Aims: Vegetation classifications are useful for a variety of management purposes as well as scientific exploration. Local classifications are common throughout the United States but only recently have been integrated into a national classification system, which is now expected for local classifications. Study Area: The Pawnee National Grasslands (PNG) in northeastern Colorado, USA, has not been classified using plot data, and is thus a gap on the baseline knowledge of the PNG plant communities that hinders impact assessment of various anthropogenic activities. Methods: Here, we use 128 plots to classify the vegetation of the PNG using a two-step process: first, classifying the PNG plots alone to characterize local uniqueness, and then employing a semi-supervised classification with an additional 64 plots from areas to the north and east of the PNG, using standard classification procedures. Results: We document on the PNG the occurrence of two Classes, three Subclasses, four Formations, five Divisions, six Macrogroups, seven Groups and eight Alliances and Associations already described in the USNVC. Conclusions: The PNG is dominated by the Bouteloua gracilis-Buchloe dactyloides Grassland Association, which we further subdivide and describe as three local subassociations. The mixed-grass concepts in the USNVC do not exist in the PNG.

Taxonomic reference: Hazlett (1998).

Syntaxonomic reference: USNVC (2016).

Abbreviations: BLM = Bureau of Land Management; CPER = Central Plains Experimental Range; ESA = Ecological Society of America; EST = Ecological Site Type; GPS UTM = Global Positioning System Universal Transverse Mercator; NEON = National Ecological Observatory Network; PNG = Pawnee National Grasslands; USNVC = United States Vegetation Classification.

Colorado, Pawnee, semi-supervised classification, shortgrass, steppe, USNVC, vegetation

Classification of vegetation provides a common language to compare communities among regions, an inventory to assess change, and a baseline for land stewardship decisions (ESA Panel 2015). Vegetation classifications are useful for: (1) documenting complex vegetation patterns, (2) developing hypotheses about processes shaping such patterns, (3) mapping vegetation and related ecosystem properties, (4) surveying, monitoring and reporting plant and animal communities, and (5) developing management and conservation strategies (De Cáceres et al. 2015). While several initial efforts toward mapping and vegetation data collection are available for the Pawnee National Grasslands, there is no plot-based classification, despite the area including the Central Range Experiment Station of the United States Agricultural Research Service and a National Ecological Observatory Network (NEON) site. Here we present a plot-based classification that follows recent standards of the United States (Jennings et al. 2009; Faber-Langendoen et al. 2014) as well as international standards (De Cáceres and Wiser 2012; De Cáceres et al. 2015).

Baker (1984) provided a preliminary list of the natural vegetation communities for the entire state of Colorado, but gave no descriptions of the communities themselves. Johnson (1987) described 13 potential natural associations (in this case, cover types) based on previous literature. Hazlett (1998) described habitats based on vegetation occurrences integrated with site abiotic characteristics, but did not use plot data and thus performed no analyses. With the multiple uses of the PNG, from grazing to missile silos to the recent oil and natural gas boom, a consistent and standards-conforming classification of vegetation communities is needed for land stewardship decisions. The Colorado Vegetation Classification Project, an effort of the Bureau of Land Management (BLM) and the Colorado Department of Wildlife (http://www.arcgis.com/home/item.html?id=893739745fcd4e05af8168b7448cda0c), produced a classification using 1993–1997 Landsat Thematic Mapper imagery that was processed using an unsupervised classification procedure. Field-gathered GPS data were used to label and group the final classes. Based on that classification on broad-based life forms, the PNG lies in the Herbaceous Riparian (only one subclass, Sedge) or Grass/Forb Rangeland, including several subclasses pertinent to the PNG: Grass Dominated Herbaceous Rangeland, Forb Dominated Herbaceous Rangeland, Grass/Forb Mix Herbaceous Rangeland, Tall-grass Prairie, Mid-grass Prairie, Short-grass Prairie, Disturbed Rangeland and Sparse Grass/Blowouts. These are general names for large-scale vegetation communities and, thus, are likely not specific enough for local land stewards.

