Page  1 ï~~2007 THE MICHIGAN BOTANIST MAPPING HERBARIUM SPECIMENS: A CASE STUDY USING LOCALITY INFORMATION FROM THE UNIVERSITY OF MICHIGAN HERBARIUM'S MICHIGAN FLORA DATABASE Rachel A. Simpson Shapiro Science Library, The University of Michigan, Ann Arbor, MI 48109-1185 Current Address: School of Natural Resources, University of Nebraska-Lincoln, Lincoln, NE 68588-0973 402-472-1319 ABSTRACT Museum specimen locality data can be visualized and analyzed using Geographic Information System (GIS) technology. Explicit and standardized references to geographic locations are required for locality information to be compatible with GIS. This paper describes the procedure used to generate information that enabled mapping of over 52,000 herbarium specimens documented in the University of Michigan Herbarium Michigan Flora database. A benefit of the procedure is the ability to process large amounts of data rapidly. Future users of these data must be cognizant of the methods by which the data were collected, especially so that they are aware of sources of locational error. The method is suitable for generating distributional data for use at a state or regional scale or for preliminary assignment of approximate locations that can be refined further as needed. KEYWORDS: geographic information systems; digital mapping; herbaria; natural history specimens; natural history collections; spatial accuracy INTRODUCTION Geographic Information System (GIS) technology permits the display and analysis of information with a spatial component. A GIS allows one to organize and query data based on geographic and other attributes, visually represent data in a variety of ways, rapidly create and update maps, access and edit databased information associated with a mapped feature, and perform spatial analyses. Explicit and standardized references (textual and/or quantitative) to geographic locations enable the use of GIS with museum specimens [e.g., for herbarium specimens of Ontario (Meades et al. 2007) and Wisconsin (Anonymous 2004; and Glen Barry and Jill Rosenberg, personal communications, 2007)]. More recent collections reference geographic coordinates captured by Global Positioning System technology, which can provide precise and accurate locational information as latitudinal and longitudinal coordinates. However, for historical collections, localities are not written according to any consistently followed convention [for examples pertaining to specimens from Michigan, see Voss (1999, 2005)]. Therefore a major hurdle in the application of GIS technology to museum specimens is translating the descriptive locality information into mappable coordinates. This paper describes the methods used to develop a spatial compo

Page  2 ï~~THE MICHIGAN BOTANIST Vol. 46 nent for a subset of specimens documented in the University of Michigan Herbarium's Michigan Flora database and briefly characterizes the resulting data. The Michigan Flora database, created by Edward G. Voss, presently Curator Emeritus, University of Michigan Herbarium, beginning in 1956 and continuing through the present, contains information from specimens of Gymnosperms (conifers and relatives) and Angiosperms (flowering plants) from Michigan that Voss examined in the process of writing the Michigan Flora (Voss 1972, 1985, 1996). The specimens were collected beginning in the early nineteenth century and are from over 20 herbaria. Until 1991 this information was manually typed on paper slips which were then filed. Since then data have been entered into a computer database program (Foxbase for Macintosh, Foxbase Software Inc., Perrysburg, Ohio) from which paper slips fully compatible in format with the typed ones are printed and inserted in the files. The records in the Michigan Flora database represent what was written by the collector on the labels. They also include annotations by specialists, such as those confirming or altering the taxonomic identification. Missing or incorrect information is often noted, which adds greatly to the quality of locational data as well. For example, for 20% of the records used in the current study, the county (or county unit; seven islands/island groups are distinguished in the data from the mainland portions of the counties) was not explicitly stated on the specimen label, but the presumed county was added as an annotation by Voss. Annotations are placed in brackets and therefore distinguished from the original information, so the integrity of that original data is maintained, as in the following locality taken verbatim from the Michigan Flora database: Mackinac along the Straites of Mackinac west of Pointe aux Chenes, Sect. 22, T 40N, R 4W. [T41N, R5W!] This investigation was based on the 56,776 computerized Michigan Flora database records available as of April 2002. These records pertain to families included in Part III of the Michigan Flora as well as new accessions in all families entered 1991-2002. The county-level distributional data are already available through the Michigan Flora (Voss 1972, 1985, 1996) and a more recently updated website (Reznicek et al. 2004). In those maps, the collection of one or more specimens from a county is represented in map form as a dot superimposed on the county (or on part of a county in the case of the seven designated islands/island groups). The work described here was undertaken in order to make finer-scale mapping and analyses more feasible and to create a system in which other benefits of GIS could be leveraged, such as access to the specimen-label data via a map interface. METHODS The major component of the work was to associate locality descriptions with mappable coordinates. Pre-processing of data included data standardization (for example of county name) and creation of new fields to hold the concatenation of the multiple fields used in the original data to hold locality information [multiple fields were used because the system was set up prior to the advent of

