Citation
Influence of the Chihuahuan biogeographic sub-element on the flora of Colorado : evidence from GIS

Material Information

Title:
Influence of the Chihuahuan biogeographic sub-element on the flora of Colorado : evidence from GIS
Creator:
Barron, Genevieve Marissa
Place of Publication:
Denver, CO
Publisher:
University of Colorado Denver
Publication Date:
Language:
English

Thesis/Dissertation Information

Degree:
Master's ( Master of integrated science)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
College of Liberal Arts and Sciences, CU Denver
Degree Disciplines:
Integrated science
Committee Chair:
Bruederle, Leo P.
Committee Members:
Moreno-Sanchez, Raphael
Islam, Melissa

Notes

Abstract:
Biogeographic patterns are a driving factor in the distribution of plant species and the amounts of suitable habitat available to them. Weber (1965) stated that southeastern Colorado is dominated by a distinct distribution of taxa that are representative of the Chihuahuan Biogeographic sub-element. The supposition by Weber leads to the question; is there evidence that the Chihuahuan sub-element is represented in Colorado flora? If so, what are the environmental limitations for those species and which areas of Colorado are likely to provide suitable habitat? To answer these questions an interdisciplinary methodology that combined Geographic Information System technology with biological niche modelling best practices, was used to address the following objectives: compile a comprehensive list of species occurring in Colorado that represent the Chihuahuan sub-element, thoroughly document the ecological niches that those species occupy within the state, and based on niche characteristics spatially model the areas of Colorado that are likely to provide suitable habitat. Results of a systematic comparison of occurrence records from online herbarium databases showed 43 taxa occurring in Colorado strongly exhibit an overall Chihuahuan distribution. Five of the taxa, whose habits represented the most commonly occurring habit types, were chosen for a habitat suitability analysis. A strong trend towards suitable habitat occupying mostly the eastern portion of the state was observed, with some areas of medium to high suitability concentrated in the southeast and southwest corners.
General Note:
n3p

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
Copyright Genevieve Marissa Barron. Permission granted to University of Colorado Denver to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.

Downloads

This item has the following downloads:


Full Text
INFLUENCE OF THE CHIHUAHUAN BIOGEOGRAPHIC SUB-ELEMENT
ON THE FLORA OF COLORADO: EVIDENCE FROM GIS by
GENEVIEVE MARIS SA BARRON B.S., Fort Lewis College, 2012
A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Master of Integrated Sciences Integrated Sciences Program
2018


This thesis for the Master of Integrated Science degree by Genevieve Marissa Barron has been approved for the Integrated Science Program by
Leo P. Bruederle, Chair Raphael Moreno-Sanchez Melissa Islam
Date: July 28, 2018
n


Barron, Genevieve Marissa (M.I.S., Integrated Sciences)
Influence of the Chihuahuan Biogeographic Sub-Element on the Flora of Colorado: Evidence from GIS
Thesis directed by Associate Professor Leo P. Bruederle
ABSTRACT
Biogeographic patterns are a driving factor in the distribution of plant species and the amounts of suitable habitat available to them. Weber (1965) stated that southeastern Colorado is dominated by a distinct distribution of taxa that are representative of the Chihuahuan Biogeographic sub-element. The supposition by Weber leads to the question; is there evidence that the Chihuahuan sub-element is represented in Colorado flora? If so, what are the environmental limitations for those species and which areas of Colorado are likely to provide suitable habitat? To answer these questions an interdisciplinary methodology that combined Geographic Information System technology with biological niche modelling best practices, was used to address the following objectives: compile a comprehensive list of species occurring in Colorado that represent the Chihuahuan sub-element, thoroughly document the ecological niches that those species occupy within the state, and based on niche characteristics spatially model the areas of Colorado that are likely to provide suitable habitat.
Results of a systematic comparison of occurrence records from online herbarium databases showed 43 taxa occurring in Colorado strongly exhibit an overall Chihuahuan distribution. Five of the taxa, whose habits represented the most commonly occurring habit types, were chosen for a habitat suitability analysis. A strong trend towards suitable habitat occupying mostly the eastern portion of the state was observed, with some areas of medium to high suitability concentrated in the southeast and southwest comers.
This form and content of this abstract are approved. I recommend its publication.
Approved: Leo P. Bruederle
m


ACKNOWLEDGMENTS
I would like to acknowledge Dr. Rafael Moreno-Sanchez for serving as a thesis committee member and technical consult for the GIS methodology. I also acknowledge Dr. Melissa Islam for serving as a thesis committee member and providing support for locating additional reference materials and offering additional viewpoints on biologically meaningful modelling concepts. I will also acknowledge Michelle Deprenger-Levin for guidance on parallels between ArcGIS and MaxEnt Modeling for Species Distributions and as well as advice on best practices for selecting variables for testing. Finally, I would like to acknowledge Dr. Leo Bruederle for all the work and help that he has provided for this thesis.
IV


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION.......................................................................9
Research Objectives...............................................................13
Methods and Materials................................................................14
Results..............................................................................15
Discussion...........................................................................17
Taxonomy..........................................................................18
Geography.........................................................................18
Ecology of the taxa...............................................................19
GIS Approach......................................................................20
Limitations.......................................................................20
Conclusion........................................................................21
Tables and Figures...................................................................22
II. MCI IE MODELING....................................................................28
Relevant Approaches to Niche Modeling................................................28
Relative Frequency................................................................29
Generalized Linear Models.........................................................30
Maxent for Presence Only Dat......................................................30
Geographic Information Systems....................................................31
Habitat Suitability Analyses Considerations.......................................32
Research Objectives...............................................................33
Methods and Materials................................................................34
Bioclim Data Processing...........................................................34
Statistical Analyses..............................................................35
Habitat Suitability Analysis......................................................35
Results..............................................................................37
Palafoxia sphacelata (Asteraceae).................................................37
Muhlenbergia arenicola (Poaceae)..................................................38
Funastrum crispurn (Apocynaceae)..................................................39
Nolina texcma (Asparagaceae)......................................................40
Muhlenbergia arenacect (Poaceae)..................................................41
Maxent Comparison.................................................................42
Discussion...........................................................................42
Abiotic and Biotic Patterns.......................................................43
v


Geographic Patterns.............................................................43
Sensitivity Analysis............................................................44
Method Comparison...............................................................44
Limitations of the Study........................................................45
Conclusion.....................................................................46
Figures...........................................................................48
REFERENCES........................................................................54
APPENDIX..........................................................................59
vi


TABLE
LIST OF TABLES
1.1 Chihuahuan species found to reach Colorado and associated EPA Level III ecoregion
classification............................................................................22
Vll


LIST OF FIGURES
FIGURE
1.1 Reference map of southwest US deserts..................................................17
2.1 Results of a habitat suitability analysis for Palafoxia sphacelata.....................49
2.2 Results of a habitat suitability analysis for Muhlenbergia arenicola...................50
2.3 Results of a habitat suitability analysis for Funastrum crispum........................51
2.4 Results of a habitat suitability analysis for Nolina texana............................52
2.5 Comparison of outputs for Palafoxia sphacelata, and Muhlenbergia arenacea.............56
viii


CHAPTERI
INTRODUCTION
Patterns of species distributions, especially those that appear to be highly restricted, have been a central focus for scientists in the environmental sciences for at least the past century (Rajakaruna, 2004). Research that has delved into finding the driving forces behind those patterns have identified climate and soil type as the primary influences of plant distributions (Rajakaruna, 2004; Cain, 1944). Broad topics in and of themselves, the efforts to understand the mechanisms that link climate and edaphic factors to plant distributions are still ongoing (Wood, 1986). To better understand these complex ecosystem processes, researchers have crossed disciplinary boundaries to adopt and incorporate methods and concepts that focus on biogeographic patterns of species distributions (Ebach, 2015).
Biogeographic patterns are shaped both by the physical characteristics of an ecosystem, as well as by the biological traits of ecosystem inhabitants (Collin and De Matinen, 2002). Abiotic factors, namely climatic weather conditions, have long been prominently linked to the distribution and range expansion of major clades across North America. Marlow and Hufford (2007) found that the aridification of northern Mexico and the western extent of the United States greatly influenced the geographic radiation of representatives of the Asteraceae, specifically Gaillardia, into the Rocky Mountain region. The biological characteristics of species inhabiting a geographic range, such as the propensity to adapt to changes to the environment, are therefore also driving forces in the biogeographic patterns of a region. Other abiotic factors, such as land bridges that connect separate biotas or mountain ranges that act as dispersal barriers between ecosystems, strongly influence biogeographic patterns, as well (Collin and De Matinen, 2002).
In the context of a floristic analysis, classifying populations based on biogeographic patterns can serve to help document species current distributions, range of expansions and conserved niches. To
9


that end, the term ‘element’ has traditionally been applied to those groups of species that occur in a previously defined biogeographic area and have similar distributions (Passalacqua, 2015).
Weber (1965) defined an element as a group of species that occupy a coincident area, and further proposed that such an element could be divided into sub-elements. His examination of the southwestern US concluded with recognition of three broad elements: Circumpolar element, Madro-Tertiary element, and Eastern-Woodland Prairie element, each with respective sub-elements. These elements are not simply species occurrences but reflect the geographic environmental conditions on the landscape.
Similarly, McLaughlin (2007) attempted to isolate floristic biogeographic elements across North America — a geographically large scale — using Principal Component Analysis of 245 local floras of Mexico, the United States, and Canada. He designated 27 floristic sub-provinces that represented discrete areas with characteristic physiography, climate, vegetation, and plant and animal life (McLaughlin, 1986). Within those provinces, ‘floristic elements’ were identified, indicating that species had nonrandom, overlapping distribution patterns, but did not consistently occur in similar ecosystem communities (McLaughlin, 1986). His analysis of the southwestern portion of the United States, including the northern boundary of Mexico, identified twelve floristic elements: The Great Basin Element, Colorado Plateau Element, Chihuahuan Element, Sonoran Element, Apachian Element, Southern Rocky Mountain-Mogollon Element, Sierra Nevada, Rocky Mountain, Peninsular, California, Vancouverian and Columbian Plateau Elements (McLaughlin, 1989). Hereafter, the use of the term ‘sub-element’ refers to Weber’s (1965) definition.
While there may be some variation in determining the boundaries of floristic elements, as they pertain to state boundaries (e.g., Colorado), there is agreement on the importance of a region’s biogeographic features on its floristic communities. Colorado is characterized by great topographic and climatic variation (Lambert and Reid, 1981). Topographic features like plateaus, basins, the southern portion of the Rocky Mountain range, and plains make up the landscape that define the borders of the state’s plant communities (Chapman et al., 2006). The diverse vegetative communities
10


that have resulted from these geographic variations are home to a variety of sensitive and, in some cases rare species.
In order to facilitate conservation and management practices for these species, attempts have been made to classify and partition land management systems based on plant communities, soil types, and other ecologically meaningful boundaries. Traditionally, the composition of an ecological community has been explained by species specific traits and the individual requirements for a species (Janzen et al., 2015). However, as the role of land managers evolves, a more robust understanding is required.
As such, the Environmental Protection Agency incorporated data on vegetation, climate, soils, land use, wildlife, hydrology, physiography and geology into their land classification systems, resulting in the recognition of ecoregions (Chapman et al., 2006). Within Colorado, the EPA designated six Level III Ecoregions: Wyoming Basin, Colorado Plateaus, Southern Rockies, Arizona/New Mexico Plateau, High Plains, and Southwestern Tablelands (Chapman et al., 2006).
Those Ecoregions are further subdivided into Level IV Ecoregions, the characteristics of which have been widely documented (Omemik and Griffith, 2014). However, the relationship between the arid regions of Colorado, and potential migration of taxa from other regions beyond the Ecoregion classification zones has not yet been thoroughly investigated.
Weber (1965) proposed that the flora of Colorado has been influenced, in part, by a sub-element of the Madro-Tertiary Element — the Chihuahuan sub-element — as evidenced by the presence of 33 species that are predominantly Chihuahuan with occurrences in Colorado (Table A. 9). Yet to date, little research has been conducted providing evidence supporting this observation.
This may be due, in part, to the fact that the generally recognized borders of the Chihuahuan Desert do not extend into the state of Colorado. Most spatial representations of the Chihuahuan Desert and the corresponding Chihuahuan sub-element depict the northern boundary reaching only as far north as the state of New Mexico. When one considers the physical boundaries that lie between the northern extent of the sub-element, and Colorado, it is not immediately obvious which, if any species of that element, might be present in the state.
11


Administrative and political boundaries have provided a generally recognized perimeter boundary for the Chihuahuan Desert. However, the boundaries of this desert, as with most ecoregions, are not entirely defined or agreed upon from the ecological viewpoint (Villareal et al., 2017, Brown, 1982, as cited in Schmidt, 1979, Morafka, 1977). Yet, McLaughlin (2007) identified the desert area as an independent floristic sub-element. The interior of the desert is located between the Sierra Madre Occidental to the West, the Sierra Madre Oriental to the east, and the Trans-Mexican Volcanic Belt at the southern end (McLaughlin, 2007). The northern extent of the desert, which intersects the Sonoran and Chihuahuan biogeographic elements (McLaughlin, 2007), is bounded by the Rio Conchos and the Deming Plains, which act as the natural geographic boundaries (Fig. 1.1). Although the Chihuahuan Desert is highly variable, both with respect to elevation and temperature ranges, its climate is characterized by summer rainfall, which provides 70-80% of the annual precipitation (Muldavin, 2002). The climate includes cold and dry winters, as well as hot summers (Villarreall et al., 2017). It is the second most floristically diverse desert in the world (Villarreall et al., 2017) — as many as 3382 plant species have been documented from this desert (Villarreall et al., 2017; Henrickson and Johnston, 2007). Approximately 67 plant families contain Chihuahuan endemics, with a total of 826 taxa (Villarreall et al., 2017). The areas of greatest richness of endemism are Coahuila, Texas, and Chihuahua, Mexico. Villarreall et al. (2017) found that the richness of a study area was not determined by the territorial proportion which they occupied within the Chihuahuan Desert. The Chihuahuan floristic element has affinities with the floras of the Great Plains biogeographic element (McLaughlin, 1986)
Since Weber (1965), few authors have considered the influence of the Chihuahuan biogeographic sub-element on the flora of Colorado. However, Clark’s (1996) floristic survey of the Mesa de Maya region of Colorado acknowledged Weber’s (1965) interpretation of the sub-element by noting the presence of several of its species present in the southeastern portion of the state. Models documenting the potential habitat of individual species from this sub-element within Colorado do not exist. However, research conducted at the species level attempts to classify individual species that represent
12


this sub-element with regard to their biogeographic patterns. Marlow and Hufford's (2007) analysis of New World Gaillardia documented the intermountain expansion of Gaillardiapulchella Foug. from its Chihuahuan origins using phylogeographic methods.
At the opposite end of the spectrum, landscape level research has sought to combine biosystematics with biogeography to determine the specific criteria and conditions that dictate large scale floristic patterns, such as those in the Chihuahuan Desert. Muldavin’s (2002) analysis of the floristic characteristics of the northern portion of the Chihuahuan Desert focused on the liminal intersection between the Chihuahuan Desert and Colorado Plateau as expressed by grassland communities. Analysis of gramminoids with qualities that indicated processes of desertification revealed that areas along the Chihuahuan Plateau that had lower elevation, provided definitive desert conditions. The findings of this study revealed that those species that were determined to be of the Chihuahuan geographic classification were one of the most restricted groups, with scrub species appearing more Chihuahuan in nature than grassland species.
Finally, extensive research addressing those factors driving ecological diversity in the Chihuahuan Desert have been compiled over the last few decades. In terms of geologic history, the Chihuahuan Desert is one of the ‘y°unger’ deserts on Earth. As such, the majority of phylogeographic research comes from its neighbors, the Mojave and Sonoran Deserts. Wood et al. (2013) examined evolutionary hot spots for genetic divergence and diversity in the Mojave, Colorado and Sonoran Deserts USA, recognizing that two major historical processes heavily shaped genetic structure in the southwestern deserts: Pleistocene climate fluctuations and pre-Pleistocene vicariance.
Research Objectives
Here I critically evaluate the influence of the Chihuahuan biogeographical sub-element on the flora of Colorado using GIS. My immediate objective was to test Weber’s hypothesis (1965) regarding the contribution of the Chihuahuan biogeographic sub-element by compiling a list of those plant species that are predominantly Chihuahuan, but also reach Colorado. Ultimately, I sought to
13


better understand this phenomenon, by exploring the taxonomy and ecology of those species identified by Weber (1965) and herein.
Methods and Materials
The mere occurrence of a species in Colorado, outside of its centralized distribution, for example from the Chihuahuan sub-element, does not necessarily constitute the influence of the respective floristic element on that region. According to Hirzel and Le Lay (2008), a more causal link can only be determined if the data meet three conditions: 1) a species must occupy mostly distinct areas; 2) niches must partially overlap; and 3) the species must occur in sympatry in some areas. Further, to accurately assess if a species truly occupies an area, additional concepts must be considered. Typically, species presence or absence from a region is determined by three constraints. First, the environmental conditions of a localized area must allow the population to grow. Second, the interactions between the target species and others must allow the target species to persist. Last, the location is accessible in relation to the abilities of the species to disperse (Hirzel and Le Lay, 2008).
To ascertain if (1) the Chihuahuan sub-element could be documented in Colorado using occurrence datasets from the Southwestern Environmental Information Network (SEINet, 2018) and the Global Biodiversity Information Facility (GBIF, 2016) and (2) the conditions proposed by Hirzel and Le Lay (2008) were met, a preliminary study was conducted. Species met the first condition of occupying a mostly distinct area if they displayed a mostly Chihuahuan Desert range distribution. Individual species distribution maps, for the species in Colorado, revealed that, in fact, some Chihuahuan species had high concentrations of occurrences in both the Chihuahuan Desert, as well as occurrences within the state of Colorado. This fulfilled the next two conditions of having partially overlapping niches and occurring in sympatry, and thus made the data eligible to be experimentally analyzed.
The methodology applied to the preliminary research, which was completed in the spring of 2017, indicated that species from this grouping met the conditions identified by Hirzel and le Lay (2008)
14


and was applied to 2326 individual taxa, grouped by family, of plants from the SEINet list of Flora of the Chihuahuan Desert Network. The initial search for taxa was performed by visually assessing distribution maps of individual taxa on the GBIF database website. If the species appeared in the Chihuahuan Desert region, but also had a broad presence globally, that species was determined to not fit the criteria of a ‘true’ Chihuahuan species and was removed from further consideration. If, however, a species appeared to have a distribution heavily centered in the Chihuahuan Desert region, it was retained for further analysis. After the GBIF determination was made, species were checked in SEINet. If the SEINet distributions showed a similar pattern to that of the GBIF occurrences, the species was kept for the next phase of assessment. Again, if the distributions on SEINet indicated a more global presence, the species were assumed to not be definitively Chihuahuan and were removed from further consideration.
The next step involved using Geographic Information Systems (GIS), specifically the ArcMap platform of ArcGIS. Point occurrence records of the ‘true’ Chihuahuan species were converted to vector shape files and overlaid with polygon representations of the southwestern US and the Chihuahuan Desert. From there it was possible to determine whether the Chihuahuan species that had been separated out actually had documented occurrences in both Colorado and the Chihuahuan Desert.
To specifically classify the habitats of species that exhibit the Chihuahuan sub-element, within Colorado, EcoRegion data Fevels III and IV were downloaded from the EPA ecoregion GIS database, compared with known presence points, and analyzed in ArcMap. The tabulated intersections matched presence points with the names of the EcoRegions III and IV in which they were contained, and the percentage of points within each ecoregion was calculated.
Results
A total of 43 Chihuahuan species representing 14 families was found to occur within Colorado (Table 1). They are: Asclepias macrotis Torr. (Apocynaceae), Funastrum crispum (Benth.) Schltr.
15


(Apocynaceae), Berlandiera lyrata Benth. (Asteraceae), Erigeron colomexicanus A.Nelson (Asteraceae), Gaillardia pulchellla Foug. var. pulchella, Palafoxia sphacelate (Nutt, ex Torr.) Cory, Thymophylla aurea (A. Gray) Greene ex Britton, (Asteraceae), Pericome caudata A. Gray (Asteraceae), Nolina texana S.Watson (Asparagaceae), Lesquerella fendleri (A. Gray) S. Watson (Brassicaceae), Physaria fendleri (A. Gray) O'Kane & Al-Shehbaz (Brassicaceae), Schoenocrambe linearifolia (A. Gray) Rollins (Brassicaceae). Aridaparviflora (A.Gray), D.R.Morgan & R.L.Hartm. (Asteraceae), Cylindropuntia imbricata (Haw.) F.M. Knuth (Cactaceae), Cucurbita foetidissima H.B.K. (Cucurbitaceae), Juniperus monosperma (Engelm.) Sarg. (Cupressaceae), Cyperus fendlerianus Boeckeler (Cyperaceae), Cyperus sphaerolepis Boeckeler (Cyperaceae), Croton texensis (Klotsch) Muell. Arg. (Euphorbiaceae), Dalea aurea C. Fraser, Dalea lanata Spreng., Dalea nana Torr., Desmanthus cooleyi (Eaton.) Torr., Hoffmannseggia drepanocarpa A. Gray (Fabaceae), Hoffmannseggia jamesii Torr. & A.Gray (Fabaceae), Mimosa borealis A. Gray (Fabaceae), Krameria lanceolata Torr. (Krameriaceae), Argemone squarrosa Greene (Papaveraceae), Andropogon saccharoidea Sw., Bouteloua barbata lag. (Poaceae), Bouteloua eriopoda (Torr.) Torr. (Poaceae), Eriochloa contracta Hitchc., Muhlenbergia arenacea (Buckl.) A.S. Hitchc. (Poaceae), Muhlenbergia arenicola Buckl. (Poaceae), Muhlenbergia fragilis Swallen (Poaceae), Muhlenbergiaporteri Scribn. (Poaceae), Muhlenbergia repens (J. Presl) Hitchc. (Poaceae), Muhlenbergia tenuifolia (Willd.) Britton, Stems, & Poggenb. (Poaceae), Sporobolus contractus A.S. Hitchc. (Poaceae), Sporobolus nealleyi Vasey (Poaceae), Cheilanthes eatonii Baker (Pteridaceae), Cheilanthes wootonii Maxon (Pteridaceae), Notholaena standleyi (Kiimmerle) Maxon (Pteridaceae).
The three families that had the greatest species diversity were Poaceae, with 13 species;
Fabaceae, with eight species; and Asteraceae, with six species. Within Poaceae, the genus Muhlenbergia was particularly well represented, accounting for six species. Within Fabaceae, the dominant genera were Dalea and Hoffsmanseggia. The Cucurbitaceae family accounted for the greatest number of occurrences, but was only represented by a single species, Cucurbita foetidissima
16


H.B.K. Two families that also had high numbers of occurrences were Asteraceae and Pteridaceae. Graminoids appeared to be the most common growth form, followed by forbs.
Chihuahuan species were present in five of out the 6 EPA Level III ecoregions, with the Wyoming Basin being the only ecoregion where species of the sub-element did not appear to be represented. The Colorado Plateau and the Arizona/New Mexico Plateau ecoregions had the lowest number of occurrences, both having fewer than ten records. In increasing order, the ecoregions that contained the highest number of occurrences were the Southern Rockies, High Plains, and Southwestern Tablelands. Over all, the southern half of the state contained the most occurrence records. While there are large clusters of occurrences in the southwestern portion of the state, the majority of occurrences are located east of the Rio Grande National Forest.
Discussion
My findings are consistent with Weber (1965), who concluded that the Chihuahuan sub-element has, in fact, influenced the flora of Colorado. Of the roughly 33 taxa that he identified as belonging to the Chihuahuan sub-element (that occur in Colorado), I found support for 23 (see Appendix Table A.9). My findings do not support the following species identified by Weber (1965): Abutilon incanum (Link) Sweet,Asclepias oenotheroides Cham. & Schlecht., Asplenium resiliens Kunze, Sapindus saponaria L., Stillingia sylvatica Gard, Engelmanniapinnatifida T.& G. and Thelesperma spp. These species appear to have other affinities with neighboring floristic elements and are not strongly Chihuahuan. For example, A. incanum, although present in both the Chihuahua Desert and Colorado, also had significant occurrences in the Sonoran, Peninsular and Coloradan floristic regions. I did not find evidence that Melampodium cinereum DC, A. Gray was of the Chihuahuan sub-element and present in Colorado. In both SEINet and GBIF, there was only one record of the species occurring in Colorado, and the coordinates were the location of the Denver Botanic Gardens. Additionally, the record was flagged for incorrect dates. Thus, it did not appear to be a reliable record and was excluded from analysis.
17


Two species identified by Weber (Eragrostis oxylepis (Torr.) Torr., Eriochloa contracta Hitchc.) were excluded in the early steps of analyses due to zero occurrence records in Colorado.
Taxonomy
Trends were observed with species diversity and numbers of occurrences. The three families that appeared to be the best represented by a diverse number of Chihuahuan species were Poaceae, Fabaceae, and Asteraceae. In contrast to the species diversity, neither Poaceae nor Fabaceae were represented with high numbers of occurrences.
The Cucurbitaceae family was the most well represented by number of occurrences. Coincidentally, the sole representative of Cucurbitaceae, Cucurbita foetidissima, was one of the original taxon that Weber identified as belonging to the Chihuahuan sub-element. Not surprisingly, Asteraceae, the largest flowering plant family, was the second greatest contributor of occurrences in this study. It also had the third greatest number of species, being represented by seven taxa. This supports Villasenor’s (1990) finding that not only does Asteraceae contribute substantially to the floristic richness of Mexico, but that Mexico may have played an important role in the diversification of the family, possibly as a secondary center of diversification
Finally, the family that was the third best represented through occurrence records was Pteridaceae, specifically Cheilanthes eatonii and Cheilanthes wootonii representing the Cheilanthoid clade. Both species have adaptations to drought stress and occur in seasonally dry environments (Hosh, 2008), which is consistent with expectations for plants of the Chihuahuan sub-element. Geography
Results differed in one prominent way from Weber’s original hypothesis. The list of identified species included species with occurrences on the Western Slope, whereas Weber separated genera that occurred west of the Rockies into the Sonoran-Great Basin elements. This may be due to differences in exclusion methods along defined floristic regions. This study did not remove a species from consideration if there were occurrences within a separate neighboring region. Rather the species was excluded if it appeared that the number of occurrences in the adjacent region was greater or equal
18


to the distribution in the study area. Nonetheless, by using the capabilities of GIS platforms to quickly create spatial representations of recorded occurrence points, I was also able to visualize that, as Weber noted, the number of species that appear to belong to this distribution pattern increase southward across the state, as do the occurrences of individual species.
Ecology of the taxa
The species that represented the Chihuahuan sub-element most strongly in Colorado appeared to be contained primarily within the boundaries of EPA Ecoregion Level IIEs Southwestern Tablelands and the High Plains ecoregions. These ecoregions are characterized by a variety of Quaternary sand substrates, sandy alluvial deposits, shales, and sandstones (Chapman et al., 2006). This is of note, as studies that have investigated the edaphic controls over floristic assemblies have identified a strong connection between gypsum substrates and certain endemic plants of the Chihuahuan Desert and Mexican high plateau (Moore et al., 2014). For instance, clades of Argemone, as well as Gaillardia (Marlowe & Hufford, 2007; Moore et al., 2014), have been documented to display gypso-philic affinities in their distributions (Schwarzbach and Kadereit, 1999; Moore et al., 2014). Gypsum, although an uncommon constituent of sand, is present in sandy deposits throughout Colorado (Adams et al., 1904) and this soil -distribution connection appears to be maintained as evidenced by this work’s recognition of Aregmone squarrosa and Gaillardiapulchella.
Graminoids and forbs appeared to be the most common plant habit type that exhibited the Chihuahuan sub-element in Colorado. This makes sense when one considers the ecoregions that these species occur in. Combined, the Southwestern Tablelands and the High Plains ecoregions are composed of predominantly prairie, grassland, and pine woodland vegetation communities, with few riparian areas (Chapman et al., 2006). Interestingly, my results paralleled the findings of Clark (1996), whose survey of the Mesa de Maya region of Colorado also described a Chihuahuan element present in the flora of southeastern Colorado. Clark reported occurrences of Notholaena standleyi, Nolina texana, and Cheilanthes eatonii, which I also identified as strongly Chihuahuan.
19


