Final Report Assessing five field sampling methods to monitor Yellowstone National Park’s northern ungulate winter range: the advantages and disadvantages of implementing a new sampling protocol Pamela G. Sikkink1, Roy Renkin2, Geneva Chong3, and Art Sikkink4 What the study revealed The five field sampling methods tested for this study differed in richness and Simpson’s Index values calculated from the raw data. How much the methods differed, and which ones were most similar to each other, depended on which diversity measure and which type of data were used for comparisons. When the number of species (richness) was used as a measure of similarity, the historic method captured significantly fewer species than the Daubenmire, modified-Whittaker, or Forest Inventory and Analysis (FIA) methods but similar numbers as the small-scale nested circular plot. When Simpson’s Index was used to compare similarities, only the large-scale modified Whittaker method showed significantly greater values than the small-scale nested circles; no differences were observed among the other methods. If frequency data instead of cover data were used to compare similarities among methods, the historical method had significantly higher evenness, skewness, and kurtosis on average than all other methods. If a correspondence measure based on diversity was used, the historic method was most similar to the Daubenmire protocol. Even though the plant communities across the northern range contained many species and each appeared distinct during sampling, only the low elevation sites (n=3) in the Gardiner Basin were significantly lower in community diversity; comparisons among the other sample locations (n=16) were not significantly different. Because diversity measures may be the same for very different communities, we used a statistical ordination technique that placed similar communities near to each other in 3-D space. The ordination shows which locations had different plant communities because it was 1 Research Ecologist, U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station, Missoula Fire Sciences Laboratory, Missoula, MT and corresponding author (psikkink@fs.fed.us) 2 Vegetation Specialist, U.S. Department of Interior, National Park Service, Yellowstone National Park,WY 3 Research Ecologist, U.S. Department of Interior, U.S. Geological Survey, Northern Rocky Mountain Science Center, Bozeman, MT 4 Volunteer (2009-2012), U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station, Missoula Fire Sciences Laboratory, Missoula, MT 1 Assessing Sampling Methods YNP 2013 Final Report based on the entire list of species at each location (and their frequency or percent cover abundance) instead of a diversity value calculated from the species list and abundance. Using ordination, the biota captured by the historic method was similar to the biota captured by contemporary methods; the most difference between methods occurred in areas where the historic transects were parallel to ecotones. Methods that sampled larger areas than the historic method were able to capture diversity of the ecotone areas. Ordination was also useful in capturing the variation in the cover of grass and annual species from one sampling year to the next. The difference in time and effort required to complete each method will be important considerations that determine future monitoring directions in the park. We recommend that historic methods continue to be employed alongside a contemporary method that meets the park’s abilities to expend time, effort, and skills for monitoring. We also recommend that efforts continue to find a relationship between data collected using historic and contemporary methods. Finally, we recommend that an interdisciplinary team of subject-matter experts be convened to assist the park in establishing explicit goals, objectives, and procedures for implementing a long-term vegetation monitoring program. The results of this study should inform such an effort. Introduction In 2002, the National Resource Council (NRC) recommended that Yellowstone National Park (YNP) find a new, robust method for monitoring vegetation in its northern ungulate winter range that could distinguish between several change factors affecting plant community composition, including grazing, climate change, and natural and anthropogenic disturbances (National Research Council 2002). To meet this challenge, several small- and large-scale field sampling methods were installed between 2009 and 2012 to evaluate the ability of each method to capture 1) species composition and 2) species coverage within diverse plant communities of the northern range. The methods included a 100-ft line-intercept (Parker 1954), which was used almost exclusively in the past, a 20 x 50 cm Daubenmire frame (Daubenmire 1959), a small circular nested plot (Barnett et al. 2007), a modified Whittaker plot (Stohlgren et al. 1995), and a multi-circle Forest Inventory and Analysis (FIA) Phase 3 plot design with 12 1-m2 quadrats (http://www.fia.fs.fed.us/program-features/basic-forest-inventory/ ). The goals for this study were similar to the goals that drove establishment of the ungulate exclosures and permanently-marked transect lines in YNP in 1957; that is, to find a sampling method that would best (1) describe the diversity in vegetation across the northern winter range and (2) detect any change in vegetation composition and structure between years and sample sites. The NRC recommendation in 2002 elevated the emphasis on detecting the causes of change, whether they were climate-driven or disturbance-driven, so the permanent transect lines and ungulate exclosures that were established in the park in 1957 have become even more critical to identifying grazing versus climatic factors in monitoring efforts. The importance of detecting cause and effect has also become important to YNP to inform management decisions on herd sizes (based on carrying capacity of the vegetation), invasive weed management, plant restoration, and park response to climate change in the 21st century. 2 Assessing Sampling Methods YNP 2013 Final Report The objectives of this study were: 1. To evaluate how each of the sampling methods captures species composition and biodiversity in plant communities of the northern ungulate winter range. 2. To evaluate whether the historic vegetation sampling method is sensitive and/or robust enough to use in interpreting vegetation-change factors in longterm data. 3. To determine the feasibility of developing a crosswalk between past and future data. 4. To determine whether one sampling method is sufficient for capturing changes in vegetation species and abundances throughout the entire northern ungulate winter range. 5. To evaluate the tradeoffs of time, effort, knowledge, and financial costs associated with each sampling method to monitor for vegetation change in the future. Background In the early- to mid-1900s, Yellowstone National Park (YNP) was the center of heated debates over whether its ungulate herds were exceeding the park’s capacity to support them and whether the vegetation resource was being irreparably degraded for future generations. To address public and neighboring rancher concerns, seasonal killings of elk and bison within the park became common practice to keep the ungulate population within predetermined limits (Barmore Jr. 2003; Yellowstone National Park 1997). The “range war” that resulted for and against the culling of these animals, and for or against natural regulation of herd sizes (Cole 1971; National Research Council 2002), led park managers to design a natural experiment to determine if there really was scientific evidence that would support the claim that the northern ungulate winter range was being grazed beyond its carrying capacity. In 1957, the new natural experiment included five range exclosures that were constructed along the length of northern winter range (Fig. 1). Their sole purpose was to attempt to answer the questions on grazing and carrying capacity of the northern range vegetation (Edwards 1957). In 1962, three more exclosures were erected to coincide with new herd-reduction measures. Inside and outside of the exclosures, park managers installed permanently marked transect lines to monitor vegetation. The historical sampling method used to evaluate vegetation change was the Parker 3-step point-intercept method (Parker 1954; Parker and Harris 1958). The Parker 3-step was rigorously tested in the mid-1950s to insure that it would adequately capture vegetation change over the five years that park managers expected the experiment to run (Parker and Harris 1958). Several transects were sampled prior to beginning the natural experiment in 1957 using the Parker 3-step method. These transects were established between 1954 and 1957 at six sites (Fig. 1) and were referred to as “free range” transects because they were open to grazing. More than 50 years after the natural experiment began, it is still running and the same sampling method is in use to monitor the vegetation. Over the past 55 years, several researchers have sampled the permanently marked transects associated with the large ungulate exclosures to evaluate the effects of ungulate grazing on the park’s sagesteppe ecosystem (Barmore Jr. 2003; Denton 1958; Denton and Kittams 1958; Houston 3 Assessing Sampling Methods YNP 2013 Final Report 1982; Sikkink 2002; Sikkink and Alaback 2006). While the historical method is simple and rapid to administer, it is perceived to lack robustness because of its minimal spatial extent and opportunistic monitoring schedule. The monitoring schedule for the northern winter range over the past 50 years has been intermittent, at best, and conducted by students, volunteers, or park employees with varying knowledge of the plants and minimal experience in vegetation monitoring. In 2002, the National Resource Council recommended that a new monitoring method be found that would better assess changes in vegetation due to climate change, grazing, or other disturbances that affect vegetation in the park (National Research Council 2002). Fig. 1. Sample sites across the northern winter ungulate range (red circles). Inset shows location of 1998 winter range boundary (stippled) (courtesy of YNP Geospatial Center), sample locations, and national forests surrounding the park that are also important for ungulate winter range. Today, there is a wealth of monitoring data for the northern range covering over 50 years. Numerous research projects exist that were based on the exclosures and permanent transects; some used the historic sampling method and some did not. Controversies still exist surrounding vegetation condition in the northern range because of current densities and distributions of the bison, elk, pronghorn and other ungulate herds in YNP (Buffalo Field Campaign 2012; Lundquist 2012). The new range wars center on whether the depletion of willow, aspens, cottonwoods, native grasses, and even sagebrush are due to changes in herd sizes. The answer to the question of whether grazing or climate is central to current vegetation changes in the modern range wars is more important than ever; and the historic exclosure data, along with a sampling approach that provides comprehensive data on biodiversity and plant community composition across the northern ungulate winter range, are critical to answering questions on change. The question on whether the historic monitoring technique is 4 Assessing Sampling Methods YNP 2013 Final Report effective at capturing those biodiversity changes and whether exclosures are effective to determine the effects of grazing is still highly debated (Stohlgren et al. 1999). When the NRC