cm) within < 30 cm height from the ground surface. We also estimated the overall cover (%) for seedling substrates for each quadrat (substrate details are given in the section Seedling surveys above). Balsam fir does not build persistent seed banks (Houle 1992), but its seedling banks can instead stagnate under closed canopies and rapidly grow under canopy gaps (Sprugel 1976, Parent, Morin, and Messier 2000; Parent, Morin, and Messier 2001; Parent and others 2003). Given these dynamics, mast years do not seem to have a strong effect on the structure of fir seedling banks (Parent and others 2003).

Seedling physiological stress: To estimate the instantaneous physiological stress that seedlings experienced in the two main substrates (moss, litter) under both canopy environments (closed canopies, gaps), we contrasted the efficiency of heat dissipation (non-photochemical quenching – NPQ, Maxwell and Johnson 2000) in seedling needles. We calculated NPQ for a subsample of 186 seedlings from the seedling bank survey to achieve comparable seedling numbers on the two dominant substrates (95 on moss, 91 on litter) and in the two canopy environments (98 under forest canopy, 88 in gaps) by selecting the most accessible ten study plots (five under the canopy, five in gaps) among the selected fir waves (see the section of Study design and field data collection).

The efficiency of heat dissipation, NPQ, is calculated from non-invasive fluorescence measurements relative to dark-adapted conditions (Murchie and Lawson 2013). Therefore, we recorded chlorophyll fluorescence on seedling needles in both dark and light conditions. Dark-adapted seedling needles (Figure S2a) were excited with a pulse of 100 steps, each lasting 1.8 μs, of an intense saturating light (20,000 μmol m^-2 s^-1) and light-adapted seedling needles (Figure S2b) were excited with this light pulse while recording ambient photosynthetically active radiation (PAR; μmol m^-2 s^-1). Using the Pulse Modulated Chlorophyll Fluorometer Model FMS2 (Hansatech Instruments Ltd, Northfork, United Kingdom), we completed these instantaneous measurements before noon between September 7 and 11, 2015.

NPQ was calculated as (F_m- F’_m)/F’_m in light-adapted samples, where F’_m is the maximum fluorescence yield in light-adapted conditions, and F_m is the maximum fluorescence yield in dark-adapted conditions (Murchie and Lawson 2013). We used a single F_m value for seedlings in equivalent conditions (substrate, canopy environment) by averaging the measurements. Long-term seedling stress was avoided by targeting only healthy seedlings and was verified with a ratio between variable and maximum fluorescence (F_v/F_m) in dark-adapted needles > 0.8 (Maxwell and Johnson 2000, Murchie & Lawson 2013) as the average Fv/Fm of dark-adapted needles of seedling established in both substrates and in both environments (n=74) was 0.84 ± 0.05 (mean ± SD). Ruling out long-term seedling stress allows us to assume that differences in efficiency of heat dissipation (NPQ) may result from short-term stress drivers (Murchie & Lawson 2013), such as climate.

Climatic stressors: To identify potentially climatically stressful conditions for tree seedlings, we monitored instantaneous microclimate during the time when seedling instantaneous physiological stress (chlorophyll fluorescence) was measured (i.e., September 7–11, 2015). Air temperature (T) and relative humidity (RH) were measured using iButton data loggers (Model DS1923; Maxim Integrated Products, Inc., Sunnyvale, California) on each plot every 15 minutes with 0.5°C resolution for temperature and 0.6% for relative humidity. Each iButton was placed at 30 cm above the ground within a handmade gill shield (~12 cm height × 9 cm diameter) to measure ambient microclimate (T, RH) while shielding the sensor against solar radiation and allowing for adequate ventilation (Tarara and Hoheisel 2007).

Exposure to mercury: We quantified total mercury in seedling needles and dominant substrates (moss, litter) on the plots used for monitoring seedling instantaneous physiological stress (see above). Assuming mercury accumulation in conifer needles over time (Blackwell and others 2014) by quantifying total mercury, we indirectly measured seedling exposure to mercury (Parent and others 2003). We selected 53 seedlings using stratified random sampling to represent the combinations of substrate (litter, moss) and canopy environment (forest canopy, gap) and collected samples of both seedling needles and associated substrates (both surface and subsurface). Sample collection followed the protocol for environmental sampling of low-level trace metals (EPA Method 1669; US EPA 1995) adapted for solid samples. Briefly, a two-member team wearing clean nitrile gloves split sampling tasks. “Clean hands” only handled the plastic bag containing the final sample. “Dirty hands” prepared sample containers, collected samples by operating pruning scissors and a soil sampling tube, cleaned these tools by rinsing them with trace metal grade HCl, and handled secondary container bags and shipping containers. All needles were collected from each seedling along with the associated seedling substrate(s). A substrate sample was collected when a single substrate was present within the 5 cm depth from the surface. Another separate substrate sample was collected if a different substrate was present within 5 cm of the substrate surface (i.e., subsurface substrate included moss, litter, or soil A horizon; Figure S1). Seedling roots and substrate sections contaminated by the adjacent substrates were discarded. Thus, a total of 125 samples were collected and transported on ice to the Center for Environmental Systems Engineering (CESE) at Syracuse University where they were kept frozen until processed. Samples were freeze-dried to a constant weight and hand homogenized to a fine powder (any twigs, stems, or rocks were removed from the samples). Homogenized samples were analyzed for total mercury concentration following the EPA Method 7473 (US EPA 2007) with an Advance Mercury Analyzer AMA 254 (Leco Corporation, Saint Joseph, Michigan); details on quality control according to the EPA Method 7473 (US EPA 2007) are described in Appendix S1 in supporting information.