Although shrubs and subshrubs are reported, graminoids are most well-represented by species from the Chihuahuan sub-element, at least in Colorado. This conclusion differs slightly from previous work (Muldavin, 2002), which suggested that scrub and shrubland species more clearly indicate a Chihuahuan element rather than grassland species. One theory that may account for this dissimilarity is that the Chihuahuan Desert itself, and Muldavin’s study area, is experiencing invasion by shrubs resulting in increased shrublands (Alvarez et al., 2011). Whereas in Colorado, although shrublands occasionally occur on the eastern side of the state, they are more prominent on the western edge. However, the interactions between shrub and grass species that result in the habitat replacement is still not well understood.
GIS Approach
Currently, two methods are frequently used to assign species to communities, as well as phylogenetic and physiographic entities. First is the method of conducting thorough comparisons between floras that have agreement in species distributions. Essentially, this method involves visually inspecting and sorting range maps of taxa (McLaughlin, 1994). Second, researchers may use 'genetic methods' to analyze lineages of species and trace them back to a common origin (Comuet et al.,
1999). The latter method serves as an advanced option for detailed analysis. The method used in this study is most akin to the comparison of regional or local floras, albeit with the somewhat novel integration of GIS. The validity is evidenced with results consistent with the findings of Weber (1965) and Clark (1996).
Limitations
Although my research was conducted with the intention of utilizing unbiased and accurate data extracted from herbarium occurrence records, there are known limitations and areas that can be identified as potential sources of error. The species occurrence datasets, which the analysis relies upon, potentially introduced geographic bias as they were derived from both herbarium records and potentially opportunistic observations instead of planned surveys that would have been the highest standard (Fourcade et al., 2014). There was inconsistent editing and assumed geographic coordinate
20


systems assigned within the records. As a result, entries included misspelled scientific names, non-accepted scientific names, and incomplete entries. To address these issues, SQL queries were used to exclude erroneous records, as well as records geo-referenced to academic institutions and other clearly commercial collection locations (i.e., Denver Botanic Gardens). Additionally, the methodology was primarily driven by spatial data. It did not take into consideration the genotypic or evolutionary concepts of individual species.
Conclusion
It's likely that by incorporating known biologic or genotypic traits into the methodology of this thesis (after spatial representations are analyzed), the accuracy of results would be greatly increased. This type of information could facilitate a more accurate understanding of whether current species distributions reflect only realized niches, or if they reflect the expansion and adaptations of a species as well. Future research should focus on phylogeography to inform patterns revealed herein. Additionally, attention should be focused on those species within Colorado that occur along the boundaries of defined floristic elements.
21


Tables and Figures
Table 1.1 Chihuahuan species found to reach Colorado, with associated EPA Level III ecoregion classification
Family CD o_ CD* GO Colorado Plateau Southern Rockies Arizona New Mexico Plateau Wyoming Basin High Plains Southwestern Tablelands
Apocynaceae Ascpelias macrotis Torr. x Funastrum crispum (Benth.) Schltr. x x
K, Asteraceae NJ Aridaparviflora (A. Gray) D.R.Morgan x x x and R.L. Hartm. Berlandiera lyrata Benth. x x
Erigeron colomexicanus A. Nelson x x x x Gaillardia pulchella Foug. var. pulchella x x x x Thymophylla aurea (A. Gray) Green ex x Britton Palafoxia sphacelata (Nutt, ex Torr) Cory x x x
Peri come caudata A. Gray
x


Table 1.1 cont’d
Family in o o_ G>' in Colorado Plateau Southern Rockies Arizona New Mexico Plateau Wyoming Basin High Plains Southwestern Tablelands
Asparagaceae Nolina texana S. Watson x x
Brassicaceae Lesquerella fendleri (A. Gray) x
Cactaceae Schoenocrambe linearifolia (A. x x x x Gray)Rollins Physaria fendleri (A. Gray) O'Kane & Al- x Shehbaz Cylindropuntia imbricata (Haw.) F.M. x x x Knuth
Cucurbitaceae Cucurbita foetidissuma H.B.K. x x x x
Cupressaceae Juniperus monosperma (Engelm.) Sarg. x x x
Cyperaceae Cyperus fendlerianus Boeckeler x x Cyperus sphaerolepis Boeckeler x
Euphorbiaceae Croton texensis (Klotsch)Muell.Arg x x


Table 1.1 cont’d
Family CD o_ CD* CO Colorado Plateau Southern Rockies Arizona New Mexico Plateau Wyoming Basin High Plains Southwestern Tablelands
Fabaceae Dalea aurea C. Fraser x Dalea lanata Spreng. x Dalea nana Torr.
K> -P^ Desmanthus cooleyi (Eaton) Branner & x Colville Hoffmannseggia drepanocarpa A. Gray x x Hoffmansegia jamesii Torr. & A. Gray x Mimosa borealis A. Gray x
Kramericeae Krameria lanceolata Torr. x x
Papaveraceae Aregmone squarrosa Greene x
Poaceae Andropogon saccharoides Sw. x x


Table 1.1 cont’d
Family
Species
Bouteloua barbata Lag. Bouteloua eriopoda (Torr.) Torr. Eriochloa contracta Hitchc.
I §
5 8
O ta
6 ^
H to
§ .a
s 8 % *
O
o
fl * o
N
*c
<
s 3
£ E
z;
60
.5 s
S "m
O 63
>>PQ
J3
E
-63
60
X
« T3 to 63 U 63
E
E 3
§ H
GO
X
to
Ol
Muhlenbergia arenacea (Buckl.) A.S. x x
Hitchc.
Muhlenbergia arenicola Buckl. x x
Muhlenbergia fragilis Swallen x
Muhlenbergiaporteri Scribn. x
Muhlenbergia repens (J. Presl) Hitchc. x
Muhlenbergia tenuifolia (Willd.)Britton, x
Stems, & Poggenb.
Sporobolus contractus A.S. Hitchc. x x x


Table 1.1 cont’d
Tl 1 m o o_ G>' in Colorado Plateau Southern Rockies Arizona New Mexico Plateau Wyoming Basin High Plains Southwestern Tablelands
Pteridaceae Sporobolus nealleyi Vasey x Cheilanthes eatonii Baker x x x Cheilanthes wootonii Maxon x x
Notholaena standleyi (Kiimmerle) Maxon x x x
NJ
Os


Figure 1 Reference map of southwest US deserts; the Mojave Desert, the Sonoran Desert, and the Chihuahuan Desert.
27


CHAPTER II
NICHE MODELING Relevant Approaches to Niche Modeling
The geographic distribution of a species is fundamentally linked to that species’ ecological niche (Peterson and Soberon, 2012). It follows then, that ecological niche models (ENMs) are a key resource in understanding complex relationships between a species’ distribution and its environment. Niche models are more frequently being used to investigate questions regarding speciation, niche evolution and cladistic ecological diversity (Warren et al., 2008). Niche models have also proven highly useful in studies of areas of endemism, and the delimitations of biogeographic units (Hausdorf, 2002), such as floristic elements.
Niche modeling is based on four facets of the niche concept: 1) niche characteristics, 2) niche interactions, 3) community-wide processes, and 4) niche evolution (Hirzel and Le Lay, 2008). Although this terminology is used across concentration areas, the literature reveals that the term niche has been assigned to cover two separate concepts. The first describes ‘niche’ as the environmental requirements needed for a species to subsist, while the second refers to ‘niche’ as the relationship between one species to others (Peterson and Soberon, 2012). These two concepts require separate and distinct tools to conduct investigations. For the purposes of my work, niche modeling refers to ‘niche’ as the term applies to environmental conditions that a species needs to populate an area.
Each of the studies mentioned in Chapter 1, as well as many others that focus on species distribution patterns, involved the use of prediction models based on associations with ecological niches (Hijmans and Elith, 2016). One such model that is frequently used is the habitat suitability model (HSM), which attempts to relate environmental variables to the likelihood of occurrence of the species (Hirzel and Le Lay, 2008).
Habitat suitability models are based heavily on niche characteristics. Also referred to as Species Distribution Models (SDM), the aim of these types of models is to estimate the similarity of the
28


conditions at any site to the conditions at the locations of known occurrences of a phenomenon (Hijmans and Elith, 2016).The steps usually taken are follows: (1) locations of species occurrence are compiled; (2) values of environmental predictor variables (i.e., climate) at these locations are collected from spatial databases; (3) those values are used to fit a model to estimate similarity to the sites of occurrence, or abundance of the species; (4) that model is used to predict the variable of interest across the focus region (Hijmans and Elith, 2016).The literature also points to this type of modeling being used to generate predictions of possible occurrences for areas of endemics (Olivier and Aranda, 2017). Although there has long been debate about the terms ENM, SDM, and HSM, in the broad context, these seemingly different terms refer to effectively the same type of analyses (Peterson and Soberon, 2012).
These predictive models aim to forecast the likelihood of a species occurrences based on environmental variables. Simultaneously, these models are often used as applications to predict outcomes such as the presence/absence of a species in a given area, or the abundance of species throughout a study area (Hirzel and Le Lay, 2008). However, some researchers contend that the relationship suggested by habitat suitability models is weak. Therefore, the strength of a habitat suitability model that is focused on the distribution-niche link is dependent upon the specific ecology of a single species, local physical constraints of the study area, and accuracy of records of historical events. (Hirzel and Le Lay, 2008).
There are a variety of different model algorithms available to modelers in the environmental sciences. The most frequently used methods present results in terms of probabilities.
Relative Frequency
Frequency data is one option used to describe the distribution of species in a community. Loehle (2012) introduced and developed a novel method for use by the National Council for Air and Stream Improvement, which compared the relative frequency of species occurrence points to that of random points (pseudo absence) to compute a frequency ratio for habitat distribution models. Specifically, the
29


Relative Frequency Function tool was developed to model predicted preferred habitat, by calculating the relative probability (frequency) of finding plant or animal species at given points within a study area. This function is viewed as an extension of the Resource Selection Probability Function (Loehle 2012, Manley et al., 2002), which is used in cases of data where only used and available points are known and can be compared to discrete resources (Lele et al., 2013).
Generalized Linear Models
The generalized linear model, or GLM, is a traditional model frequently used by biologists before the advent of Maxent. It is mathematically comparable to a multiple regression analysis and has been used extensively in SDMs because of its strong statistical foundation in modeling ecological relationships (Elith et al., 2006). GLMs have been widely used in studies where binary data are used (e.g., wildlife biology). Results are interpreted by looking at the regression coefficients of each variable, which can be useful for explanatory SDMs. The advantage of this type of model is that it is an extension of a linear model, so it allows for non-linear relationships in the data and, thus, does not force non linear data into non-natural scales (Guisan et al., 2002).
MaxEnt for Presence Only Data
The statistical modeling software Maxent (Phillips et al., 2006) is currently considered one of the most accurate and reliable options for biological researchers investigating ecological processes (Elith et al., 2010). One of its main performance highlights is its ability to accurately model potential species’ distributions using ‘presence only’ datasets. For example, this software has been used to build niche models using the evolutionary history of desert species to understand population expansions (Graham et al., 2013). It has also been used to predict geographic distributions of a variety of plant species for conservation purposes (Phillips et al., 2005). The program is widely used because of ease of use especially for researchers who are not necessarily trained in standard GIS or computer science platforms, as they can conduct multi-criteria model analysis (Fourcade et al., 2014). Despite
30


the program’s popularity, many aspects such as the statistical functions of the program are not well or completely understood by its users (Yackulic et al., 2012).
Geographic Information Systems
Coupled with the increasing availability of open source data and advances in technology, GIS has become a highly effective tool for studying and representing ecological trends (Dark and Bram,
2007). Tools and models created within GIS applications have recently been used to model changes in the representations of both plant and animal species distributions. For example, Olivier and Aranda (2017) used niche analysis to model a little known species of grasshopper with DIVA-GIS by comparing point occurrence data with 19 bioclimatic variables from the WorldClim database. Their results revealed that GIS models can effectively predict geographic distributions for species using presence only data, while having limited background knowledge of a species’ biological traits or affinities.
Some of the most commonly used tools and procedures for GIS based niche modeling have been compiled by Brown (2014), including the Weighted Linear Combination (WLC) method (Maczewski, 2004). The WLC, also referred to as simple additive weighting, is considered the most straightforward option for modeling landscape processes in GIS. It is based on the concept of a weighted average, where the modeler directly assigns “weights of relative importance” to each attribute map layer. A ratio scale is then constructed, with values corresponding from least to most preferable alternatives. Products are summed over all the attributes and the score of the alternative with the highest overall score is chosen. The results of this technique are used to derive composite maps.
The two most critical elements that must be carefully addressed to accurately implement the WLC model are assigning weights to attribute maps and implementing procedures for deriving the attribute maps (Malczewski, 2000). Simply, to have confidence in the accuracy of the final composite
31


map, the attribute maps (which are over-layed to produce the final results) must represent rankings and weights that have been calculated as accurately and objectively as possible.
Appropriately deriving values for different attributes within a layer is particularly challenging for ecological managers (e.g., the annual precipitation range 10 to 15mm is most likely to be linked to species occurrence therefore its assigned value is 10, versus the annual precipitation range of 7 to 9mm is least likely therefore its assigned value is 1). One method that has been suggested is the value function approach (Hepner, 1984, Hobbs, 1980, Lai and Hopkins, 1989, Keisler and Sundell, 1997), which converts different levels of an attribute into scores. The scores are then related to a preference scale reflective of the decision makers interests. A common technique used to derive the value scores is to follow the mid-value method (Keeney, 1980, Lai and Hopkins, 1989). Users identify the maximum and minimum values of an attribute, then the midpoint value of the interval between the maximum and minimum attribute values is calculated. Once the midpoint value is determined, quarter points can be calculated between the minimum and the midpoint, and then between the midpoint and the maximum. The procedure can be repeated, accuracy increasing with points added.
Habitat Suitability Analyses Considerations
A caveat that comes with all environmental models is that results can have high variability depending on parameters and input data. This is particularly true with the increasing use of GIS applications, such as ArcGIS, QGIS and Grass GIS. Today, well intentioned modelers have difficulty avoiding the many pitfalls, unintentional or otherwise, associated with applying modeling methods without fully comprehending the mechanisms involved.
Second, nearly all environmental models suffer from the Modifiable Areal Unit Problem (Dark and Bram, 2007). The Modifiable Areal Unit Problem (MAUP) occurs as a result of setting artificial partitions to define an area of analysis. These boundaries are often necessary for the simple reason that focus areas or areas of interest must be bounded or contained to some standardly recognized area.
32


Finally, the choice of model determines model outcomes (Gutierrez et al., 2013), therefore, model selection should match research objectives. Ideally, the model would be capable of using multicriteria decision-making methods that would account for potentially conflicting parameters, future expert options, or more species-specific findings that could be inserted into model parameters. It was important to select a model that would provide easily interpreted, but highly informative results.
There are few comparable studies that specifically attempt to link or model the influence of the Chihuahuan biogeographic sub-element on the flora of Colorado. However, more literature exists documenting similar research on the California Floristic Province (Baldwin, 2014). These studies detailed the framework for methods that involved the use of multiple regression analyses to link occurrence hotspots to climatic variables (Richerson and Lum, 1980).
Although geoprocessing tools exist in the ArcGIS platform that can conduct regression analysis, such as the Ordinary Least Squares (OLS) and Geographically Weighted Regression tool (GWR), those tools are not appropriate for this study. The OLS tool, and similarly the GWR tool, are classic linear methods that perform poorly when relationships between the dependent variable and the explanatory variables are nonlinear (Esri ArcGIS), which is often the case in ecology. For these reasons, the approach of combining a modified GLM structure for variable analysis with the Weighted Linear Combination method was taken.
Research Objectives
My objective was to characterize the niche for those Chihuahuan species that are present in the flora of Colorado (based on analyses in Chapter 1) using GIS. I first identified the environmental factors that were influential to a species overall distribution, by testing their effect on species presence through statistical analyses. The final goal was to broadly enumerate a set of criteria based on the environmental conditions that contribute to overall distributions and then create spatial representations of those criteria within Colorado. Thus, areas within Colorado that can provide suitable habitat would be identified.
33


Methods and Materials
To gain a broad and practical insight into the variety of floristic niches that appear to be influenced by flora of the Chihuahuan sub-element, five species were chosen from the list compiled in Chapter 1 for niche modeling. The first goal when selecting the model species was to include species that had an almost exclusively Chihuahuan distribution, with little overlap to other elements. Consideration was also given towards previous recognition of a species in association with the Chihuahuan sub-element, for example by Weber (1965) or Clark (1996). Next, species needed to encompass a variety of plant habit types. Ultimately, Funastrum crispum (Benth.) Schltr., Palafoxia sphacelata (Nutt, ex Torr) Cory, Nolina texana S. Watson, Muhlenbergia arenacea (Buckl.) A.S. Hitchc. andMuhlenbergia arenicola Buckl. were selected.
BioClim Data Processing
Data preparation followed best practices outlined by the SDM Toolbox (Brown, 2014). Bioclimatic variables provided by the Worldclim database served as the independent variables (Fick et al., 2017). These 19 variables (Appendix Table A.l) were selected as predictor/independent variables for this work as they are commonly used in ecological models as experimental variables. To incorporate the most up to date information possible, the ‘Future’ dataset, which interpolates data from 1970 to 2000, was used for this analysis.
Next, data was processed to mitigate latitudinal bias. Most models, including Maxent, assume equal cell size across the data layer. However, the WorldClim Bioclim datasets are not projected, and as such, there is varying cell area in the rasters. To correct for this, the layers were projected from WGS84 to the NAD (1983) 2011 Albers Equal Area Projection. Additionally, a bias grid was constructed to correct for latitudinal bias during the selection of background/pseudo absence points and projected to match.
Pseudo absence points were generated using a pre-made python script (Dilts 2015) and merged with known occurrence points. Then points were selected to include only those completely within the
34


bias file polygon for a complete occurrence dataset per species. Last, the data points were spatially rarefied using a pre-made python script tool supplied from the SDM toolbox (Brown, 2014). Actual presence location points were dummy coded as 1, and the pseudo absence points as 0 in the attribute tables.
The Sample tool in ArcMap was then used to collect the values of BioClim data at each XY location of occurrence and pseudo absence points and displayed them in a table. The next step was the statistical analysis of the data, using SPSS, and inputting compiled Sample tool result tables. Statistical Analyses
To ensure the key environmental variables that affect these species were identified, and not arbitrarily excluded during dimension reduction, Principal Component Analysis was first utilized inputting all 19 variables, with a loading cutoff value of .5. This cutoff level was imposed both to attempt to mitigate for the inclusion of highly correlated variables as well as ensure dimension reduction so only the most important variables were examined further. By selecting a large number of individual variables, and then synthesizing or grouping the variables into factor groups, the important variables were not excluded. This was chosen as the first step in statistical analysis as it focuses on those co-variates that account for the greatest percent of variance for the species (Cruz-Cardenas et al., 2014). Co-variates from the component that accounted for the most variance were included in the next round of analysis if they met the load value cutoff.
Those co-variates were then tested with stepwise binomial regression analysis. This was performed so that it would be possible to understand the levels of influence each variable had on the dependent variable (species presence) as well as thoroughly mitigate the inclusion of highly correlated variables.
Habitat Suitability Analysis
The GIS Weighted Linear Combination Method was subsequently used to perform a modified habitat suitability assessment within Colorado for the five selected species.
35


The beta coefficients calculated in the regression analysis were normalized to a 0-1 scale value and served as the factor weights. Finally, the alternative attribute values and corresponding preference scores were determined by inputting the data from each co-variate selected as an analysis layer into the Relative Frequency Function Tool (RFF). The suitability scores generated from the RFF Tool were averaged, classified (binned) by natural breaks groups. A grouping was considered a natural break if the average for an individual attribute was distinct from other values. For example, if the average RFF score for 10 degrees Celsius calculated to 7, and no other average score in the temperature range of 9-15 calculated to 7, that was a natural break. If the resulting averages were the same across consecutive attributes, they were grouped together. From there, the average scores were normalized to values corresponding to a 1-10 scale. Those served as the alternative attributes and preference values.
After the core steps of this process were completed, sensitivity analyses were conducted for all model outputs. The purpose of this was to define how changes in a model parameter would affect the overall model outputs, and to some degree indicate the level of uncertainty involved in the measurement of that parameter (Wainwright, 2013). The confidence levels associated with the selection of analysis layers through PCA, and factor weight values determined from regression analysis were high enough that testing their sensitivity to different parameters was not considered necessary. However, the greatest uncertainty stemmed from the novel methods implemented to classify alternative attributes and score them. As such, the sensitivity analyses tested the results when the alternative attributes were classified using average midpoints and quartile values instead of classifying them primarily as natural break value groups.
Finally, to further compare the accuracy of the WLC habitat suitability outputs, results were compared with a MaxEnt model output. All defaults of MaxEnt parameters were accepted and data was prepped following step by step best practices (Young et al., 2018). To run MaxEnt v 3.3 through ArcMap, the python script tool MaxEnt Script Tool for ArcGIS (Donoghue, 2013) was downloaded and run in ModelBuilder.
36


Results
In all five models, the percent variance attributed to these predictor variables was over 90%, indicating that these variables successfully included factors that were important to the presence or absence of these species. Yet, when all statistical assumptions were addressed and accounted for (i.e., potentially high correlations between variables or symptoms of multicollinearity, etc.) the number of variables that independently influenced the presence or absence of a species was usually less than ten. Palafoxia sphacelata (Asteraceae)
Two principle components accounted for a cumulative 94.620% of variance. The first component accounted for the greatest percent of variance (67.284%) and was used for niche modeling (the second component accounted for the remaining 27.335%). When the 0.5 loading value cutoff was applied to the first component, 10 co-variates were retained for further testing. Of those, the regression analysis showed that four (Table A.5) were the most useful with Bio 3 (Isothermality) having the greatest effect on species presence. Bio 2 (Mean Diurnal Range) subsequently had the next greatest effects on species presence for the target species.
The results of the Relative Frequency Function (RFF) tool showed that for Bio 3 (Isothermality), the bin that yielded the highest average suitability score (6) was the 36 to 38.9° C grouping. For Bio 2 (Mean Diurnal Range), the grouping that had the highest associated suitability score (6) was 19 to 22.9° C.
The full range of the distribution of Palafoxia sphacelata stretched from Chihuahua, Mexico to northern Colorado USA (Fig. 2.1). The highest suitability level was calculated to a maximum of approximately 5.77 and a minimum value of 1.55 where the minimum represented areas that had the lowest suitability but still met suitable habitat requirements. This species appears to follow the trend of the most suitable habitat appearing on the eastern and southern extents of Colorado. The EPA Level II ecoregions that contained the areas of highest suitability were (in descending order)
Southern Rockies, High Plains, Arizona/New Mexico Plateaus, and Southwestern Tablelands.
37


Within those ecoregions, the following counties had the areas of highest suitability: Weld, Morgan, Logan, Sedgwick, Phillips, Yuma, Washington, Adams, Arapahoe, Douglas, Elbert, Kit Carson, Cheyenne, Lincoln, El Paso, Teller, Park, Chaffee, Fremont, Pueblo, Crowley, Kiowa, Prowers,
Bent, Otero, Custer, Saguache, Alamosa, Huerfano, Las Animas, Baca, Costilla Conejos. Muhlenbergia aretticola (Poaceae)
Two principle components accounted for a cumulative 94.884% of variance. The first component accounted for the greatest percent of variance (64.441%) and was used for niche modeling. (The second component accounted for the remaining 30.442%.) When the 0.5 cutoff level was applied to the first component, six co-variates were retained for further testing: (Bio 11, 9, 7, 6, 4,3) A binary logistic regression analysis showed that all six co-variates contained useful information (Appendix Table A.3). Of those, Bio 11 (Mean Temperature of the Coldest Quarter) had the greatest effect on species presence, followed by Bio 6 (Min Temperature of the Coldest Month). The Relative Frequency Function (RFF) output for Bio 11 (Mean Temperature of the Coldest Quarter) showed that the bin with the highest suitability (7) was the 11 to 12.9 ° C range, and for Bio 6 (Min Temperature of the Coldest Month) it was the range of 0 to 4.9° C.
M. arenicola occurs from Mexico City, Mexico to Colorado (Fig. 2.2). The highest suitability level was calculated to a maximum of approximately 6.4 and a minimum value of 2.03, where the minimum represented areas that had the least likelihood of suitability but still met suitable habitat requirements.
The EPA Level III EcoRegions that contained the areas of highest suitability were (in descending order): Southern Rockies, High Plains, Southwestern Tablelands, and Colorado Plateaus. Within those ecoregions, the following Colorado counties had the areas of highest suitability: Boulder, Broomfield, Denver, Arapahoe, Kit Carson, Cheyenne, Lincoln, El Paso, Fremont, Pueblo, Crowley, Kiowa, Prowers, Bent, Otero, Huerfano, Las Animas, Baca, Southwestern Montezuma. Suitability values within Colorado ranged from 3.2 to 5.3.
38


