In south-eastern Utah, USA, we conducted three multi-part studies. (i) The first study examined macro-scale heterogeneity in lichen and moss cover and composition among several nearby sites, and tested for microsite-scale heterogeneity in habitat characters between adjacent +LMC and –LMC patches within these sites. (ii) The second study investigated the heterogeneity in lichen–moss community composition among five micro-aspects of soil crust pedicels. (iii) The third study quantified characteristics of Collema-occupied habitat at microsite, micro-aspect and organismal scales. We then synthesized all the results.
study 1: macrosite- and microsite-scale sampling
Eight study sites were selected within a single macrosite, to represent considerable variation in south-eastern Utah's soil texture and chemistry. Soils ranged widely in CaCO3 content (0·3–12·9%) and texture (60·8–89·7% sand). Calcium carbonate is important because it immobilizes nutrients (P, Mn and others) via adsorbtion onto its surface. Vascular plant communities were variable among sites in this and the following studies, the most common communities being dominated by Achnatherum hymenoides (Roemer & J.A. Schultes) Barkworth and Hesperostipa comata (Trn. & Rupr.) Barkworth, Pinus edulis Engelm. and Juniperus osteosperma (Torr.) Little, Artemisia tridentate Nutt. and Coleogyne ramosisimma Torr. We selected sites from 1352 to 1831 m altitude and within 70 km of Moab, Utah, to restrain climatic variability. Historical precipitation records for the macrosite logged in Moab, two sites in Canyonlands National Park and Arches National Park were 22·9 cm, 22·9 cm, 21·6 cm and 22·1 cm, respectively (Western Regional Climate Center 2005). Six sites had been undisturbed for at least 35 years while the other two showed little evidence of recent disturbance, possibly because of low amounts and poor quality of cattle forage or lack of water resources.
At each of these eight sites, we sampled five pairs of 25 × 25-cm quadrats (just three pairs at a site with low variability in crust cover within quadrat type) for a total of 75 quadrats. We believe this sample size was adequate to capture much of the spatial heterogeneity in lichen–moss crust cover at the macro- and microsite scales, because we intentionally built important edaphic gradients into the site-selection scheme and intentionally selected maximal +LMC/–LMC contrasts in the quadrat placement. Quadrat size was selected to accommodate the size of the smallest +LMC microsites. Each pair represented one +LMC and one –LMC microsite. Paired quadrats, separated by < 1 m in their respective microsites, were positioned non-randomly to capture the greatest possible within-microsite contrast. Biological soil crust cover in the vicinity of each quadrat pair averaged approximately the mean of that in the immediate vicinity. Categorization as +LMC or –LMC was relative to typical conditions within a site; as a result –LMC often contained some lichens and mosses (average cover 13% compared with 55% for +LMC). To keep microclimatic regimes approximately constant, we placed quadrats only in interspaces between shrubs. Within each quadrat, percentage cover was estimated visually to the nearest 5% for each moss and lichen species and for mineral crusts (whitish crust of unknown chemical composition), with additional observations of < 5% recorded as 2·5%. We used the field-based lichen key by McCune & Rosentreter (1995) and Flowers (1973) to identify lichens and mosses, respectively. For specimens not well covered in the keys, we consulted with Drs John Spence, Lloyd Stark (mosses) and Roger Rosentreter (lichens). Lichen nomenclature follows Esslinger (2004), except Aspicilia aspera which follows Hafellner, Nimis & Tretiach (2004), and moss nomenclature follows Zander (2004). Mean percentage cover values were estimated separately for +LMC and –LMC microsites at each site. Cover was computed as the percentage of available habitat, excluding litter and rock (by definition unavailable to soil crust organisms). This transformation did not strongly alter analysis results, as cover of unavailable habitat averaged only c. 8%. Voucher specimens from this and following studies (numbers BOWKER2005 :1–21) were deposited at the Boise State University (Idaho) Department of Biology herbarium.
After percentage cover was recorded, samples from the top 1 cm of soil were collected from at least 10 haphazardly positioned surface soil cores (2·5 cm diameter) in each quadrat, and compiled into a composite sample for that quadrat. Soils were sifted through a 2-mm sieve to remove rocks and litter. Lichens and mosses were also carefully removed at this time to allow us to distinguish between nutrients attributable to high lichen and/or moss cover, and those incorporated in or adherent to lichens and mosses. Within each site, composite soil samples were created by combining soils within microsite type (+LMC vs. –LMC).
Samples were homogenized and split into duplicate sets of equal mass. One set of samples was sent to the Brigham Young University Soils laboratory (Provo, UT, USA) and analysed for sodium bicarbonate extractable P and K (Ka), exchangeable cations [K, sodium (Na), calcium (Ca), magnesium (Mg)], micronutrients [Zn, copper (Cu), iron (Fe), Mn], electrical conductivity (EC; a measure of salts more soluble than gypsum) and acid-neutralizing potential (ANP). ANP is a measure of CaCO3 and other agents that buffer soil acids. Two forms of K were measured: Ka (the amount of K in solution), and Ke (the amount of K held electrostatically on colloids). Because some lichens are N-fixing organisms that strongly influence soil levels, we chose not to measure N. We analysed the second soil split at the USGS Earth Surface Processes laboratory (Denver, CO, USA) for percentage CaCO3, particle size distribution and magnetic properties. We used a laser-light scattering method (capable of measuring particles from 0·49 to 2000 µm) to determine particle size as a percentage of total volume. We employed a combination of magnetic and reflected light petrographic methods to measure and confirm the following magnetic properties: (i) magnetic susceptibility (MS), a measure of all magnetic material but predominantly magnetite, and (ii) hard isothermal remnant magnetization (HIRM), an approximate measure of haematite (King & Channel 1991). Our study area included sedimentary parent materials only; therefore, the presence of these igneous-derived magnetic minerals indicated input of non-local eolian dust (Reynolds et al. 2001).
study 2: micro-aspect-scale sampling
To test for affinity between lichen–moss crust species assemblages and five micro-aspects, we selected three sites in Canyonlands National Park (c. 50 km from Moab). All sites were relatively flat, having similar climate and soils derived primarily from the Cedar Mesa Formation. Soils ranged from sandy loams to loamy sands and differed in depth and vegetative cover.