recommended using a more rigorous method to monitor vegetation change, few studies were available that compared biodiversity captured by the historic method with biodiversity captured by other sampling methods (Stohlgren et al. 1998). Methods Field data Data were collected from sampling methods that included the historic Parker 3-step, a small circular plot with nested quadrats (Barnett et al. 2007), 20 small Daubenmire frames (Daubenmire 1959) , a modified Whittaker plot (Stohlgren et al. 1995), and a set of four circular plots containing nested quadrats designed for the Forest Inventory and Analysis (FIA) program (http://www.fia.fs.fed.us/program-features/basic-forestinventory/). Each method covered a different spatial scale, but all were centered on a historic, permanently-marked transect so the methods covered the same area and could be repeated in any future monitoring program (Fig. 2). E. Forest Inventory and Analysis Plot -shown 1/2 scale (12 microplots, 4 subplots) D. Modified Whittaker Plot (10 microplots, 3 subplots) C. Daubenmire Plot (20 frames = 20 microplots) B. Nested Circular Plot (3 microplots, 1 subplot) A. Parker Cluster (3 transect lines combined = 300 microplots; if 2 transects, microplots=200) Fig. 2. Schematic layout of a Parker transect line (A) overlain by each contemporary sampling method. The center rebar of the 100-ft (30.48 m) Parker transect line (lower level) is used to center each method. The Parker method encompasses 0.03 m2 analyzed/transect. Contemporary methods include a nested circular plot (B) encompassing 3 m2 analyzed/plot; a set of 20 Daubenmire frames (C) with 2 m2 analyzed/plot; a modified Whittaker plot (D) with 10 m2 analyzed/plot and a species list 5 Assessing Sampling Methods YNP 2013 Final Report constructed with additional subplots and the entire plot area; and a Forest Inventory and Analysis (FIA) plot (E) with 12 m2 analyzed/plot. Groups of one to three Parker transect lines with similar grazing potential, slope, and aspect were designated as “clusters” in 1958. Whittaker plots were centered on the center rebar of the Parker line but rotated to align with the environmental gradient. FIA layer is diagrammed at ½ scale to fit within the block. Time to sample each location was recorded to gauge total time needed to monitor each site using a particular method. Time to complete included (1) setup, (2) sampling and (3) total effort (i.e., removal of tapes, flags, etc.). Hikes to and from sample locations were not included in total efforts. Four of the methods were sampled almost exclusively by one field crew from 2009 to 2012. A second crew sampled all of the small circular plots with nested quadrats in 2009 and the two FIA plots at the Blacktail exclosure in 2010. 2010 and 2012 data for the small circular nested plots were pulled from the center FIA circle sampled in those years. No samples were collected in 2011. Historic Parker 3-step The Parker 3-step (hereafter referred to as the Parker method, Parker transects, or historic method) is the historic monitoring technique for the northern ungulate winter range of Yellowstone National Park. It is a modified point-intercept method devised to sample vegetation and substrate characteristics along a permanently marked 100-ft line (30.48 m) (Parker 1954). Unlike strict point-intercept sampling, a loop was used for each measurement instead of a point and the modification was an attempt to add area to the sampling method. Vegetation was sampled in the Parker method along a 100-ft line using a ¾ in. (1.9 cm) diameter loop. The loop was lowered to the ground at each foot mark along the 100-ft line and plants whose basal portion fell within the loop were recorded as “hits”. The “hits” could consist of shared species if plants are small and the basal portions of both fell within the loop. If vegetation was not present within the loop, the “hit” was recorded as substrate. Substrate elements included bare ground, pavement (rocky hardpan surface), rock (>7.6 cm), moss and lichen combined, and litter. Substrate elements were never recorded as a shared hit. Live elements (i.e., moss and lichen) took precedence over the non-living substrate elements when recording. For the non-living substrate, the element that covered the most area within the loop was recorded. If shrubs extended over the transect line at a foot mark, the shrub species was recorded as “overstory.” Overstory was not a factor in data analysis for the Parker transects in this study. Each Parker transect was originally established as part of a “cluster” group (Fig. 2A). Clusters had similar grazing potentials, slopes, and aspects. They existed inside and outside of exclosures and were comprised of one to three transects per cluster depending on location. For this study, all data for the Parker method were summed and averaged by the cluster group. Clusters were also summed and averaged for the “free-range” transects by combining the multiple transects for each sample location. 6 Assessing Sampling Methods YNP 2013 Final Report Parker transect data were most appropriate for calculating frequency or percentages of species and substrate, although coverage calculations were possible. In this study, frequency and cover data for the Parker method were equivalent. Each loop was considered a “microplot” and no sub-plots existed. The total area sampled along each Parker transect was 0.32 ft2 (0.03 m2). Circular plot with nested quadrats The small circular plot is 7.23 m in diameter and contained three 1 x 1 m sampling frames (quadrats) within its circumference (Fig. 2B). Each of the 1 x 1 m sampling frames within the circle were placed 4.57 m from the circle center at 30o, 150o, and 270o off due north (Fig. 2b). To sample the frames, visual estimates to the nearest 1% were made for all species and substrate elements. In addition to estimates of cover within the sampling frames, visual estimates of cover were made to the nearest 1% for each species within the entire circle. The full circle covered 168.2 m2 and was considered a subplot and used mainly for the species tallies in this study. The 1x1 m quadrats sampling frames were considered “microplots” and used in all data analyses. The total area sampled in microplots of the small circular plot was 3 m2. Daubenmire small frame This sampling method used a 0.20 x 0.50 m sampling frame placed along one side of the historical transect (Daubenmire 1959). Frames were started at zero with the long edge aligned along the 100-ft tape (Fig. 2C). Subsequent frames were placed every five feet along the transect line for a total of 20 samples. Within each frame, percent cover of all species and substrate elements were visually estimated using a 1-6 cover scale. Estimates of plant and substrate coverage often totaled more than 100% because species overlapped. Only the vegetation portion within the frame was estimated if the base or foliage extended beyond the frame edge. In addition to the substrate variables required by the historic method, the Daubenmire method required cover estimates for persistent litter. All Daubenmire estimates were converted to the mid-point of each of the six classes to establish the cover values for all species and substrate elements. Each plot frame was considered as a “microplot” for this study and there were no sub-plots. The total area sampled along each Daubenmire transect was 2 m2. Modified Whittaker plot The modified Whittaker plot is a common contemporary sampling method covering 50 x 20 m (Stohlgren et al. 1995). It is comprised of 10 0.5 x 2 m, two 2 x 5 m, and one 5 x 20 m subplot nested within the 20 x 50 m plot. Percent cover of all species and substrate was visually estimated to the nearest 1% within the 0.5 x 2 m plots (Fig. 2D). In addition to the substrate variables required by the historic method, the modified Whittaker method required cover estimates for persistent litter. Only presence/absence data were collected for plants within the 2 x 5 m subplot, the 5 x 20 m subplots, and the full 20 x 50 m plot, all of which were used for species counts in this study. Each small 0.5 x 2 m subplot was considered a “microplot”; all of the larger plots were considered sub-plots. The area sampled for the entire method 7 Assessing Sampling Methods YNP 2013 Final Report was 1000 m2. The area sampled by the microplots and used in all statistical analyses was 10 m2. Forest Inventory and Analysis (FIA) plot The FIA plot consists of four circular plots with nested quadrats (Fig. 2E). The method is currently being used across the United States for inventory and monitoring of forests, shrublands, and grasslands. FIA plots for this study were established according to standard FIA methods for Phase 3 sampling (http://www.fia.fs.fed.us/program-features/basic-forest-inventory/). Species and substrate were visually estimated for the 1 x 1 m quadrats and the entire 7.28 m diameter circles to the nearest 1% as described above. In addition to substrate cover estimates required by the historic Parker 3-step method, cover estimates in the FIA plots included dead wood and dung. The total area sampled for this method was 672.7 m2. The full circles were considered subplots. The 12 1 x 1 m quadrats were considered “microplots” for analysis in this study. FIA samples covered a total area of 12 m2. Data analysis Analyses for this study were conducted on frequency data, percent cover data, and on several biodiversity measures calculated from the frequency and cover data including richness5, Simpson’s diversity index6, Shannon’s diversity index 7, and evenness8. Statistical analyses used the microplot data; unique species analyses used the entire plot area. Differences between methods primarily focused on richness (i.e., the number of different species represented in a field sample or ecological community) and Simpson’s diversity index (i.e., a measure of the likelihood that two randomly chosen plants from a sample unit or ecologic community would be different species) (McCune and Grace 2002). Data sets included the following: Frequency data of species and substrate were computed for each sample site based on the number of times a species or substrate occurred in all of the microplots that comprise a method. Percent cover of each species or substrate was calculated as an average of all cover estimates from the microplots that comprise a method. 5 The diversity measure general form is , where D=diversity, p=proportion of individuals belonging to species I, and s=number of species; if a=0, then D o = number of species in a sample unit or richness. Therefore, richness is just a count of species. 6 If a=2 in the general diversity form, then results in the diversity form of Simpson’s index, which is the likelihood that two randomly chosen individuals from a sample unit will be different species. 7 The Shannon-Wiener Index quantifies the uncertainty, or surprise, associated with correctly predicting which species will be drawn next in a string of draws. It is calculated as in PCOrd, which essentially is the log of the number of species of equal abundance. 8 Evenness is a measure of the relative abundance of the different species making up the richness of a sample unit. It quantifies how equal the species abundances are. As calculated in PCOrd, evenness is Pielou’s J or , where H’ is the Shannon-Wiener Index and S is the average species richness. 