Funastrum crispum (Apocynaceae)
Two principle components accounted for a cumulative 92.998% of variance. The first component accounted for the greatest percent of variance (67.376%) and was used for niche modeling (the second component accounted for the remaining 25.622%).
When the 0.5 loading value cutoff was applied to the first component, nine co-variates were retained for further testing. The regression analysis showed that five co-variates (Table A.4) had the greatest effect on species presence for Funastrum crispum, with Bio 2 (Mean Diurnal Range (Mean of monthly (max temp-min temp))and Bio 16 (Precipitation of Wettest Quarter) having the greatest impact.
The results of the Relative Frequency Function (RFF) showed that for Biol6(Precipitation of Wettest Quarter), the bin whose attributes had the highest suitability score (8) was the 281 to 331,9mm range. For Bio 2 (Mean Diurnal Range (Mean of monthly (max temp-min temp), the bin associated with the highest suitability score (8) was the 9 to 11.9° C range.
The full range of the distribution of F. crispum stretched from Aguascalientes, Mexico to Colorado, USA (Figure 2.3). The highest suitability level was calculated to a maximum of approximately 7.9 and a minimum value of 2.0 where the minimum represented areas that had the lowest suitability but still met suitable habitat requirements. Within Colorado, there were small non continuous patches of areas with high suitability. These were located clustered along, and to the west of the Continental Divide encompassing two ecoregions. The EPA Level III EcoRegions that contained the areas of highest suitability were the Southern Rockies and the Colorado Plateaus. The following counties within those ecoregions that had the areas with the highest suitability: Alamosa, Boulder, Chaffee, Clear Creek, Costilla, Custer, Delta, Dolores, Eagle, El Paso, Fremont, Garfield, Gilpin, Gunnison, Hinsdale, Huerfano, Jackson, Jefferson, La Plata, Lake, Larimer, Las Animas, Mesa, Moffat, Montezuma, Ouray, Park, Pitkin, Pueblo, Rio Blanco, Routt, Saguache, San Juan, San Miguel, Summit, and Teller.
39


Nolina texana (Asparagaceae)
Three principle components accounted for a cumulative 98.885% of variance. The first component accounted for the greatest percent of variance (73.824%) and was used for niche modeling (the second component accounted for 19.13%, the third accounted for 5.9%). When a 0.5 loading value cutoff was applied to the first component, eight co-variates were retained for further testing (Bio 19, Bio 17, Bio 15, Bio 14, Bio 13, Bio 12, Bio 3, Bio 2).
The regression analysis for that subset of co-variates showed that seven co-variates were the most useful (Table A.6). Bio 2 (Mean Diurnal Range (mean of monthly (max temp-min temp)) and Bio 15 (Precipitation Seasonality) had the greatest positive effect on species presence for Nolina texana. Then RFF tool output showed that for Bio 2 (Mean Diurnal Range (mean of monthly (max temp-min temp)), the bin whose attributes had the highest suitability score (9) was the 10 to 12.9° C range . For Bio 15 (Precipitation Seasonality), the bin whose attributes had the highest suitability score (10) was the 31 to 41.9 mm range.
The full range of the distribution of N. texana stretched from San Luis Potosi, Mexico to Colorado, USA(Fig. 2.4). The highest suitability level was calculated to a maximum of approximately 9.06 and a minimum value of 1.93 where the minimum represented areas that had the lowest suitability but still met suitable habitat parameters. The suitability values within the state of Colorado range from 1 to 10. Although portions of Colorado scored as high suitability, most areas in Colorado were moderate to low and were only found in one ecoregion. The EPA Level III Ecoregion that contained the areas of highest suitability was the Southern Rockies. Nolina texana, appears to have habitat requirements found mostly in the eastern half of Colorado. Within that ecoregion, the following counties had the highest suitability scores: Alamosa, Boulder, Chaffee, Clear Creek, Costilla, Custer, El Paso, Fremont, Huerfano, Jefferson, Las Animas, Larimer, Park, Pueblo, Saguache and Teller.
40


Muhlenbergia arenacea (Poaceae)
Two principle components accounted for a cumulative 96.15% of variance. The first component accounted for the greatest percent of variance (71.024%) and was used for niche modeling (the second component accounted for the remaining 25.129%). When the 0.5 loading value cutoff was applied to the first component, ten co-variates were retained for further testing: Bio 5, Bio8, BiolO, Biol2, Biol3, Bio 14, Biol6, Bio 17, Bio 18, and Biol9. A binary logistic regression analysis showed that six co-variates contained the most useful information (Appendix Table A.7). Of those, Bio 18 (Precipitation of Warmest Quarter) had the greatest effect on species presence, followed by Bio 16 (Precipitation of Wettest Quarter). The Relative Frequency Function (RFF) output for Bio 18 (Precipitation of Warmest Quarter) showed that the bin with the highest suitability score (6) was the 145 mm to 166.9 mm range, for Bio 16 (Precipitation of Wettest Quarter) it was the range of 20.3 mm to 26.69 mm.
Muhlenbergia arenacea occurs from Zacatecas, Mexico north to Colorado (Fig. 2.5). The highest suitability level was calculated to a max of approximately 6.02 and a minimum value of 3.9, where the minimum represented areas that had the least likelihood of suitability but still met suitable habitat requirements.
The EPA ecoregions that contained the areas of highest suitability were (in descending order): High Plains, Southern Rockies, Southwestern Tablelands, and Arizona/New Mexico Plateau. Within those ecoregions, the following Colorado counties had the areas of highest suitability: Adams, Alamosa, Arapahoe, Baca, Bent, Chaffee, Cheyenne, Costilla, Crowley, Custer, Denver, Douglas, El Paso, Elbert, Fremont, Huerfano, Kiowa, Larimer, Las Animas, Lincoln, Morgan, Otero, Park, Prowers, Pueblo, Saguache, Washington, and Weld. Suitability values within Colorado ranged from 2.27 to 6.9. Counties on the eastern half of the state (Las Animas, Huerfano, Pueblo, Otero, Bent, Crowley, Lincoln, Kiowa, Prowers, Morgan, Adams, Weld, and Arapaho) were revealed to harbor much of the most suitable habitat (Fig. 2.5). The results of the habitat suitability analysis also reveal
41


that seven counties were almost entirely composed of suitable habitat and could likely sustain populations of Miihlenbergia ctrectncect: Adams, Arapahoe, Custer, Huerfano, Morgan, Park and Weld counties.
MaxEnt Comparison
When comparing MaxEnt outputs with statistical results derived from PCA and regression analyses, there were similar findings for all five of the species modeled. Areas that scored medium to high suitability from the WLC method were also generally areas of suitability, according to MaxEnt outputs.
For all modeled species, calculations produced from MaxEnt which identified co-variates of greatest importance, resulted in findings that corresponded to my statistical calculations. For F. crispum, both methods revealed that Bio 16 (Precipitation of Wettest Quarter) had a high contribution to species presence. For P. sphacelata, both methods identified Bio 3 (Isothermality) as having high influence on species presence. Similarly, for M. cirenicolci, both methods calculated Bio 6 (Min Temperature of Coldest Month), and Bio 11 (Mean Temperature of Coldest Quarter), to be the variables with the greatest impact on species presence. For A texcma, both methods identified Bio 15(Precipitation Seasonality) and Bio 2 (Mean Diurnal Range) as the having the greatest impacts. Finally, both methods revealed that forM ctrenacect, Bio 18 (Precipitation of Warmest Quarter), and Bio 16 (Precipitation of Wettest Quarter) were important to predicting species presence.
Discussion
A GIS based model was used to model the climatic niche for five species of the Chihuahuan subelement that are present in Colorado. In so doing, I also compare my results with those of software more commonly used by ecologists (i.e., MaxEnt), thereby assessing their validity. In several ways it is not unlike other niche model studies. However, it moved beyond the standard format of results, which are usually composed of a list of predictor variables that are important in explaining species distributions. Here I was able to obtain results that included: a list of predictor variables that impact
42


species presence, their relative levels of influence, subdivided groupings of individual variable characteristics ranked from least to most suitable, and visual representations of alternative results based on changes in parameterizations.
Abiotic and Biotic Patterns
Two variables appeared to have the greatest impacts on three of the modeled species. First, the variable Bio 2 (Mean Diurnal Range) appears to link species presence and climatic conditions. It had the greatest influence on both Nolina texana, Funastrum crispum and Palafoxia sphacelata. Second, Bio 16 (Precipitation of Wettest Quarter) was calculated to have great impact on both F. crispum and M. arenacea. It’s possible that these environmental factors play a key role in the presence of plants of the Chihuahuan biogeographic sub-element in Colorado.
Geographic Patterns
Although low to medium high suitability is distributed across Colorado, small pockets of high suitability occur, particularly in the southwestern and southeastern portions of the state. The differences between individual species niches were noticeably apparent. First, while most of the species had strong levels of suitable habitat in the southeast comer of Colorado, the suitability of the rest of the state was not as uniform. For some species nearly the entire state contained at least some level of suitable habitat, (Palafoxia sphacelata). Others such as Muhlenhergia arenicola very clearly excluded the northwestern comer and only met suitable habitat criteria in the eastern portion of Colorado. The eastern portion of the state is composed mostly of mixed grass shmblands (Rondeau et al. 2013), which is consistent with the observation that graminoids account for the majority of habit types of the identified species.
Interestingly, majority of suitable habitat in Colorado was contained within the Southern Rockies EPA Level III EcoRegion. Although the majority of actual occurrences were found to occur in the High Plains and Southwestern Tablelands EcoRegions (Chapter I) it appears that species of this element may also find suitable habitat westward of their current known locations. N. texana and F.
43


crispum demonstrate this unique westerly affinity most clearly. The three other species outputs do indicate areas of suitable habitats that extend into the actual Rocky Mountain range. However, the most suitable areas in Colorado for TV texana and F. crispum are only slivers of land, almost exclusively located along the Rocky Mountain chain. This suggests there is a possible connection with the climatic conditions produced in pockets of mountainous areas with the presence of these species.
Sensitivity Analysis
From the five sensitivity analyses, three outputs (Palafoxia sphacelata, Nolina texana, Muhlenbergia arenacea) appeared to have somewhat inverse results from the original Natural Breaks grouping method. The results for P. sphacelata show an area in northeast Colorado, Weld county, as highly suitable habitat, but the same county appears as low suitability in the sensitivity outputs (Fig. 2.1). In two of the outputs (Muhlenbergia arenicola, Funastrum crispum), the sensitivity results showed a less conservative classification of suitable habitat. Areas within Baca county that appeared to be low suitability for F. crispum in the natural grouping method were classified as moderate in the sensitivity analysis (Fig. 2.3). However, upon further examination of the sensitivity outputs with actual occurrence points, the sensitivity outputs more often than not, place known occurrence points in areas of least suitability. It is possible that in the instances where results were impacted by altering parameters, results might have been more congruent if additional midpoint breaks had been added. This could help ensure that small variations in data are not lost in large quartile groupings.
In the case of these five species, the sensitivity analysis revealed that by applying pressure to the preference scale parameter the actual range of suitability scores were not dramatically impacted, but the placement and geographic classification of suitable habitat had great variation.
Method Comparison
Ecologists generally agree that using MaxEnt for species distribution models produces highly reliable and accurate results (Aguirre-Gutierrez et al., 2013). However, users inevitably require the
44


use of additional software packages such as R, ArcGIS, QGIS, DIVA-GIS or Grass GIS for data processing. This aspect of the use of MaxEnt seems redundant, as both R, ArcGIS, Q and Grass GIS have statistical computing capabilities in addition to their cartographic functions. Additionally, despite the accuracy associated with MaxEnt, its black box style processes do not easily allow for finding and correcting errors entered into the workflow, whereas with my process, it was possible to backtrack if needed.
In this work, the majority of MaxEnt outputs compared well with result outputs from the Weighted Linear Combination method. Any differences stemmed from a few different parameters. For example, suitability values associated with the WLC outputs are normalized to a 1 to 10 scale. Outputs from MaxEnt are given in log odds values. Second, with the WLC, areas that did not meet the suitable habitat requirements were assigned a NoData value and had no graphical representation. MaxEnt, outputs still assign values to areas that have near zero suitability. As a result, MaxEnt outputs display values and corresponding color to the entire study area whereas in the WLC, areas that might be considered not suitable are not included in final outputs (Fig. 2.6).
Limitations of the Study
Several known limitations of the Weighted Linear Combination Method for habitat suitability analyses have been addressed in similar research (Malczewski, 2002, Drobne and Lisec, 2009), and are applicable to this work. First, the model relies on the quality of the data used. True presence-true absence studies have been shown to yield the highest performance capability when used in environmental modeling. Here, presence only datasets were modified to function as ‘presence-pseudo absence.’ While the methods used to create the ‘background’ sample points followed standard industry procedures (Barbet-Massin et al., 2012), the results of the analysis should not be interpreted to represent those of a real-world inventory of true presence and true absence data.
The bioclimatic datasets although provided by a reputable database, are collected using interpolation methods, which also impact results. Additionally, the 19 bioclimatic variables provided by the WorldClim database, while readily available and frequently used as predictor variables in
45


SDMs (Cruz-Cardenas et al., 2014, Aguirre-Gutierez et al., 2013), are not necessarily the most relevant variables for floristic studies. Desert plant distributions have been found to be strongly impacted by soil characteristics (Laport et al., 2013). Variables such as soil moisture, soil type, or radiation levels might serve as better predictor variables, but the data are not as easily accessible (Forester, 2012).
Next, the regression analysis that was used to select the variables to include as raster layers in the habitat suitability analysis accounted for less than the cumulative variance originally extracted from the first components analyzed through PCA. This means that there are still unknown factors that account for large portions of the variance. Future studies that incorporate additional environmental variables into regression analyses could produce more detailed and comprehensive results.
Finally, and most significant, the ‘Assign Ratio Value Scale of Preferability to attribute values’ step in the WLC method has notable impact on the final outcome for areas of potential habitat. For this study, the preference scale was assembled using averages of the RFF suitability scores for attribute ranges, as at this time, there are few if any records of expert opinions regarding the preferences of the modeled species. Although it is common for ecological researchers to narrow down which environmental factors influence species occurrence or presence distributions, there is little literature further subdividing the individual variables into classes of preferability. One method frequently used by modelers of the GIS community is a modified Delphi Technique in combination with the Analytical Hierarchy Process (AHP). However, due to time and resource constraints, as well as the immediate objectives of this study, scale rank by panel survey was not feasible. However, if a small scale and targeted species approach was used, it would be possible to collect this information. Here, the first steps have been taken to attempt to categorize ‘within variable’ conditions that can dictate the suitability of an area for an individual species.
Conclusion
The landscape of Colorado is heterogeneous, supporting species representing several floristic elements and sub-elements, including the Chihuahuan biogeographic sub-element (Weber, 1965). By
46


spatially analyzing the species comprising this sub-element, it was possible to (1) identify their ecological niches and (2) locate areas within Colorado that could serve as suitable habitat. This study has contributed to a better understanding of the influence of the Chihuahuan sub-element on the flora of Colorado. It also offers an opportunity: to consider multi-disciplinary methodology that incorporates intra-variable attributes to identify relationships between ecological niches and geographic distributions.
47


Figures
Fig. 2.1. Results of a habitat suitability analysis for Palafoxia sphacelata revealing (a) suitable habitat
in the southwestern United States and Mexico and (b) Colorado suitability contrasted with (c) results
from a sensitivity analysis.
48


a
Fig. 2.2. Results of a habitat suitability analysis for Muhlenbergict arenicola reveal( a) suitable habitat in the southwestern United States and Mexico and (b) Colorado suitability contrasted with (c) results from a sensitivity analysis.
49


a
a Funastrum crispum
Suitability Index Score
--------T High : 7.90345
Low: 2.05517
b
c
Fig. 2.3. Results of a habitat suitability analysis for Funastrum crispum reveal (a) suitable habitat in
the southwestern United States and Mexico and (b) Colorado suitability contrasted with (c) results
from a sensitivity analysis.
50


Fig. 2.4. Results of a habitat suitability analysis for Nolina texcma reveal (a) suitable habitat in the southwestern United States and Mexico and (b) Colorado suitability contrasted with (c) results from a sensitivity analysis.
51


a
Fig. 2.5 Results of a habitat suitability analysis for Muhlenbergict ctrenacect revealing (a) suitable
habitat in the southwestern United States and Mexico and (b) Colorado suitability 'contrasted with (c)
results from a sensitivity analysis.
52


Figure 2. 6 Comparison of outputs for (top) Palafoxia sphacelata, and (bottom) Mnhlenbergict cirenaceci) from both WLC and MaxEnt methods.
53


REFERENCES
Adams. G. et al. 1904. Gypsum Deposits in the United States. United States Geological Survey.
Aguirre-Gutierrez, J., Carvalheiro, U. G., Polce, C., Uoon, E. E., Raes, N., Reemer, M., &
Biesmeijer, J. C. 2013. Fit-for-purpose: Species distribution model performance depends on evaluation criteria - Dutch Hoverflies as a case study. PLoS ONE, 8(5). doi: 10.1371/joumal.pone.0063708
Alvarez, L. J., H. E. Epstein, J. Li, and G. S. Okin. 2011. Spatial patterns of grasses and shrubs in an arid grassland environment. Ecosphere 2(9): 103. doi: 10.1890/ES11-00104.1
Baldwin B.G. 2014. Origins of plant diversity in the California Floristic Province. The Annual Review of Ecology, Evolutions, and Systematics. 45: 347-369. Doi: 10.1146/annuren-ecolsys-110512-135847.
Barbet-Massin, M., Jiguet, F., Albert, C. and Thuiller, W. 2012. Selecting pseudo-absences for species distribution models: how, where and how many?. Methods in Ecology and Evolution, 3(2), pp.327-338.
Brown, D.E. 1982. “Chihuahuan Desertscrub.” Desert Plants, vol. 4, arizona. openrepository.com/handle/1015 0/5 50974.
Brown J. 2014. SDMtoolbox: a Python-Based GIS Toolkit for Landscape Genetic, Biogeographic and Species Distribution Model Analyses. Methods in Ecology and Evolution.5:694-700.
Caffrey M, Doemer J. 2012. A 7000 year record of environmental change, Bear Lake, Rocky Mountain National Park, USA. Physical Geography. 33(5): 438-456.
Cain, S. A., 1944, Foundations of plant geography: New York, NY, Harper Bros.
Chapman, S.S., Griffith, G.E., Omemik, J.M., Price, A.B., Freeouf, J., and Schrupp, D.L. 2006. Ecoregions of Colorado (color poster with map, descriptive text, summary tables, and photographs): Reston, Virginia, U.S. Geological Survey (map scale 1:1,200,000).
Collin, R., & De Maintenon, M. 2002. Integrative approaches to biogeography: Patterns and processes on land and in the sea. Integrative and Comparative Biology, 42(5), 911-912. doi: 10.1093/icb/42.5.911
Cruz-Cardenas, G., Lopez-Mata, L., Villasenor, J. L., & Ortiz, E. 2014. Potential species distribution modeling and the use of principal component analysis as predictor variables. Revista Mexicana De Biodiversidad, 85(1), 189-199. doi: 10.7550/rmb.36723
Dark, S. J., & Bram, D. 2007. The modifiable areal unit problem (MAUP) in physical geography. Progress in Physical Geography, 31(5), 471-479. doi: 10.1177/0309133307083294
54


Dilts, T.E. (2015) Generate Random Points Around Points Tool for ArcGIS 10.3.1. University of Nevada Reno. http://www.arcgis.com/home/item.html?id=e43b0fde28914bl2b36c9725467c0189
Donoghue J.2013. A Maxent Script Tool for ArcGIS.
Drobne, S. & Lisec A. 2009. Multi-attribute decision analysis in GIS: Weighted linear combination and Ordered weighted averaging. Informatica, 33, 459-474.
Elith, J., Phillips, S. J., Hastie, T., Dudik, M., Chee, Y. E., & Yates, C. J. (2010). A statistical explanation of MaxEnt for ecologists. Diversity and Distributions, 17(1), 43-57. doi: 10.1111/j. 1472-4642.2010.00725.x
Elith, J., H. Graham, C., P. Anderson, R., Dudik, M., Ferrier, S., Guisan, A., J. Hijmans, R., Huettmann, F., R. Leathwick, J., Lehmann, A., Li, J., G. Lohmann, L., A. Loiselle, B., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., McC. M. Overton, J., Townsend Peterson, A., J. Phillips, S., Richardson, K., Scachetti-Pereira, R., E. Schapire, R., Soberon, J., Williams, S., S. Wisz, M. and E. Zimmermann, N. 2006. Novel methods improve prediction of species’ distributions from occurrence data. Ecography, 29(2),129-151.
Ebach, M. (2015). The multidisciplinary nature of biogeography. Australian Systematic Botany, 28(3), 79.
Encyclopedia Britannica. (2018). Pteridaceae | plant family, [online] Available at: https://www.britannica.com/plant/Pteridaceae [Accessed 14 Mar. 2018],
Fick, S.E. & R.J. Hijmans, 2017. Worldclim 2: New 1-km spatial resolution climate surfaces for global land areas. International Journal of Climatology.
Forester, B. R. (2012) "Integrating ensemble species distribution modeling and statistical phylogeography to inform projections of climate change impacts on species distributions" WWU Masters Thesis Collection. 202.http://cedar.wwu.edu/wwuet/202
Fourcade Y., Engler J.O., Rodder D., and Secondi J .2014. Mapping species distributions with MAXENT using a geographically biased sample of presence data: A performance assessment of methods for correcting sampling bias. PLoS ONE 9(5): e97122. https: //doi. org /10.13 71 /j oumal .pone .0097122
Guisan, A., Edwards, T. C., & Hastie, T. 2002. Generalized linear and generalized additive models in studies of species distributions: Setting the scene. Ecological Modelling, 157(2-3), 89-100. doi: 10.1016/s03 04-3 800(02)00204-1
GBIF.org (29th February 2016) GBIF Occurrence Download https://doi.org/10.15468/dl.ywhpmz
Graham A. 2011. The age and diversification of terrestrial New World ecosystems through Cretaceous and Cenozoictime. American Journal of Botany. 98: 336-351.
55


Graham M.R., Jaeger J.R., Prendini L., and Riddle B.R. 2013. Phylogeography of Beck’s Desert Scorpion, Paruroctonus becki, reveals Pliocene diversification in the Eastern California Shear Zone and post- glacial expansion in the Great Basin Desert. Molecular Phylogenetics and Evolution 69: 502-513.
Henriques-Silva R., Lindo Z., Peres-Neto P.R. 2013. A community of metacommunities: exploring patterns in species distributions across large geographical areas. Ecology, Ecological Society of America. 94(3):627-639.
Hijmans R, Elith J. 2016.Species Distribution Modeling. Species Distribution Modeling — R Spatial. Geospatial and Farming Systems Research Consortium (GFC).rspatial.org/sdm/index.html.
Hirzel A.H., Le Lay G.2008. Review: Habitat suitability modelling and niche theory. Journal of Applied Ecology. 45(5): 1372-1381. Web of Science, http://www.jstor.org/stable/2014410516.
Janzen T, Haegeman B., Etienne R.S. 2015.A sampling formula for ecological communities with multiple dispersal syndromes. Journal of Theoretical Biology. 387:258-261.
Laport, Robert G., et al. “Ecological Niche Modeling Implicates Climatic Adaptation, Competitive Exclusion, and Niche Conservatism among Larrea Tridentata Cytotypes in North American desertsl,2.” The Journal of the Torrey Botanical Society, vol. 140, no. 3, 2013, pp. 349-363., doi: 10.315 9/torrey-d-13-00009.1
Lele, S. R., Merrill, E. H., Keim, J., & Boyce, M. S. (2013). Selection, use, choice and occupancy: Clarifying concepts in resource selection studies. Journal of Animal Ecology, 82(6), 1183-1191. doi: 10.1111/1365-2656.12141
Loehle, C. 2012. Relative frequency function models for species distribution modeling. Ecography, 35(6), 487-498. doi:10.1111/j.l600-0587.2012.07389.x
Malczewski, J. 2000. On the Use of Weighted Linear Combination Method in GIS: Common and Best Practice Approaches. Transactions in GIS, 4(1), 5-22. doi: 10.1111/1467-9671.00035
Marlowe, K., Hufford, L. (2007). Taxonomy and biogeography of Gaillardia (Asteraceae): A phylogenetic analysis. Systematic Botany, 32(1), 208-226. doi: 10.1600/036364407780360229
McLaughlin, S., 2007. Tundra to Tropics: The floristic plant geography of North America. Fort Worth Texas. Botanical institute of Texas.
Mclaughlin, S. 1994. Floristic plant geography: The classification of floristic areas and floristic elements. Progress in Physical Geography, 18(2), 185-208. doi: 10.1177/030913339401800202.
McLaughlin S. 1986. Floristic analysis of the Southwestern United States. Great Basin Naturalist. scholarsarchive.byu.edu/gbn/vol46/iss 1/5/.
Moore, M. J., J. F. Mota, N. A. Douglas, H. Flores Olvera, and H. Ochoterena. 2014. The ecology, assembly, and evolution of gypsophile floras. Pp. 97-128 in N. Rajakaruna, R. Boyd, and T. Harris
56


(eds.), Plant Ecology and Evolution in Harsh Environments. Nova Science Publishers, Hauppauge,
NY.
Muldavin E.H. 2002. Some floristic characteristics of the northern Chihuahuan Desert: A search for its northern boundary. Taxon. 51(3):453-462. International Association for Plant Taxonomy. Web of Science, doi: 10.2307/1554858.
Neiswenter S.A., B. R. Riddle. 2010. Diversification of the Perognathus flavus species group in emerging arid grasslands of western North America. Journal of Mammalogy. 91: 348-362.
Olivier R.D.S., Aranda R. 2017. Potential geographic distribution niche modeling based on bioclimatic variables of three species of Temnomastax Rehn and Rehn, 1942 (Orthoptera: Eumastacidae). Journal of Natural History. 51(21-22): 1197-1208. Web of Science, doi: 10.1080/00222933.2017.1324647
Olson, D.M., E. Dinerstein, E.D. Wikramanayake, N.D. Burgess, G.V.N. Powell, E.C. Underwood, J.A. DAmico, I. Itoua, H.E. Strand, J.C. Morrison, C.J. Loucks, T.F. Allnutt, T.H. Ricketts, Y. Kura, J.F. Lamoreux, W.W. Wettengel, P. Hedao, and K.R. Kassem. Terrestrial Ecoregions of the World:
A New Map of Life on Earth (PDF, 1.1M) BioScience 51:933-938
Omemik, J.M., Griffith G.E. 2014. Ecoregions of the conterminous United States: evolution of a hierarchical spatial framework. - PuhMed - NCBI. [online] Ncbi.nlm.nih.gov. Available at: https://www.ncbi.nlm.nih.gov/pubmed/25223620
PassalacquaN.G. 2015. On the definition of element, chorotype and component in biogeography. Journal of Biogeography, onlinelibrary.wiley.com/doi/10.1111/jbi. 12473.
Peterson, A. T. 2001. Predicting species geographic distributions based on ecological niche modeling. The Condor, 103(3), 599. doi:10.1650/0010-5422(2001)103[0599:psgdbo]2.0.co;2
Peterson, A.T., & Soberon, J. 2012. Species distribution modeling and ecological niche modeling: Getting the concepts right. Natureza & Conscrvacao. 10(2), 102-107. doi: 10.4322/natcon.2012.019
Phillips S.J., Dudik M., Schapire R.E. Maxent software for modeling species niches and distributions (Version 3.4.1). http://biodiversitvinformatics.amnh,org/open source/maxent/.
Phillips, S.J., Anderson, R.P, & Schapire, R.E. 2006. Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190(3-4), 231-259. doi:10.1016/j.ecolmodel.2005.03.026
RajakarunaN. 2004: The edaphic factor in the origin of plant species. International Geology Review, 46:5, 471-478
Richerson P.J, Lum K.L. 1980. Patterns of plant species diversity in California: relation to weather and topography. Am. Nat. 116:504-36
57