The Vegetation Subcommittee of the Federal Geographic Data Committee has developed a standard for vegetation classification in the United States (FGDC 2008), as well as descriptions of the approach (Jennings et al. 2009; Faber-Langendoen et al. 2009; Franklin et al. 2012; Faber-Langendoen et al. 2014), and the resulting United States National Vegetation Classification (USNVC) was released in February of 2016 (http://usnvc.org/website-launch/). The USNVC has already been successfully used to develop state-and-transition models of landscape change (Kudray and Cooper 2005) by standardizing the definition of states, develop habitat suitability maps and high-quality vegetation maps essential for biodiversity stewardship and research (Evens and Keeler-Wolf 2014), and improve the sharing of vegetation information among agencies for intra- and interagency management, such as mapping of vegetation and fuels in the LandFire program (https://my.usgs.gov/eerma/data/index/4f4e486ee4b07f02db50bea7).

Classification systems around the world are being developed and used for such purposes (Bruelheide and Chytrý 2000; Rodwell 2006), but small-scale, unconnected classifications within and among countries, and in the United States, within and among governmental units, have been the bane of developing regional classifications and the identification of community concepts over the range of their occurrence. Such is the problem in many areas of the United States and a standardized effort is needed to both corroborate USNVC concept descriptions and fill in the holes of the USNVC. Peet and Roberts (2013) define nine primary components of vegetation classification: 1) project planning, 2) data acquisition, 3) data preparation, 4) community entitation, 5) cluster assessment, 6) community characterization, 7) community determination, 8) classification integration, and 9) classification documentation. The advent of the USNVC has changed how researchers in the US approach these components; specifically, regarding classification integration recognizing that integration may also affect the iterative process of entitation and assessment. Because the USNVC concept descriptions are meant to cover the range of characteristics of a community concept, while collected data are potentially from a restricted area such as a park (as is the case in this study), documenting variations on that concept that are specific to the location may be beneficial to local stewards. However, that does not suggest the community concept itself be changed, as currently accepted concepts should only be modified after careful reflection (Jennings et al 2009; Peet and Roberts 2013).

An important element of any classification is the heterogeneity of the landscape, such that many different vegetation types may be found in a small geographic area. Further, one of the main uses of such classifications is mapping that provides information to stakeholders to make stewardship decisions (ESA Panel 2015), and this mapping level tends to be at the Macrogroup scale of the USNVC (combinations of moderate sets of diagnostic plant species and diagnostic growth forms that reflect biogeographic differences; FGDC 2008). While we fully expect the Great Plains Shortgrass Prairie to dominate the PNG landscape, we also expect to find more arid (e.g., Arid West Interior Freshwater Marsh) and more mesic types (e.g., Great Plains Flooded Forest).

The objective of this research was to develop a plot-based vegetation classification of the natural and semi-natural vegetation communities in the Pawnee National Grasslands in accordance with the USNVC. We followed standard procedures for data acquisition, used a variety of multivariate analyses for community entitation and determination, and integrated our community concepts with those of the USNVC, following the standards of Peet and Roberts (2013) and De Cáceres et al. (2015). We predicted that vegetation would be strongly affected by topography, especially slope positions that affect moisture levels, and that repeating patterns of vegetation communities would be found throughout the PNG landscape (i.e., community concepts would be recognizable).

The Pawnee National Grasslands (PNG), administered by the USDA Forest Service, covers 79,876 ha in Weld County, Colorado, between 40°36’ and 41°00’ N latitude and between 103°34’ and 104°48’ W longitude (Figure 1). The grasslands are a mosaic pattern of private and public lands; both are used for grazing, oil and gas extraction, and house below-ground nuclear missiles. Included within the PNG are the Central Plains Experimental Range (CPER; 6057 ha), a research area administered by the Agricultural Research Service (now also part of the National Ecological Observation Network, NEON) and the Shortgrass Steppe Long-term Ecological Research site (now maintained by Colorado State University).