Page  3 ï~~2007 THE MICHIGAN BOTANIST computer databases, and the paper slips used had fixed amounts of space for each field. Furthermore, when a computer database was eventually adopted, the fields also had limitations on the number of characters. Thus, in order to ensure that all the text from the labels be included in the database, text sometimes is placed in multiple fields (E.G. Voss, personal communication).] Following data preparation, the specific methods used depended on the type of information available in the locality. When available a U.S. Public Land Survey (PLS) section coordinate was extracted and standardized. It was then associated with a latitude and longitude for the centroid, the geographical center as calculated using GIS software, of the section, using a reference data set for the state maintained by the Michigan Department of Natural Resources. For the majority of records, however, coordinates were assigned by associating the specimen record with a geographic feature found in an electronic gazetteer. The electronic gazetteer used was the Geographic Names Information System (GNIS), maintained by the U.S. Geological Survey (U.S. Geological Survey 1981). Each feature in this gazetteer is associated with an x and y coordinate in units of decimal degrees. This coordinate is called the 'primary point.' For some feature types (e.g., streams and valleys) additional coordinates are included as well. The primary point of a linear feature depends on the type of feature. For streams, for example, the primary point is the stream's mouth. For areal features, the primary point is the approximate geographic center, with some exceptions. For example, "the primary point of a populated place is the center of the original place such as the city or town hall, main post office, or town square regardless of changes over time (U.S. Geological Survey 1981)." The subset of GNIS data pertaining to Michigan comprised 32,056 unique features. The selection of a feature in the electronic gazetteer was accomplished using the database software application 4D (4th Dimension, San Jose, CA), customized to permit a combination of automated and manual matching of each specimen locality with place name(s) for the relevant county in the reference gazetteer. Scripts that run from within the application find any matches in the reference database. The scripts rely on a simple system of pattern matching. Each entry in the GNIS for the county listed for a particular specimen is compared with the locality itself, and if any string of text in the locality description exactly matches a given entry in the GNIS, that entry is flagged as a potential match. This automated candidate match selection occurs for the entire set of specimen data and requires no user input. Once that step is completed, the user can review and modify the selection(s) for each record by means of a simple interface. For example, when multiple matches for a particular locality occur, the user can select the most appropriate one with a computer-mouse click. When no match is found, or if none of those found automatically are appropriate, the user can easily retrieve the list of places for the county (or other counties if desired) and select an appropriate match if one exists. This feature is helpful for example when there are different versions or spellings of a place name. Of the 22,000 unique localities (i.e., a unique combination of county and place) included in the specimen data, the system found one or more candidate matches automatically for 15,000. Whether any matches were found by the automated system or not, each record for which there was no PLS coordinate was briefly reviewed. The review was supported by a GIS (using ArcView 3.2, ESRI, Redlands, California) customized for the purpose, in which gazetteer localities could be displayed with county outlines, rivers, roads, and topographic maps. If, for example, a place name is used more than once within a county, the GIS could be used to quickly find the location with which the specimen is best matched, provided that the specimen label has the location information necessary to distinguish among options. State atlases and other references (Hanes and Hanes 1947, Romig 1972, Voss 1978) were consulted as well. Data were reviewed in order by county; this facilitated applying information about localities gleaned from other specimen records. When multiple place names were available, georeferencing was to the nearest named place. For example, for the locality "South Manitou Island, South End of Lake Florence" in Leelanau County, 'Florence Lake' was selected as the locality (despite the fact that it did not appear as a candidate match because of the difference in the order of the words 'Florence' and 'Lake' between the GNIS and the specimen record). Name changes, missing names, and spelling variations required investigation. For streams, the automated system related a stream name to its default location, the mouth. This set of coordinates was selected unless the locality specifically referenced the headwaters. Because streams generally cover a large linear distance, alternate feature types were used when these were expected to more accurately reflect the actual location. For example, if a stream and a town along a stream were named (e.g., 'along the banks of the Huron River at South Rockwood'), the town was generally selected.