Rondeau, R., Pearson, K. and Kelso, S. 2013. Vegetation response in a Colorado grassland-shrub
community to extreme drought: 1999-2010. The American Midland Naturalist, 170(1), pp.14-25.
Ryberg WA, Fitzgerald L.A. 2016. Landscape composition, not connectivity, determines metacommunity structure across multiple scales. Ecography (Copenhagen).39(10):932-941. nri.tamu.edu/publications/peer-reviewed-17 publications/2016/landscape-composition-notconnectivity-determines-metacommunitystructure-across-multiple-scales/.
Villarreal-Quintanilla JA, Bartolome-Hemandex JA, Estrada-Castillon E,
Ramirex-Rodriguez H, Martinez-Amador SJ. 2017. The endemic element of the Chihuahuan Desert vascular flora. Acta Botanica Mexicana 118: 65-96. Print.
Villasenor, J. 1990. The genera of Asteraceae endemic to Mexico and adjacent regions. Aliso, 12(4), 685-692. doi: 10.5642/aliso.19901204.04
Wainwright, J, Mulligan M. 2013. Chapter 2 Modelling and Model Building. Environmental Modelling: Finding Simplicity in Complexity. Wiley-Blackwell. 2:7-24.
Warren, DL., Glor, RE., & Turelli, M. 2008. Environmental niche equivalency versus conservatism: Quantitative approaches to niche evolution. Evolution, 62(11), 2868-2883. doi: 10.1111/j.1558-5646.2008.00482.x
Weber WA. 1965. Plant geography in the Southern Rocky Mountains, in the Quaternary of the United States: Princeton, N.J., Princeton Univ. Press, p. 453- 468.
Weber WA, Wittmann RC. 2001. Colorado Flora: Western Slope, 3rd ed. University Press of Colorado, Niwot.
Wood DA, Vandergast AG, Barr KR, Inman DR, Esque CT, Nussear KE, Fisher NR. 2013. Comparative phylogeography reveals deep lineages and regional evolutionary hotspots in the Mojave and Sonoran Deserts. Diversity and Distributions. 19(7):722-737. doi: 10.1111/ddi. 12022
Woodward FI. 1986. Climate and Plant Distribution. Cambridge University Press.
Yackulic, Charles B., et al. “Presence-Only Modelling Using MAXENT: When Can We Trust the Inferences?” Journal of Applied Ecology, Wiley/Blackwell (10.1111), 21 Nov. 2012, besjoumals.onlinelibrary.wiley.com/doi/full/10. Ill l/2041-210x. 12004.
Young, N., Carter, L. and Evangelista, P. (2018). A MaxEnt Model v3.3.3e Tutorial (ArcGIS vlO). http://ibis.colostate.edu/webcontent/ws/coloradoview/tutorialsdownloads/a_maxent_model_v7.pdf [Accessed 14 Mar. 2018],
58


APPENDIX
Table A. 1 Original Bioclim dataset source and properties.
File Name Variable Source Original Resolution/ Projection Notes
Biol Annual Mean Temperature WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_2 Mean Diurnal Range (Mean of monthly(max temp-min temp) WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_3 Isothermality (Bio2/Bio7)(*100) WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_4 Temperature Seasonality (stand, dev *100) WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_5 Max Temperature of Wannest Month WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_6 Min Temperature of Coldest Month WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_7 Temperature Annual Range (Bio5-Bio6) WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_8 Mean Temperature of Wettest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_9 Mean Temperature of Driest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
BiolO Mean Temperature of Wannest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Biol 1 Mean Temperature of Coldest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_12 Annual Precipitation WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_13 Precipitation of Wettest Month WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_14 Precipitation of Driest Month WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_15 Precipitation Seasonality) Coefficie nt of Variation) WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous USA Albers
59


File Name Variable Source Original Resolution/ Projection Notes
Bio_16 Precipitation of Wettest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_17 Precipitation of Driest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_18 Precipitation of Wannest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ ContiguousUS AAlbers
Bio_19 Precipitation of Coldest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous USA Albers
Table A. 2 Sources and properties of boundary polygons.
File Name Layer Source Original Resolution/Projection Notes
Co_eco_14 EPA Ecoregions Enviromnental Protection Agency North American 1983, USA Contiguous Albers Equal Area Conic USGS
Terrestrial Ecoregions of the World Chihuahuan Desert World Wildlife Fund/Data Basin GCSWGS1984
USA States Generalized Southwest U.S. State Boundary Esri, TomTom, US Census Bureau, US Dept, of Coimnerce GCSWGS 1984 Shapefile created from selection of CO, NM, NV, TX, OK, CA, UT
State Boundaries of Mexico Mexico Esri GCSWGS1984 Merged with US Southwest shapefile
COcounties CO Counties Esri, US Census Bureau GCSWGS1984
60


Table A. 3 Variables, attribute preference scores, and associated factor weights used in the WLC
calculations for Muhlenbergia arenicolct.
Layer Attributes Muhlenbergia arenicola Preference Score Factor Weight
Bio 11 (-9) to (-2) 2 0.331183
(-1.9) to (-1.0) 3
0 to 5.9 6
6 to 10.9 5
11 to 12.9 7
13 to 20.9 1
Bio7 17 to 24.9 4 0.22509
25 to 29.9 6
30 to 46.9 5
Bio6 (-19) to (-11.0) 3 0.331183
(-10.0) to (-1) 5
0 to 4.9 6
5 to 5.9 2
6 to 13.9 1
Bio4 150 to 353.9 6 0.009319
354 to 404.9 7
405 to 506.9 4
507 to 1011 5
Bio3 31 to 41.9 3 0.103226
42 to 61.9 1
62 to 65.9 4
66 to 75.9 6
61


Table A. 4 Variables, attribute preference scores, and associated factor weights used in the WLC
calculations for Funastrum crispum.
Layer Attributes Funastrum crispum Preference Score Factor Weight
Bio 19 19 to 62.9 4 0.034482759
63 to 73.9 2
74 to 84.9 3
85 to 128.9 6
129 to 139.9 5
140 to 261.9 7
Bio 18 11 to 214.9 4 0.027586207
215 to 265.9 7
266 to 629.9 6
Bio 16 26 to 76.9 2 0.055172414
77 to 127.9 3
128 to 280.9 5
281 to 331.9 8
332 to 678.9 6
Bio 12 63 to 163.9 1 0.017241379
164 to 264.9 2
265 to 770.0 5
771 to 1073 6
Bio 2 9 to 11.9 8 0.865517241
12 to 16.9 6
17 to 19.9 4
20 to 22.9 2
62


Table A. 5 Variables, attribute preference scores, and associated factor weights used in the WLC
calculations for Palafoxia sphacelata.
Layer Attributes Palafoxia sphacelata Preference Score Factor Weight
Bio 16 93 to 147 7 0.058122
148 to 193 4
194 to 373 3
Bio 15 17 to 49.9 1 0.14158
50 to 71.9 5
72 to 82.9 7
83 to 115.9 4
Bio 12 100 to 199 8 0.010432
200 to 299 7
300 to 599 4
600 to 699 2
700 to 1075 1
Bio3 32 to 35.9 2 0.438152
36 to 38.9 6
39 to 47.9 5
48 to 49.9 4
50 to 57.9 3
Bio 2 10 to 13.9 1 0.351714
14 to 14.9 3
15 to 15.9 4
16 to 18.9 5
19 to 22.9 6
63


Table A. 6 Variables, attribute preference scores, and associated factor weights used in the WLC
calculations for Nolina texana.
Layer Attributes Nolina texana Preference Score
Bio 19 10 to 29.9 5
30 to 51.9 4
52 to 62.9 3
63 to 84.9 2
85 to 128.9 7
129 to 175.9 9
Bio 15 31 to 41.9 10
42 to 52.9 3
53 to 63.9 1
64 to 74.9 5
75 to 85.9 6
86 to 118.9 4
Bio 2 10 to 12.9 9
13 to 15.9 6
16 to 16.9 5
17 to 18.9 3
19 to 22.9 2
Factor Weight 0.021994135
0.065982405
0.91202346
64


Table A. 7 Variables, attribute preference scores, and associated factor weights used in the WLC
calculations for Muhlenbergia arenacea.
Layer Attributes Muhlenbergia arenacea Preference Score Factor Weight
Bio 18 79 to 144.9 5 0.41014402
145 to 166.9 6
167 to 299.9 4
Bio 16 80 to 202.9 5 0.293675642
203 to 226.9 6
227 to 334 4
Bio 13 2 to 7.9 7 0.010644959
8 to 13.9 5
14 to 40.9 4
Bio 12 192 to 237.9 3 0.247964934
238 to 344.9 5
345 to 395.9 6
396 to 708 4
Bio 10 9 to 18.9 3 0.026299311
19 to 20.9 6
21 to 21.9 7
22 to 29.9 4
Bio 8 9 to 16.9 3 0.011271133
17 to 21.9 6
22 to 28.9 4
65


Table A. 8 EPA Ecoregion Level III, IV, and plant habit type for each species identified.
SCIENTIFIC NAME FAMILY LEVEL III KEY LEVEL IV KEY PLANT HABIT
Andropogon saccharoides Sir. Poaceae 25 High Plains 26 Southwestern Tablelands 25d Flat to Rolling Plains 26f Mesa de Maya/Black Mesa Gramminoid
Argemone squarrosa Papaveraceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Forb
Asclepias macrotis Torr. Apocynaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands 26g Purgatoire Hills and Canyons Subshrub
Berlandiera lyrata Asteraceae 25 High Plains 26 Southwestern Tablelands 25c Moderate Relief Plains 25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa Forb
Bouteloua barbata Lag. Poaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Gramminoid
Bouteloua eriopoda (Torr.) Torr. Poaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands 26g Purgatoire Hills and Canyons Gramminoid
Cheilanthes eatonii Pteridaceae 21 Southern Rockies 25 High Plains 26 Southwestern Tablelands 2Id Foothill Shrublands 25c Moderate Relief Plains 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons 25c Moderate Relief Plains 26f Mesa de Maya/Black Mesa Forb
Cheilanthes wootonii Klaxon Pteridaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands 26g Purgatoire Hills and Canyons Forb
Cucurbita foetidissima Cucurbitaceae 20 Colorado Plateaus 21 Southern Rockies 25 High Plains 26 Southwestern Tablelands 20c Semiarid Benchlands and Canyonlands 21c Crystalline Mid-Elevation Forests 2Id Foothill Shrublands 2 If Sedimentary Mid-Elevation Forests 25b Rolling Sand Plains Forb
25c Moderate Relief Plains
66


SCIENTIFIC FAMILY LEVEL III KEY LEVEL IV KEY PLANT HABIT
NAME
25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons 26h Pinyon-Juniper Woodlands and Savannas 26j Foothill Grasslands 26k Sandsheets
Cyperus Cyperaceae 21 Southern 21b Crystalline Subalpine Gramminoid
fendlerianus Rockies Forests
26 Southwestern 21c Crystalline Mid-
Tablelands Elevation Forests 2Id Foothill Shrublands 2 If Sedimentary Mid-Elevation Forests 26j Foothill Grasslands
Dalea lanata var. Fabaceae 26 Southwestern 26g Purgatoire Hills and Forb
lanata Tablelands Canyons
Desmanthus Fabaceae 25 High Plains 25b Rolling Sand Plains Subshrub
cooleyi (Eaton) Branner & Coville
Erigeron Asteraceae 20 Colorado 20c Semiarid Benchlands Forb
colomexicanus Plateaus and Canyonlands
21 Southern 21c Crystalline Mid-
Rockies Elevation Forests
22 Arizona/New Mexico Plateau 2Id Foothill Shrublands
25 High Plains 21e Sedimentary Subalpine Forests 2 If Sedimentary Mid-Elevation Forests 22b San Luis Alluvial Flats and Wetlands 25d Flat to Rolling Plains 251 Front Range Fans 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons 26j Foothill Grasslands
Funastrum crispum Apocynaceae 21 Southern 2Id Foothill Shrublands Forb
Rockies 26 Southwestern 2Id Foothill Shrublands
Tablelands
2Id Foothill Shrublands
67


SCIENTIFIC
NAME
FAMILY
LEVEL III KEY
LEVEL IV KEY
PLANT HABIT
26f Mesa de Maya/Black Mesa
26f Mesa de Maya/Black Mesa
Krameria lanceolata Krameriaceae 25 High Plains 26 Southwestern Tablelands 25c Moderate Relief Plains 25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa Forb
Mimosa borealis A. Gray Fabaceae 25 High Plains 25c Moderate Relief Plains Shrub
Muhlenbergia porteri Scribn. Poaceae 26 Southwestern Tablelands 26g Purgatoire Hills and Canyons Gramminoid
Muhlenbergia repens (J.Presl) Hitchc. Poaceae 21 Southern Rockies 21c Crystalline Mid-Elevation Forests Gramminoid
Nolina texana Asparagaceae 25 High Plains 26 Southwestern Tablelands 25d Flat to Rolling Plains 26f Mesa de Maya/Black Mesa Gramminoid
Notholaena standleyi Pteridaceae 25 High Plains 26 Southwestern Tablelands 25d Flat to Rolling Plains 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons Forb
Physaria fendleri Brassicaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands 26g Purgatoire Hills and Canyons 26h Pinyon-Juniper Woodlands and Savannas Forb
Schoenocrambe Brassicaceae 20 Colorado 20a Monticello-Cortez Forb
linearifolia Plateaus 21 Southern Rockies 22 Arizona/New Mexico Plateau 26 Southwestern Tablelands Uplands 20c Semiarid Benchlands and Canyonlands 2Id Foothill Shrublands 2 If Sedimentary Mid-Elevation Forests Subshrub
21h Volcanic Mid-Elevation Forests 22a San Luis Shrublands and Hills
26e Piedmont Plains and Tablelands 26h Pinyon-Juniper Woodlands and Savannas
68


SCIENTIFIC NAME FAMILY LEVEL III KEY LEVEL IV KEY PLANT HABIT
Sporobolus nealleyi Vasey Poaceae 20 Colorado Plateaus 20c Semiarid Benchlands and Canyonlands Gramminoid
Thymophylla aurea Asteraceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands 26g Purgatoire Hills and Canyons Forb
Palafoxia sphacelata Asteraceae 21 Southern Rockies 25 High Plains 26 Southwestern Tablelands 21c Crystalline Mid-Elevation Forests 25b Rolling Sand Plains 25c Moderate Relief Plains 25d Flat to Rolling Plains 251 Front Range Fans 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26k Sandsheets Forb
Muhlenbergia arenacea Poaceae 21 Southern Rockies 26 Southwestern Tablelands 2Id Foothill Shrublands 26e Piedmont Plains and Tablelands 26g Purgatoire Hills and Canyons 26h Pinyon-Juniper Woodlands and Savannas Gramminoid
Muhlenbergia arenicola Poaceae 25 High Plains 26 Southwestern Tablelands 25b Rolling Sand Plains 25c Moderate Relief Plains 25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons 26k Sandsheets Gramminoid
Croton texensis Euphorbiaceae 25 High Plains 26 Southwestern Tablelands 25b Rolling Sand Plains 25c Moderate Relief Plains 25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands Forb
Cylindropuntia imbricata (Haw.) F.M.Knuth Cactaceae 20 Colorado Plateaus 21 Southern Rockies 26 Southwestern Tablelands 20c Semiarid Benchlands and Canyonlands 21c Crystalline Mid-Elevation Forests 2Id Foothill Shrublands Shrub
21h Volcanic Mid-Elevation Forests
69


SCIENTIFIC
NAME
FAMILY
LEVEL III KEY
LEVEL IV KEY
PLANT HABIT
26e Piedmont Plains and Tablelands 26g Purgatoire Hills and Canyons 26j Foothill Grasslands
Dalea aurea Fabaceae 25 High Plains 25c Moderate Relief Forb/Herb
C.Fraser Plains
Dalea aurea Pursh Fabaceae 26 Southwestern 25c Moderate Relief Plains 26e Piedmont Plains and Forb/Herb
Tablelands Tablelands 26j Foothill Grasslands
Dalea nana Torr. Fabaceae 25 High Plains 25b Rolling Sand Plains Forb/Herb
ex A. Gray Gaillardia Asteraceae 21 Southern 21c Crystalline Mid- Forb/Herb
pulchella Foug. Rockies Elevation Forests
22 Arizona/New Mexico Plateau 2Id Foothill Shrublands
25 High Plains 2 If Sedimentary Mid-Elevation Forests
26 Southwestern 21h Volcanic Mid-
Tablelands Elevation Forests 22c Salt Flats 25c Moderate Relief Plains 25d Flat to Rolling Plains 251 Front Range Fans
26 Southwestern 26e Piedmont Plains and
Tablelands Tablelands 26i Pine-Oak Woodlands 26j Foothill Grasslands
Juniperus Cupressaceae 20 Colorado 20b Shale Deserts and Tree
monosperma Plateaus Sedimentary Basins
(Engelm.) Sarg. 21 Southern 20c Semiarid Benchlands
Rockies and Canyonlands
26 Southwestern 20e Escarpments
Tablelands 2Id Foothill Shrublands 2 If Sedimentary Mid-Elevation Forests 26e Piedmont Plains and Tablelands
Lesquerella Brassicaceae 26 Southwestern 26e Piedmont Plains and Forb/Herb
fendleri (A. Gray) Tablelands Tablelands
S.Watson 26g Purgatoire Hills and Canyons
70


236
Family
Figure A. 1 Chart comparing the number of occurrences by taxonomic family.
71


350
300
aj
u 250 aj
3
U
u
O
200
150
295



i: LI

5 â–  2
di
.a
£
3
100
50
â„¢ 2
o E
-a a)
ro -s=
1— +-»
O 3
° 5
O
(N
(N
£
JTO CL CL -C
o .5P
<
CM
ID
CM
Ecoregion Level
Figure A. 2 Chart of comparison of number of occurrences per Ecoregion Level III.
Apocynaceae EcoRegion Representations
â–  21 Southern Rockies 21d Foothill Shrublands
â–  26 Southwestern Tablelands 26e Piedmont Plains and Tablelands
â–  26 Southwestern Tablelands 26f Mesa de Maya/Black Mesa
â–  26 Southwestern Tablelands 26g Purgatoire Hills and Canyons
Figure A. 3 Proportion of Apocynaceae per Ecoregion Level III.
72


Asparagaceae Ecoregion Representations
â–  25 High Plains â–  26 Southwestern Tablelands
Figure A. 4 Proportion of Asoaragaceae per Ecoregion Level III.
73


Asteraceae Ecoregion Representations
3.26086956522 .2.17391304348 2.17391304348 ___ | /
15.21739130430
11.95652173910
16.30434782610
gig
10.86956521740
_9.78260869565
21.73913043480
2.17391304348 1.08695652174
1.08695652174
.2.17391304348
â–  20 Colorado Plateaus 20c Semiarid Benchlands and Canyonlands Asteraceae
â–  21 Southern Rockies 21c Crystalline Mid-Elevation Forests Asteraceae
â–  21 Southern Rockies 21d Foothill Shrublands Asteraceae
â–  21 Southern Rockies 21e Sedimentary Subalpine Forests Asteraceae
â–  21 Southern Rockies 21f Sedimentary Mid-Elevation Forests Asteraceae
Figure A. 5 Proportion of Asteraceae per Ecoregion Level III.
74


Brassicaceae
8.33333333333
â–  26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Brassicaceae
â–  26 Southwestern Tablelands 26g Purgatoire Hills and Canyons Brassicaceae
â–  26 Southwestern Tablelands 26h Pinyon-Juniper Woodlands and Savannas Brassicaceae
Figure A. 6 Proportion of Brassicaceae per Ecoregion Level III.
75


â–  20 Colorado Plateaus 20c Semiarid Benchlands and Canyonlands Cucurbitaceae
â–  21 Southern Rockies 21c Crystalline Mid-Elevation Forests Cucurbitaceae
â–  21 Southern Rockies 21d Foothill Shrublands Cucurbitaceae
â–  21 Southern Rockies 21f Sedimentary Mid-Elevation Forests Cucurbitaceae
â–  25 High Plains 25b Rolling Sand Plains Cucurbitaceae
â–  25 High Plains 25c Moderate Relief Plains Cucurbitaceae
â–  25 High Plains 25d Flat to Rolling Plains Cucurbitaceae
â–  26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Cucurbitaceae
â–  26 Southwestern Tablelands 26f Mesa de Maya/Black Mesa Cucurbitaceae
Figure A. 7 Proportion of Cucurbitaceae per Ecoregion Level III.
2.54237288136
76


Cyperaceae Ecoregions
6.66666666667
26.66666666670
6.66666666667.
6.66666666667

53.33333333330
â–  21 Southern Rockies 21b Crystalline Subalpine Forests Cyperaceae
â–  21 Southern Rockies 21c Crystalline Mid-Elevation Forests Cyperaceae
â–  21 Southern Rockies 21d Foothill Shrublands Cyperaceae
â–  21 Southern Rockies 21f Sedimentary Mid-Elevation Forests Cyperaceae
â–  26 Southwestern Tablelands 26j Foothill Grasslands Cyperaceae
Figure A. 8 Proportion of Cyperaceae per Ecoregion Level III.
Fabaceae
â–  25 High Plains 25b Rolling Sand Plains Fabaceae
â–  25 High Plains 25c Moderate Relief Plains Fabaceae
â–  26 Southwestern Tablelands 26g Purgatoire Hills and Canyons Fabaceae
Figure A. 9 Proportion of Fabaceae per Ecoregion Level III.
77


Krameriaceae Ecoregion Representations
â–  25 High Plains 25c Moderate Relief Plains Krameriaceae
â–  25 High Plains 25d Flat to Rolling Plains Krameriaceae
â–  26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Krameriaceae 26 Southwestern Tablelands 26f Mesa de Maya/Black Mesa Krameriaceae
Figure A. 10 Proportion of Krameriaceaee per Ecoregion Level III.
Poaceae Ecoregion Representations
7.69230769231
â–  20 Colorado Plateaus 20c Semiarid Benchlands and Canyonlands Poaceae
â–  21 Southern Rockies 21c Crystalline Mid-Elevation Forests Poaceae
â–  25 High Plains 25d Flat to Rolling Plains Poaceae
â–  26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Poaceae
â–  26 Southwestern Tablelands 26f Mesa de Maya/Black Mesa Poaceae
â–  26 Southwestern Tablelands 26g Purgatoire Hills and Canyons Poaceae
Figure A. 11 Proportion of Poaceae per Ecoregion Level III.
78


Pteridaceae
5.76923076923
â–  21 Southern Rockies 21d Foothill Shrublands Pteridaceae
â–  25 High Plains 25c Moderate Relief Plains Pteridaceae
â–  25 High Plains 25d Flat to Rolling Plains Pteridaceae
â–  26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Pteridaceae
â–  26 Southwestern Tablelands 26f Mesa de Maya/Black Mesa Pteridaceae
â–  26 Southwestern Tablelands 26g Purgatoire Hills and Canyons Pteridaceae
Figure A. 12 Proportion of Pteridaceae per Ecoregion Level III.
79


Figure A. 13 RFF outputs of co-variates of Muhlebnergict arenicolct.
80


Figure A. 14 RFF output for variables associated with Palafoxia sphacelata.
81


Figure A. 15 RFF outputs for co-variates associated with Funastnim crispum.
82


BI02 BI03
BI015 BI017
BI019
Figure A. 16 RFF outputs for co-variates associated with Nolina texcma.
83


Figure A. 17 RFF outputs for co-variates associated with Muhlenbergia arenacect.
84


Table A. 9 List of species that support Weber's original list identified as plants of the Chihuahuan sub
element.
Species Identified by Weber Identified through GIS
Abutilon inccmum (Link) Sweet Andropogron saccharoides Sw. X
Argemone squarrosa Greene X
Ascelpias macrotis Torr. X
Asclepias oenotheroides Cham. & Schlecht. Asplenium resiliens Kunze Berlcmdiera lyrata Benth. X
Bouteloua barbata Lag. X
Bouteloua. eriopoda (Torr) Torr. X
Cheilcmthes eatonii Baker X
C. wootonii Maxon X
Croton texensis (Klotsch) Muell. Arg. X
Cucurbita foetidissima H.B.K X
Dalea lanata Spreng. X
Dalea nan a Torr. X
Desmanthus cooleyi (Eaton) Trek Engehnanniapinnatifida T.& G. Eragrostis oxylepis (Torr.) Torr. X
Eriochloa contracta Hitch.
Gaillardiapulchella Foug. X
Hoffinanseggia densiflora Benth. X
Hoffinanseggia drepanocarpa A. Gray X
Hoffinanseggia jamesii T&G. X
Juniperus mono sperm a (Engelm.) Sarg. X
Krameria spp. X
Melampodium cinereum DC.
Mimosa borealis A. Gray X
Palafoxia spp. X
Pericome caudata A. Gray X
Sapindus saponaria L.
Sarcostemma crispurn Benth X
Stillingia sylvatica Gard.
Thelesperma spp.
85


Figure A. 18 Flowchart documenting GIS methods to build models for habitat suitability models.
86


Figure A. 18 contd.
87


Environmental Variable Environmental Variable
Figure A. 18 contd.
Jackknife of regularized training gain for Muhlenbergiaarenacea
bio10 bio12 bio13 bio16 bio18 bio8
Without variable â–  With only variable â–  With all variables â– 
0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5
regularized training gain
bio11 b i o 3 bio4 b i o 6 b i o 7
Jackknife of regularized training gain for Muhlenbergiaarenicola
Without variable â–  With only variable â–  With all variables â– 
0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90
regularized training gain
88


Environmental Variable Environmental Variable Environmental Variable
Jackknife of regularized training gain for Palafoxiasphacelata
regularized training gain
Without variable â–  With only variable â–  With all variables â– 
Jackknife of regularized training gain for Nolina texana
bio12 bio1 4 bio1 5 bio17 bio1 9 bio2 b i o 3
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4
regularized training gain
Without variable With only variable With all variables
Figure A. 19 MaxEnt jackknifed training gains for modeled species.
89


Full Text

PAGE 1

INFLUENCE OF THE CHIHUAHUAN BIOGEOGRAPHIC SUB ELEMENT ON THE FLORA OF COLORAD O: EVIDENCE FROM GIS by GENEVIEVE MARISSA BARRON B.S. , Fort Lewis College, 2012 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Ma ster of Integrated Sciences Integrated Sciences Program 2018

PAGE 2

! ! ii ! This thesis for the Master of Integrated Science degree by Genevieve Marissa Barron h as been approved for the Integrated Science Program b y Leo P. Bruederle , Chair Raphael Moreno Sanchez Melissa Islam Date: July 28 , 2018

PAGE 3

! ! iii ! Barron, Genevieve Marissa (M.I.S. , Integrated Sciences) Influence of the Chihuahuan Biogeographic Sub Element on the Flora of Colorado: Evidence from GIS Thesis directed by Associate Professor Leo P. Bruederle ABSTRACT Biogeographic patterns are a driving factor in the distribution of plant species and the amounts of suitable habitat available to them. Weber (1965) stated that southeastern Colorado is dominated by a distinct distribution of taxa that are representative of the Chihuahuan Biogeographic sub element. The supposition by Weber leads to the question; is there evidence that the Chihuahuan sub e lement is represented in Colorado flora? If so, what are the environ mental limitations for those species and which areas of Colorado are likely to provide suitable habitat? To answer these questions an interdisciplinary methodology that combined Geographic Information System technology with biological niche modelling best practices, was used to address the following objectives: compile a comprehensive list of species occurring in Colorado that represent the Chihuahuan sub element, thoroughly document the ecological niches that those species occupy within the state, and ba sed on niche characteristics spatially model the areas of Colorado that are likely to provide suitable habitat. Results of a systematic comparison of occurrence records from online herbarium databases showed 4 3 taxa occurring in Colorado strongly exhib it an overall Chihuahuan distribution. Five of the taxa, whose habits represented the most commonly occurring habit types, were chosen for a habitat suitability analysis. A strong trend towards suitable habitat occupying mostly the eastern portion of t he state was observed, with some areas of medium to high suitability concentrated in the southeast and southwest corners. This form and content of this abstract are approved. I recommend its publication. Approved: Leo P. Bruederle

PAGE 4

! ! iv ! ACKNOWLEDGMENTS I would like to acknowledge Dr. Rafael Moreno Sanchez for serving as a thesis committee member and technical consult for the GIS methodology. I also acknowledge Dr. Melissa Islam for serving as a thesis committee member and providing support for locating additional reference materials and offering additional viewpoints on biologically meaningful modelling concepts. I will also a cknowledge Michelle Deprenger Levin for guidance on parallels between ArcGIS and MaxEnt Modeling for Species Distributions and as well as advice on best practices for selecting variables for testing. Fin ally, I would like to acknowledge Dr. Leo Bruederle for all the work and help that he has provided for th is thesis.