8 Assessing Sampling Methods YNP 2013 Final Report Biodiversity measures were calculated within PCOrd (McCune and Mefford 1999) from the frequency or cover species data obtained from microplots within each method. When locations were sampled in multiple years over the three sample years, diversity measures were calculated separately for each sample year to incorporate year-to-year changes within the data set. The five sampling methods, and their separation by year, resulted in 147 distinct method/location/year records for each analysis. Data were analyzed for species characteristics at each location and differences among methods, locations, and plant communities9. For all comparisons, the Parker transect data were analyzed by grouping individual transects into their originally designated clusters of one to three transect lines. Species and substrate frequency and cover values for the clusters were formed by averaging the values from individual lines that formed each respective cluster. Data distribution, tests for normality, homogeneity of variance, and significance tests among groups were tested within SAS 9.3 (SAS Institute Inc. 2008) using non-parametric analyses. Tests for normal distribution were conducted within SAS on groups of unequal size using the Shapiro-Wilk test (Shapiro and Wilk 1965). Homogeneity of variance was tested with Brown and Forsythe’s test (Brown and Forsythe 1974), because it is more robust for non-normal distributed data (SAS Institute Inc. 2008) than either Levene (Levene 1960) or Bartlett tests (Bartlett 1937; Snedecor and Cochran 1989). KruskalWallis Rank Sum tests (Kruskal and Wallis 1952) were conducted within SAS to determine if groups were significantly different. Groups that tested positive for significant differences in their means with Kruskal-Wallis tests were further examined with Dunn’s Pairwise Multiple Comparison tests (Dunn 1961) in SAS (Elliot and Hynan 2011) to determine which groups within the methods and location tests were different. All statistical analyses were considered significant if p<0.05. Descriptions of species characteristics The difference between the number of species captured in the microplots and the number of species captured by an entire method was described using tallies of unique species. The tallies gave a better picture of the ability of a method to assess species richness on the landscape because the sub-plot data from each method, which were comprised of presence-absence data or estimates of species cover for whole plots, was ignored in all PCOrd calculations of biodiversity. Therefore, the actual number of species captured by each sampling method was under represented in most statistical analyses. Sample data and information on species names and phenologic characteristics were stored in linked tables and queries within an Access database (Microsoft Corporation 2010). Filters were used to output information on species 9 A plant community is an assemblage or association of populations of two or more different species occupying the same geographical area. 9 Assessing Sampling Methods YNP 2013 Final Report characteristics by location, method, or both. The information on species characteristics by method and locations was summarized in tables and bar charts. Tests for differences among methods To compare how the five sampling methods detected values of biodiversity, the diversity calculations were analyzed separately by location and sample year. A Skillings-Mack statistic for non-parametric, unbalanced, incomplete (i.e., missing data) block sample designs was used to test for differences among the methods. The statistic was run within SAS using procedures by Cunningham (2010). Tests were run on the 2009 sample methods as a group; and on the 2010-2012 methods as a group. Within each statistical test, a weighted sum of centered ranks was calculated that indicated the rank order of observations, such as which had the highest richness or Simpson’s Index values and which methods were ranked closest to each other for each test. Box plots were constructed to visually show the range and differences in means among the five methods. Diversity measures were also compared among the historic method and contemporary methods using graphical correspondence measures. Plot diversity measures obtained from the Parker method were plotted against the same diversity obtained from a contemporary method on a scatterplot where a perfect fit between Parker diversity measures and a contemporary method was marked by a 1:1 correspondence line. The plots were analyzed for the amount of scatter and the distance from the correspondence line to determine similarity of method results. The correspondence line was also compared to a linear regression line through the data points for each method to determine whether there was a relationship that might be exploited to relate historic and contemporary methods. Tests for differences among locations Only Parker 3-Step samples were used to test whether there were differences in diversity between the inside and outside of exclosures or the free range areas. The Parker transects were sampled each year of the study, both inside and outside of exclosures, and by the same crew; therefore, the variation in the plots was theoretically due mainly to year-to-year differences in vegetation that could be detected with this method. Parker transect data were coded by location (inside exclosures, outside exclosures, free range, etc.) and processed together using Kruskal-Wallis tests to detect differences in group means. 1958 and 1962 exclosures were combined for the inside samples at each respective location. Tests for differences among communities Because plant communities can have similar numbers of species in various locations but very different community compositions, diversity measures alone can be inadequate to show the variation in species that would need to be monitored across the northern ungulate winter range. To determine whether the biota of the northern winter range divided into distinct biotic communities and whether different sampling methods captured the community composition similarly, an ordination was conducted using the percent cover data for the biota. Ordinations use a distance measurement to calculate how similar samples are in n-dimensional 10 Assessing Sampling Methods YNP 2013 Final Report space. Samples that plot close together in n-dimensional space are more similar in composition than those that are separated by great distance (McCune and Grace 2002). For this analysis, a non-metric multidimensional scaling (NMS) ordination was chosen (Kruskal 1964; Mather 1976). A number of combinations of species data (from all methods, locations, and years) and substrate data (using all Parker 3-step categories, slope, elevation, and aspect) were tested for the NMS, but the best representation of the biotic community was obtained when the ordination was run with only species data. The relationships of the substrate and the environmental variables with the species assemblages were explored with correlation analysis. NMS was run within PCOrd using Sorensen (Bray-Curtis) distance and the built-in automated process (see McCune and Grace (2002) for a full description of the automated process). One analysis incorporated cover data from all locations, all methods, and all years. A second analysis incorporated just Parker transect data from all locations and all years to show how one method could vary from one sampling year to the next. The data set for all methods consisted of 147 records with 205 species; the data set for the Parker clusters consisted of 125 records with 113 species. Prior to running each ordination, the raw frequency or percent cover data were transformed with a simple logarithmic transformation that put the data on similar relative scales but still preserved rank orders. Results During the three years of sampling, data were collected from 125 individual Parker transect lines that were grouped into 44 individual clusters, 36 nested circular plots, 32 Daubenmire plots, 8 modified Whittaker plots, and 6 FIA plots. We summarized the sampling results in several ways. First, we described the general characteristics of plant communities for the northern winter range as a whole. Second, we grouped and summarized biodiversity data by sample method to examine how well each sampling method detected the variety of species present in the winter range. Third, we grouped all data to summarize species presence and abundance inside and outside of exclosures and for the free range areas. Finally, we used ordination graphs for individual plots to summarize similarities in species composition and abundance within the plant communities and their changes over time. Statistical summaries for each group of data sets are available in Appendix A (Tables A1A4). Regardless of how the data were grouped, distribution analyses, Shapiro-Wilk, and Brown-Forsythe tests showed that most of the diversity measures for each group were normally distributed and that they had distinct differences in group variances (Appendix A, Table A5-A6). Within each data set, however, one or two elements were found to have non-normally distributed data or equitable variances, which resulted in a decision to use non-parametric tests to determine if group differences were statistically significant. The groups that were not normally distributed or had equitable variances are discussed in their applicable sections below. General characteristics of the northern winter range biota The total number of species captured by all microplots and sub-plots for all sample methods during the three-year study was 205. This tally of species was not intended to 11 Assessing Sampling Methods YNP 2013 Final Report be a comprehensive list of flora present within the park or the northern winter range. It only includes the biota encountered during three years of sampling in a restricted number of locations. Of the number of unique species tallied at each sample location, forbs comprised 55 - 65%. Grasses comprised about 20% of the total numbers. The proportions of grasses, forbs, and shrubs were fairly consistent across all locations. Although mosses and lichens were an important component to ecological function in all sample areas, they were not individually identified for this study; rather they were grouped together as a moss/lichen entity during sampling and there is not a true measure of their diversity within the sampled areas. Differences among methods Sampling time The five methods explored in this study took significantly different times to complete (p<0.05). The average times for set up and sampling for each method were: Historic method (clusters): Nested circular plot: Daubenmire (20 frames): Modified Whittaker: Full FIA (4 circular plots): <1.5 hr 2 hr 3 hr 5 hr 7 hr None of the sample times included hiking to the locations, finding the permanent markers used as center points for each method, or identifying unknown plants during the cover estimation process. All unknowns were given descriptions, collected, and identified in an office after sampling was complete. A two-person crew was used for all methods. One crew had 6 years’ of experience working with plants of the northern range and two of the sampling methods; the second had no previous experience with the local vegetation but extensive experience using the nested circular plot method in other geographic areas including sage-steppe. Species list The total number of species captured by each sample method is shown in Table 1. The methods that covered large spatial scales (i.e., the Whittaker and FIA plots) captured at least 50% more species than either the Parker clusters or the nested circle but they also occasionally sampled vegetation that was very different from what was intended to be monitored by the historic method. Lamar In and Junction Butte In (In=inside exclosure) were prime examples of this problem. Because of the angles and distance off center that the circular plots