Sagebrush_grazing_data

Data from an ecological field sampling and modelling effort. These data were used in the analyses presented in a manuscript entitled "Livestock grazing reduces perennial grass abundance but does not affect richness or composition in big sagebrush plant communities" being submitted, and are here presented to comply with open data standards. Metadata is available in the manuscript, the methods from the manuscript are copied below, and additional metadata available from the corresponding author upon reasonable request.

Study Area We conducted this study in big sagebrush-dominated plant communities in the upper Green River Basin of Wyoming, USA, in a ~4,000 km2 area. Big sagebrush plant communities exhibit considerable variability in both functional type and total species richness across their geographic extent in Wyoming (Jordan et al. 2020). Therefore, we constrained the geographic extent of our study area to maximize homogeneity in plant community composition, and soils, while representing variability in precipitation and temperature at the basin scale. All of the study sites are currently grazed by cattle during the summer, and are also grazed by mule deer (Odocoileus hemionus), pronghorn (Antilocapra americana), and feral horses (Equus caballus). All sites have a semiarid climate with mean annual temperature ranging from 1.6 to 3.7 C and mean annual precipitation between 191 and 349 mm. Sites ranged from 2040 to 2284 m in elevation. Soil texture was similar across sites (Appendix Figure A.1). In addition to big sagebrush, the dominant plant functional types are perennial bunchgrasses and perennial forbs. The dominant perennial bunchgrass species in the study area are Sandberg’s bluegrass (Poa secunda), Indian ricegrass (Achnatherum hymenoides), needle and thread (Hesperostipa comata), and bottlebrush squirreltail (Elymus elymoides). The most common perennial forbs are Hood’s phlox (Phlox hoodii), pussytoes (Antennaria spp.), and milk vetch (Astragalus spp.). The most common non-native species was desert alyssum (Alyssum desertorum). We identified plants that did not possess adequate identifiable species characteristics to genus or functional type. Plant species nomenclature and nativity are based on the USDA PLANTS database (USDA, NRCS 2017). Sampling Design and Site Selection To identify the location of livestock watering sources, we obtained a database of artificial watering locations on Bureau of Land Management lands from BLM Field Offices within the study area. Our use of artificial watering points that were consistent in size allowed us to 1) control for the total area of livestock concentration around water relative to natural water bodies which are more variable in size and shape, 2) avoid confounding from topographically-mediated vegetation gradients that often exist around natural watering sources, and 3) obtain records of how long each artificial watering source had been in place. We chose 17 watering sources (hereafter “sites”) on level uplands (slope < 2°) with similar soil types and vegetation dominated by big sagebrush. Because distance from water studies capture historical grazing pressure, we determined how long each watering point had been in place. Our artificial watering sources had an average age of 46 years (SD=16) and 88% of watering sources were in place for at least two decades. The exceptions were one watering source for which we were unable to obtain a construction date and a second that had been in place for eight years. We sampled plant communities along the grazing intensity gradient within each site by randomly selecting plot locations within each of four distance bins that radiated outward from each watering source: 30-75, 75-300, 300-500, and 500-700 m (Adler et al. 2005, Adler and Hall 2005). Other studies using distance from water gradients have used distances comparable to ours (Eldridge and Whitford 2009, Manthey and Peper 2010, Sasaki et al. 2011, Sternberg 2012). We established a 100 m2 plot in each of these distance bins by randomly choosing a distance within each bin, resulting in four plots per site and a total of 68 plots across all sites. We avoided the denuded, trampled area immediately adjacent to the watering source (<30 m). All field data were collected between May and June 2017 during peak phenological activity for most species. Vegetation Data In each plot, we recorded presence and estimated percentage cover for each plant species. We also recorded estimates for other cover groups: dead shrubs, woody debris, lichen, animal dung, litter, biotic soil crusts, bare ground, and rocks. Additionally, we recorded aspect, elevation, and slope at the plot origin. Because previous studies have found that bunchgrasses can be sensitive to grazing (Rickard et al. 1975, Anderson and Holte 1981), we characterized grazing effects on perennial bunchgrasses. We collected 40 randomly selected bunchgrass individuals within each plot and recorded their identity, measured their basal diameter and height, and collected all aboveground biomass, excluding aboveground crowns. We dried samples to a constant weight at 40° C, and recorded weights to the nearest tenth of a gram for each individual. Soils We collected three soil samples with an auger at 10 cm increments to a depth of 30 cm in each plot for a total 12 samples per site. We used a gravimetric method modified from Bouyoucos (1935) to quantify soil texture. We measured soil salinity using the NRCS 2009 protocols (Burt 2009) with an Oakton Waterproof EcoTestr EC High Conductivity Tester Pocket Meter. Presence of Cattle and Feral Horses We estimated dung density at each plot as a second measurement of grazing intensity among plots (Tate et al. 2003). We recorded dung density by counting all dung pats within our plots, separately for cattle and feral horse dung, if present. Analysis We conducted all analyses in R v. 4.1.1 (R Core Team 2021). To determine the relationship between our two proxies for grazing intensity- dung density and distance from water- we modeled dung density vs. distance from water with a nonlinear model using generalized least squares. To aid interpretation of this relationship, we calculated the pseudo-R2 for the model as the summed square of residuals divided by the summed squared difference to the mean. For plant communities, we quantified total species and functional type richness in each plot. To evaluate functional type richness, we assigned each species to a functional type: bunchgrass, rhizomatous grass, annual forb, perennial forb, or shrub. We excluded succulents (represented by a single species) from the analysis, as this species was not abundant at any site. To quantify similarity in species composition across plots, we calculated Jaccard’s Coefficient on square-root transformed species percentage cover data, which results in a larger contribution of low abundance plant species (which are common in big sagebrush plant communities) to the similarity index than untransformed cover (McCune and Grace 2002). To visualize the differences between plots, we used nonmetric multidimensional scaling (NMDS), the most appropriate indirect ordination technique for plant community data (McCune and Grace 2002). We used the metaMDS function in the vegan package (Oksanen et al. 2020), for an ordination solution with 5 dimensions and 100 random starts. To identify how species compositional changes differed among sites and with grazing intensity, we overlaid convex hulls for both sites and distance from water bins using the ordihull function in the vegan package (Oksanen et al. 2020). We followed the same procedure to determine the similarity of bunchgrass species composition within plots across distance from water sampling bins. For each plot, we summed biomass for all bunchgrass individuals belonging to the same species, which served as the measure of abundance. We then calculated a Jaccard’s Coefficient based on the biomass for bunchgrass species. To test for significant differences in species composition in plots across grazing intensity gradients, we used multi-response permutation procedure (MRPP; McCune and Grace 2002). MRPP tests the null hypothesis of no difference between two or more defined groups, either by differences in the mean or the spread of variables and is a nonparametric analog to multivariate analysis of variance (MANOVA). We implemented MRPP using function mrpp in the vegan package (Oksanan et al. 2020) based on Jaccard’s Coefficient calculated using square-root transformed cover values for each species to test for differences in species composition between distance from water bins. To evaluate the effects of grazing intensity on species richness, functional type richness, total vegetation cover, total grass cover, bunchgrass biomass, bunchgrass height, and bunchgrass basal area, we fit linear mixed effects models using function lmer in package lme4 (Zuur et al. 2009, Bates et al. 2015) to examine changes in these variables over the distance from water and dung density gradients. In these models, we included random intercepts to account for site-level effects and for non-independence of samples within each site, and treated either distance from watering source or cow dung density (proxies for grazing intensity) as a fixed effect.