Figure 1. 

Location of Pawnee National Grasslands (PNG). Inset includes NatureServe Ecoregions of study area and additional plot data locations and studies: Classification of Natural Riparian/Wetland Plant Associations for Colorado (CWRC, throughout CO), Fort Laramie National Historic Site (FLNHS), Agate Fossil Beds National Monument (AFBNM), and Devil’s Tower national Monument (DTNM).

Climate is continental, but large air masses from maritime areas may move across the area. Crabb (1981) reported an average air temperature of -2°C during the winter with an average daily minimum temperature of -10°C; during summer months, average air temperature is 21°C with an average daily maximum temperature of 31°C.

The Pawnee National Grasslands also lie in the rainshadow of the Rocky Mountains to the west. Mean annual precipitation for the study area is 305–380 mm; average annual snowfall is 102 mm (Crabb 1981). Wind-driven snow often accumulates on leeward sides of hills (typically southeastern sides), around shrubs, and near roads; meltdown, especially in rocky or sandy soil, results in water penetration to greater depths at these locations (Hazlett 1998). The PNG lies within Kuchler’s (1964) Shortgrass Steppe, dominated by C₄ grasses, and two of his four potential natural vegetation types may occur on the PNG: the overwhelmingly dominant Bouteloua-Buchloe Type and the Artemisia-Schizachyrium Type on deep sandy soils. The Shortgrass Steppe is typically dominated by graminoids (> 60%) with less than 20% cover of succulents, dwarf shrubs, and herbaceous dicots (Laurenroth 2008). Classifications of portions of the PNG, e.g. the Central Plains Experimental Range, suggest only a handful of vegetation community types (Moir and Trlica 1976). The PNG falls in the Loamy Plains (Atriplex canescens/Bouteloua gracilis-Pascopyrum smithii) Ecological Site Type (EST), part of the Central High Plains (https://esis.sc.egov.usda.gov/Welcome/pgESDWelcome.aspx). The EST classification includes discrete biological and physical factors that denote specific vegetation/soil/physical characteristics that respond similarly to management and disturbance. In addition, Hazlett (1998) differentiated six habitat types on the Pawnee: (1) open steppe (> 80% of study area), (2) sandy soils (~5%), (3) breaks and barrens (<2%), (4) cliffs and ravines (<2%), (5) riparian (~5%), and (6) roadsides and disturbed soils (< 5 %).

In general, the elevation of the Colorado Piedmont, an uplifted Cretaceous shale physiography that includes the PNG, declines from the mountain foothills toward the east at a rate of about 2 m km^-1; the highest elevation is 1,935 m in the northwestern portion near the “Chalk Bluffs” and the lowest elevation is 1,310 m in the southeastern portion around South Pawnee Creek. Most of the soils on the Pawnee National Grassland are shallow to deep loams that are well drained (Crabb 1981). Over most of the area is a loamy, wind-mixed veneer layer of soil of varying depths. These soils are underlain by a variable pattern of shale and sandstone bedrock materials. Barren rock or gravel areas of shale and sandstone can be exposed when erosive wind removes upper layers of soil. In addition, past tectonics and water erosion have exposed ravine “break” areas with rock exposed on the sides of the ravine. Sandy soils occur along stream terraces and on leeward sides of some hills (Hazlett 1998).

Swale areas often have finer textured soils than ridgetops, as mobile soil particles, such as silt and clay, have eroded from higher topographic positions and have been deposited in lower areas. This difference in soil texture is sometimes reflected by a greater abundance of Buchloe dactyloides in swales. In addition, some drainages, playas, and riparian areas have an accumulation of salts on or near the surface and thus host alkaline-tolerant plant species. Maps and detailed descriptions of the soil series types that occur in this study area can be found in Crabb (1981).