were placed, two of the FIA circles were situated in forested communities or shrub areas that were very different in composition than the areas sampled by the other two FIA circles or than areas sampled by the other methods used in this study. As a result, the number of unique species captured at these two locations appears very high and contains species that are uncharacteristic for these particular grass and shrub communities (Table 3). Each method captured approximately the same numbers of species of shrubs, grasses, and moss/lichen lifeform species (Table 2). The nested circle and Daubenmire captured the highest number of unique species using similar numbers of plots. The Parker method captured a similar number of species but with 50% more plots. Forbs were 12 Assessing Sampling Methods YNP 2013 Final Report Table 1. Total number of unique species captured by each of the five sampling methods across the northern winter range separated by location and sample year. Where double exclosures exist, “IN” includes species numbers from inside the 1958 and 1962 exclosures combined. “Out” are matching locations to IN found outside the exclures. “Free” indicates free-range plots not associated with an exclosure. Totals include basal and overstory species from each method’s entire area. Sample years include 2009 (09), 2010 (10), and 2012 (12). Parker Year 0 9 1 0 1 2 Nested Circle 09 10 1 2 Daubenmire Whittaker 09 09 12 12 FIA 10 Blacktail In 25 23 34 44 14 1 38 36 34 Blacktail Out 24 28 30 40 211 39 36 37 Gardiner In 25 20 29 42 34 36 Gardiner Out 15 21 25 28 Junction Butte In 22 29 29 Junction Butte Out 18 23 30 Lamar In 30 34 39 Lamar Out 15 18 25 24 Mammoth In 27 29 49 42 Mammoth Out 20 18 34 35 Geode Creek Free 31 28 37 36 Blacktail Hill Free 24 24 27 29 Landslide Creek Free 12 18 18 27 23 Game Ranch Free 20 14 17 29 23 27 30 Lamar Horseshoe Free Specimen Ridge Free 22 22 20 19 21 33 34 24 -- 21 36 31 27 27 28 12 33 2 30 53 45 35 59 632 29 35 42 1 Nested plot data from 2010 and 2012 were constructed by using data from the center circle of each FIA plot. The FIA plots were measured by different personnel than measured the nested plots in 2009. For 2010 and 2012, only one center circle for inside and outside locations were measured, whereas four center circles were measured inside the Blacktail exclosure in 2009 and two were measur ed outside, which explains the large discrepancy in species numbers between 2009 and 2010. 2 Samples include forest and ecotone communities in part. 13 Assessing Sampling Methods YNP 2013 Final Report Table 2. Unique species captured by each of the five sampling methods identified by lifeform. Sample locations include only those areas where both the Parker method and a contemporary method were sampled in the same year. Data includes all basal and overstory species captured within each method’s entire area. Sample years include 2009, 2010, and 2012. Method Parker Clusters n 44 Nested Circle 30 Nested Circle (FIA centers) 6 Daubenmire 32 Whittaker 8 FIA 6 Conifer Tree Deciduous tree 1 1 Evergreen Shrub Deciduous Shrub Grass Forb Lichen Total Species 2 8 26 73 1 110 3 14 28 89 1 136 6 15 45 1 67 2 11 28 93 1 1 136 1 13 24 67 1 2 108 2 10 24 68 2 107 Total 205 Moss captured best by the Daubenmire method; although Whittaker and FIA methods would probably have been much higher than any of the methods if their sample numbers (n) were equivalent. The numbers of unique native and non-native species captured by each method had the same general relationships among methods as the lifeforms had (Table 3). Daubenmire and nested circle captured slightly more species in both the non-native and native types than the other methods. The nested plot extracted from the FIA center had the lowest number of unique species. In some cases, the low number of species for the extracted method was because the plots were done in areas of low diversity (e.g., Game Ranch and Landslide Creek); but in other areas, it may simply relate to the low sample number overall compared to the nested circles done in 2009. For the Whittaker and FIA plots, lower than expected total species for these large plots probably reflects the low number of total field samples completed for this method compared to the small-scale samples. The total number of non-native species encountered during sampling was 17. Nonnative species were found only within the grass and forb lifeforms; and they utilized a full range of lifecycles. The highest numbers of non-native species were found on the mudflats and disturbed areas near Gardiner; however, they were also found at every other location to varying degrees. The most commonly encountered perennial nonnatives included Bromus inermus (smooth brome), Camelina microcarpa (littlepod false flax), and Taraxacum officinale (common dandelion). The non-native grasses Bromus tectorum (cheatgrass), B. japonicus (Japanese brome), and Agropyron triticeum (annual wheatgrass) were concentrated mainly in the Gardiner sites. However, isolated pockets of Bromus tectorum and Bromus inermus were also found at Lamar Horseshoe along with the small annual mustard, Alyssum desertorum (desert madwort). Distribution and spread of three non-native, invasive species across the northern winter range (i.e., Bromus tectorum, Alyssum desertorum, Alyssum alyssoides, and Agropyron triticeum are covered in a separate report by Chong et al. (2011). 14 Assessing Sampling Methods YNP 2013 Final Report Table 3. Unique species captured by each of five sampling methods identified by origin. Data includes all basal and overstory species captured within each method’s entire area. Sample years include 2009, 2010, and 2012. Method n Number of locations sampled Parker Clusters 44 16 12 90 8 110 Nested Circle 30 16 14 117 5 136 Nested Circle (FIA centers) 6 4 6 57 4 67 Daubenmire 32 16 13 109 14 136 Whittaker 8 4 10 93 5 108 FIA 6 4 12 87 8 107 Total 205 Non-native species Native species Unknown 1 origin Total unique species 1 Unknown origin includes species identified to Genus only or plant specimens that were not identifiable in either the field or lab because of deterioration. Diversity comparisons Within the cover data, richness varied from a mean of 15.5 species in the nested circle plots to 30.7 species in the large FIA plots; mean Simpson’s Index values ranged from 0.76 in the nested circles to 0.84 in the Whittaker plots (see summary statistics for all the diversity measures using cover data in Appendix A (Table A-1)) . Simpson’s Index represents a combination of richness and evenness that indicates how evenly individuals and species are distributed among samples; higher Simpson’s index values indicate both greater richness and evenness. Richness values were lowest in the Parker clusters and nested circles (Fig. 3A), but Parker clusters had the greatest spread and highest Simpson’s Index values (0.46-0.92; see Fig. 3B). Distribution of all richness data was normal except for the nested circle method (Shapiro-Wilk p=0.004) (Appendix A Table A-5). Whittaker, and FIA plot data were normally distributed for the Simpson’s Diversity Index; Parker cluster, nested circles, and Daubenmire plots were not. The only analyses that had distinct differences in variance for the five methods was Simpson’s Index using cover data (Brown-Forsythe homogeneity of variance p=0.335). All others did not have significant differences in variance, which indicated that the variances within each method were no different than would be expected from random samples drawn from a population (Appendix A Table A-6). 15 Assessing Sampling Methods YNP 2013 Final Report A. B. Fig. 3. Box plots showing range of values for A) richness and B) Simpson’s Index for each sampling method using only cover data. Data encompasses multiple locations across the northern ungulate winter range and multiple sample years for each method. Boxes show the 25th percentile (lower line), median (middle line), and 75th percentile (upper line), and the whiskers show the limits for 1.5 times the interquartile range. The Skillings-Mack non-parametric tests showed significant differences between the three sampling methods used in 2009 (see Appendix A Table A-7). The weighted sum values from the Skillings-Mack tests on richness and Simpson’s Index indicate that (1) the Parker and nested plots are most similar and (2) the Daubenmire method captured the most richness but had the lowest Simpson’s Index (Table 4). For all five methods tested between 2010 and 2012, richness was significantly different for all of the methods but Simpson’s Index was not. The weighted sums again showed that the Parker and nested were similar and Daubenmire was quite different. They also showed that the large-scale plots were similar to each other and that they captured the most richness. Table 4. Weighted sum of center-ranked values from Skillings-Mack tests to determine how groups differ among methods using only cover data. All test statistics were significant except 2010-2012 Simpson’s Index. p-values <0.05 were considered significant. Weighted Sums Response Parker Nested Daubenmire Whittaker FIA Richness 2009 -18.19 -31.31 49.5 - - Simpson's Index 2009 17.32 -25.25 7.91 - - -12.3 -10.57 0 13.21 6.51 -7.21 -1.41 2.51 9.66 0.41 Richness Simpson's Index 16 Year 20102012 20102012 Assessing Sampling Methods YNP 2013 Final Report Correspondence between the historic and contemporary methods With the Parker cluster showing statistically significant differences with the Daubenmire, Whittaker, and FIA methods, scatterplots were constructed to see how far off of a 1:1 correspondence line with contemporary methods the Parker cluster plots were. The scatterplots showed that all methods except the nested circle plotted consistently above the 1:1 correspondence line using richness measures (Fig. 4); but, when plotted using Simpson’s Index data, the plots fall on both sides of the correspondence line in an unpredictable pattern (Fig. 5). Most of the plots were quite highly scattered, except when there were very few samples (i.e., Whittaker or FIA plots). When linear regression lines were plotted, most of the regression lines were very different from the correspondence line. The exception was for Daubenmire richness. It plotted nearly parallel to the 1:1 line but offset above it (Fig. 4). All other regression lines were quite different than the correspondence line. 17 Assessing Sampling Methods YNP 2013 Final Report A. B. C. D. Fig. 4. Correspondence of historic method (Parker clusters) and four contemporary sampling methods using the richness diversity measure and cover data. Pairings include: A) Parker cluster vs. nested circle; B) Parker cluster vs. Daubenmire; C) Parker cluster vs. Whittaker; and D) Parker cluster vs. FIA. The solid black line is the 1:1 correspondence line. The colored dashed line shows the linear regression through each data set; and the colored dotted line shows the area of 95% prediction. Regression line r2 values equaled A) 0.403; B) 0.704; C) 0.182; and D) 0.246. 18 Assessing Sampling Methods YNP 2013 Final Report A. B. C. D. Fig. 5. Correspondence of historic method (Parker clusters) and four contemporary sampling methods using the Simpson’s Index diversity measure and cover data. Pairings include: A) Parker cluster vs. nested circle; B) Parker cluster vs. Daubenmire; C) Parker cluster vs. Whittaker; and D) Parker cluster vs. FIA. The solid black line is the 1:1 correspondence line. The colored dashed line shows the linear regression through each data set; and the colored dotted line shows the area of 95% prediction. Regression line r2 values equaled A) 0.236; B) 0.527; C) 0.913; and D) 0.083. NOTE: The clumping of a small number of points for the regression in “C” results in a 95% prediction line that is perpendicular to the 1:1 line, resulting in a deceptively high r2. 