Page  4 ï~~THE MICHIGAN BOTANIST Vol. 46 If specimen collection dates are available in addition to location, spatiotemporal analyses and mapping may be carried out. Therefore year of collection was extracted from the field in the original data which held date information exactly as written on the specimen. Simple tools available in Microsoft Excel (such as pattern-searching and filtering) made it possible to extract century and decade/year of the decade. For records for which century was not explicitly given, a century was assigned based on other dated records of the collector, botanical literature, and on the Harvard University Herbaria's online database of botanical collectors ( cms-wb/botanist_index.html). Last, in order to make some inferences about sampling effectiveness and distributions of numbers of species, a grid with cells 30 km by 30 km was superimposed on the resulting data. The number of specimens and of species was calculated for each grid cell and divided by the area of the land circumscribed by the grid cell. Locations of major herbaria in Michigan cited in Voss (1985, 1996) as having been sources of specimens examined were also superimposed on the data, using the relevant GNIS coordinates. RESULTS AND DISCUSSION The overall result was that 52,967, 92%, of the 56,776 records were georeferenced to locations more specific than county. Township, range, and section (PLS coordinates) were parsed and georeferenced for 18,530 records (35% of the georeferenced records). Of the remaining records, a mappable feature name was associated with 34,437 records (65%). Mapped locations are shown in Fig. 1. For a technical description of the spatial data written according to the GIS community standard, the Federal Geographic Data Committee's Content Standard for Digital Geospatial Metadata, see Simpson 2006. More than half of the records were from specimens collected after 1950 and another third from 1900 to 1950. A century, decade, and year was assigned to or standardized for 96% of the records; in the original data a four-digit year was explicitly provided for only 37% of the records. The figure of 96% could undoubtedly be improved further with treatment by botanists familiar with the collectors and the specimens. Records which were not georeferenced range from those that would be impossible to georeference to better-than-county-level accuracy (or impossible to do so with confidence) to those that could easily be georeferenced with additional time. In the former category are records for which a county but no place description was given. In other cases, a place description was given, but a county was not named (either by the collector or in an annotation by Voss). In other cases, the collector's 'locality' was actually a habitat description and would be impossible to locate, e.g., 'plentiful in moist meadow' and 'in shallow water.' A few records provided names of places that appeared to be in a county other than that which the collector named. Unless a correction to the named county was given by Voss, coordinates were not assigned to these places. For still other records, multiple places by that name exist within the named county. In the course of this investigation many such records were annotated, e.g., a record for 'Little Lake' in Marquette County includes a note indicating that there are 3 places (all lakes) by that name in 2 different parts of the county, and on the previous day the collector has a record from 12 and 16 miles away from each loca

Page  5 ï~~2007 THE MICHIGAN BOTANIST 2007 THE MICHIGAN BOTANIST 0 100 - Miles * Estimated location 0 County boundary FIGURE 1. Locations associated with all Michigan specimens mapped to better than county-level accuracy. tion. With a review of the collector's field notes, if available, it might be possible to determine the correct location in cases such as these. In other cases place names were used which were not found in the various resources used, but with further research they might well be found. For example, the location of 'Peter White Camp, western part of Alger County' for a 1916 record is now easily discoverable through Voss 2005, published after the work described here was concluded. For about a quarter of the unmapped records annotations were added which might assist those interested in adding to the number of georeferenced records. Some unmapped records contained specific location information for which georeferencing was not supported by the methods described here. For example, many of the unmapped records did not include PLS coordinates or a place name, but did include road intersection/mileage data (e.g., "two miles N. of jct. of Mich.