GIS techniques have been shown to be useful in determining distribution of plant and animal communities (Rotenberry et al 2006; Sangermano and Eastman 2006). The initial phase of this project used GIS map layers to develop an ecological land type classification that was subsequently used to stratify field plots (Kupfer and Franklin 2000). Map layers included elevation, bedrock geology, and soil classification obtained from the State of Colorado (http://coloradogeologicalsurvey.org/geologic-mapping/gis-data/). Plots (see below) were positioned within all 100 m elevation zones (1300–1800 m, which also was essentially an east to west gradient) and on all major parent materials (dune sand, gravel, sandstone, shale). We examined geology, soils, and topographic factors in an attempt to place plots in all environments (i.e., land types) of the Pawnee National Grasslands. Some noticeable trends are important (Figure 2). The western portion of the Pawnee is dominated by Cretaceous shales and the eastern portion by Tertiary sandstone; the eastern portion also contains some quaternary gravel and sand. There is also a general gradient in elevation, decreasing from west to east.

Figure 2. 

GIS maps of aspect, elevation, slope, soil type and vegetation type of the Pawnee National Grasslands (PNG). The two polygons represent the east and west sections of the PNG. Beers et al. (1966) transformation was used for aspect, ranging from 0 (SW) to 2 (NE); elevation ranges from 1309–1800 in 100 m intervals; slope ranges from low (0–10%) to medium (10–30%) to steep (>30%); vegetation and soild are based on previous classifications (see text).

We used plot data to document the occurrence of two USNVC Classes, three Subclasses, four Formations, five Divisions, six Macrogroups, seven Groups and eight Alliances and Associations on the PNG, ranging from mesomorphic tree vegetation (i.e., Populus woodlands along riparian zones) to mesomorphic shrub and herb vegetation dominated by the wide-ranging shortgrass steppe species Bouteloua gracilis and Buchloe dactyloides. The latter is the matrix of the landscape with fragments of more mesic conditions nested within, ranging from standing water locations (e.g.. farm ponds) dominated by Schoenoplectus pungens or Sporobolus airoides under greater salinity, to Carex, Juncus, Eleocharis, and Pascopyrum smithii dominance in swales with varying levels of periodic moisture during the growing season.

Our plot-groups relate to those outlined by Hazlett (1998). Our Buchloe dactyloides-Pascopyrum smithii Steppe and Bouteloua gracilis-Buchloe dactyloides Steppe local subassociations together match his Open Steppe and Sandy Soils habitats, and our Rhus trilobata-Ribes cernuum Shrubland association matches his Cliffs and Ravines habitat. We suggest that his Breaks and Barrens habitat relates to our Rhus trilobata/Schizachirium scoparium-Bouteloua spp. Outcrop local subassociation, and that the remainder of our vegetation concepts relate to his Riparian habitat. For the latter, we clearly defined a number of different vegetation types within his one habitat, which is not surprising due to the azonal nature of more mesic locations (Faber-Langendoen et al. 2014). Baker (1984) appeared to take a strong splitter approach with grasslands and developed several associations from the many possible dominants at small scales (< 10 m²). For Bouteloua-dominated types, he recognizes two, similar to our two local subassociations; Bouteloua gracilis Shortgrass Prairie and Bouteloua gracilis-Buchloe dactyloides Shortgrass Prairie, but also types like the Hordeum jubatum Plains Grassland. The unique barrens and outctrops are noted by associations such as the Arenaria hookeri Barrens and Rhus trilobata-Ribes cereum /Schizachyrium scoparium Shrub Association, but also at least two mixed prairie associations; Stipa comata Mixed Prairie and Schizachyrium scoparium Mixed Prairie. As did we, he also recognized several mesic types, including Juncus balticus Wetland, Carex nebrascensis-Juncus balticus Wetland, Carex nebrascensis-Catabrosa aquatica-Juncus balticus Spring Wetland, Eleocharis palustris Wetland, Sporobolus airoides Salt Meadow, and Distichlis spicata var. stricta Salt Meadow, as well as several Populus deltoides Forest/Woodland associations that are not clearly related to those on the PNG. There are two considerations with these comparisons. First, the previous studies are expert-based classifications and not plot-based. Further, at least for Baker (1984), that classification was for the entire state of Colorado, although we still believe he split concepts too finely compared to the current USNVC. Regardless, direct comparisons are difficult.