19 Assessing Sampling Methods YNP 2013 Final Report Differences among locations Richness and Simpson’s Index measures on Parker transect cover data were highly variable at each location. The mean number of species for the Parker Cluster cover data ranged from a low of 12 species in plots outside of the Gardiner exclosures to a high of 29 species at Geode Creek (GO and Y6 in Appendix A, Table A-4). The greatest variability in richness occurred in the transect lines outside of the Blacktail exclosure (Fig. 6); whereas the most variability in Simpson’s Index measures occurred in two free range plots, Y3 and Y4, located near Gardiner (Fig. 7). All richness data were normally distributed except for the data set from Landslide Creek (Y3) (Shapiro-Wilk p<0.0001 in Appendix A, Table A-5). All Simpson’s Index data were normally distributed except for the plots inside the Blacktail exclosure (Shapiro-Wilk p=0.02). Each group had distinct differences in variance for both richness and Simpson’s Index (Brown-Forsythe Homogeneity of variance p=.318 and 0.685, respectively). Numbers of samples at each location ranged from two to 10 (Appendix A Table A-4). Fig. 6. Richness results by location using only the Parker 3-step data. Boxes show the 25th percentile (lower line), median (middle line), and 75th percentile (upper line), and the whiskers show the limits for 1.5 times the interquartile range. BI=Blacktail Inside Exclosures; BO=Blacktail Outside Exclosures; GI=Gardiner Inside Exclosures; GO=Gardiner Outside Exclosures; JI=Junction Butte Inside Exclosure; JO=Junction Butte Outside Exclosures; LI=Lamar Inside Exclosures; LO= Lamar Outside Exclosures. Y-Series = free-range plots pre-construction of 1957 exclosures (Y1=Geode Creek, Y2=Blacktail Hill, Y3=Landslide Creek, Y4=Game Ranch, Y5=Lamar Horseshoe, Y6=Specimen Ridge). 20 Assessing Sampling Methods YNP 2013 Final Report Fig. 7. Simpson’s Diversity Index results by location using only the Parker 3-step data. Boxes show the 25th percentile (lower line), median (middle line), and 75th percentile (upper line), and the whiskers show the limits for 1.5 times the interquartile range. BI=Blacktail Inside Exclosures; BO=Blacktail Outside Exclosures; GI=Gardiner Inside Exclosures; GO=Gardiner Outside Exclosures; JI=Junction Butte Inside Exclosure; JO=Junction Butte Outside Exclosures; LI=Lamar Inside Exclosures; LO= Lamar Outside Exclosures. Y-Series = free-range plots pre-construction of 1957 exclosures (Y1=Geode Creek, Y2=Blacktail Hill, Y3=Landslide Creek, Y4=Game Ranch, Y5=Lamar Horseshoe, Y6=Specimen Ridge). Groupings on location were significantly different in both richness (Kruskal-Wallis p=0.0025) and Simpson’s Index (Kruskal-Wallis p= 0.0001) (Appendix A, Table A-8). However, the differences were significant only between the Gardiner Out location and Geode Creek free range using richness values; and Geode Creek and Landslide Creek locations using Simpson’s Index values. There were no statistical differences between any of the exclosures and their outside plots, or among the individual free range plots, using the data collected between 2009 and 2012. 21 Assessing Sampling Methods YNP 2013 Final Report Differences among communities Ordination using all data The ordination diagram formed by using data from all methods and all years showed four important results. First, the diagram showed that methods were intermixed throughout ordination space (Fig. 8), indicating that sample method was secondary to community composition in placing plots across the ordination. Second, for most of the samples, Parker transects were in the same area of the diagram as the contemporary methods. The most variation in sample-methods positions was for Mammoth, Lamar, and Geode Creek, which were all locations of high species diversity. Third, even though the diversity analyses had only two locations significantly different from each other using cover data, several sample locations plotted at extreme edges of 3-D ordination space. This indicated that the locations were actually quite different in community composition, which better reflected the impressions formed during field sampling than the results obtained by reducing community data to richness or Simpson’s Index. Finally, plots sampled inside and outside of exclosures were not greatly separated in ordination space, indicating they were more similar to each other than they were to other locations. The most important species separating plots in the NMS included Festuca idahoensis (Idaho fescue), Artemisia tridentata (big sagebrush), Poa secunda (Sandberg’s bluegrass), Phlox hoodii (Hood’s phlox), Antennaria microphylla (Rosy Pussytoes), and Erigeron corymbosus (Foothill Daisy). The first separation of plots was based on the presence or absence of Festuca idahoensis. F. idahoensis was present to some extent in all areas except Gardiner, so Gardiner In and Out, Landslide Creek, and Game Ranch separated from all other sample locations mainly on the absence of this particular perennial grass. Gardiner plant communities were also separated from other areas based on their unique shrubs, Krascheninnikovia lanata (winterfat), and Atriplex nuttallii (saltbrush), and abundant non-native annuals. Artemisia tridentata was quite important in dividing the ordination space. In the Parker transects, however, A. tridentata and other shrubs were considered as overstory species and overstory species were not included in microplot analyses. Unlike the other methods which reflected shrub presence, the Parker microplots only registered shrubs if there was a direct hit on the base of the shrub or a seedling that fell within the loop that could be counted as a hit. The factors that appear to be driving each of the NMS axes in Fig. 8 were the same factors that are important to many rangelands. Axis NMS 1, which explained 44% of the variation in the ordination, appeared to be driven mostly by evapotranspiration. From the low elevation areas near Gardiner to areas located in the shadows of hills or adjacent to coniferous forest stands for cooling effects, NMS 1 ranged from dry, hot, low elevation conditions on the left to moister, more shaded, and usually higher elevation areas on the right. Using the species associations along the axis, the lowest NMS1 values (i.e., the Gardiner area plots) consisted of plots with cactus, annual grasses and forbs with shallow root systems, and shrubs that were common in very dry environments. Most other areas consisted of plants that needed more moisture, such as Idaho fescue and large perennial forbs. Evapotranspiration integrates the 22 Assessing Sampling Methods YNP 2013 Final Report Fig. 8. 3-dimensional NMS diagram showing all samples from all years and methods. The plots were separated by Free Range and Exclosure areas for clarity. Both diagrams use exactly the same axes and increments so samples are in the same positions they would be in if all were plotted on the same diagram. Note that the plots are mostly separated by location, not method, in both diagrams; and that the historic Parker clusters are not widely separated from most other methods. NMS Axis 1 r2=0.44; NMS Axis 2 r2=0.28; NMS Axis 3 r2=0.17 (Total variance in the data explained = 89%). Possible drivers for distribution along the axes are given in the text. 23 Assessing Sampling Methods YNP 2013 Final Report temperature, precipitation, rocky soil, and elevation conditions into a single entity that appeared to explain the locations of plots along NMS1 fairly well. NMS 2 explained 28% of the variation in the ordination and plot distribution could also be attributed to a common driver in rangelands. Plots along NMS2 appeared to be driven by soil conditions, specifically texture and depth. At the lower end of NMS2 were thin, young, rocky soils characterized by the plots at Specimen Ridge. Soil map layers obtained from Yellowstone National Park’s Spatial Analysis Center showed the soils to be mollisols and bedrock. At the upper end of this axis were plots from Geode Creek. Geode Creek had more developed mollisols than Specimen Ridge; they were less rocky texture, moister, and supportive of several species indicative of a productive site. The fine-grained, mudflow deposits (inceptisols) near Gardiner differed from areas on either end of NMS 2 that were located on mollisols or mollisols/bedrock. Because soils were not specifically sampled and characterized in this study, their direct effects on the ordination could not be fully evaluated but vegetation was distinct on each type. NMS 3 explained 17% of the variation in the ordination. It appeared to be some effect associated with the exclosures themselves. Free-range plots and plots outside of the exclosures were mainly located at the lower end of the axis. Plots inside the exclosures were, in general, located at the upper end. These effects were not obvious in the field. Visually, most of the exclosures looked like they had about the same types of vegetation and cover on the inside as out. The effects may be due to cover reductions from grazing, less plant vigor due to lack of grazing, soil compaction on the outside of exclosures that affected plant growth or soil moisture retention, soil enrichment of nutrients due to animal droppings, or a number of other effects that cannot be determined from the substrate or environmental characteristics collected for this study. The distribution of inside vs. outside plots was, however, distinctive so more study needs to be conducted to more clearly determine cause(s). Ordination using Parker transect data The NMS was fairly sensitive to changes in composition and/or abundance from year to year. Using Parker data only, plots that were sampled in 2010 and 2012 experienced minimal changes in community characteristics so the NMS plots for those locations moved very small distances in the NMS space. When locations were sampled in 2009 and not again until 2012, the distances that the samples moved in NMS space were greater than those with the shorter sampling schedule (Fig. 9). A few of the sites with longer movements also had different numbers of transects sampled by the Parker method each year but, for the most part, the ordination highlights variations in biota not variations in sample numbers. The ordination using only the Parker Clusters distributes plots similarly to the ordination that uses all plots and all methods shown in Fig. 8. Because each axis from the two ordinations is derived from independent analyses, however, the drivers for each should not be interpreted as identical. Using only the Parker samples, NMS Axis 1 has hot, dry, low elevation plots on the right; and cooler, moister, more shaded, and higher elevation plots on the left (Fig. 9). NMS2 is distributed along the same soil gradient as described for Fig. 8. A third axis was not created during the ordination process using only Parker data. This should not be interpreted as Parker transects not exhibiting exclosure effects. 