PAGE 5

! ! v ! TABLE OF CONTENTS CHAPTER I. INTRODUCTION ................................ ................................ ................................ ........................... 9 Research Objectives ................................ ................................ ................................ ..................... 13 Methods and Materials ................................ ................................ ................................ ..................... 14 Results ................................ ................................ ................................ ................................ .............. 15 Discussion ................................ ................................ ................................ ................................ ........ 17 Taxonomy ................................ ................................ ................................ ................................ ..... 18 Geography ................................ ................................ ................................ ................................ .... 18 Ecology of the taxa ................................ ................................ ................................ ....................... 19 GIS Approach ................................ ................................ ................................ ............................... 20 Limitations ................................ ................................ ................................ ................................ .... 20 Conclusion ................................ ................................ ................................ ................................ .... 21 Tables and Figures ................................ ................................ ................................ ............................ 22 II . NICHE MODELING ................................ ................................ ................................ ....................... 28 Relevan t Approaches to Niche Modeling ................................ ................................ ......................... 28 Relative Frequency ................................ ................................ ................................ ....................... 29 Generalized Linear Models ................................ ................................ ................................ .......... 30 Maxent for Presence Only Dat ................................ ................................ ................................ ..... 30 Geographic Information Systems ................................ ................................ ................................ . 31 Habitat Suitability Analyses Consideratio ns ................................ ................................ ................ 32 Research Objectives ................................ ................................ ................................ ..................... 33 Methods and Materials ................................ ................................ ................................ ..................... 34 Bioclim Da ta Processing ................................ ................................ ................................ .............. 34 Statistical Analyses ................................ ................................ ................................ ....................... 35 Habitat Suitability Analysis ................................ ................................ ................................ .......... 35 Results ................................ ................................ ................................ ................................ .............. 37 Palafoxia sphacelata (Asteraceae) ................................ ................................ ............................... 37 Muhlenbergia arenicola (Poaceae) ................................ ................................ .............................. 38 Funastrum crispum (Apocynaceae) ................................ ................................ ............................. 39 Nolina texana (Asparagaceae) ................................ ................................ ................................ ...... 40 Muhlenbergia arenacea (Poaceae) ................................ ................................ ............................... 41 Maxent Comparison ................................ ................................ ................................ ..................... 42 Discussion ................................ ................................ ................................ ................................ ........ 42 Abiotic and Biotic Pattern s ................................ ................................ ................................ ........... 43

PAGE 6

! ! vi ! Geographic Patterns ................................ ................................ ................................ ...................... 43 Sensitivity Analysis ................................ ................................ ................................ ...................... 44 Method Compa rison ................................ ................................ ................................ ..................... 44 Limitations of the Study ................................ ................................ ................................ ............... 45 Conclusion ................................ ................................ ................................ ................................ .... 46 Figures ................................ ................................ ................................ ................................ .............. 48 REFERENCES ................................ ................................ ................................ ................................ . 54 APPENDIX ................................ ................................ ................................ ................................ ...... 59

PAGE 7

! ! vii ! LIST OF TABLES TABLE 1.1 Chihuahuan species found to reach Colorado and associated EPA Level III ecoregion classification. ................................ ................................ ................................ ................................ 22 !

PAGE 8

! ! viii ! LIST OF FIGURES FIGURE 1. . . ...17 2. 1 Results of a habitat suitability analysis for Palafoxia sp hacelata . ................................ ................ 49 ! 2. 2 Results of a habitat suitability analysis for Muhle nbergia arenicola ................................ ............. 50 ! 2. 3 Results of a habitat suitability analysis for Funastrum crispum . ................................ .................... 51 ! 2. 4 Results of a habitat suitability analysis for Nolina texana . ................................ ........................... 52 ! 2.5 Comparison of outputs for Palafoxia sphacelata , and Muhlenbergia arenacea ... .. !

PAGE 9

! ! 9 ! CHAPTER I INTRODUCTION Pa tterns of s pecies distributions , especially those that appear to be highly restricted, have been a central focus for scientists in the environmental sciences for at least the past century ( Rajakaruna , 2004) . Research that has delved into finding the drivin g forces behind those patterns have identified climate and soil type as the primary influencers of plant distributions (Rajakaruna , 2004; Cain , 1944) . Broad topics in and of themselves, the efforts to understand the mechanisms that link climate and edaphic factors to plant distributions are still ongoing (Wood , 1986 ). To better understand these complex ecosystem processes, researchers have cross ed disciplin ary boundaries to adopt and incorporate methods and concepts that focus on biogeographic patterns of species distributions (Ebach , 2015). Biogeographic patterns are shaped both by the physical characteristics of an ecosystem, as well as by the biological traits of ecosystem inhabitants (Collin and De Matinen , 2002 ). Abiotic factors , namely climatic weat her conditions , have long been prominently linked to the distribution and range expansion of major clades across North America. Marlow and Hufford (2007) found that the aridification of northern Mexico and the western extent of the United States greatly in fluenced the geographic radiation of r epresentatives of the Asteracea e , s pecifically Gaillardia , into the Rocky Mountain region. The biological characteristics of species inhabiting a geographic range, such as the propensity to adapt to changes to the envi ronment, are therefore also driving forces in the biogeographic patterns o f a region. Other a biotic factors , such as land bridges that connect separate biotas or mountain ranges that act as dispersal barriers between ecosystems, strongly influence biogeogr aphic patterns , as well (Collin and D e Matinen , 2002 ). In the context of a floristic analysis, c lassifying populations based on biogeographic patterns can serve to help document specie s current distribution s , range of expansion s and conserved niche s . To

PAGE 10

! ! 10 ! that end, t has traditionally been applied to those groups of species that occur in a previously defined biogeographic area and have similar distributions (Passalacqua , 2015). Weber (1965) defined an element as a group of species that occ upy a coincident area, and further proposed that such an element could be divided into sub element s . His examination of the southwestern US concluded with recognition of three broad elements: Circumpolar element, Madro Tertiary element, and Eastern Woodlan d Prairie element, each with respective sub elements. These elements are not simply species occurrences but reflect the geographic environmental conditions on the landscape. S imilarly, McLaughlin (2007) attempted to isolate floristic biogeograph ic elements across North America a geographically large scale using Principal Component Analysis of 245 local floras of Mexico, the United States, and Canada. He designated 27 floristic sub provinces that represented discrete areas with characteristic physiography, climate, vegetation, and plant and animal life (McLaughlin , species had nonrandom, overlapping distribution patterns , but did not consistently occur in simil ar ecosystem communities (McLaughlin , 1986). His analysis of the southwestern portion of the United States , including the northern boundary of Mexico, identified twelve floristic elements: The Great Basin Element, Colorado Plateau Element, Chihuahuan Eleme nt, Sonoran Element, Apachian Element, Southern Rocky Mountain Mogollon Element , Sierra Nevada, Rocky Mountain, Peninsular, California, Vancouverian and Columbian Plateau Elements ( McLaughlin , 1989) . Hereafter, th e use of the term While there may be some variation in determining the boundaries of floristic elements , as they pertain to state boundaries (e.g., Colorado) , there is agreement on the importance of a biogeographic features on its floristic communities. Colorado is characterized by great topographic and climatic variation (Lambert and Reid , 1981). Topographic features like plateaus, basins, the s outhern portion of the Rocky Mountain range, and plains make up the la ndscape that define the , 2006). The diverse v egetative communities

PAGE 11

! ! 11 ! that have resulted from these geographic variations are home to a variety of sensitive and , in some cases rare species. In order to facilitate conservation and management practices for these species, attempts have been made to classify and partition land management systems based on plant communities, soil types, and other ecologically meaningful boundaries. Traditionall y, the composition of an ecological community has been explained by species specific traits and the individual requirements for a species (Janzen et al. , 2015). However, as the role of land managers evolves, a more robust understanding is required. As such , the Environmental Protection Agency incorporated data on vegetation, climate, soils, land use, wildlife, hydrology, physiography and geology into their land classification systems, resulting in the recognition of ecoregions (Chapman et al. , 2006) . Within Colorado, the EPA designated six Level III Ecoregions: Wyoming Basin, Colorado Plateaus, Southern Rockies, Arizona/New Mexico Plateau, High Plains, and Southwestern Tablelands (Chapm a n et al. , 2006 ). Those Ecoregions are further subdivided into Level IV Ecoregions , t he characteristics of which have been widely documented ( Omernik and Griffith , 201 4 ) . However, the relationship between the arid regions of Colorado, and potential migration of taxa from other regions beyond the Ecoregion classi fication zones ha s not yet been thoroughly investigated . Weber ( 1965 ) proposed that the flora of Colorado has been influenced, in part, by a sub element of the Madro Tertiary Element the Chihuahuan sub e lement as evidenced by the presence of 33 speci es that are predominantly Chihuahuan with occurrences in Colorado (Table A. 9) . Yet to date, little research has been conducted providing evidence supporting th is observation. This may be due, in part, to the fact that the generally recognized borders of the Chihuahuan Desert do not extend into the state of Colorado. M ost spatial representations of the Chihuahuan Desert and the corresponding Chihuahuan sub element depict the northern boundary reaching only as far north as the state of New Mexico. When one considers the physical boundaries that lie between the northern exten t of the sub element, and Colorado, it is not immediately obvious which, if any species of that element , might be present in the state.

PAGE 12

! ! 12 ! Administrative and political boundaries have provided a generally recognized perimeter boundary for the Chihuahuan D esert . However, the boundaries of this desert, as with most ecoregions, are not entirely defined or agreed upon from the ecological viewpoint ( Villareal et al . , 2017, Brown , 1982 , as cited in Schmidt , 1979 , Morafka , 1977 ) . Yet, McLaughlin (2007) identified the desert area as an independent floristic su b element . The interior of the desert is located between the Sierra Madre Occidental to the West, the Sierra Madre Oriental to the east, and the Trans Mexican Volcanic Belt at the southern end (McLaughlin , 2007). The northern extent of the desert, which intersects the Sonoran and Chihuahuan biogeographic elements (McLaughlin , 2007), is bounded by the Rio Conchos and the Deming Plains, which act as the natural geographic boundaries (Fig . 1. 1) . Although the Chihuahuan Desert is highly variable, both w ith respect to elevation and temperature ranges, its climate is characterized by summer rainfall, which provides 70 80% of the annual precipitation (Muldavin , 2002). The climate includes cold and dry winters, as well as hot summers ( Villarreal l et al. , 2 017). It is the second most floristically diverse desert in the world ( Villarreal l et al ., 2017) as many as 3382 plant species have been documented from th is d esert ( Villarreal l et al. , 2017; Henrickson and Johnston , 2007). Approximately 67 plant famil ies contain Chihuahuan endemics, with a total of 826 taxa ( Villarreal l et al. , 2017). The areas of greatest richness of endemism are Coahuila, Texas, and Chihuahua , Mexico . Villarreal l et al. (2017) found that the richness of a study area was not determine d by the territorial proportion which they occupied within the Chihuahuan Desert. The Chihuahuan floristic element has affinities with the floras of the Great Plains biogeographic element (McLaughlin , 1986) Since Weber (1965), few autho rs have considered the influence of the Chihuahuan b iogeographic sub e lement on the flora of Colorado. However, region of Colorado acknowledged element by notin g the presence of several of its species present in the southeastern portion o f the state. M odels documenting the pot ential habitat of individual species from this sub element within Colorado do not exist . However, research conducted at the species level attempt s to classify individual species that represent

PAGE 13

! ! 13 ! this sub element with regard to their biogeographic patterns. Marlow and Hufford's (2007) analysis of New World Gaillardia do cumented the intermountain expansion of Gaillardia pulchella Foug . from its Chihuahuan origins using phylogeographic methods . At the opposite end of the spectrum, landscape level research has sought to combine biosystematics with biogeography to determine the specific criteria and conditions that dictate large 2 ) analysis of the floristic characteristics of the northern portion of the Chihuahuan Desert focused on the liminal i ntersection between the Chihuahuan Desert and Colorado Plateau as expressed by grassland communities. Analysis of gramminoids with qualities that indicated processes of desertification revealed that areas along the Chihuahuan Plateau that had lower elevati on, provided definitive desert conditions. The findings of this study revealed that those species that were determined to be of the Chihuahuan geographic classification were one of the most restricted groups, with scrub species appearing more Chihuahuan in nature than grassland species. Finally, extensive research addressing th os e factors driving ecological diversity in the Chihuahuan Desert have been compiled over the last few decades. In terms of geologic history, the research comes from its neighbors, the Mojave and Sonoran Deserts. Wood et al . (2013) examined evolutionary hot spots for genetic divergence and diversity in the Mojave, Colorado and Sonoran Deserts USA, recognizing that two major historical processes heavily shaped genetic structure in the southwestern deserts: Pleistocene climate fluctuations and pre Pleistocene vicariance . Research Objectives Here I critically evaluate the influence of the Chihuahuan b iogeographical sub e lement on the flora of Colorado using GIS. My immediate objective was to (1965) regarding the contribution of the Chihuahuan b iogeographic sub e lement by compil ing a list of those plant species t hat are predominantly Chihuahuan , but also reach Colorado. Ultimately, I sought to

PAGE 14

! ! 14 ! better understand this phenomenon, by exploring the taxonomy and ecology of those species identified by Weber (1965) and herein . Methods and Materials The mere occurrence o f a species in Colorado , outside of its centralized distribution , for example from the Chi huahuan sub element , does not necessarily constitu t e the infl uence of the respec ti ve floris ti c element on that region. According to Hirzel and Le Lay (2008), a mor e causal link can only be determined if t he data meet three conditions: 1) a species must occupy mostly distinct areas ; 2) niches must partially overlap ; and 3) t he species must occur in sympatry in some areas . Further, to accurately assess if a species truly occupies an area, additional concepts must be considered. Typically, species presence or absence from a region is determined by three constraints. First, the environmental conditions of a localized area must allow the population to grow. Second, the interactions between the target species and others must allow the target species to persist. Last, the location is accessible in relation to the abilities of the species to disperse (Hirzel and Le Lay , 2008). T o ascertain if (1) the Ch ihuahuan sub element could be documented in Colorado using occurrence datasets from the Southwestern Environmental Information Network (SEINet , 2018 ) and the Global Biodiversity Information Facility (GBIF , 2016 ) and (2) the conditions proposed b y Hirzel and Le Lay (2008) were met , a preliminar y study was conducted . S pecies met the first c ondition of occupying a mostly distinc t area if they displayed a mostly Chihuahuan Desert range distribution. Individual species distribution maps, for the species in Colorado, revealed that , in fact , some Chihuahuan species had high concentrations of occurrences in both the Chihuahuan Desert , as well as occurrences within the state of Colorado. This fulfilled the next two condition s of having partially overlapping niches and occurring in sympatry, and thus made the data eligible to be experimentally analyzed. The methodology applied to the p reliminary research, which was completed in the spring of 2017, indicated that species from this grouping met the conditions identified by Hirzel and le Lay (2008)

PAGE 15

! ! 15 ! and was applied to 2326 individual taxa, grouped by family, of plants from the SEINet list of Flora of the Chihuahuan Desert Network . The initial sea rch for taxa was performed by visually assessing distribution maps of individual taxa on the GBIF database website. If the species appeared in the Chihuahuan Desert region, but also had a broad presence globally, that species was determined to not fit the however, a species appeared to have a distribution heavily centered in the Chihuahuan Desert region, it was retained for further analysis. After the GBIF determination was made, species were checked in SEINet. If the SEINet distributions showed a similar pattern to that of the GBIF occurrences , the species was kept for the next phase of assessment . Again , if the distributions on SEINet indicated a mo re global presence, the species were assumed to not be definitively Chihuahuan and were removed from further consideration. The next step involved using Geographic Information Systems (GIS) , specific ally the ArcMap were convert ed to vector shape files and over laid with polygon representations of the southwestern US and the Chihuahuan Desert . From there it was possible to det ermine whether the Chihuahuan species that had been separated out actually had documented occurrences in both Colorado and the Chihuahuan Desert. To specifically classify the habitats of species that exhibit the Chihuahuan sub element, within Colorado, Ec oRegion data Levels III and IV were downloaded from the EPA ecoregion GIS database , compared with known presence points, and analyzed in ArcMap. The tabulated intersections matched presence points with the names of the EcoRegions III and IV in which they w ere contained , and the percentage of points within each ecoregion was calculated . ! R esults A total of 4 3 Chihuahuan species representing 1 4 famil i es was found to occur within Colorado (Table 1). They are: Asclepias macrotis Torr . ( Apocynaceae) , Fun astrum crispum (Benth.) Schltr.

PAGE 16

! ! 16 ! (Apocyn aceae ) , Berlandiera lyrata Benth. (Asteraceae) , Erigeron colomexicanus A.Nelson (Asteraceae), Gaillardia pulchellla Foug. var. pulchella , Palafoxia sphacelate (Nutt. ex Torr.) Cory, Thymophylla aurea (A. Gray) Gre ene ex Britton, (Asteraceae) , Pericome caudata A. Gray (Asteraceae) , Nolina texana S.Watson (Asparagaceae) , Lesquerella fendleri (A. Gray) S. Watson (Brassicaceae) , Physaria fendleri (A. Gray) O'Kane & Al Shehbaz (Brassicaceae) , Schoenocrambe linearifo lia (A. Gray) Rollins (Brassicaceae) , Arida parviflora (A.Gray), D.R.Morgan & R.L.Hartm . (Asteraceae) , Cylindropuntia imbricat a (Haw.) F.M. Knuth (Cactaceae) , Cucurbita foetidissima H.B.K. (Cucurbitaceae) , Juniperus monosperma (Engelm.) Sarg. (Cupressa ceae) , Cyperus fendlerianus Boeckeler (Cyperaceae) , Cyperus sphaerolepis Boeckeler (Cyperaceae) , Croton texensis (Klotsch) Muell. Arg. (Euphorbiaceae) , Dalea aurea C. Fraser , Dalea lanata Spreng. , Dalea nana Torr. , Desmanthus cooleyi (Eaton.) Torr. , H offmannseggia drepanocarpa A. Gray (Fabaceae) , Hoffmannseggia jamesii Torr. & A.Gray (Fabaceae ) , Mimosa borealis A. Gray (Fabaceae) , Krameria lanceolata Torr. (Krameriaceae) , Argemone squarrosa Greene (Papaveraceae) , Andropogon saccharoidea S w. , Bouteloua barbata lag. (Poaceae) , Bouteloua eriopoda (Torr.) Torr. (Poaceae) , Eriochloa contracta Hitchc. , Muhlenbergia arenacea (Buckl.) A.S. Hitchc. ( Poaceae) , Muhlenbergia arenicola Buckl. (Poaceae) , Muhlenbergia fragilis Swallen (Poaceae) , Muhlenbergia porter i Scribn. (Poaceae) , Muhlenbergia repens ( J. Presl) Hitchc. ( Poaceae) , Muhlenbergia tenuifolia (Willd.) Britton, Sterns, & Poggenb. (Poaceae) , Sporobolus contractus A.S. Hitchc. (Poaceae) , Sporobolus nealleyi Vasey (Poaceae) , Ch eilanthes eatonii Baker (Pteridaceae) , Cheilanthes wootonii Maxon (Pteridaceae) , Notholaena standleyi (KŸmmerle) Maxon (Pteridaceae). The three families that had the greatest species diversity were Poaceae , with 13 species; Fabaceae, with eight species; an d Asteraceae , with six species . Within Poaceae, the genus Muhlenbergia was particularly well represented , accounting for six species. Within Fabaceae, the dominant genera were Dalea and Hoffsmanseggia . The Cucurbitaceae family accounted for the greatest nu mber of occurrences , but was only represented by a single species , Cucurb i ta foetidissim a

PAGE 17

! ! 17 ! H.B.K . T wo families that also had high numbers of occ urrences were Asteraceae and Pteridaceae. Graminoids appeared to be the most common growth form, followed by forb s. Chihuahuan s pecies were present in five of out the 6 EPA Level III ecoregions, with the Wyoming Basin being the only ecoregion where species of the sub element did not appear to be represented. The Colorado Plateau and the Arizona/New Mexico Plateau ecoregions had the lowest number of occurrences, both having fewer than ten records. In increasing order, the ecoregions that contained the highest number of occurrences were the Southern Rockies, High Plains, and Southwestern Tablelands. O ve r all, the southern half of the state contained the most occurrence records . While there are large clusters of occurrences in the southwestern por t ion of the state, the majority of occurrences are located east of the Rio Grande National Forest . Discussion My findings are consistent with Weber (1965), who concluded that the Chihuahuan sub element has , in fact, influenced the flora of Colorado. Of the roughly 33 taxa that he identified as belonging to the Chihuahuan sub element ( that occur in Colorado), I found support for 2 3 ( see Appendix Table A.9) . M y findings do not support the following species identified by Weber (1965): Abu tilon inca n um (Link) Sweet , Asclepias oenotheroides Cham. & Schlecht . , ! Asplenium r esiliens Kunze , Sapindus saponaria L . , Stillingia sylvatica Gard, Engelmannia pinnatifida T.& G. and Thelesperma spp . These species appear to have other affinities with neighboring floristic elements and are not strongly Chihuahuan. For example, A. incanu m , although present in both the Chihuahua Desert and Colorado, also had significant occurrences in the Sonoran, Peninsular and Coloradan floristic regions. I did not find evidence that Melampodium cinere um DC, A. Gray was of the Chihuahua n sub element and present in Colorado. In both SEINet and GBIF, there was only one record of the species occurring in Colorado, and the coordinates were the location of the Denver Botanic Gardens. Additionally, the record was flagged for incorrect dates. T hus, it did not appear to be a reliable record and was excluded from analysis.

PAGE 18

! ! 18 ! T wo species identified by Weber ( E r agrostis oxylepis (Torr.) Torr. , Eriochloa contracta Hitchc. ) were excluded in the early steps of analyses due to zero occurrence records in Colorado. Taxonomy T rends were observed with species diversity and numbers of occurrences. The three families that appeared to be the best represented by a diverse number of Chihuahuan species were Poaceae, Fabaceae, and Asteraceae. In contrast to the species diversity, neither Poaceae nor Fabaceae were represented with high numbers of occurrences. ! The Cucurbitaceae f amily was the most well represent ed by number of occurrences. Coincidentally, the sole representative of Cucurbitaceae, Cucurbita foetidissima, was one of the original tax on that Weber identified as belonging to the Chihuahuan sub element. Not surprisingly, Asteraceae, the largest flowe ring plant family, was the second greatest contributor of occurrences in this study. It also had the third greatest number of species , being represented by s even taxa. This ceae contribute substantially to the floristic richness of Mexico, but that Mexico may have played an important role in the diversification of the family, possibly as a secondary center of diversification ! Finally, the family that was the third best represe nted through occurrence records was Pteridaceae , specifically Cheilanthes eatonii and Cheilanthes wootonii representing the Cheilanthoid clade . Both species have adaptations to drought stress and occur in seasonally dry environments (Hosh , 2008), which i s consistent with expectations for plants of the Chihuahuan sub element . Geography R species included species with occurrences on the Western Slope , whereas Weber separated genera that occurred west of the Rockies into the Sonoran Great Basin elements. This may be due to differences in exclusion methods along defined floristic regions. This study did not remove a species from consideration if there were occurrences within a separate neighboring region. Rather the species was excluded if it appeared that the number of occurrences in the adjacent region was greater or equal

PAGE 19

! ! 19 ! to the distribution in the study area. Nonetheless, by using the capabilities of GIS platform s to quickly create spatial representations of recorded occurrence points, I was also able to visualize that, as Weber noted, the number of species that appear to belong to this distribution pattern increase southward across the state , as do the occurrence s of i n dividual species . Ecology of the taxa The species that represented the Chihuahuan sub element most strongly in Colorado appeared to be contained primarily s and the High Pl ains ecoregion s . These ecoregions are characterized by a variety of Quaternary sand substrates , sandy alluvi al deposits , shales, and sandstones (Chapman et al. , 2006). This is of note, as studies that have investigated the edaphic controls over floristic a ssemblies have identified a strong connection between gypsum substrates and certain endemic plants of the Chihuahuan Desert and Mexican high plateau (Moore et al . , 2014). For instance, clades of Argemone , as well as Gaillardia ( Marlowe & Hufford, 2007; Moo re et al. , 2014 ) , have been documented to display gypso philic affinities in their distributions (Schwarzbach and Kadereit , 1999; Moore et al. , 2014 ). Gypsum, although an uncommon constituent of sand, is present in sandy deposits through out Colorado ( Ad ams et al. , 1904 ) and th is soil distribution connection appears to be maintained as evidenced by th is recognition of Aregmone squarrosa and Gaillardia pulchella . Graminoids and forbs appear ed to be the most common plant habit type that exhibit ed th e Chihuahuan sub element in Colorado. This makes sense when one considers the ecoregions that th ese species occur in . Combined, the Southwestern Tablelands and the High Plains ecoregions are composed of predominantly prairie, grassland, and pine woodland v egetation communities, with few riparian areas (Chapman et al. , 200 6 ) . Interestingly, my results paralleled the findings of Clark (1996), who se survey of the Mesa de Maya region of Colorado also described a Chihuahuan element present in the flora of southeastern Colorado. Clark reported occurrences of Notholaena standleyi , Nolina texana , and Cheilanthes eatonii , which I also identified as strongly Chihuahuan.