Page  6 ï~~THE MICHIGAN BOTANIST Vol. 46 95 & U.S. 2" in Dickinson County) and thus can be suitably mapped with other methods. Records that provided a county with township and range numbers, but no section number, were not mapped, but certainly could be. Furthermore, township, range, and section information is written in many different ways by collectors, most but not all of which were accounted for in the automated methods used. For example, the township number was not captured for a record for which the township was written as "T.24.N." because the patterns searched for did not include those with periods between the township number and the direction. Spatial accuracy varies among the georeferenced records greatly, but one indicator of accuracy is the type of geographic entity referenced. For those records associated with a PLS section coordinate, the location will usually be accurate to within 0.72 miles (the sum of the distance from the center of a one-square-mile section to a section corner and the accuracy of the PLS reference data). This error estimate assumes that the PLS coordinate extracted is the true one. A precise error estimate would need to be calculated for each section and specimen individually, for several reasons. First, in cases where more than one section was stated, the first one was referenced arbitrarily, but this may or may not be the best representation of the true location. Second, sections adjacent to state boundaries and shoreline are not necessarily square. Third, while the intent of the Public Land Survey was for a section to be a standard size, in fact there is some variation because of surveying errors and adjustments (Stewart 1935). For those records for which the mapped location is that of a gazetteer place name, the spatial accuracy varies considerably. Unlike Survey sections, the sizes of towns (for example) are non-standard, and the sizes, boundaries, and centers of towns can change considerably over time. In addition, localities are often described as being a given distance from a geographic feature, and that distance may be an actual distance, a driving distance, or an estimated distance, and may be from the edge of a feature such as a town, the center of town, or some other reference point now impossible to ascertain. Generalizations about the magnitude of inaccuracy, however, may be made by considering the feature types of the place names matched with specimen localities. The top 10 feature types referenced, accounting for 96% of the gazetteer-based records, were, in order of frequency, populated place, civil division, lake, island, park, bay, stream, school, cape, and locale. Over half of gazetteer-based records were matched to a 'populated place (47%)' or a 'civil division (14%).' Of the top 10, or in fact of all, feature types referenced, 'stream (a category that includes both rivers and streams)' is expected to have the potential for yielding the least accurate location, because in some cases a stream or a section of stream in a county extends the distance of a county. Three percent (907) of the specimens were matched to the 'stream' feature type. For over 4,800 records the only available locality information was a township name and county, and for these the feature type was 'civil division.' The accuracy of the assigned location for many of these specimens will be approximately 8.5 miles (distance from the center of a township to a township corner). However, some changes have occurred in township extents, boundaries, and names, so a review of these records is warranted before making conclusions about the accuracy of specific records. The feature type assigned has been retained in the data, so that users can eliminate or work with selected feature types

Page  7 ï~~2007 THE MICHIGAN BOTANIST according to their needs in terms of accuracy or otherwise. A comparatively small source of positional inaccuracy is that of the GNIS itself, which is described as having no more than 10% of the points tested in error by more than 40 feet (U.S. Geological Survey 1981). A greater problem with the GNIS is that it is not a comprehensive database of places in the state. Certainly botanists could add to the list of lakes called 'Deer Lake,' for example. For such common place names, specimen locality has undoubtedly in a few cases been misassigned. There are a number of ways in which the spatial accuracy, the distance between the mapped location and the best possible assessment of the true location, can be improved with further research. For example, a simple improvement could be made by reviewing the records that reference a township, range, and section to extract more precise locational information, such as quarter-section, when it is available. A review of the data would also reveal