24 Assessing Sampling Methods YNP 2013 Final Report They may be integrated within axes NMS1 and/or NMS2 because of the limited number of samples in the three- year dataset. Fig. 9. NMS derived from Parker Cluster samples only. Arrows show changes in plant composition and/or abundance between sample years for the same locations. Samples were taken in 2009, 2010, and 2012. NMS Axis 1 r2= 0.55; NMS Axis 2 r2=0.32. The movement of the same clusters of plots within the ordination in Fig. 9 appeared to be caused mainly by variations in the year-to-year frequencies of the different grass and sedge species, but variations in the frequency of annuals (particularly Alyssum desertorum), and the timing of sampling in relation to vegetation development (i.e. some years we were on site when Lewisia rediviva was visible and other years we were not) were also important. There were no major plots where major influxes of new invasive species explained movements along each axis. Summary A major finding of this study is that the historic Parker transect sampling method, if grouped as clusters, is as effective at capturing species richness and abundance as the small-scale contemporary methods tested in this study. The Parker clusters are most effective in areas where the vegetation communities are fairly homogeneous; and even the large-scale plot methods have results similar to the historic Parker clusters if the samples come from homogeneous landscapes. Alternately, if the landscape is composed of several different types of plant communities and/or ecotones that are 25 Assessing Sampling Methods YNP 2013 Final Report located within short distances of the historic transects, there was much more variation between results of the contemporary and historic methods. Examples of such highly varied landscapes were found at the Mammoth, Lamar, and Junction Butte exclosures where the methods covering large spatial-scales, and even the smaller-scale nested circles, included data from ecotones that spanned grasslands and forests or shrublands and forests, which were not normally described using a line transect or microplot along a line. If a large-scale method is chosen for future monitoring effort in these diverse areas, there will be little correlation with historic data because they sample different vegetation communities. To meet some monitoring objectives for the park, however, these ecotones might be important for some purposes and large-scale methods would be justified. A second important finding of this study is that diversity calculations were ineffective at making connections between any historic and contemporary data sets. Too many sites have the same calculated richness and/or Simpson’s index values even though they are derived from very unique communities. The results gave statistically insignificant differences in sample data between locations when they were, in fact, quite different. The uniqueness of each site, and the ability of each method to capture its characteristics, was lost unless the full complement of species and their abundances were used to describe the sampled communities. Monitoring goals for YNP will be critical to make decisions on which monitoring method is most appropriate for future needs. The broad, long-term goals of the park are to monitor for vegetation change and to partition change among biotic and abiotic factors. The NRC directives are to find a method that can detail trends in biodiversity, plant spatial distribution, or both, in enough detail to determine cause and effect. Capturing species numbers and/or rare plants over time and defining their change mechanisms is a very different objective than discerning changes in forage species or cover for the ungulate herds and their underlying causes. Finding a single, cost-effective monitoring method to do both objectives will be difficult. More specific objectives will need to be formulated before any sampling method (or methods) can be recommended for a new monitoring program. In the meantime, each sampling method does have specific advantages and disadvantages to its use. The Parker historic method is easy to learn and takes the least amount of time to do. It requires little equipment to implement. It is as effective as the contemporary methods if used in homogeneous landscapes but does not capture species composition and abundance from ecotones or other diverse landscapes as well. The ordination of the Parker data showed little difference in the plant communities inside and outside of the exclosures, but the ordination encompassed only three years of data and all samples were collected under drought conditions when vegetation was stressed. The Parker method is also backed by 55 years of historic data, which is a very important legacy for correlations with climate change. No other method has this legacy data. The nested circular plot covers more spatial area than the Parker transects, but takes 25% longer to do. Estimating percentages to 1% accuracy for the entire 168.2 m2 circle is difficult without abundant experience, although this accuracy is not a normal component for data collection using this method. The mean richness and Simpson’s 26 Assessing Sampling Methods YNP 2013 Final Report Index values calculated from the microplots were the lowest of any method; but, using species tallies from both the microplots and entire circle, it did capture 40-50% more species than the historic method. Although there were no statistically significant differences between it and the Parker cluster data, correspondence and linear regression lines between it and the historic method were very low, indicating that it might not be appropriate for cross-walk correlations between it and the historic data. The Daubenmire method using 20 microplots improves on the historic method because it captures percent cover values for biota and substrate and it creates a more extensive species list to track changes in species over time. It is the most time intensive of the small-scale methods with low setup times but long times for estimating cover of all species and substrate elements in the 20 microplots. The method requires little equipment, and consistent estimates of cover percent for each species are easily made. It samples more area than the Parker transects, but much less than any other contemporary method. Richness measures using cover data correlated very closely with the historic method as did an offset linear regression line. In most areas, Daubenmire samples correlated closely with the historic method using full biotic composition and abundance (i.e., they plotted close together on the NMS diagrams), probably because both methods stayed on or adjacent to the historic line. There was also no significant difference between the Daubenmire and large spatial scale plots (Whittaker and FIA). It captured as many species as the Whittaker plots did using the entire 20 x 50 m area, which might make it a good compromise between small- and large-scale methods to monitor for changes in cover and changes in richness over time. Daubenmire legacy data exists on the historical transects only for 1994 (Prather and Prather 1994; Utley and Stringfield 1994) and minimally for 2002 (Sikkink 2002), but neither of these data sets are probably comprehensive enough to correlating variations in vegetation with climate change or grazing. The Whittaker method is time intensive for both set-up and percent-cover estimations of species and substrate elements. Because species and substrate cover estimates are made within rectangular portions, however, it is like working a huge Daubenmire plot with half as many microplots. Unlike the Daubenmire plots where cover is estimated more quickly within broad cover classes, cover within the Whittaker plots is estimated to within the nearest 1% so the method captures highly detailed data for species and substrate variables. It did not capture the highest number of species for this study, but this is probably a reflection of the limited number of plots that could be established by one crew. One problem with using the full modified-Whittaker method is that, even though presence-absence data are easy to collect from the sub-plots, they are in such a different format than the microplot data that it is difficult to tie both together in data analyses. Because this method was not statistically different than either the Daubenmire or FIA methods, the decision to make it a preferred monitoring method will depend on whether the highest priority for monitoring is large spatial coverage or limited sample times. The large spatial area covered by this method is an asset if changes in cover are the priority because movement or disappearance of communities would have more probability of being captured using a larger plot. However, the reality of sampling 27 Assessing Sampling Methods YNP 2013 Final Report numerous Whittaker plots during a season, or even staggered over many seasons, may be unfeasible with limited personnel or budgets. The FIA method is the most time consuming of the methods. It captures the most species; and, because of its large size, it sometimes samples completely different vegetation communities in forests or ecotones. Like the nested circle, estimating cover to 1% accuracy for the full circles (four in total) was difficult without a wealth of experience. The method did not capture any more species than the Whittaker plot for the extra time it took to implement nor did it correlate with the historic method using correlation analysis or linear regression. It is also likely that the differences in vegetation communities affected the ordination analysis, and this influence will be examined in future analysis. Because of the long times necessary to complete this method, personnel might also be exposed to safety issues during the sampling process (i.e., bison, bears, and wolves), especially in the free range areas and areas outside of exclosures. Recommendations The analyses presented here clearly demonstrate that no single method of vegetation sampling maximizes inferences that are desired from the data. Each method has its own strengths and constraints, depending on the specific question or metric being analyzed. The analysis, however, does offer two options for consideration. The first option is for maintaining the long-term integrity of the database and the historic experimental design, including retaining the established exclosures. The second option is for developing a long-term, broad-scale vegetation monitoring protocol for the park. These options result in the following recommendations: 1) Retain the sites and historic methods currently employed to monitor vegetation in YNP, and further explore the reported correlation between species richness/abundance (Parker clusters vs. Daubenmire) and the temporal trends of species cover among and within sites. Within this report, we state that plant frequency is identical to percent cover using the Parker methodology, and that a linear, additive relationship appears to exist with Daubenmire cover richness. Exploring this relationship further and subjecting the Parker and Daubenmire cover data to a time series analysis would allow for an assessment of temporal changes in species cover among and within sites, including the effect of exclosure. Even though Daubenmire cover data is categorical, analysis using the mid-points of the cover classes should be sufficiently robust to allow for detection of real and significant changes in cover richness. The analysis would further reveal, for example, the arrival/presence of any non-native species that contribute to temporal changes in vegetation community structure. The outcome of the analysis identified above would further support a recommendation to retain the long-term ungulate exclosures and the sampling protocols that were established at the time of their construction. The historical experimental design is useful to monitor for vegetation trends and inform management of the ungulate influences; but, with more than 50 years of protection from grazing/browsing, it also adds power to detect any climate-induced changes in vegetation that are anticipated under future scenarios of global climate change. Even if broad-scale inference is 28 Assessing Sampling Methods YNP 2013 Final Report