PAGE 20

! ! 20 ! Although shrubs and subshrubs are reported , graminoids are most well represented b y species from the Chihuahuan sub element , at least in Colorado. This conclusion differs slightly from previous work (Muldavin , 2002) , which suggested that scrub and shrubland species more clearly indicate a Chihuahuan element rather than grassla nd species . One theory that may account for this dissimilarity is experienc ing invasion by shrub s resulting in increased shrub lands (Alvarez et al. , 2011). Whereas in Colorado, a ltho ugh shrub land s occasionally occur on the eastern side of the state, they are more prominent on the western edge. However, the interaction s between shrub and grass species that result in the habitat replacement is still not well understoo d . GIS Approach Currently, two methods are frequently used to assign species to communities , as well as phylogenetic and physiographic entities. First is the m ethod of conducting thorough comparisons between floras that have agreement in species distributions . Essentially, this method involves visually inspecting and sorting range maps of taxa (McLaughlin , 1994). Second, researchers may us e 'genetic methods' to analyze lineages of species and trace them back to a common origin (Cornuet et al. , 1999). The latter method serves as a n advanced option for detailed analysis. The method used i n this study is most akin to the comparison of regional or loc al floras, albeit with the some what novel integration of GIS. The validity is evidenced with results consistent with the findings of Weber (1965) and Clark ( 1996 ). ! Limitations Altho ugh my research was conducted with the intention of utilizing unbiased and accurate data extracted from h erbarium occurrence records , there are known limitations and a reas that can be identified as potential sources of error. T he species occurrence datasets, which the analysis relies upon , potentially introduced geographic bias as they were derived from both herbarium records and potentially opportunistic observations instead of planned survey s that would have been the highest standard (Fourcade et al. , 2014). There was inconsistent editing and assumed geographic coordinate

PAGE 21

! ! 21 ! systems assigned within the records. As a result, entries included misspelled scientific names, non accepted scientific names, and incomplete entries. To address these issues, SQL queries were used to exclude erroneous records, as well as records geo referenced to academic institutions and other clearly commercial collection locations (i.e. , Denver Botanic Gardens) . Addit ionally, the methodology was primarily driven by spatial data . It did not take into c onsideration the ge notypic or evolutionary concepts of individual species . Conclusion It's likely that by incorporating known biologic or genotypic traits into the methodology of this thesis ( after spatial r epresentations are analyzed ) , the accuracy of results would be greatly increased. This type of information could facilitate a more a ccurate understand ing of whether current species distributions reflect only realized niches, or if th ey reflect the expansio n and adaptations of a species as well . Future research should focus on phylogeography to inform patterns revealed herein . Additionally, attention should be focused on those species within Colorado that occur along the boundaries of defined floristic elem ents .

PAGE 22

! ! ! 22 Tables and Figures Table 1 .1 Chihuahuan spe cies found to reach Colorado, with associated EPA Level III ecoregion classification . Table Family Species Colorado Plateau Southern Rockies Arizona New M exico Plateau Wyoming Basin High Plains Southwestern Tablelands Apocynaceae Ascpelias macrotis Torr. x Funastrum crispum (Benth.) Schltr. x x Asteraceae Arida parviflora (A. Gray) D.R.Morgan and R.L. Hartm. x x x Berlandiera lyrata Benth. x x Erigeron colomexicanus A. Nelson x x x x Gaillardia pulchella Foug. var. p ulchella x x x x Thymophylla aurea (A. Gray) Green ex Britton x Palafoxia sphacelata (Nutt. ex Torr) Cory x x x Pericome caudata A . Gray x

PAGE 23

! ! ! 23 Table Family Species Colorado Plateau Southern Rockies Arizona New M exico Plateau Wyoming Basin High Plains Southwestern Tablelands Asparagaceae Nolina texana S. Watson x x Brassicaceae Lesquerella fendleri (A. Gray) S. Watson x Schoenocrambe linearifolia (A. Gray)Rollins (A. Gray) Rollins x x x x Physaria fendleri (A. Gr ay) O'Kane & Al Shehbaz x Cactaceae Cylindropuntia imbricata (Haw.) F.M. Knuth x x x Cucurbitaceae Cucurbita foetidissuma H.B.K. H.B.K . x x x x Cupressaceae Juniperus monosperma (Engelm.) Sarg. x x x Cyperaceae Cyperus fendlerianus Boeckeler Boeckeler x x Cy p erus sphaerolepis Boeckeler x Euphorbiaceae Croton texensis (Klotsch)Muell.Arg (Klotsch)Muell.Arg . x x

PAGE 24

! ! ! 24 Table Family Species Colorado Plateau Southern Rockies Arizona New M exico Plateau Wyoming Basin High Plains Southwestern Tablelands Fabaceae Dalea aurea C. Fraser C. Fraser x Dalea lanata Spreng. Spreng. x Dalea nana Torr. Torr. Ex A. Gray Desmanthus co o leyi ( Eaton) Branner & Colville x Hoffmannseggia drepanocar pa A. Gray x x Hoffmansegia jamesii Torr. & A. Gray x Mimosa borealis A. Gray A.Gray x Kramericeae Krameria lanceolata Torr. Torr. x x Papaveraceae Aregmone squarrosa Greene Greene x Poaceae Andropogon saccharoides Sw. x x

PAGE 25

! ! ! 25 Table Family Species Colorado Plateau Southern Rockies Arizona New M exico Plateau Wyoming Basin High Plains Southwestern Tablelands Bouteloua barbata Lag. x Bouteloua eriopoda ( Torr.) Torr. Eriochloa contracta Hitchc. x Muhlenbergia arenacea (Buckl .) A.S. Hitchc . x x Muhlenbergia arenicola Buckl. x x Muhlenbergia fragilis Swallen x Muhlenbergia porter i Scribn. x Muhlenbergia repens ( J. Presl) Hitchc. x Muhlenbergia tenuifolia (Willd.)Britton, Sterns, & Poggenb. x Sporobolus contractus A.S. Hitchc. x x x

PAGE 26

! ! ! 26 Table Family Species Colorado Plateau Southern Rockies Arizona New M exico Plateau Wyoming Basin High Plains Southwestern Tablelands Sporobolus nealleyi Vasey x Pteridaceae Cheilanthes eatonii Baker x x x Cheilanthes wootoni i Maxon x x Notholaena standleyi (KŸmmerle) Maxon x x x !

PAGE 27

! 27 ! ! Figure 1 Reference map of southwest US deserts; th e Mojave Desert, the Sonoran Desert, and the Chihuahuan Desert.

PAGE 28

! ! 28 ! CHAPTER II NICHE MODELING Relevant Approaches to Niche Modeling The geographic distribution of a species is fundamentally link (Peterson and Soberon , 2012) . It follows then, that ecological niche models (ENMs) are a key resource in understanding complex relationships between a species distribution and its environment. Niche models are more frequently being used to investigate questions regarding speciation, niche evolution and cladistic ecological diversity ( Warren et al. , 2008 ). Niche models have also proven highly useful in studies of areas of endemism, and the de limitations of biogeographic units (Hausdorf, 2002), such as floristic elements. Niche modeling is based on four facets of the niche concept: 1) niche characteristics, 2) niche interactions, 3) community wide processes, and 4) niche evolution (Hirzel an d Le Lay , 2008). Although this terminology is used across concentration areas, the literature reveals that the term niche e environmental requirements needed for a species t o subsist , between one species to others (Peterson and Soberon , 2012). These two concepts require separate and distinct tools to conduct investigations. For the purposes of my work, niche model ing ref as the term applie s to environmental conditions that a species needs to populate an area. Each of the studies mentioned in Chapter 1 , as well as many others that focus on species distribution patterns, involved the use of prediction models based on associations with ecological niches (Hijmans and Elith , 2016). One such model that is frequently used is the habitat suitability model (HSM), which attempts to relate environmental variables to the likelihood of occurrence of the species (Hirzel a nd Le Lay , 2008). Habitat suitability models are based heavily on niche characteristics. Also referred to as Species Distribution Models (SDM) , the aim of these types of mo dels is to estimate the similarity of the

PAGE 29

! ! 29 ! conditions at any site to the conditions at the locations of known occurrence s of a phenomenon (Hijmans and Elith , 2016) . The steps usually taken are follow s : (1) locations of species occurrence are compiled; (2) values of environmental predictor variables (i.e. , climate) at these loc ations are collected from spatial databases; (3) those values are used to fit a model to estimate similarity to the sites of occurrence, or abundance of the species; (4) t hat model is used to predict the variable of interest across the focus region (Hijman s and Elith , 2016). The literature also points to this type of modeling being used to generate predictions of possible occurrences for areas of endemics (Olivier and Aranda , 2017). Although there has long been debate about the terms ENM, SDM, and HSM, in the broad context, these seemingly different terms refer to effectively the same type of analyses (Peterson and Soberon , 2012). These p redictive models aim to forecast the likelihood of a species occurrences based on environmental variables. Simult aneously, these models are often used as applications to predict outcomes such as the presence/absence of a species in a given area, or the abundance of species throughout a study area (Hirzel and Le Lay , 2008). However, some researchers contend that the r elationship suggested by habitat suitability models is weak. Therefore, the strength of a habitat suitability model that is focused on the distribution niche link is dependent upon the specific ecology of a single species, local physical constraints of the study area, and accuracy of records of historical events. (Hirzel and Le Lay , 2008). There are a variety of different model algorithms available to modelers in the environmental sciences . T he most frequently used methods present results in terms of proba bilities . Relative Frequency Frequency data is one option used to describe the distribution of species in a community . Loehle (2012) introduced and developed a novel method for use by the National Council for Air and Stream Improvement , which compared t he relative frequency of species occurrence points to that of random points (pseudo absence) to compute a frequency ratio for habitat distribution models. Specifically, the

PAGE 30

! ! 30 ! Relative Frequency Function tool was developed to model predicted preferred habit at, by calculating the relative probability (frequency) of finding plant or animal species at given points within a study area. This function is viewed as an extension of the Resource Selection Probability Function (Loeh l e 2012 , Manley et al ., 2002), which is used in cases of data where only used and available points are known and can be compared to discrete resources (Lele et al. , 2013 ) . Generalized Linear Models The generalized linear model, or GLM, is a traditional model frequently used by biologi sts before the advent of Max e nt. It is mathematically comparable to a multiple regression analysis and ha s been used extensively in SDMs because of its strong statistical foundation in modeling ecological relationships (Elith et al. , 2006). GLMs h ave been widely used in studies where binary data ar e used (e.g., wildlife biology) . Results are interpreted by looking at the regression coefficients of each variable, which can be useful for explanatory SDMs. The advantage of this type of model is th at it is an extension of a linear model, so it allows for non linear relationships in the data and , thus , does not force non linear data into non natural scales (Guisan et al. , 2002 ). MaxEnt for Presence Only Dat a The statistical modeling software Maxent ( Phillips et al. , 2006 ) is currently considered one of the most accurate and reliable options for biological researchers investigating ecological processes (Elith et al. , 2010). One of it s main performance highlights is its ability to accurately model pote ntial species For example, t his software has been used to build niche models using the evolutionary history of desert species to understand population expansions (Graham et al . , 2013). It has also been used to predict geographic distribution s of a variety of plant species for conservation purposes (Phillips et al . , 2005). The program is widely used because of ease of use especially for researchers who are not necessarily trained in standard GIS or computer science platforms , as they can conduct multi criteria model analysis (Fourcade et al. , 2014) . Despite

PAGE 31

! ! 31 ! are not well or completely understood by its users (Yackulic et al. , 2012 ) . Geographic Information Systems Coupled with the increasing availability of open source data and advances in technology, GIS has become a highly effective tool for studying and re presenting ecological trends (Dark and Bram , 2007) . Tools and models created within GIS applications have recently been used to model changes in the representations of both plant and animal species distribution s . For example, Olivier and Ar anda (2017) used niche analysis to model a little known species of grasshopper with DIVA GIS by comparing point occurrence data with 19 bioclimatic variables from the WorldClim database. Their results revealed that GIS models can effectively predict geogra phic distributions for species using presence only data, while having limited background knowledge of a species biological traits or affinities. Some of the most commonly used tools and procedures for GIS based niche modeling have been compiled by Brown (2014) , including the Weighted Linear Combination (WLC) method (Maczewski , 2004). The WLC , also referred to as simple additive weighting, is considered the most straightforward option for modeling landscape processes in GIS . It is based on the concept of a weighted average , where the modeler directly assigns weights of relative importance to each attribute map layer. A ratio scale is then constructed, with values corresponding from least to most preferable alternatives. P roducts are summed over all th e attributes and the score of the alternative with the highest overall score is chosen. The results of this technique are used to derive composite maps. The two most critical elements that must be carefully addressed to accurately implement the WLC mo del are assigning weights to attribute maps and implementing procedures for deriving the attribute maps (Malczewski , 2000). Simply, to have confidence in the accuracy of the final composite

PAGE 32

! ! 32 ! map, the attribute maps (which are over la yed to produce the final results) must represent rankings and weights that ha ve been calculated as accurately and objectively as possible. Appropriately deriving values for different attributes within a layer is particularly challenging for ecological managers ( e .g. , the annual precipitat ion range 10 to 15mm is most likely to be linked to species occurrence ther efore its assigned value is 10, versus the annual precipitation range of 7 to 9mm is least likely therefore its assigned value is 1) . One method that has been suggested is the value function approach (Hepner , 1984, Hobbs , 1980, Lai and Hopkins , 1989, Keisler and Sundell , 1997) , which converts different levels of an attribute into scores . The scores are then related to a preference scale reflective of the decision makers interests. A common technique used to derive the value scores is to follow the mid value method (Keeney , 1980, Lai and Hopkins , 1989). User s identify the maximum and minimum values of an attribute, then the midpoint val ue of the interval between the max imum and minimum attribute values is calculated. Once the midpoint value is determined, quarter points can be calculated between the minimum and the midpoint, and then between the midpoint and the maximum. The procedure ca n be repeated, accuracy increasing with points added . ! Habitat Suitability Analyses Considerations A caveat that comes with all environmental models is that results can have high variability depending on parameters and input data. This is particularly true with the increasing use of GIS applications, such as ArcGIS, QGIS and Grass GIS. Today, well intentioned modelers have difficulty avoiding the many pitfalls, unintentional or otherwise, associated with applying modeling methods without fully comprehending the mechanisms involved. Second, nearly all environmental models suffer from the Modifiable Areal Unit Problem (Dark and Bram , 2007) . The Modifiable Areal Unit Problem (MAUP) occurs as a result of setting artificial partitions to define an area of analys is. These boundaries are often necessary for the simple reason that focus areas or areas of interest must be bound ed or contained to some standardly recognized area .

PAGE 33

! ! 33 ! Finally, the choice of model determines model outcom es ( Gutierrez et al . , 2013 ) , t herefore, model selection should match research objectives. Ideally, the model would be capable of using multi criteria decision making methods that would account for po tentially conflicting parameters , future expert options, or more species specific findings that could be inserted into model parameters. It was important to select a model that would provide easily interpreted, but highly informative results. There are fe w comparable studies that specifically attempt to link or model the influence of the Chihuahuan b iogeographic sub e lement on the flora of Colorado . However, more literature exists documenting similar research on the Californi a Floristic Province (Baldwin , 2014). Th ese studies detailed the framework for methods that involved the use of multiple regression analyses to link occurrence hotspots to climatic variables (Rich er son an d Lum , 1980) . Although geoprocessing tools exist in the ArcGIS platform that can conduct regression analysis , such as the Ordinary Least Squares (OLS) and Geographically Weighted Regression tool (GWR) , those tools are not appropriate for this study. The OLS tool , and similar ly the GWR tool, are classic linear methods that perform poorly when relationships between the dependent variable and the explanatory variables are nonlinear (Esri ArcGIS), which is often the case in eco logy . For these reasons, the approach of combining a modified GLM structure for variable analysis with the Weighted Linear Combination method was taken . Research Objectives My objective was to characterize the niche for those Chihuahuan species that are present in the flora of Colorado (based on analyses in Chapter 1) using GIS . I first identified the environmental factors that were influential to a species overall distribution , by testing their effect on species presenc e through statistical analyses . The final goal was to broadly enumerate a set of criteria based on the environmental conditions that contribute to overall distributions and t hen create spatial representations of those criteria within Colorado. Thus, areas within Colorado that can provide suitable habitat would be identified .

PAGE 34

! ! 34 ! Methods and Materials To gain a broad and practical insight into the variety of floristic niches that appear to be influenced by flora of the Chihuahuan sub elemen t, five species were chosen from the list compiled in Chapter 1 for niche modeling. The first goal when selecting the model species was to include species that had an almost exclusively Chihuahuan distribution, with little overlap to other elements. Consid eration was also given towards previous recognition of a species in association with the Chihuahuan sub element, for example by Weber (1965) or Clark (1996). Next, species needed to encompass a variety of plant habit types. Ultimately, Funastrum crispum (B enth.) Schltr., Palafoxia sphacelata (Nutt. ex Torr) Cory, Nolina texana S. Watson, Muhlenbergia arenacea (Buckl.) A.S. Hitchc. a nd Muhlenbergia arenicola Buckl. were selected . ! BioClim Data Processing Data preparation followed best practices outlined by the SDM Toolbox (Brown , 2014 ) . Bioclimatic variable s provided by the Worldclim database served as the independent variables (Fick et al. , 2017) . The se 19 variables (Appendix Table A .1 ) were selected as predictor/independent variables for this work as they are commonly used in ecological models as experimental variables . To dataset, which interpolates data from 1970 to 2000, was used for this analysis. Next, data was processed to mitigate latitudinal bias. Most models, including Ma x e nt, assume equal cell size across the data layer. However, the WorldClim Bioclim datasets are not projected, and as such, there is varying cell area in the rasters. To correct for this, the layers wer e projected from WGS84 to the NAD (1983) 20 1 1 Albers Equal Area Projection. Additionally, a bias grid was constructed to correct for latitudinal bias during the selection of background/pseudo absence points and projected to match. Pseudo absence points we re generated using a pre made python script (Dilts 2015) and merged with known occurrence points . Then points were selected to include only those completely within the

PAGE 35

! ! 35 ! bias file polygon for a complete occurrence dataset per species. Last, the data points w ere spatially rarefied using a pre made python script tool supplied from the SDM toolbox (Brown , 2014) . Actual presence location points were dummy coded as 1, and the pseudo absence points as 0 in the attribute tables . T he Sample tool in ArcMap was then used to collect the values of BioClim data at each XY location of occurrence and pseudo absence points and display ed them in a table . The next step was the statistical analysis of the data , using SPSS, and inputting compiled Sample tool result tables . S tatistical Analyses To ensure the key environmental variables that affect these species were identified, and not arbitrarily excluded during dimension reduction, Principal Component Analysis was first utilized inputting all 19 variables, with a loadin g cutoff value of .5. This cutoff level was imposed both to attempt to mitigate for the inclusion of highly correlated variables as well as ensure dimension reduction so only the most important variables were examined further. By selecting a large number o f individual variables, and then synthesizing or grouping the variables into factor groups, the important variables were not excluded. This was chosen as the first step in statistical analysis as it focuses on those co variates that account for the greates t percent of variance for the species (Cruz Cardenas et al. , 2014) . Co variates from the component that accounted for the most variance were included in the next round of analysis if they met the load value cutoff. Th ose co variates were then te sted with stepwise binomial regression a nalysis. This was performed so that it would be possible to understand the levels of influence each variable had on the dependent variable (species presence) as well as thoroughly mitigate the inclusion of highly cor related variables . Habitat Suitability Analysis T he GIS Weighted Linear Combination Method was subsequently used to perform a modified habitat suitability assessment within Colorado for the five selected species.

PAGE 36

! ! 36 ! The beta coefficients calculated in the r egression analysis were normalized to a 0 1 scale value and served as the factor weights. Finally, the alternative attribute values and corresponding preference scores were determined by inputting the data from each co variate selected as an analysis layer into the Relative Frequency Function Tool (RFF) . The suitability scores generated from the RFF Tool were averaged, classified (binned) by natural breaks groups. A grouping was considered a natural break if the average for an individual attribute was disti nct from other values. For example, if the average RFF score for 10 degrees Celsius calculated to 7, and no other average score in the temperature range of 9 15 calculated to 7, that was a natural break. If the resulting averages were the same across conse cutive attributes, they were grouped together. From there, the average scores were normalized to values corresponding to a 1 10 scale. Those served as the alternative attributes and preference values. After the core steps of this process were completed , se nsitivity analyses were conducted for all model outputs. The purpose of this was to define how changes in a model parameter would affect the overall model outputs, and to some degree indicate the level of uncertainty involved in the measurement of that par ameter (Wainwright , 2013). The confidenc e levels associated with the selection of analysis layers through PCA, and factor weight values determined from regression analysis were high enough that testing their sensitivity to different parameters was not considered necessary. However, the greatest uncertainty stemmed from the novel methods implemented to classify alternative attributes and score them. As such, the sensitivity analyses tested the results when the alternative attributes were classified using average midpoints and quartile values inste ad of classifying them primarily as natural break value groups. Finally, to further compare the accuracy of the WLC habitat suitability outputs, results were compared with a MaxEnt model output. All defaults of MaxEnt parameters were accepted and data was prepped following step by step best practices (Young et al . , 2018). To run MaxEnt v 3.3 through ArcMap, the python script tool MaxEnt Script Tool for ArcGIS (Donoghue , 2013) was downloaded and run in Mode l Builder .

PAGE 37

! ! 37 ! R esults In all five models, the percent variance attributed to these predictor variables was over 90%, indicating that the se variables successfully included factors that were important to the presence or absence of t hese species. Yet, when all statistical assumptions were addressed and accounted for ( i.e. , potentially high correlations between variables or symptoms of multicollinearity , etc. ) the number of variables that independently influenced the presence or ab sence of a species was usually less than ten. ! Palafoxia sphacelata (Asteraceae) Two principle components accounted for a cumulative 94.620% of variance. The first component accounted for the greatest percent of variance (67.284%) and was used for niche mo deling (the second component accounted for the remaining 27.335%). When the 0.5 loading value cutoff was applied to the first component, 10 co variates were retained for further testing. Of those, the regression analysis showed that four (Table A.5) were t he most useful with Bio 3 (Isothermality) having the greatest effect on species presence. Bio 2 (Mean Diurnal Range) subsequently had the next greatest effects on species presence for the target species. The results of the Relative Frequency Function (RFF ) tool showed that for Bio 3 (Isothermality), the bin that yielded the highest average suitability score (6) was the 36 to 38.9¡ C grouping. For Bio 2 (Mean Diurnal Range), the grouping that had the highest associated suitability score (6) was 19 to 22.9¡ C. The full range of the distribution of Palafoxia sphacelata stretched from Chihuahua, Mexico to northern Colorado USA (Fig. 2.1). The highest suitability level was calculated to a max imu m of approximately 5.77 and a minimum value of 1.55 where the mini mum represented areas that had the lowest suitability but still met suitable habitat requirements. This species appears to follow the trend of the most suitable habitat appearing on the eastern and southern extents of Colorado. The EPA Level II ecoregions that contained the areas of highest suitability were (in descending order) Southern Rockies, High Plains, Arizona/New Mexico Plateaus, and Southwestern Tablelands .

PAGE 38

! ! 38 ! Within those ecoregions, the following counties had the areas of highest suitability: Weld, Morgan, Logan, Sedgwick, Phillips, Yuma, Washington, Adams, Arapahoe, Douglas, Elbert, Kit Carson, Cheyenne, Lincoln, El Paso, Teller, Park, Chaffee, Fremont, Pueblo, Crowley, Kiowa, Prowers, Bent, Otero, Custer, Saguache, Alamosa, Huerfano, Las Animas, Ba ca, Costilla Conejos. Muhlenbergia arenicola (Poaceae) Two principle components accounted for a cumulative 94.884% of variance. The first component accounted for the greatest percent of variance (64.441%) and was used for niche modeling. (The second compo nent accounted for the remaining 30.442%.) When the 0.5 cutoff level was applied to the first component, six co variates were retained for further testing: (Bio 11, 9, 7, 6, 4,3) A binary logistic regression analysis showed that all six co variates contain ed useful information (Appendix Table A.3). Of those, Bio 11 (Mean Temperature of the Coldest Quarter) had the greatest effect on species presence, followed by Bio 6 (Min Temperature of the Coldest Month). The Relative Frequency Function (RFF) output for B io 11 (Mean Temperature of the Coldest Quarter) showed that the bin with the highest suitability (7) was the 11 to 12.9 ¡ C range, and for Bio 6 (Min Temperature of the Coldest Month) it was the range of 0 to 4.9¡ C. M. arenicola occurs from Mexico City, Mexico to Colorado (Fig. 2.2). The highest suitability level was calculated to a maximum of approximately 6.4 and a minimum value of 2.03, where the minimum represented areas that had the least likelihood of suitability but still met suitable habitat requi rements. The EPA Level III EcoRegions that contained the areas of highest suitability were (in descending order): Southern Rockies, High Plains, Southwestern Tablelands, and Colorado Plateaus . Within those ecoregions, t he following Colorado counties had t he areas of highest suitability: Boulder, Broomfield, Denver, Arapahoe, Kit Carson, Cheyenne, Lincoln, El Paso, Fremont, Pueblo, Crowley, Kiowa, Prowers, Bent, Otero, Huerfano, Las Animas, Baca, Southwestern Montezuma. Suitability values within Colorado ra nged from 3.2 to 5.3.