Data analyses and hypothesis testing

To test our hypothesis H1, whether the moss layer had a positive effect on tree seedling density, we used a multivariate approach by fitting generalized linear models (GLM). We modeled the maximum count of fir seedlings per square meter for two seedling height classes, small (≥5 and ≤ 10 cm tall) and large (>10 cm and ≤ 25 cm tall) at the plot level (n=40) with a Poisson error distribution and ‘log’ link function. These two height classes allowed us to target seedlings established in, at least, the last decade (Parent and others 2003). The full model predictor set included the percentage moss cover, four uncorrelated forest structural metrics (P>0.05 and rho < |0.300|; see Appendix S2)—canopy openness, fir sapling density, fir tree density, and density of other tree species—and the interaction between the moss cover and canopy openness. Because the forest structural metrics accounted for the structural differences among the plots, neither random nor categorical effects by canopy environment (i.e., gap vs. canopy) were included in the model; consistently including plot as a random effect did not change the model selection. We did not include the cover of litter (the other dominant substrate) in the model to avoid collinearity since moss and litter covers were highly correlated (Pearson correlation coefficient = -0.538, p<0.001). We fitted all possible models and the best model was selected using delta AIC corrected for small sample size (DAICc; Anderson and Burnham 2002); models within DAICc<2 were considered equivalent.

To test our hypothesis H2, whether the moss layer had positive effects on seedling recruitment into the larger size class, we calculated an index of seedling recruitment potential. The index was calculated as a demographic ratio between large (>10 and ≤ 25 cm tall) and small seedlings (≥5 and ≤10 cm tall) for those quadrats where both seedling height classes were present. The differences in seedling recruitment potential between the two dominant substrates were analyzed with the Kruskal-Wallis tests as data did not follow a normal distribution.

To test our hypothesis H3, whether tree seedling stress varied by substrate, we performed several comparisons. We assessed the difference of the instant desiccation potential of the air between canopy environments, the difference of instant physiological stress of seedlings between substrates within each canopy environment, and the interaction between substrate and canopy environment on the instant physiological stress. Instant microclimate and instant physiological stress, synchronized by their time record, were positively correlated (Pearson’s correlation coefficient between instant air desiccation potential – VPD – and efficiency of heat dissipation – NPQ –: rho = 0.21, p<0.001, n = 186). Field measurements of temperature (T) and relative humidity (RH) were combined to calculate instant vapor pressure deficit (VPD), a measure of the desiccation potential (Will and others, 2013). VPD was calculated as the difference between saturation vapor pressure (SVP) and actual vapor pressure (AVP) at a given temperature following Ward and Trimble (2003), where SVP = exp (16.78 × T - 116.9)/ T + 237.3, and AVP = SVP × RH/100. Seedling physiological stress was estimated as non-photochemical quenching (NPQ, see section Seedling physiological stress). Since NPQ data were non-normal (Shapiro-Wilk: W = 0.40, P < 0.001) and variance homogeneity was not met for either substrate (Bartlett: x2(1) = 112.9, P < 0.001) or canopy environment (Bartlett: x2(1) = 11.2, P < 0.001), we used non-parametric Kruskal-Wallis test followed by Friedman test (Friedman 1937) for the interaction between substrate and canopy environment.

Finally, to assess whether exposure to mercury differs between seedling substrates (H4), we tested the differences in total mercury concentrations (THg) both in (i) seedling needles (by substrate and canopy environment) and (ii) among the sampled substrates (moss, litter, and soil by canopy environment). Unlike instant physiological stress, exposure to mercury is cumulative as mercury concentration in plant tissues increases over time (Driscoll and others 2013). Samples from quadrats were averaged at the plot level for each sample type (seedling needles, moss, litter, and soil). Since seedling total mercury concentrations met the assumptions of normality (Shapiro-Wilk: W=0.92; P=0.150) and variance homogeneity (between substrates, Bartlett: x2(1) = 0.00; P = 0.958; and between environments Bartlett: x2(1) = 0.43, P = 0.512), we used two-way ANOVA to test whether tree seedling total mercury differed by substrate in the forest as a whole and between canopy environments, while considering also the interaction between substrate and canopy environment. Finally, we tested for the differences in total mercury among seedling substrates using a Kruskal-Wallis test since substrate total mercury concentrations were not normally distributed (Shapiro-Wilk: W = 0.90; P = 0.01) and their variances were not homogeneous (x2(1) = 12.03, P = 0.003). We evaluated the main effect of canopy environment on the total mercury concentration in each substrate using one-way ANOVA since the data for each substrate separately complied with normality (Shapiro-Wilk: W<0.93; P>0.05).

All analyses were performed in R (R Core Team, 2018) with the ‘stats’ package as well as with additional specific packages; ‘lme4’ (Bates and others 2015), ‘MuMIn’ (Barton 2020), and ‘arm’ (Gelman and Su 2020) for model fit and selection.