constrained by the existing experimental design (see below), “manipulation” by means of exclosures is the only rigorous means of determining cause-and-effect relationships between vegetation change, grazing, and climatic trends. A secondary recommendation, which goes along with the recommendation to retain the ungulate exclosures and historic sampling protocols, would be to continue to explore the correlations between historic and new sampling protocols. The correlations explored in this study, and in past studies, are based on sample sizes that are too small to construct convincing equations and/or relationships for a “bridge” or crosswalk between historic and current data. For this reason, the need to continue sampling with the historic sampling method is ongoing. With several more years of data collection using the historic method alongside a contemporary sampling method, AND collecting data under a variety of climatic conditions, building a crosswalk between past and present may become more feasible. The nature of the Parker data focuses change detection on common species. Shrubs are rarely included, even though they may be common in the community. Rare species are usually not factored at all as are nonnative species early in their establishment. In other words, sampling only common species inside and outside of exclosures may not be able to discern changes beyond those expected by background variation in moisture (Coughenour et al. 1994). However, the ordination of recent Parker data presented here (Figure 9) suggests a way that the different methods may be compared and the long-term data set re-analyzed to look for magnitude and direction of change between pairs of sampling years. So, our “bridge” may just be a recommendation to re-analyze historic and contemporary data using ordination techniques. 2) Develop a broad-scale vegetation monitoring program. The free-ranging transects (n = 15 transects in clusters of 2-3 at 6 sites) and exclosures still in existence today (n = 8 at 5 locations) were established to be coincident with planned elk herd reductions (195758, which did not occur, and 1961-62, which did occur) and not as part of a larger goal to maximize the spatial inferences to be drawn. Sites were deliberately selected to represent a variety of vegetation types and topo-edaphic conditions, and many transects within sites were purposely placed on ridgetops or upper slopes where harsh conditions and exposure to grazing was maximized (D. Houston, 1978 letter to files N1433). Standard methods of vegetation measurement (Parker transects for herbaceous strata, sagebrush, aspen, and willow belts, etc.) and analysis from that era were employed in an attempt to describe vegetation change. These early transects were established to detect vegetation changes between two permanently marked points due to grazing – not to describe the vegetation community of the site as a whole (Canfield 1941). Barmore Jr. (2003) reiterates this philosophy when he says that “the [Parker] method is not appropriate as an index or estimate of absolute measures of vegetation… but more appropriate for detecting vegetation differences over time and between sites.” He goes on to say that accuracy for the method can be increased by using more transects and grouping species. Because the early range studies used transects primarily to detect vegetation change due to grazing, little consideration was given to powerful and unbiased quantitative measures – balanced experimental design, stratification and replication, explicit hypothesis-driven field sampling, and statistical analyses- that would determine the exact information being pursued and the optimal methods to acquire and analyze the data. Today, the experimental design employed to 29 Assessing Sampling Methods YNP 2013 Final Report monitor for long-term vegetation change would encompass these modern requirements. Recognizing these potential biases and shortcomings, it is desirable to develop a broad-based vegetation monitoring program, not only for the sagebrushsteppe system of the park referenced here, but to encompass all major vegetation zones within the park. Therefore, it is recommended that an interdisciplinary team of subjectmatter experts be convened to assist the park in establishing explicit goals, objectives, and procedures for implementing a long-term vegetation monitoring program. The information and analyses provided in this report on sampling protocols, cost-benefit, and site conditions would inform such an effort. For reasons identified above, it is expected that historic vegetation monitoring sites/methods would be retained as part of a long-term vegetation monitoring program, and that basic and applied research objectives would be a component of the vegetation monitoring program to provide further insight into cause-and-effect relationships for vegetation change. Acknowledgements This study was conducted under Research Joint Venture Agreements between Yellowstone National Park, the U.S. Forest Service (09-IA-11221637-252) and the U.S. Geological Survey (1580090009). We sincerely thank the Yellowstone National Park Foundation for providing a portion of the funding that made this project possible. We also thank Kathi Irvine (statistician, USGS) and Scott Baggett (statistician, RMRS Forest Service), for valuable statistical discussions and guidance; Jared Woolsey, Ben Chemel, and Rebecca Saunders for field sampling the nested circle plots and two FIA plots; and Christie Hendrix, Stacy Gunther, and John Klaptowski for logistics and assistance during field sampling. 30 Assessing Sampling Methods YNP 2013 Final Report Literature cited Barmore Jr., William J. 2003. Ecology of ungulates and their winter range in northern Yellowstone National Park - research and synthesis 1962-1970, Mammoth Hot Springs, WY: Yellowstone Center for Resources, Yellowstone National Park, WY. 526 p. Barnett, D.T.; Stohlgren, Thomas J.; Jarnevich, C.S.[and others]. 2007. The art and science of weed mapping. Environmental Monitoring and Assessment. 132: 235252. Bartlett, M.S. 1937. Properties of sufficiency and statistical tests. Proceedings of the Royal Statistical Society. Series A 160: 268-282. Brown, M.B.; Forsythe, A.B. 1974. Robust tests for equality of variances. Journal of the American Statistical Association. 69: 364-367. Buffalo Field Campaign 2012. Overpopulation of Wild Bison Fact Sheet: Yellowstone National Park Buffalo - Are there too many? Wild Rockies.org. Canfield, R.H. 1941. Application of the line-interception method in sampling range vegetation. Journal of Forestry. 39(4): 388-394. Chong, G.W.; Barnett, D. ; Chemel, B. ; Renkin, R. ; Sikkink, P. 2011. Vegetation Monitoring to Detect and Predict Vegetation Change: Connecting Historical and Future Shrub/Steppe Data in Yellowstone National Park. In: Andersen, C. 10th Biennial Scientific Conference on the Greater Yellowstone Ecosystem Questioning Greater Yellowstone’s Future: Climate, Land Use, and Invasive Species; October 11–13, 2010; Mammoth Hot Springs Hotel, Yellowstone National Park. Yellowstone National Park, WY, and Laramie, WY: Yellowstone Center for Resources and University of Wyoming William D. Ruckelshaus Institute of Environment and Natural Resources: 84-92. Cole, G. C. 1971. An ecological rationale for the natural and artifical regulation of native ungulates in parks. Trans. North American Wildlife Conference. 36: 417-425. Coughenour, M.B.; Singer, F.J.; Reardon, J. 1994. The Parker transects revisited: Longterm herbaceous vegetation trends on Yellowstone's northern winter range. In: Despain, D. G. Plants and their environments: Proceedings of the first biennial scientific conference on the greater Yellowstone ecosystem; Sept. 16-17, 1991; Yellowstone National Park, Wyoming: USDA National Park Service Rocky Mountain Region Yellowstone National Park: 73-95. Cunningham, Michael. 2010. Statistics and Data Analysis Paper 275: A nonparametric method to assess treatment effects for unbalanced designs using SAS/IML. In: SAS Global Forum; 11-14 April 2010; Seattle, WA: 1-12. Daubenmire, R. 1959. A canopy-coverage method of vegetation analysis. Northwest Science. 33(1): 43-64. Denton, Gail 1958. Sage-belt transect field data from the big-game exclosures (Mammoth and Gardiner areas), unpublished data Yellowstone National Park Archive Box N-333. Denton, Gail; Kittams, W.J. 1958. Sage-belt transect field data from the big-game exclosures (Lamar and Blacktail outside area), unpublished data Report type. Yellowstone National Park Archives Dunn, O.J. 1961. Multiple comparisons among means. Journal of the American Statistics Association. 56: 52-64. Edwards, Harold O. 1957. Departmental letter to Yellowstone National Park park personnel on exclosure construction and rationale, p. 1. 31 Assessing Sampling Methods YNP 2013 Final Report Elliot, Alan C.; Hynan, Linda S. 2011. A SASR macro implementation of a multiple comparison post hoc test for a Kruskal-Wallis analysis. Computer Methods and Programs in Biomedicine. 102(1): 75-80. Houston, D.B. 1982. The Northern Yellowstone Elk: Ecology and Management, New York: MacMillan. 474 p. Kruskal, W.H. 1964. Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis. Psychometrika. 29: 115-129. Kruskal, W.H.; Wallis, W.A. 1952. Use of ranks in one-criterion variance analysis. Journal of the American Statistical Association. 47(260): 583-621. Levene, H. 1960. Robust tests for equality of variance. In: Olkin, I.; Ghurye, S.G.; Hoeffeling, W.; Madow, W.G.; Mann, H.B. Contributions to probability and statistics. Stanford, CA: Stanford University Press: 278-292. Lundquist, Laura 2012. Bison population exceeds Yellowstone Park limit. Bozeman Daily Chronicle, Bozeman, MT. Mather, P.M. 1976. Computational methods of multivariate analysis in physical geography, London: J Wiley & Sons. 532 p. McCune, Bruce; Grace, James B. 2002. Analysis of ecological communities, Gleneden Beach, Oregon: MjM Software Design. 300 p. McCune, Bruce; Mefford, M.J. 1999. PC-Ord. MjM Software Design, Gleneden Beach, Oregon. Microsoft Corporation. 2010. Microsoft Access 2010 for Windows. National Research Council. 2002. Ecological dynamics on Yellowstone's northern range, Washington, D.C.: National Academy Press. Parker, Kenneth W. 1954. A method for measuring trend in range condition on national forest ranges with supplemental instructions for measurement and observation of vigor, composition and browse. Report type. Washington, D.C.: U.S. Forest Service. 37 p. Parker, Kenneth W.; Harris, Robert W. 1958. The 3-step method for measuring condition and trend of forest ranges: a resume of its history, development, and use. In: Techniques and methods for measuring understory vegetation; October 1958; Tifton, Georgia: USDA Forest Service Southern and Southeastern Forest Experiment Stations: 55-69. Prather, Joy; Prather, Rick 1994. unpublished Daubenmire data, YNP Northern Range Exclosures and Free Range Plots, Yellowstone National Park archives. SAS Institute Inc. 2008. SAS for Windows version 9.3, Cary, NC. Shapiro, S.S.; Wilk, M.B. 1965. An analysis of variance test for normality (complete samples). Biometrika. 52(3-4): 591-611. Sikkink, Pamela G. 2002. unpublished field data, western Montana and Yellowstone National Park northern ungulate winter range, Missoula, MT. Sikkink, Pamela G.; Alaback, Paul B. 2006. Through the historical lens: An examination of compostiional change in Yellowstone's Bunchgrass Communities, 1958-2002. In: Biel, Alice Wondrak. Greater Yellowstone Public Lands: A century of discovery, hard lessons, and bright prospects. Proceedings of the 8th biennial scientific conference on the Greater Yellowstone Ecosystem; Oct. 17-19, 2005; Mammoth Hot Springs Hotel, Yellowstone National Park, WY: Yellowstone Center for Resources, Yellowstone National Park, WY: 148-158. Snedecor, George W.; Cochran, William G. 1989. Statistical Methods, 8th. ed., Ames, Iowa: Iowa State University Press. 