PAGE 39

! ! 39 ! Funastrum crispum (Apocynaceae) Two principle components accounted for a cumulative 92.998% of variance. The first component accounted for the greatest percent of variance (67.376%) and was used for niche modeling (the second compone nt accounted for the remaining 25.622%). When the 0.5 loading value cutoff was applied to the first component, nine co variates were retained for further testing. The regression analysis showed that five co variates (Table A.4) had the greatest effect on species presence for Funastrum crispum, with Bio 2 (Mean Diurnal Range (Mean of monthly (max temp min temp))and Bio 16 (Precipitation of Wettest Quarter) having the greatest impact. The results of the Relative Frequency Function (RFF) showed that for Bio1 6(Precipitation of Wettest Quarter), the bin whose attributes had the highest suitability score (8) was the 281 to 331.9mm range. For Bio 2 (Mean Diurnal Range (Mean of monthly (max temp min temp), the bin associated with the highest suitability score (8) was the 9 to 11.9¡ C range. The full range of the distribution of F. crispum stretched from Aguascalientes, Mexico to Colorado, USA (Figure 2.3). The highest suitability level was calculated to a max imum of approximately 7.9 and a minimum value of 2.0 whe re the minimum represented areas that had the lowest suitability but still met suitable habitat requirements. Within Colorado, there were small non continuous patches of areas with high suitability. These were located clustered along, and to the west of th e Continental Divide encompassing two ecoregions . The EPA Level III EcoRegions that contained the areas of highest suitability were the Southern Rockies and the Colorado Plateaus . The following counties within those ecoregions that had the areas with the highest suitability: Alamosa, Boulder, Chaffee, Clear Creek, Costilla, Custer, Delta, Dolores, Eagle, El Paso, Fremont, Garfield, Gilpin, Gunnison, Hinsdale, Huerfano, Jackson, Jefferson, La Plata, Lake, Larimer, Las Animas, Mesa, Moffat, Montezuma, Ouray, Park, Pitkin, Pueblo, Rio Blanco, Routt, Saguache, San Juan, San Miguel, Summit, and Teller.

PAGE 40

! ! 40 ! Nolina texana (Asparagaceae) Three principle components accounted for a cumulative 98.885% of variance. The first component accounted for the greatest percent of variance (73.824%) and was used for niche modeling (the second component accounted for 19.13%, the third accounted for 5.9%). When a 0.5 loading value cutoff was applied to the first component, eight co variates were retained for further testing (Bio 19, Bio 17, Bio 15, Bio 14, Bio 13, Bio 12, Bio 3, Bio 2). The regression analysis for that subset of co variates showed that seven co variates were the most useful (Table A.6). Bio 2 (Mean Diurnal Range (mean of monthly (max temp min temp)) and Bio 15 (Preci pitation Seasonality) had the greatest positive effect on species presence for Nolina texana . Then RFF tool output showed that for Bio 2 (Mean Diurnal Range (mean of monthly (max temp min temp)), the bin whose attributes had the highest suitability score ( 9) was the 10 to 12.9¡ C range . For Bio 15 (Precipitation Seasonality), the bin whose attributes had the highest suitability score (10) was the 31 to 41.9 mm range. The full range of the distribution of N. texana stretched from San Luis Potosi, Mexico to Colorado, USA(Fig. 2.4). The highest suitability level was calculated to a max imum of approximately 9.06 and a minimum value of 1.93 where the minimum represented areas that had the lowest suitability but still met suitable habitat parameters. The suitabi lity values within the state of Colorado range from 1 to 10. Although portions of Colorado scored as high suitability, most areas in Colorado were moderate to low and were only found in one ecoregion . The EPA Level III Ecoregion that contained the areas of highest suitability was the Southern Rockies . Nolina texana , appears to have habitat requirements found mostly in the eastern half of Colorado. Within that ecoregion, t he following counties had the highest suitability scores: Alamosa, Boulder, Chaffee, Cl ear Creek, Costilla, Custer, El Paso, Fremont, Huerfano, Jefferson, Las Animas, Larimer, Park, Pueblo, Saguache and Teller.

PAGE 41

! ! 41 ! Muhlenbergia arenacea (Poaceae) Two principle components accounted for a cumulative 9 6.15 % of variance. The first component accoun ted for the greatest percent of variance ( 71.024 %) and was used for niche modeling (the second component accounted for the remaining 25.129%). When the 0.5 loading value cutoff was applied to the first component, ten co variates were retained for further t esting : Bio 5 , Bio 8 , Bio 10 , Bio 12 , Bio 13 , Bio 14, Bio1 6, Bio 17, Bio 18, and Bio1 9 . A binary logistic regression analysis showed that six co variates contained the most useful information (Appendix Table A. 7 ). Of those, Bio 18 (Precipitation of Warmest Q uarter) had the greatest effect on species presence , followed by Bio 16 (Precipitation of Wettest Quarter) . The Relative Frequency Function (RFF) output for Bio1 8 ( Precipitation of Warmest Quarter) showed that the bin with the highest suitability sco re (6) was the 145 mm to 166.9 mm range , for Bio 16 ( Precipitation of Wettest Quarter) it was the range of 20.3 mm to 26.69 mm. Muhlenbergia arenacea occurs from Zacatecas , Mexico north to Colorado (Fig. 2.5). The highest suitability level was calculated t o a max of approximately 6.02 and a minimum value of 3.9 , where the minimum represented areas that had the least likelihood of suitability but still met suitable habitat requirement s. The EPA ecoregions that contained the areas of highest suitability we re (in descending order): High Plains, Southern Rockies, Southwestern Tablelands, and Arizona/New Mexico Plateau . Within those ecoregions, t he following Colorado counties had the areas of highest suitability: Adams, Alamosa, Arapahoe, Baca, Bent, Chaffee, Cheyenne, Costilla, Crowley, Custer, Denver, Douglas, El Paso, Elbert, Fremont, Huerfano, Kiowa, Larimer, Las Animas, Lincoln, Morgan, Otero, Park, Prowers, Pueblo, Saguache, Washington, and Weld. Suitability values within Colorado ranged from 2.27 to 6 .9. Counties on the eastern half of the state (Las Animas, Huerfano, Pueblo, Otero, Bent, Crowley, Lincoln, Kiowa, Prowers, Morgan, Adams, Weld, and Arapaho) were revealed to harbor much of the most suitable habitat (Fig. 2.5 ). The results of the habit at suitability analysis also reveal

PAGE 42

! ! 42 ! that seven counties were almost entirely composed of suitable habitat and could likely sustain populations of Muhlenbergia areancea : Adams, Arapahoe, Custer, Huerfano, Morgan, Park and Weld countie s. MaxEnt Comparison When comparing MaxEnt outputs with statistical results derived from PCA and regression analyses, there were similar findings for all five of the species modeled. A re as that scored medium to high suitability from the WLC method were also generally areas of suitability , according to MaxEnt outputs. For all modeled species, calculations produced from MaxEnt which identified co variates of greatest importance, resulted i n findings that corresponded to my statistical calculations. For F. crispum, both methods revealed that Bio 16 (Precipitation of Wettest Quarter) had a high contribution to species presence. For P. sphacelata , both methods identified Bio 3 ( Isothermality) as having high influence on species presence. Similarly, for M. arenicola , both methods calculated Bio 6 (Min Temperature of Coldest Month), and Bio 11 (Mean Temperature of Coldest Quarter), to be the variables with the greatest impact on species presence. For N. texana , both methods identified Bio 15(Precipitation Seasonality) and Bio 2 (Mean Diurnal Range) as the having the greatest impacts. Finally, both methods revealed that f or M. arenacea , Bio 18 (Precipitation of Warmest Quarter) , and Bio 16 (Preci pitation of Wettest Quarter) were important to predicting species presence . ! Discussion A GIS based model was used to model the climatic niche for five species of the Chihuahuan sub element that are present in Colorado. In so doing, I also compare my results with those of software more commonly used by ecologists (i.e., MaxEnt), thereby assessing their validity. In several ways it is not unlike other niche model studies. However, it moved beyond the standard format of results, which are us ually composed of a list of predictor variables that are important in explaining species distributions . Here I was able to obtain results that include d : a list of predictor variables that impact

PAGE 43

! ! 43 ! species presence, their relative levels of influence, subdivi ded groupings of individual variable characteristics ranked from least to most suitable, and visual representation s of alternative results based on changes in parameterizations . Abiotic and Biotic Patterns Two variables appeared to have the g reatest impacts on three of the modeled species. First, the variable Bio 2 (Mean Diurnal Range ) appears to link species presence and climatic conditions. It had the greatest influence on both Nolina texana, Funastrum crispum and Palafoxia sphacelata . Secon d, Bio 16 (Precipitation of Wettest Quarter) was calculated to have great impact on both F. crispum and M. arenacea . th ese environmental factor s play a key role in the presence of plants of the Chihuahuan b iogeographic sub elem ent in Colorado. Geographic Patterns Although low to medium high suitability is distributed across Colorado , small pockets of high suitability occur , particularly in the southwest ern and southeast ern portions of the state. The differences between individu al species niches were noticeably apparent. First, while most of the species had strong levels of suitable habitat in the southeast corner of Colorado, the suitability of the rest of the state was not as uniform. For some species nearly the entire state co ntained at least some level of suitable habitat, ( Palafoxia sphacelata ). Others such as Muhlenbergia ar enicola very clearly excluded the northwestern corner and only met suitable habitat criteria in the eastern portion of Colorado. The eastern portion of t he state is composed mostly of mixed grass shrublands (Rondeau et al. 2013) , which is consistent with the observation that graminoids account for the majority of habit types of the identified species . Interestingly, majority of suitable habitat in C olorado was contained within the Southern Rockies EPA Level III EcoRegion. Although the majority of actual occurrences were found to occur in the High Plains and Southwestern Tablelands EcoRegions (Chapter I) it appears that species of this element may als o find suitable habitat westward of their current known locations. N. texana and F.

PAGE 44

! ! 44 ! crispum demonstrate this unique westerly affinity most clearly. The three other species outputs do indicate areas of suitable habitats that extend into the actual Rocky Mountain range. However, the most suitable areas in Colorado for N. texana and F. crispum are only slivers of land, almost exclusively located along the Rocky Mountain chain. This suggests there is a possible connection with the climatic conditions produc ed in pockets of mountainous areas with the presence of these species . Sensitivity Analysis From the five sensitivity analyses, three outputs ( Palafoxia sphacelata , Nolina texana , Muhlenbergia arenacea ) appeared to have somewhat inverse results from the original Natural Breaks grouping method. The results for P. sphacelata show an area in northeast Colorado, Weld county, a s highly suitable habitat, but the sam e county appears as low suitability in the sensitivity outputs (Fig. 2.1). In t wo of the outpu ts ( Muhlenbergia arenicola, Funastrum crispum ) , the sensitivity results showed a less conservative classification of suitable habitat. Areas within Baca county that appeared to be low suitab ility for F. crispum in the natural grouping method were classifie d as moderate in the sensitivity analysis (Fig. 2.3) . Howeve r, upon further examination of the sensitivity outputs with actual occurrence points, the sensitivity outputs more often than not, place known occurrence points in areas of least suitability. It is possible that in the instances where results were impacted by altering parameters, results might have been more congruent if additional midpoint breaks had been added. This could help ensure that small variations in data are not lost in large quartil e groupings. In the case of these five species, the sensitivity analysis revealed that by applying pressure to the preference scale parameter the actual range of suitability scores were not dramatically impacted, but the placement and geographic classif ication of suitable habitat had great variation . Method Comparison Ecologists generally agree that using MaxEnt for species distribution models produces highly reliable and accurate results (Aguirre Gutierrez et al. , 2013). However, users inevit ably require the

PAGE 45

! ! 45 ! use of additional software packages such as R, ArcGIS, QGIS, DIVA GIS or Grass GIS for data processing. This aspect of the use of MaxEnt seems redundant, as both R, ArcGIS, Q and Grass GIS have statistical computing capabilities in additio n to their cartographic functions. Additionally, despite the accuracy associated with MaxEnt, it s black box style processes do not easily allow for finding and correcting errors entered into the workflow, whereas with my process, it was possible to backtr ack if needed . In this work, the majority of MaxEnt outputs compared well with result outputs from the Weighted Linear Combination method. Any differences stemmed from a few different parameters. For example, suitability values associated with t he WLC outputs are normalized to a 1 to 10 scale. Outputs from MaxEnt are given in log odds values. Second, with the WLC, areas that did not meet the suitable habitat requirements were assigned a NoData value and had no graphical representation. MaxEnt, outputs still assign values to areas that have near zero suitability. As a result, MaxEnt outputs display values and corresponding color to the entire study area whereas in the WLC, areas that might be considered not suitable are not included in fi nal outputs (Fig. 2.6) . Limitations of the Study Several known limitations of the Weighted Linear Combination Method for habitat suitability analyses have been addressed in similar research (Malcze wski , 2002, Drobne and Lisec , 2009) , and are applicable t o this work. First, the model relies on the quality of the data used. True presence true absence studies have been shown to yield the highest performance capability when used in environmental modeling. Here, presence only datasets were modified to function standard industry procedures (Barbet Massin et al. , 2012), the results of the analysis should not be interpreted to represent those of a real wor ld inventory of true presence and true absence data. The bioclimatic datasets although provided by a reputable database, are collected using interpolation methods, which also impact results. Additionally, the 19 bioclimatic variables provided by the WorldC lim database, while readily available and frequently used as predictor variables in

PAGE 46

! ! 46 ! SDMs (Cruz Cardenas et al. , 2014, Aguirre Gutierez et al. , 2013), are not necessarily the most relevant variables for flor istic studies. Desert plant distributi ons have been found to be strongly impacted by soil characteristics (Laport et al. , 2013). Variables such as soil moisture, soil type, or radiation levels might serve as better predictor variables, but the data are not as easily accessible (Forester , 2012) . Next, the regression analysis that was used to select the variables to include as raster layers in the habitat suitability analysis accounted for less than the cumulative variance originally extracted from the first components analyzed through PCA. Thi s means that there are still unknown factors that account for large portions of the variance. Future studies that incorporate additional environmental variables into regression analyses could produce more detailed and comprehensive results. Finally, and step in the WLC method has notable impact on the final outcome for areas of potential habitat. For this study, the preference scale was assembled using averages of the R FF suitability scores for attribute ranges , as at this time, there are few if any records of expert opinions regarding the preferences of the modeled species. Although i t is common for ecological researchers to narrow down which environmental factors influ ence species occurrence or presence distributions , t here is little literature further subdividing the individual variables into classes of preferability. One method frequently used by modelers of the GIS community is a modified Delphi Technique in combinat ion with the Analytical Hierarchy Process ( AHP ) . However, due to time and resource constraints, as well as the immediate objectives of this study, scale rank by panel survey was not feasible. However, if a small scale and targeted species approach was used , it would be possible to collect this information. Here, the first steps have been taken to attempt to categorize variable conditions that can dictate the suitability of an area for an individual species. Conclusion The landscape of Colorado is heterogeneous, supporting species representing several floristic elements and sub elements, including the Chihuahuan biogeographic sub element (Weber , 1965). By

PAGE 47

! ! 47 ! spatially analyzing the species comprising this sub element, it was possible to (1) identify th eir ecological niches and (2) locate areas within Colorado that could serve as suitable habitat. This study has contributed to a better understanding of the influence of the Chihuahuan sub element on the flora of Colorado. It also offers an oppo rtunity: to consider multi disciplinary methodology that incorporates intra variable attributes to identify relationships between ecological niches and geographic distributions . !

PAGE 48

! ! 48 ! Figures Fig. 2 .1. Results of a habitat suitability analys is for Palafoxia sphacelat a revealing (a) suitable habitat in the southwestern United States and Mexico and (b) Colorado suitability contrasted with (c) results from a sensitivity analysis. a c b

PAGE 49

! ! 49 ! Fig. 2 .2. Results of a habitat suitability analysis for Muhlenbergia arenicola reveal( a) suitable habitat in the southwestern United States and Mexico and (b) Colorado suitability contrasted with (c) results from a sensitivity analysis. b a c

PAGE 50

! ! 50 ! Fig . 2. 3. Results of a habitat suitability analysis f or Funastrum crispum reveal (a) suitable habitat in the southwestern United States and Mexico and (b) Colorado suitability contrasted with (c) results from a sensitivity analysis. b c a

PAGE 51

! ! 51 ! Fig . 2. 4. Results of a habitat suitability analysis for Nolina texana reveal (a) suitable habitat in the southwestern United States and Mexico and (b) Colorado suitability contrasted with (c) results from a sensitivity analysis. a b c

PAGE 52

! ! 52 ! ! ! ! Fig . 2 .5 Results of a habitat suitability analysis for Muhlenbergia arenacea revealing (a) suitable habitat in the southwestern United States and Mexico and (b) Colorado suitability ` contrasted with (c) results from a sensitivity analysis. a b c

PAGE 53

! ! 53 ! Figure 2. 6 Comparison of outputs f or (top) Palafoxia sphacelat a , and (botto m ) Muhlenbergia arenacea ) from both WLC and MaxEnt methods .

PAGE 54

! ! 54 ! REFERENCES Adams. G. et al. 1904. Gypsum Deposits in the United States. United States Geological Survey. Aguirre GutiŽrrez, J., Carvalheiro, L. G., Polce, C., Loon, E. E., Raes, N., R eemer, M., & Biesmeijer, J. C. 2013. Fit for p urpose: Species d istribution m odel p erformance d epends on e valuation c riteria Dutch Hoverflies as a c ase s tudy. PLoS ONE, 8(5). doi:10.1371/journal.pone.0063708 Alvarez, L. J., H. E. Epstein, J. Li, and G. S . Okin. 2011. Spatial patterns of grasses and shrubs in an arid grassland environment. Ecosphere 2(9):103. doi:10.1890/ES11 00104.1 Baldwin B . G. 2014. Origins of plant diversity in the California Floristic Province. The Annual Review of Ecology, Evolutio ns, and Systematics. 45: 347 369. Doi: 10.1146/annuren ecolsys 110512 135847. Barbet Massin, M., Jiguet, F., Albert, C. and Thuiller, W. 2012. Selecting pseudo absences for species distribution models: how, where and how many?. Methods in Ecology and Evol ution, 3(2), pp.327 338. Brown, D . E. 1982. arizona.openrepository.com/handle/10150/550974. Brown J. 2014. SDMtoolbox: a Python Based GIS Toolkit for Landscape Genetic, Biogeographic and Species Distributi on Model Analyses. Methods in Ecology and Evolution.5:694 700. Caffrey M, Doerner J. 2012. A 7000 y ear r ecord of e nvironmental c hange, Bear Lake, Rocky Mountain National Park, USA. Physical Geography. 33(5): 438 456. Cain, S. A.,1944, Foundations of plant geography: New York, NY, Harper Bros. Chapman, S.S., Griffith, G.E., Omernik, J.M., Price, A.B., Freeouf, J., and Schrupp, D.L. 2006 . Ecoregions of Colorado (color poster with map, descriptive text, summary tables, and photographs): Reston, Virginia, U.S. Geological Survey (map scale 1:1,200,000). Collin, R., & De Maintenon, M. 2002. Integrative a pproaches to b iogeography: Patterns and p rocesses on l and and in the s ea. Integrative an d Comparative Biology, 42(5), 911 912. doi:10.1093/icb/42.5.911 Cruz C‡rdenas, G., L—pez Mata, L., Villase–or, J. L., & Ortiz, E. 2014. Potential species distribution modeling and the use of principal component analysis as predictor variables. Revista Mex icana De Biodiversidad, 85(1), 189 199. doi:10.7550/rmb.36723 Dark, S. J., & Bram, D. 2007. The modifiable areal unit problem (MAUP) in physical geography. Progress in Physical Geography, 31(5), 471 479. doi:10.1177/0309133307083294

PAGE 55

! ! 55 ! Dilts, T.E. (2015) Ge nerate Random Points Around Points Tool for ArcGIS 10.3.1. University of Nevada Reno. http://www.arcgis.com/home/item.html?id=e43b0fde28914b12b36c9725467c0189 Donoghue J .2013. A Maxent Script Tool for ArcGIS . Drobne, S . & Lisec A. 2009 . Multi att ribute d ecision a nalysis in GIS: Weighted l inear c ombination and Or dered w eighted av eraging. Informatica , 33 , 459 474. Elith, J., Phillips, S. J., Hastie, T., Dud’k, M., Chee, Y. E., & Yates, C. J. (2010). A statistical explanation of MaxEnt for ecologists. Div ersity and Distributions, 17(1), 43 57. doi:10.1111/j.1472 4642.2010.00725.x Elith, J., H. Graham, C., P. Anderson, R., Dud’k, M., Ferrier, S., Guisan, A., J. Hijmans, R., Huettmann, F., R. Leathwick, J., Lehmann, A., Li, J., G. Lohmann, L., A. Loiselle, B., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., McC. M. Overton, J., Townsend Peterson, A., J. Phillips, S., Richardson, K., Scachetti Pereira, R., E. Schapire, R., Sober—n, J., Williams, S., S. Wisz, M. and E. Zimmermann, N. 2006. Novel methods im occurrence data. Ecography, 29(2),129 151. Ebach, M. (2015). The multidisciplinary nature of biogeography. Australian Systematic Botany, 28(3) , 79. Encyclopedia Britannica. (2018). Pteridaceae | plant famil y. [online] Available at: https://www.britannica.com/plant/Pteridaceae [Accessed 14 Mar. 2018]. Fick, S.E. & R.J. Hijmans, 2017. Worldclim 2: New 1 km spatial resolution climate surfaces for global land areas. International Journal of Climatology. Forest er, B. R. (2012) "Integrating ensemble species distribution modeling and statistical phylogeography to inform projections of climate change impacts on species distributions" WWU Masters Thesis Collection. 202.http://cedar.wwu.edu/wwuet/202 Fourcade Y . , En gler J . O . , Ršdder D . , and Secondi J . 2014 . Mapping s pecies d istributions with MAXENT u sing a g eographically b iased s ample of p resence d ata: A p erformance a ssessment of m ethods for c orrecting s ampling b ias. PLoS ONE 9(5): e97122. https://doi.org/10.1371/jou rnal.pone.0097122 Guisan, A., Edwards, T. C., & Hastie, T. 2002. Generalized linear and generalized additive models in studies of species distributions: Setting the scene. Ecological Modelling, 157(2 3), 89 100. doi:10.1016/s0304 3800(02)00204 1 GBIF.org (29th February 2016) GBIF Occurrence Download https://doi.org/10.15468/dl.ywhpmz Graham A. 2011. The age and diversification of terrestrial New World ecosystems through Cretaceous and Cenozoic time. American Journal of Botany. 98: 336 351.

PAGE 56

! ! 56 ! Graham M . R . , Jaeger J . R . , Prendini L . , and Riddle B . Desert Scorpion, Paruroctonus becki, reveals Pliocene diversification in the Eastern California Shear Zone and post glacial expansion in the Great Basin Desert. Molecular Phylogenet ics and Evolution 69: 502 513. Henriques Silva R . , Lindo Z . , Peres Neto P.R. 2013. A community of metacommunities: exploring patterns in species distributions across large geographical areas. Ecology, Ecological Society of America. 94(3):627 639. Hijmans R, Elith J. 2016.Species Distribution Modeling. Species Distribution Modeling R Spatial. Geospatial and Farming Systems Research Consortium (GFC).rspatial.org/sdm/index.html. Hirzel A . H . , Le Lay G.2008. Review: Habitat s uitability m odelling and n iche t heory. Journal of Applied Ecology. 45(5): 1372 1381. Web of Science. http://www.jstor.org/stable/2014410516. Janzen T, Haegeman B . , Etienne R . S. 2015.A s ampling f ormula for e cological c ommunities with m ultiple d ispersal s yndromes. Journal of Theoretical B iology. 387:258 261. Exclusion, and Niche Conservatism among Larrea Tridentata Cytotypes in North American ciety, vol. 140, no. 3, 2013, pp. 349 363., doi:10.3159/torrey d 13 00009.1 Lele, S. R., Merrill, E. H., Keim, J., & Boyce, M. S. (2013). Selection, use, choice and occupancy: Clarifying concepts in resource selection studies. Journal of Animal Ecology, 8 2(6), 1183 1191. doi:10.1111/1365 2656.12141 Loehle, C. 2012. Relative frequency function models for species distribution modeling. Ecography, 35(6), 487 498. doi:10.1111/j.1600 0587.2012.07389.x Malczewski, J. 2000. On the Use of Weighted Linear Combina tion Method in GIS: Common and Best Practice Approaches. Transactions in GIS, 4(1), 5 22. doi:10.1111/1467 9671.00035 Marlowe, K., Hufford, L. (2007). Taxonomy and b iogeography of Gaillardia (Asteraceae): A p hylogenetic a nalysis. Systematic Botany, 32(1 ), 208 226. doi:10.1600/036364407780360229 McLaughlin, S., 2007. Tundra to Tropics: The f loristic p lant g eography of North America. Fort Worth Texas. Botanical institute of Texas. Mclaughlin, S. 1994. Floristic plant geography: The classification of flor istic areas and floristic elements. Progress in Physical Geography, 18(2), 185 208. doi:10.1177/030913339401800202 . McLaughlin S. 1986. Floristic a nalysis of the Southwestern United States. Great Basin Naturalist. scholarsarchive.byu.edu/gbn/vol46/iss1/5/ . Moore, M. J., J. F. Mota, N. A. Douglas, H. Flores Olvera, and H. Ochoterena. 2014. The ecology, assembly, and evolution of gypsophile floras. Pp. 97 128 in N. Rajakaruna, R. Boyd, and T. Harris