Within each site, three haphazardly selected microsites consisted of 10 soil crust pedicels that were sampled on five micro-aspects. Only pedicels with well-developed micro-aspects were sampled. Using a measuring tape, we sampled five miniature line intercept transects on each pedicel representing each micro-aspect [NNW, SSE, ENE, west-south-west (WSW), TOP; n= 450 total line intercept transects]. All transects paralleled the major axis of the pedicels (c. NNW–SSE) and were placed as follows: (i) NNW and SSE, centred on the major axis, running from base to top in the vertical plane; (ii) ENE and WSW, parallel to the major aspect, halfway between base and top in the horizontal plane; (iii) TOP, centred on the major axis, running from NNW to SSE. Transect length varied with the size and shape of the pedicel but generally ranged from 3 to 10 cm. Distance intercepted by each moss and lichen species was recorded and converted to percentage cover for that given line intercept. Data from the 10 replicate pedicels within each possible site × microsite × micro-aspect combination were averaged prior to analysis (to reduce the large number of zero values encountered on individual pedicels). This procedure resulted in 45 sets of mean cover values (3 sites × 3 microsites × 5 micro-aspects).
study 3: microsite-, micro-aspect-, and organismal-scalecollemastudy
To determine which environmental variables were correlated with Collema cover at multiple small spatial scales (microsite, micro-aspect and organismal scales), we collected a series of 60 c. 200-g composite soil samples. Composite samples consisted of six subsamples from different pedicels wherein the top 3–5 mm of soil was shaved off with a knife and collected. Each composite collection was made from two adjacent microsites at haphazardly selected locations at a single site in Canyonlands. One of the microsites contained no Collema (10 composite samples) while Collema was found in the paired sample (50 composite samples, allocation of replicates detailed below). In the microsite with Collema, soils were collected independently from pedicels with no Collema (10 composite samples), low Collema (5–10% cover; 20 composite samples) and high Collema (> 10% cover; 20 composite samples) cover. For all sampled pedicels, replication was evenly split between ENE and WSW micro-aspects. On pedicels with Collema, paired samples were taken from Collema patches (Collema and the soil under it) and immediately adjacent areas lacking Collema (within 1–2 mm) on both micro-aspects. For each collection, the top 3–5 mm of soil was sampled. Samples were thoroughly homogenized, and then sent to Brigham Young University (BYU) for analysis of N, P and Ka, exchangeable cations and micronutrients.
Hypotheses about macro-scale lichen and moss distribution (in study 1) were tested using methods reviewed in McCune & Grace (2002). Non-metric multidimensional scaling (NMDS) was used in conjunction with the Bray–Curtis distance measure to ordinate species and plot vectors of environmental variables. To test for a significant correlation between community data and environmental variables, and to obtain a measure of ‘effect size’, we used the Mantel test first as a whole model test, then as a post-hoc test to determine which individual variables probably accounted for significance. Here and in all the following multivariate tests, all variables were relativized to their maxima to equalize their influence upon the analysis, and species with only one occurrence were omitted (McCune & Grace 2002).
Hypotheses regarding soil properties on microsite scales (study 1) were tested using permanova (Anderson 2001). To focus the analysis on between-microsite rather than among-site differences, we aligned median values of each response variable within each block to zero. Because whole model significance was detected, post-hoc ‘design matrix’ Mantel tests (McCune & Grace 2002) were used to characterize further differences in the individual soil parameters between the two types of microsite (+LMC and –LMC).
We analysed lichen–moss crust communities across micro-aspects (study 2) using a combination of permanova and indicator species analysis. Indicator species analysis generates a percentage of perfect indication value (IV; ranging from 0 to 1) for each variable–group combination and employs a Monte Carlo test to determine the probability of obtaining an IV that equals or exceeds the calculated value. Kruskal–Wallis tests in conjunction with post-hoc Tukey HSD tests were used to test for significant differences in abundance of functional groups and species richness (based upon the finest taxonomic level we could attain) across micro-aspects.
Soil nutrient data from study 3 were analysed using two manova models. The first compared nutrient abundances across microsites with and without Collema, and among micro-aspects, and the interaction of the two factors. The second compared nutrients on pedicels with high and low Collema cover, different micro-aspects and in adjacent habitat with and without Collema individuals, and all interactions. Post-hoc univariate anovas were used to determine which variables probably accounted for significance. Statistics were conducted in pc-ord version 4 (1999 MJM Software Design), jmp in 4·0 (2002 SAS Institute) and permanova version 6 (2005 M. J. Anderson). In study 2, the sequential Bonferroni correction was applied so that table-wise α= 0·05. We did not apply a correction for multiple comparisons in studies 1 and 3 because there was a high degree of intercorrelation of response variables. When response variables are highly correlated, Bonferroni-type corrections tend to be too conservative (Manley 2001).