32 Assessing Sampling Methods YNP 2013 Final Report Stohlgren, Thomas J.; Bull, Kelly A.; Otsuki, Yuka. 1998. Comparison of Rangeland Vegetation Sampling Techniques in the Central Grasslands. Journal of Range Management. Allen Press and Society for Range Management. 51(2): 164-172. Stohlgren, Thomas J.; Falkner, M.B.; Schell, L.D. 1995. A modified-Whittaker nested vegetation sampling method. Vegetatio. 117: 113-121. Stohlgren, Thomas J.; Schell, L.D.; Vanden Heuvel, B. 1999. How grazing and soil quality affect native and exotic plant diversity in Rocky Mountain grasslands. Ecological Applications. 9: 45-64. Utley, S.; Stringfield, D. 1994. unpublished Daubenmire data, YNP Northern Range Exclosures and Free Range Plots, Yellowstone National Park archives Yellowstone National Park. 1997. Yellowstone's northern range: complexity and change in a wildland ecosystem, Mammoth Hot Springs, Wyoming: National Park Service. 148 p. 33 Assessing Sampling Methods YNP 2013 Final Report Appendix A: Analysis Results Section 1.1 Summary statistics Table A-1a. Summary statistics for diversity measures by sample methods using cover data. S=Richness; E=Evenness; D’= Simpson’s Diversity Index. Method Variable Mean Std Dev Minimum Maximum n Parker 3-Step Cover S 17.32 5.00 6.00 31.00 65 E 0.77 0.11 0.43 0.93 65 D` 0.81 0.10 0.46 0.92 65 Skewness 6.35 1.73 3.71 10.29 65 Kurtosis 47.48 25.70 14.45 108.50 65 S 15.47 4.26 9.00 29.00 36 E 0.70 0.12 0.42 0.88 36 D` 0.76 0.12 0.48 0.90 36 Skewness 7.21 2.28 3.46 11.04 36 Kurtosis 61.60 35.59 13.22 125.65 36 S 24.22 6.12 14.00 36.00 32 E 0.66 0.10 0.37 0.83 32 D` 0.80 0.10 0.49 0.90 32 Skewness 7.08 1.81 4.40 10.91 32 Kurtosis 58.15 29.03 18.89 123.20 32 S 29.88 8.11 22.00 43.00 8 E 0.69 0.04 0.64 0.75 8 D` 0.84 0.03 0.80 0.89 8 Skewness 5.90 0.98 4.67 7.51 8 Kurtosis 38.93 14.08 23.07 64.22 8 S 30.67 9.83 22.00 49.00 6 E 0.65 0.07 0.50 0.70 6 D` 0.82 0.09 0.65 0.90 6 Skewness 6.02 1.37 4.31 8.49 6 40.90 19.08 19.75 76.79 6 Nested Circle Cover Daubenmire Cover Whittaker Cover FIA Cover Kurtosis 34 Assessing Sampling Methods YNP 2013 Final Report Table A-1b. Summary statistics for diversity measures by sample methods using frequency data. S=Richness; E=Evenness; D’= Simpson’s Diversity Index. Method Variable Mean Std Dev Minimum Maximum N Parker 3-step Freq S 17.32 5.00 6.00 31.00 65 E 0.76 0.10 0.45 0.93 65 Nested Circle Freq Daubenmire Freq Whittaker Freq FIA Freq 35 D` 0.81 0.09 0.48 0.92 65 Skewness 6.29 1.67 3.58 10.23 65 Kurtosis 46.52 24.75 13.27 107.69 65 S 15.47 4.25 9.00 29.00 36 E 0.96 0.00 0.94 0.98 36 D` 0.91 0.02 0.86 0.95 36 Skewness 3.20 0.91 1.09 4.67 36 Kurtosis 10.62 6.21 -0.07 23.29 36 S 24.21 6.11 14.00 36.00 32 E 0.89 0.02 0.81 0.94 32 D` 0.92 0.02 0.87 0.95 32 Skewness 3.48 0.69 2.36 4.97 32 Kurtosis 12.45 5.76 4.65 25.53 32 S 29.87 8.11 22.00 43.00 8 E 0.92 0.02 0.88 0.95 8 D` 0.94 0.01 0.93 0.96 8 Skewness 2.39 0.50 1.62 3.16 8 Kurtosis 5.13 2.71 1.82 9.46 8 S 30.66 9.83 22.00 49.00 6 E 0.91 0.02 0.89 0.95 6 D` 0.94 0.01 0.92 0.97 6 Skewness 2.51 0.76 1.36 3.55 6 Kurtosis 6.39 4.54 0.95 13.57 6 Assessing Sampling Methods YNP 2013 Final Report Table A-2. Summary Statistics for locations within the northern ungulate winter range using richness cover data. Richness Parker Cover Location Blacktail Inside BI Blacktail Outside Gardiner Inside Gardiner Outside Junction Butte Inside Junction Butte Outside Lamar Inside Lamar Outside Mammoth Inside Mammoth Outside Free: Geode Creek Free: Blacktail Hill Free: Landslide Creek Free: Game Ranch Free: Lamar Horseshoe Free: Specimen Ridge BO GI GO JI JO LI LO MI MO Y1 Y2 Y3 Y4 Y5 Y6 36 Mean 15.7 Std Dev 4.4 n 9 18.8 14.5 12.2 16.6 15.6 22.5 15.0 25.0 18.0 29.0 22.5 16.0 16.0 20.7 22.5 7.3 3.6 2.9 2.7 2.4 1.3 1.4 1.4 1.4 2.8 0.7 3.5 3.0 1.5 2.1 5 10 6 5 5 4 2 2 2 2 2 3 3 3 2 Assessing Sampling Methods YNP 2013 Nested Cover Std Mean Dev n 2.7 5 12.4 16.3 12.6 11.5 17.0 15.5 19.7 15.0 18.0 14.0 21.0 17.0 14.0 16.0 17.0 22.0 6.4 2.1 3.5 3.5 0.7 8.1 . 4.2 1.4 . . . 4.2. 4.2. . 3 5 2 3 2 3 1 2 2 1 1 1 2 2 1 Daubenmire Cover Std Mean Dev n 3.6 4 18.8 22.5 18.4 17.0 25.3 24.0 32.7 24.0 31.5 28.5 36.0 29.0 23.0 23.0 21.0 31.0 6.4 2.1 0.0 4.5 0.0 2.5 . 2.1 4.9 . . . . . . 2 5 2 3 2 3 1 2 2 1 1 1 1 1 1 Whitaker Cover Std Mean Dev n . 1 30 FIA Cover Std Mean Dev . 24 n 1 30 27 22 41 . . . . 1 1 1 1 30 . 1 33 . 1 43 . 1 49 . 1 23 23 . . 1 1 22 26 . . 1 1 Final Report Table A-3. Summary statistics for areas within the northern ungulate winter range using Simpson’s Diversity Index cover data. Location Blacktail Inside Blacktail Outside Gardiner Inside Gardiner Outside Junction Butte Inside Junction Butte Outside Lamar Inside BI BO GI GO JI JO LI Parker Cover Std Mean Dev n 0.06 0.79 9 0.09 0.82 5 0.05 10 0.83 0.07 0.76 6 0.06 0.84 5 0.02 0.87 5 0.90 0.01 Simpson's Diversity Index Nested Cover Daubenmire Cover Std Std Mean Dev n Mean Dev n 0.14 5 0.10 4 0.69 0.77 0.17 3 0.04 2 0.79 0.85 0.17 5 0.12 5 0.77 0.73 0.02 2 0.02 2 0.80 0.76 0.07 3 0.01 3 0.79 0.84 0.05 2 0.05 2 0.80 0.85 0.80 0.02 2 2 2 2 0.63 0.68 0.77 0.85 . 0.08 0.00 . 4 0.87 0.03 3 1 2 2 1 0.76 0.83 0.85 0.87 . 0.04 0.02 . 1 2 2 1 3 Whitaker Cover Std Mean Dev . 0.83 . 0.89 . 0.80 . 0.80 n 1 1 1 1 FIA Cover Std Mean Dev . 0.81 . 0.84 n 1 1 0.86 . 1 0.85 . 1 0.85 . 1 0.90 . 1 Lamar Outside Mammoth Inside Mammoth Outside Free: Geode Creek LO MI MO Y1 0.84 0.84 0.86 0.91 0.00 0.02 0.01 0.00 Free: Blacktail Hill Free: Landslide Creek Free: Game Ranch Y2 Y3 0.87 0.56 0.01 0.10 2 3 0.87 0.48 . . 1 1 0.86 0.49 . . 1 1 Y4 0.62 0.11 3 0.77 0.10 2 0.69 . 1 0.87 . 1 0.65 . 1 Free: Lamar Horseshoe Free: Specimen Ridge Y5 0.87 0.01 3 0.79 0.08 2 0.88 . 1 0.85 . 1 0.85 . 1 Y6 0.90 0.02 2 0.90 . 1 0.90 . 1 37 Assessing Sampling Methods YNP 2013 Final Report Table A-4: Summary statistics for locations sampled by Parker 3-step only. BI=Blacktail Inside Exclosures; BO=Blacktail Outside Exclosures; GI=Gardiner Inside Exclosures; GO=Gardiner Outside Exclosures; JI=Junction Butte Inside Exclosure; JO=Junction Butte Outside Exclosures; LI=Lamar Inside Exclosures; LO= Lamar Outside Exclosures. Y-Series = free-range plots pre-construction of 1957 exclosures (Y1=Geode Creek, Y2=Blacktail Hill, Y3=Landslide Creek, Y4=Game Ranch, Y5=Lamar Horseshoe, Y6=Specimen Ridge). Location BI BO GI GO JI JO LI LO MI 38 Variable Mean StdDev Minimum Maximum n S 15.67 4.39 10.00 22.00 9 E 0.74 0.06 0.64 0.83 9 H 2.01 0.18 1.70 2.23 9 D` 0.79 0.06 0.67 0.84 9 S 18.80 7.26 11.00 27.00 5 E 0.76 0.10 0.60 0.89 5 H 2.18 0.43 1.59 2.52 5 D` 0.82 0.09 0.67 0.89 5 S 14.50 3.63 6.00 20.00 10 E 0.81 0.07 0.70 0.93 10 H 2.12 0.25 1.67 2.41 10 D` 0.83 0.05 0.74 0.89 10 S 12.17 2.93 7.00 16.00 6 E 0.74 0.08 0.66 0.88 6 H 1.82 0.29 1.47 2.20 6 D` 0.76 0.07 0.70 0.87 6 S 16.60 2.70 14.00 20.00 5 E 0.78 0.08 0.68 0.87 5 H 2.18 0.23 1.85 2.47 5 D` 0.84 0.06 0.73 0.89 5 S 15.60 2.41 13.00 19.00 5 E 0.85 0.04 0.82 0.91 5 H 2.33 0.13 2.17 2.46 5 D` 0.87 0.02 0.85 0.89 5 S 22.50 1.29 21.00 24.00 4 E 0.86 0.04 0.82 0.90 4 H 2.67 0.07 2.61 2.74 4 D` 0.90 0.01 0.89 0.92 4 S 15.00 1.41 14.00 16.00 2 E 0.80 0.00 0.80 0.80 2 H 2.16 0.07 2.11 2.21 2 D` 0.84 0.00 0.83 0.84 2 S 25.00 1.41 24.00 26.00 2 Assessing Sampling Methods YNP 2013 Final Report MO Y1 Y2 Y3 Y4 Y5 Y6 39 E 0.76 0.04 0.73 0.79 2 H 2.45 0.09 2.38 2.52 2 D` 0.84 0.02 0.83 0.85 2 S 18.00 1.41 17.00 19.00 2 E 0.82 0.02 0.80 0.83 2 H 2.36 0.12 2.27 2.45 2 D` 0.86 0.01 0.85 0.87 2 S 29.00 2.83 27.00 31.00 2 E 0.82 0.02 0.80 0.83 2 H 2.75 0.02 2.73 2.76 2 D` 0.91 0.00 0.91 0.91 2 S 22.50 0.71 22.00 23.00 2 E 0.78 0.00 0.78 0.78 2 H 2.43 0.03 2.40 2.45 2 D` 0.87 0.01 0.87 0.88 2 S 16.00 3.46 12.00 18.00 3 E 0.50 0.07 0.43 0.57 3 H 1.39 0.29 1.08 1.65 3 D` 0.56 0.10 0.46 0.66 3 S 16.00 3.00 13.00 19.00 3 E 0.54 0.12 0.47 0.68 3 H 1.48 0.22 1.34 1.74 3 D` 0.62 0.11 0.55 0.75 3 S 20.67 1.53 19.00 22.00 3 E 0.79 0.04 0.76 0.83 3 H 2.40 0.07 2.31 2.45 3 D` 0.87 0.01 0.86 0.88 3 S 22.50 2.12 21.00 24.00 2 E 0.85 0.04 0.82 0.88 2 H 2.65 0.19 2.51 2.78 2 D` 0.90 0.02 0.89 0.92 2 Assessing Sampling Methods YNP 2013 Final Report Section 1.2 Tests for distribution and variance within data Table A-5. Shapiro-Wilk p-values from tests for normality. Data were diversity calculations based on frequency and percent cover analyses from five sampling methods from locations throughout the northern ungulate winter range. Table is divided by (1) method (upper portion), where data was grouped solely on method and could include several locations and years; (2) location (middle portion), where each location may have been sampled by several different methods and in multiple years; and (3) Parker locations only (lower portion). Red = data not normally distributed (i.e., if p-values <0.05, reject the null hypothesis that a sample came from a normally distributed population). BI=Blacktail Inside Exclosures; BO=Blacktail Outside Exclosures; GI=Gardiner Inside Exclosures; GO=Gardiner Outside Exclosures; JI=Junction Butte Inside Exclosure; JO=Junction Butte Outside Exclosures; LI=Lamar Inside Exclosures; LO= Lamar Outside Exclosures. Y-Series = free-range plots pre-construction of 1957 exclosures (Y1=Geode Creek, Y2=Blacktail Hill, Y3=Landslide Creek, Y4=Game Ranch, Y5=Lamar Horseshoe, Y6=Specimen Ridge). By Location Richness (Cover) Simpson’s Index (Cover) By Method Richness (Frequency) Simpson’s Index (Frequency) Richness (Cover) Simpson’s Index (Cover) Parker 0.23 Nested 0.004 Daubenmire 0.17 Whittaker 0.10 FIA 0.15 <0.001 0.0032 0.08 0.63 0.60 0.23 0.00 0.17 0.10 0.15 <0.0001 0.00 0.00 0.66 0.06 BI BO GI GO JI JO LI LO MI MO Y1 Y2 Y3 Y4 Y5 Y6 0.36 0.26 0.10 0.29 0.17 0.78 0.97 1.00 1.00 1.00 1.00 1.00 <0.0001 1.00 0.64 1.00 0.02 0.10 0.37 0.06 0.16 0.24 0.07 1.00 1.00 1.00 1.00 1.00 0.82 0.17 0.21 1.00 Table A-6. Brown-Forsythe test results for homogeneity of variance By Method Richness (Frequency) Simpson’s Index (Frequency) Richness (Cover) Simpson’s Index (Cover) 40 F-value 2.95 n 147 p-value 0.0223 8.36 147 <0.0001 2.95 147 0.0223 1.15 147 0.3353 Assessing Sampling Methods YNP 2013 Final Report By Location Richness (Frequency) Simpson’s (Frequency) Richness (Cover) Simpson’s Index (Cover) F-value N p-value 0.3186 1.20 65 ----- 65 ------ 1.20 65 0.3186 .72 65 0.6853 Section 1.3 Statistical tests for differences among groups Table A-7. Skillings-Mack results for differences among methods using cover data. Block = site, treatment = method, and response = richness or Simpson’s Index. The weighted sum of centered ranks shows which methods have the higher or lower ranked observations. P-values are considered significant if p<0.05. Weighted Sums Year SkillingsMack value Response Scale Richness Simpson's Index Small 2009 42.388 Small 2009 11.356 Richness Simpson's Index Mixed Mixed 20102012 20102012 p-value Parker Nested Daubenmire Whittaker FIA 6.25 E-10 -18.19 -31.31 49.5 - - 0.003 17.32 -25.25 7.91 - - 18.762 0.0009 -12.3 -10.57 0 13.21 4.07 0.397 6.51 -7.21 -1.41 2.51 9.66 0.41 Table A-8. Kruskal-Wallis results for richness and Simpson’s Index using frequency and cover data from Parker data only. Parker 1958 and 1962 exclosures insides were combined for location analyses (n=15). P-values considered significant if p<0.05. Differences By (Parker) Location 41 Variable Richness Data Type Frequency N 65 Chi-square 34.9802 DF Pr>Chi Square 15 0.0025 Simpson’s Index Richness Simpson’s Index Frequency Cover Cover 65 65 65 43.0360 34.9802 43.6033 15 15 15 Assessing Sampling Methods YNP 2013 0.0002 0.0025 0.0001