PAGE 57

! ! 57 ! (eds.), Plant Ecology and Evolution in Harsh Environments. Nova Science Publishers, Hauppauge, NY. Muldavin E . H. 2002. Some f loristic c haracteristics of the n orthern Chihuahuan Desert: A s earch for i ts n orthern b oundary. Taxon. 51(3):453 462. International Association for Plant Taxonomy. Web of Science. doi:10.2 307/1554858. Neiswenter S . A . , B. R. Riddle. 2010. Diversification of the Perognathus flavus species group in emerging arid grasslands of western North America. Journal of Mammalogy. 91: 348 362. Olivier R . D . S . , Aranda R. 2017. Potential g eographic d istri bution n iche m odeling b ased on b ioclimatic v ariables of t hree s pecies of Temnomastax Rehn and Rehn, 1942 (Orthoptera: Eumastacidae). Journal of Natural History. 51(21 22): 1197 1208. Web of Science. doi:10.1080/00222933.2017.1324647 Olson, D.M., E. Diners tein, E.D. Wikramanayake, N.D. Burgess, G.V.N. Powell, E.C. Underwood, J.A. D'Amico, I. Itoua, H.E. Strand, J.C. Morrison, C.J. Loucks, T.F. Allnutt, T.H. Ricketts, Y. Kura, J.F. Lamoreux, W.W. Wettengel, P. Hedao, and K.R. Kassem. Terrestrial Ecoregions o f the World: A New Map of Life on Earth (PDF, 1.1M) BioScience 51:933 938 Omernik, J . M . , Griffith G . E. 201 4 . Ecoregions of the conterminous United States: evolution of a hierarchical spatial framework. PubMed NCBI . [online] Ncbi.nlm.nih.gov. Available at: https://www.ncbi.nlm.nih.gov/pubmed/25223620 Passalacqua N . G. 2015. On the d efinition of e lement, c horotype and c omponent in b iogeography. Journal of Biogeography. onlinelibrary.wiley.com/doi/10.1111/jbi.12473. Peterson, A. T. 2001. Predicting s peci es g eographic d istributions b ased o n e cological n iche m odeling. The Condor, 103(3), 599. doi:10.1650/0010 5422(2001)103[0599:psgdbo]2.0.co;2 Peterson, A . T., & Sober—n, J. 2012. Species d istribution m odeling and e cological ni che m odeling: Getting the c once pts r ight. Natureza & Conserva‹o, 10(2), 102 107. doi:10.4322/natcon.2012.019 Phillips S . J . , Dud’k M . , Schapire R . E. Maxent software for modeling species niches and distributions (Version 3.4.1). http://biodiversityinformatics.amnh.org/open_source/maxent/ . Phillips, S . J., Anderson , R . P, & Schapire, R . E. 2006. Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190(3 4), 231 259. doi:10.1016/j.ecolm odel.2005.03.026 Rajakaruna N. 2004: The e daphic f actor in the o rigin of p lant s pecies. International Geology Review, 46:5, 471 478 Richerson P . J, Lum K . L. 1980. Patterns of plant species diversity in California: relation to weather and top ography. Am. Nat.116:504 36

PAGE 58

! ! 58 ! Rondeau, R., Pearson, K. and Kelso, S. 2013. Vegetation r esponse in a C olorado g rassland shrub c ommunity to e xtreme d rought: 1999 2010. The American Midland Naturalist, 170(1), pp.14 25. Ryberg WA, Fitzgerald L.A. 2016. Lands cape c omposition, n ot c onnectivity, d etermines m etacommunity s tructure across m ultiple s cales. Ecography (Copenhagen).39(10):932 941. nri.tamu.edu/publications/peer reviewed 17 publications/2016/landscape composition notconnectivity determines metacommunit ystructure across multiple scales/. Villarreal Quintanilla JA, Bartolome Hernandex JA, Estrada Castillon E, Ramirex Rodriguez H, Martinez Amador SJ. 2017. The e ndemic e lement of the Chihuahuan Desert v ascular fl ora. Acta Botanica Mexicana 118 : 65 96. Pri nt. Villase–or, J. 1990. The g enera of Asteraceae e ndemic to Mexico and a djacent r egions. Aliso, 12(4), 685 692. doi:10.5642/aliso.19901204.04 Wainwright, J, Mulligan M. 2013. Chapter 2 Modelling and Model Building. Environmental Modelling: Finding Simpl icity in Complexity. Wiley Blackwell. 2:7 24. Warren, DL., Glor, RE., & Turelli, M. 2008. Environmental n iche e quivalency v ersus c onservatism: Quantitative a pproaches t o n iche e volution. Evolution, 62(11), 2868 2883. doi:10.1111/j.1558 5646.2008.00482.x Weber WA. 1965. Plant geography in the Southern Rocky Mountains, in the Quaternary of the United States: Princeton, N.J., Princeton Univ. Press, p. 453 468. Weber WA, Wittmann RC. 2001. Colorado Flora: Western Slope, 3rd ed. University Press of Colorado , Niwot. Wood DA, Vandergast AG, Barr KR, Inman DR, Esque CT, Nussear KE, Fisher NR. 2013. Comparative p hylogeography r eveals d eep l ineages and r egional e volutionary h otspots in the Mojave and Sonoran Deserts. Diversity and Distributions. 19(7):722 737. d oi:10.1111/ddi.12022 Wood ward FI. 1986. Climate and Plant Distribution. Cambridge University Pres s. Only Modelling Using MAXENT: When Can We Trust the besjournals.onlinelibrary.wiley.com/doi/full/10.1111/2041 210x.12004. Young, N., Carter, L. and Evangelista, P. (2018). A MaxEnt Model v3.3.3e Tutorial (ArcGIS v10) . http://ibis.colostate.edu/webcontent/ws/coloradoview/tutorialsdownloads/a_maxent_model_v7.pdf [Accessed 14 Mar. 2018].

PAGE 59

! ! 59 ! APPENDIX Table A. 1 Original Bioclim dataset source and properties ! " File Name Variable Source Original Resolution/ Projection Notes Bio_1 Annual Mean Temperature WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_2 Mean Diurnal Range (Mean of monthly(max temp min temp) WorldC lim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_3 Isothermality (Bio2/Bio7)(*100) WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_4 Temperature Seasonality (stand. dev *100) WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_5 Max Temperature of Warmest Month WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_6 Min Temperature of Coldest Month WorldClim No projection WG S 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_7 Temperature Annual Range (Bio5 Bio6) WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_8 Mean Temperature of Wettest Quarter WorldClim No projection WGS 84 Pr ojected to NAD_1983_2011_ Contiguous_USA_Albers Bio_9 Mean Temperature of Driest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_10 Mean Temperature of Warmest Quarter WorldClim No projection WGS 84 Proj ected to NAD_1983_2011_ Contiguous_USA_Albers Bio_11 Mean Temperature of Coldest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_12 Annual Precipitation WorldClim No projection WGS 84 Projected to NAD_1983_201 1_ Contiguous_USA_Albers Bio_13 Precipitation of Wettest Month WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_14 Precipitation of Driest Month WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA _Albers Bio_15 Precipitation Seasonality(Coefficie nt of Variation) WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers

PAGE 60

! ! 60 ! File Name Variable Source Original Resolution/ Projection Notes Bio_16 Precipitation of Wettest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ C ontiguous_USA_Albers Bio_17 Precipitation of Driest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Bio_18 Precipitation of Warmest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA _Albers Bio_19 Precipitation of Coldest Quarter WorldClim No projection WGS 84 Projected to NAD_1983_2011_ Contiguous_USA_Albers Table A. 2 Sources and properties of boundary polygons. File Name Layer Source Original Res olution/Projection Notes Co_eco_14 EPA Ecoregions Environmental Protection Agency North American 1983 , USA Contiguous Albers Equal Area Conic USGS Terrestrial Ecoregions of the World Chihuahuan Desert World Wildlife Fund /Data Basin GCS_WGS_1984 USA S tates Generalized Southwest U.S. State Boundary Esri , TomTom, US Census Bureau, US Dept. of Commerce GCS_WGS 1984 S hapefile created from selection of CO, NM, NV, TX, OK, CA, UT State Boundaries of Mexico Mexico Esri GCS_WGS_1984 Merged with US Southwest s hapefile CO_counties CO Counties Esri , US Census Bureau GCS_ WGS _1984

PAGE 61

! ! 61 ! Table A. 3 Variables, attribute preference scores, and associated factor weights used in the WLC calculations for Muhlenbergia arenicola . Muhlenbergi a arenicola Layer Attributes Preference Score Factor Weight Bio 11 ( 9) to ( 2) ( 1.9) to ( 1.0) 0 to 5.9 6 to 10.9 11 to 12.9 13 to 20.9 2 3 6 5 7 1 "#$$%%&$ ! Bio7 17 to 24.9 25 to 29.9 30 to 46.9 4 6 5 "#''(") ! Bio6 ( 19) to ( 11.0) ( 10.0) to ( 1) 0 to 4.9 5 to 5.9 6 to 13.9 3 5 6 2 1 "#$$%%&$ ! Bio4 150 to 353.9 354 to 404.9 405 to 506.9 507 to 1011 6 7 4 5 "#"")$%) ! Bio3 31 to 41.9 42 to 61.9 62 to 65.9 66 to 75.9 3 1 4 6 "#%"$''* !

PAGE 62

! ! 62 ! Table A. 4 Variables, attribute preference scores, and associated factor weights used in the WLC calculations for Funastrum crispum . Funastrum crispum Layer Attributes Preference Score Factor Weight Bio 19 19 to 62.9 63 to 73.9 74 to 84.9 85 to 128.9 129 to 139.9 140 to 261.9 4 2 3 6 5 7 0.034482759 Bio 18 11 to 214.9 215 to 265.9 266 to 629.9 4 7 6 0.027586207 Bio 16 26 to 76.9 77 to 127.9 128 to 280.9 281 to 331.9 332 to 678.9 2 3 5 8 6 0.055172414 Bio 12 63 to 163.9 164 to 264.9 265 to 770.0 771 to 1073 1 2 5 6 0.017241379 Bio 2 9 to 11.9 12 to 16.9 17 to 19.9 20 to 22.9 8 6 4 2 0.865517241

PAGE 63

! ! 63 ! Table A. 5 Variables, attribute preference scores, and associated factor weights used in the WLC calculations for Palafoxia sphacelata . Palafoxia sphacelata Layer Attributes Preference Score Factor Weight Bio 16 93 to 147 148 to 193 194 to 373 7 4 3 "#"(&%'' ! Bio 1 5 17 to 49.9 50 to 71.9 72 to 82.9 83 to 115.9 1 5 7 4 "#%+%(& ! Bio 1 2 100 to 199 200 to 299 300 to 599 600 to 699 700 to 1075 8 7 4 2 1 "#"%"+ $' ! Bio 3 Bio 2 32 to 35.9 36 to 38.9 39 to 47.9 48 to 49.9 50 to 57.9 10 to 13.9 14 to 14.9 15 to 15.9 16 to 18.9 19 to 22.9 2 6 5 4 3 1 3 4 5 6 "#+$&%(' ! "#$(%,%+ !

PAGE 64

! ! 64 ! Table A. 6 Variables, attribute preferen ce scores, and associated factor weights used in the WLC calculations for Nolina texana . Nolina texana Layer Attributes Preference Score Factor Weight Bio 19 10 to 29.9 30 to 51.9 52 to 62.9 63 to 84.9 85 to 128.9 129 to 175.9 5 4 3 2 7 9 0.021994135 Bi o 15 31 to 41.9 42 to 52.9 53 to 63.9 64 to 74.9 75 to 85.9 86 to 118.9 10 3 1 5 6 4 0.065982405 Bio 2 10 to 12.9 13 to 15.9 16 to 16.9 17 to 18.9 19 to 22.9 9 6 5 3 2 0.91202346

PAGE 65

! ! 65 ! Table A. 7 Variables, attribute preference scores, and associated factor weights used in the WLC calculations for Muhlenbergia arenacea . Muhlenbergia arenacea Layer Attributes Preference Score Factor Weight Bio 18 79 to 144.9 145 to 166.9 167 to 299.9 5 6 4 0.4101440 2 Bio 16 80 to 202.9 203 to 226.9 227 to 334 5 6 4 0.293675642 Bio 13 Bio 12 Bio 10 Bio 8 2 to 7.9 8 to 13.9 1 4 to 40.9 192 to 237.9 238 to 344.9 345 to 395.9 396 to 708 9 to 18.9 19 to 20.9 21 to 21.9 22 to 29.9 9 to 16.9 17 to 21.9 22 to 28.9 7 5 4 3 5 6 4 3 6 7 4 3 6 4 0.010644959 0.247964934 0.026299311 0.011271133

PAGE 66

! ! 66 ! Table A. 8 EPA Ecoregion Level III, IV, and plant habit type for each species identified. SCIENTIFIC NAME FAMILY LEVE L III KEY LEVEL IV KEY PLANT HABIT Andropogon saccharoides Sw. Poaceae 25 High Plains 25d Flat to Rolling Plains Gramminoid 26 Southwestern Tablelands 26f Mesa de Maya/Black Mesa Argemone squarrosa Papaveraceae 26 Southwestern Tablelands 26 e Piedmont Plains and Tablelands Forb Asclepias macrotis Torr. Apocynaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Subshrub 26g Purgatoire Hills and Canyons Berlandiera lyrata Asteraceae 25 High Plains 25c Moderate Relief Plains Forb 26 Southwestern Tablelands 25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa Bouteloua barbata Lag. Poaceae 26 Southwestern Tablelands 26e Piedmont P lains and Tablelands Gramminoid Bouteloua eriopoda (Torr.) Torr. Poaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Gramminoid 26g Purgatoire Hills and Canyons Cheilanthes eatonii Pteridaceae 21 Southern Rockies 21d Foo thill Shrublands Forb 25 High Plains 25c Moderate Relief Plains 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons 25c Moderate Relief P lains 26f Mesa de Maya/Black Mesa Cheilanthes wootonii Maxon Pteridaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Forb 26g Purgatoire Hills and Canyons Cucurbita foetidissima Cucurbitaceae 20 Colorado Plateaus 20c Semiarid Benchlands and Canyonlands Forb 21 Southern Rockies 21c Crystalline Mid Elevation Forests 25 High Plains 21d Foothill Shrublands 26 Southwestern Tablelands 21f Sedimentary Mid Elevation Forests 25b Rolling Sand Plains 25c Moderate Relief Plains

PAGE 67

! ! 67 ! SCIENTIFIC NAME FAMILY LEVE L III KEY LEVEL IV KEY PLANT HABIT 25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons 26h Pinyon Juniper Woodlands and Savannas 26j Foothill Grasslands 26k Sandsheets Cyperus fendlerianus Cyperaceae 21 Southern Rockies 21b Crystalline Subalpine Forests Gramminoid 26 Southwestern Tablelands 21c Crystalline Mid Elevation Forests 21d Foot hill Shrublands 21f Sedimentary Mid Elevation Forests 26j Foothill Grasslands Dalea lanata var. lanata Fabaceae 26 Southwestern Tablelands 26g Purgatoire Hills and Canyons Forb Desmanthus cooleyi (Eaton) Branner & Coville Fa baceae 25 High Plains 25b Rolling Sand Plains Subshrub Erigeron colomexicanus Asteraceae 20 Colorado Plateaus 20c Semiarid Benchlands and Canyonlands Forb 21 Southern Rockies 21c Crystalline Mid Elevation Forests 22 Arizona/N ew Mexico Plateau 21d Foothill Shrublands 25 High Plains 21e Sedimentary Subalpine Forests 21f Sedimentary Mid Elevation Forests 22b San Luis Alluvial Flats and Wetlands 25d Flat to Rolling Plains 25l Front Range F ans 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons 26j Foothill Grasslands Funastrum crispum Apocynaceae 21 Southern Rockies 21d Foothill Shrublands Forb 26 Southwestern Tablelands 21d Foothill Shrublands 21d Foothill Shrublands

PAGE 68

! ! 68 ! SCIENTIFIC NAME FAMILY LEVE L III KEY LEVEL IV KEY PLANT HABIT 26f Mesa de Maya/Black Mesa 26f Mesa de Maya/Black Mesa Krameria lanceolata Krameriaceae 25 High Plains 25c Moderate Relief Plains Forb 26 Southwestern Tablelands 25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa Mimosa borealis A.Gray Fabaceae 25 High Plains 25c Moderate Relief Plains Shrub Muhlenbergia porteri Scribn. Poace ae 26 Southwestern Tablelands 26g Purgatoire Hills and Canyons Gramminoid Muhlenbergia repens (J.Presl) Hitchc. Poaceae 21 Southern Rockies 21c Crystalline Mid Elevation Forests Gramminoid Nolina texana Asparagaceae 25 High Plains 25d Flat to Rolling Plains Gramminoid 26 Southwestern Tablelands 26f Mesa de Maya/Black Mesa Notholaena standleyi Pteridaceae 25 High Plains 25d Flat to Rolling Plains Forb 26 Southwestern Tablelands 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons Physaria fendleri Brassicaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Forb 26g Purgatoire Hills and Canyons 26h Pinyon Juniper Woodlands and Savannas Schoenocrambe linear ifolia Brassicaceae 20 Colorado Plateaus 20a Monticello Cortez Uplands Forb Subshrub 21 Southern Rockies 20c Semiarid Benchlands and Canyonlands 22 Arizona/New Mexico Plateau 21d Foothill Shrublands 26 Southwestern Tablela nds 21f Sedimentary Mid Elevation Forests 21h Volcanic Mid Elevation Forests 22a San Luis Shrublands and Hills 26e Piedmont Plains and Tablelands 26h Pinyon Juniper Woodlands and Savannas

PAGE 69

! ! 69 ! SCIENTIFIC NAME FAMILY LEVE L III KEY LEVEL IV KEY PLANT HABIT Sporobolus nealleyi Vasey Poacea e 20 Colorado Plateaus 20c Semiarid Benchlands and Canyonlands Gramminoid Thymophylla aurea Asteraceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Forb 26g Purgatoire Hills and Canyons Palafoxia sphacelata Aster aceae 21 Southern Rockies 21 c Crystalline Mid Elevation Forests Forb 25 High Plains 25b Rolling Sand Plains 26 Southwestern Tablelands 25c Moderate Relief Plains 25d Flat to Rolling Plains 25l Front Range Fans 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26k Sandsheets Muhlenbergia arenacea Poaceae 21 Southern Rockies 21d Foothill Shrublands Gramminoid 26 Southwestern Tablelands 26e Piedmont Plains and Tableland s 26g Purgatoire Hills and Canyons 26h Pinyon Juniper Woodlands and Savannas Muhlenbergia arenicola Poaceae 25 High Plains 25b Rolling Sand Plains Gramminoid 26 Southwestern Tablelands 25c Moderate Relief Plains 25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands 26f Mesa de Maya/Black Mesa 26g Purgatoire Hills and Canyons 26k Sandsheets Croton texensis Euphorbiaceae 25 High Plains 25b Rolling Sand Plains Forb 26 Southwestern Tablelands 25c Moderate Relief Plains 25d Flat to Rolling Plains 26e Piedmont Plains and Tablelands Cylindropuntia imbricata (Haw.) F.M.Knuth Cactaceae 20 Colorado Plateaus 20c Semiarid Benchlands and Canyonlands Shrub 21 Southern Rockies 21c Crystalline Mid Elevation Forests 26 Southwestern Tablelands 21d Foothill Shrublands 21h Volcanic Mid Elevation Forests

PAGE 70

! ! 70 ! SCIENTIFIC NAME FAMILY LEVE L III KEY LEVEL IV KEY PLANT HABIT 26e Piedmont Plains and Tablelands 26g Purgatoire Hills and Canyons 26j Foothill Grasslands Dalea aurea C.Fraser Fabaceae 25 High Plains 25c Moderate Relief Plains Forb/Herb Dalea aurea Pursh Fabaceae 25c Moderate Relief Plains Forb/Herb 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands 26j Foothill Grasslands Dalea nana Torr. ex A.Gray Fabaceae 25 High Plains 25b Rolling Sand Plains Forb/Herb Gaillardia pulchella Foug. Asteraceae 21 Southern Rockies 21c Crystalline Mid Elevation Forests Forb/Herb 22 Arizona/New Mexic o Plateau 21d Foothill Shrublands 25 High Plains 21f Sedimentary Mid Elevation Forests 26 Southwestern Tablelands 21h Volcanic Mid Elevation Forests 22c Salt Flats 25c Moderate Relief Plains 25d Flat to Rolling Plains 25l Front Range Fans 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands 26i Pine Oak Woodlands 26j Foothill Grasslands Juniperus monosperma (Engelm.) Sarg. Cupressaceae 20 Colorado Plateaus 20b Shale Deserts a nd Sedimentary Basins Tree 21 Southern Rockies 20c Semiarid Benchlands and Canyonlands 26 Southwestern Tablelands 20e Escarpments 21d Foothill Shrublands 21f Sedimentary Mid Elevation Forests 26e Piedmont Plains and T ablelands Lesquerella fendleri (A.Gray) S.Watson Brassicaceae 26 Southwestern Tablelands 26e Piedmont Plains and Tablelands Forb/Herb 26g Purgatoire Hills and Canyons

PAGE 71

! ! 71 ! Figure A. 1 Chart comparing th e number of occurrences by taxonomic family.

PAGE 72

! ! 72 ! ! Figure A. 2 Chart of comparison of number of occurrences per Ecoregion Level III. Figure A. 3 Proportion of Apocynaceae per Ecoregion Level III. ( &" ' %%% ')( " (" %"" %(" '"" '(" $"" $(" '"!!-./.012.!3/145167 '%!!8.64950:!;.<=>57 ''!!?0>@.:1AB5C!D5E><.!3/14516 '(!!F>G9!3/1>:7 '*!!8.649C57450:!H1I/5/1:27 B6JI50!.K!L<<6005:<57 M<.05G>.:!N5O5/!PPP

PAGE 73

! ! 73 ! " ! Figure A. 4 Proportion of Asoaragaceae per Ecoregion Level III.

PAGE 74

! ! 74 ! ! Figure A. 5 Proportion of Asteraceae per Ecoregion Level III.

PAGE 75

! ! 75 ! ! Figure A. 6 Proportion o f Brassicaceae per Ecoregion Level III.

PAGE 76

! ! 76 ! ! Figure A. 7 Proportion of Cucurbitaceae per Ecoregion Level III.

PAGE 77

! ! 77 ! ! Figure A. 8 Proportion of Cyperaceae per Ecoregion Level III. " ! Figure A. 9 Proportion of Fabaceae per Ecoregion Level III. '(#""""""""""" ("#""""""""""" '(#""""""""""" Q1I1<515 '(!!F>G9!3/1>:7!'(I!!;.//>:G!81:2!3/1>:7!Q1I1<515 '(!!F>G9!3/1>:7!'(5K!3/1>:7!Q1I1<515 '*!!8.649C57450:!H1I/5/1:27!'*G!!360G14.>05!F>//7!1:2!-1:R.:7!Q1I1<515

PAGE 78

! ! 78 ! ! Figure A. 10 Proportion of Krameriaceaee per Ecoregion Level III. " ! Figure A. 11 Proportion of Poaceae per Ecoreg ion Level III. '*#*********," %$#$$$$$$$$$$" '"#""""""""""" +"#""""""""""" S01J50>1<515!M<.05G>.:!;5T0575:414>.:7 '(!!F>G9!3/1>:7!'(5K!3/1>:7!S01J50>1<515 '(!!F>G9!3/1>:7!'(2!!Q/14!4.!;.//>:G!3/1>:7!S01J50>1<515 '*!!8.649C57450:!H1I/5/1:27!'*5!!3>52J.:4!3/1>:7!1:2!H1I/5/1:27!S01J50>1<515 '*!!8.649C57450:!H1I/5/1:27!'*K!!D571!25!D1R1AU/1<=!D571!S01J50>1<515 ,#*)'$",*)'$% ,#*)'$",*)'$% ,#*)'$",*)'$% $"#,*)'$",*)'" ,#*)'$",*)'$% $&#+*%($&+*%(" 3.1<515!M<.05G>.:!;5T0575:414>.:7 '"!!-./.012.!3/145167!'"10>2!U5:<9/1:27!1:2!-1:R.:/1:27!3.1<515 '%!!8.64950:!;.<=>57!'%:5!D>2VM/5O14>.:!Q.05747!3.1<515 '(!!F>G9!3/1>:7!'(2!!Q/14!4.!;.//>:G!3/1>:7!3.1<515 '*!!8.649C57450:!H1I/5/1:27!'*5!!3>52J.:4!3/1>:7!1:2!H1I/5/1:27!3.1<515 '*!!8.649C57450:!H1I/5/1:27!'*K!!D571!25!D1R1AU/1<=!D571!3.1<515 '*!!8.649C57450:!H1I/5/1:27!'*G!!360G14.>05!F>//7!1:2!-1:R.:7!3.1<515

PAGE 79

! ! 79 ! ! Figure A. 12 Proportion of Pteridaceae per Ecoregion Level III. %)#'$",*)'$"&" (#,*)'$",*)'$ (#,*)'$",*)'$ (#,*)'$",*)'$ (,#*)'$",*)'$" (#,*)'$",*)'$ 3450>21<515 '%!!8.64950:!;.<=>57!'%2!!Q..49>//!8906I/1:27!3450>21<515 '(!!F>G9!3/1>:7!'(5K!3/1>:7!3450>21<515 '(!!F>G9!3/1>:7!'(2!!Q/14!4.!;.//>:G!3/1>:7!3450>21<515 '*!!8.649C57450:!H1I/5/1:27!'*5!!3>52J.:4!3/1>:7!1:2!H1I/5/1:27!3450>21<515 '*!!8.649C57450:!H1I/5/1:27!'*K!!D571!25!D1R1AU/1<=!D571!3450>21<515 '*!!8.649C57450:!H1I/5/1:27!'*G!!360G14.>05!F>//7!1:2!-1:R.:7!3450>21<515

PAGE 80

! ! 80 ! Figure A. 13 RFF outputs of co variates of Muhlebnergia arenicola ! "

PAGE 81

! ! 81 ! ! Figure A. 14 RFF output for variables associated with Palafoxia sphacelata .

PAGE 82

! ! 82 ! ! Figure A. 15 RFF outputs for co variates associated with Funastrum crispum .

PAGE 83

! ! 83 ! ! Figure A. 16 RFF outputs for c o variates associated with Nolina texana .

PAGE 84

! ! 84 ! ! Figure A. 17 RFF outputs for co variates associated with Muhlenbergia arenacea . "

PAGE 85

! ! 85 ! Table A. 9 List of species that support Weber's original list identified as plants of the Chihuahuan sub element. Species Identified by Weber Identified through GIS Abutilon incanum (Link) Sweet Andropogron saccharoides Sw. x Argemone squarrosa Greene x Ascelpias macrotis Torr. x A sclepias oenoth eroides C h am. & Schlecht. Asplenium resiliens Kunze Berlandiera lyrata Benth. x Bouteloua barbata Lag. x B outeloua . eriopoda (Torr) Torr. x Cheilanthes eatonii Baker x C. wootonii Maxon x Croton texensis (Klotsch) Muell. Arg. x Cucurbita foetidiss ima H.B.K x Dalea lanata Spreng. x D alea nana Torr. x Desmanthus cooleyi (Eaton) Trel. x Engelmannia pinnatifida T.& G. Eragrostis oxylepis (Torr .) Torr. Eriochloa contracta Hitch. Gaillardia pulchella Foug. x Hoffmanseggia densiflora Benth. x H offmanseggia d r epanocarpa A. Gray x H offmanseggia jamesii T&G. x Juniperus monosperma (Engelm.) Sarg. x Krameria spp. x Melampodium cinereum DC. Mimosa borealis A. Gray x Palafoxia spp. x Pericome caudata A. Gray x Sapindus saponaria L. Sar costemma crispum Benth x Stillingia sylvatica Gard. Thelesperma spp.

PAGE 86

! ! 86 ! ! Figure A. 18 Flowchart documenting GIS methods to build models for habitat suitability models.

PAGE 87

! ! 87 ! Figure A . %&!<.:42 .

PAGE 88

! ! 88 ! Figure A . %&!<. :42#!

PAGE 89

! ! 89 ! Figure A. 1 9 MaxEnt jackknifed training gains for modeled species.