We propose local subassociations that may be helpful for land stewardship, but not as a change to the Bouteloua gracilis-Buchloe dactyloides Grassland Association concept. Our limited geographic reference for this concept does not allow any major changes, but that same geographic size suggests local subassociations may exist (Jennings et al. 2009). These groups have clear characteristic species and environments that may be of interest for conservation management.

Our ‘semi-supervised’ classification was successful in that it let us classify several rarer (in our dataset) plots. The ability to compare previously-classified plots with unknown plots (Tichý et al. 2014) in the same analysis allowed for a much better entitation and cleared up nearly all of our questions from the Pawnee-only analysis, and such analyses are needed to improve all future local classification efforts. One major conclusion from this analysis is that the mixed-grass concepts in the USNVC do not exist in the PNG. While the Colorado vegetation map suggests these communities are part of the PNG landscape, we argue that the vegetation composition and structure as a whole are different and should be considered so as the lines demarking the Shortgrass Steppe Ecoregion suggest (Sayre et al. 2009).

There are of course limitations to our study and this classification exercise. First, while the plot data are solid, the low number of plots (n=101+27) for the area of the PNG is a concern. Especially for the types where we have little data, additional plots are warranted. Further, while we thoroughly traversed the PNG looking for different vegetation associations, we may have missed certain associations that occur in the PNG, notably the four-wing saltbush (Atriplex canescens) lowlands as well as purposefully ignoring ruderal communities that are generally restricted to roadsides and highly disturbed sites in the PNG (Kotanen et al. 1998). The occurrence of the four-wing saltbush type seems to be rare, mainly on low-lying areas of private lands in the northeastern corner of the PNG (pers. obs.) and perhaps due to coarser soils (Dodd et al. 2002) or grazing intensity (Cibils et al. 2000; Hart 2001), or simply previous disturbance (Coffin et al. 1996; Augustine et al. 2017).

Finally, a thorough assessment of the abiotic characteristics of these sites is warranted, since soil texture (Dodd and Lauenroth 1997; Dodd et al. 2002) and moisture (Boutton et al. 1980) are known to affect vegetation community composition and structure on the PNG but were not examined on a site-specific basis here. While abiotic factors would not affect our plot-based vegetation classification, environmental data would be useful for interpreting the vegetation patterns.

Finally, we make a plea here that all vegetation scientists with full-species plot data place those data into VegBank or another public database. While we were able to relate some of our more mesic concepts to plots from other studies, little plot data existed for the typical shortgrass steppe communities dominated by Bouteloua species. Our data represent a small geographic fraction of the area this concept covers and a regional analysis would be beneficial for the PNG and the USNVC (Palmquist et al. 2016).

All data are in VegBank (http://www.vegbank.org).

SBF collected and analyzed data and wrote initial paper; MS collected and analyzed data and edited paper; JS helped with analyses and edited paper.

We honor thank Don Hazlett for helping find some of the unique vegetation locations, his insight to Pawnee National Grasslands, and his help with identification of plant specimens – his presence will be greatly missed. We thank Bruce Hoagland and James Van Kley for their friendly reviews of earlier versions of this manuscript. This research was partially funded by NSF-ROA grant and a Provost Grant from the University of Northern Colorado.

• Scott Franklin (Corresponding author, scott.franklin@unco.edu), ORCID: https://orcid.org/0000-0003-3922-8857
• Michael Schiebout (mschiebout@uu.edu)
• Jozef Šibik (jozef.sibik@savba.sk), ORCID: https://orcid.org/0000-0002-5949-862X