cases where the most specific locality information was actually in the habitat field [either because the locality information would not all fit in the locality field or because the information serves both as a habitat description or locality (e.g., 'Evans field' or 'shore of Mud Lake' (examples per communication by E.G. Voss)]. In this particular case of the Michigan Flora data, consulting the specimen itself will in certain cases result in the ability to more accurately place the locality. This is because until recently latitude/longitude and Universal Transverse Mercator coordinates were not entered (E.G. Voss, personal communication). However, because there is no indication in the database of whether such data can be found on the label, retrieving this information would probably be seen as worthwhile only in particular cases. Another example would be to take advantage of the distance and bearing information sometimes provided, e.g., 'six miles northwest of the town of Pinckney.' Currently such a locality is mapped at the GNIS location for Pinckney (populated place). However, a GIS can be used to calculate a new coordinate for the given distance and bearing (or distance and direction along a route) from the GNIS coordinates. A set of such records can be processed as a group when the locational information is appropriately formatted. Furthermore, with on-screen review and editing of individual points in a GIS environment, particularly if done by specialists familiar with the localities, the accuracy can certainly be improved even further. Finally, given the volume of records addressed, there will be oversights and errors, and it is expected that these will be discovered and corrected by users of the data over time. Some errors will be easily discovered, while others more difficult to sort out. Errors that are the result of incorrect or misleading labels will be perhaps most difficult to discover and correct. For example, there are 15 specimens identified as being from Ingham County and north or northeast of the Agricultural College (now Michigan State University). While the county is the correct one for the College, some of these specimens were actually collected over the border in Clinton County (E.G. Voss, personal communication). Similarly, there are over 300 records of collections by C.K. Dodge catalogued as being from St. Clair County and "near Port Huron." These records of course become associated with the coordinates for Port Huron, St. Clair County. However, some of these specimens were collected not only from a different county, but from a different country: they were collected across the river in Canada (E.G. Voss, personal communication). In summary, these

Page  8 ï~~THE MICHIGAN BOTANIST Vol. 46 THE MICHIGAN BOTANIST Vol. 46 0 100 Miles Date of collection0" O 1911-1920O 0 O 1921-1940 0 O 1941 - 1960 d O o 1961 - 1980O * 1981 - 1996 L I County boundary FIGURE 2. Distribution of mapped localities for specimens of Centaurea stoebe (=Centaurea maculosa, spotted knapweed). Specimens mapped are from (or were at the time the information about them was captured in the Michigan Flora database) the private herbarium of William R. Overlease, the private herbarium of Edward G. Voss, and from herbaria at the following institutions: Albion College, Alma College, Andrews University, Central Michigan University, Cornell University, Cranbrook Institute of Science, Harvard University, Isle Royale National Park, Michigan State University, Missouri Botanical Garden, Morton Arboretum, Northern Michigan University, Seney National Wildlife Refuge, University of Michigan, University of Michigan Biological Station, University of Notre Dame, Wayne State University, and Western Michigan University. methods assume that the county named was indeed the county in which the collection was made. Review of the data in map form (for example, mapped by species, or mapped by collector and classified by date) by individuals familiar with the expected distributions will facilitate detection of some errors. Georeferencing rate after the initial automated matching of place names by the database was 170 per hour (N=5 time periods monitored, SD=94). By comparison, an entirely manual method used to georeference mammal collections was one tenth as fast, 17 (SD=8) records per hour (Stein and Wieczorek 2004).

Page  9 ï~~2007 THE MICHIGAN BOTANIST.2. 3{ 4, A A o R * Specimen locality,.... Riveri... < Counties in which species. has been collected }\. Lj Countyboundary FIGURE 3. Distribution of mapped localities for specimens of Solidago houghtonii (Houghton's goldenrod), superimposed on the county-level distribution of specimens. Mapped specimens from western Cheboygan County represent a cultivated individual or individuals (E. G. Voss, pers. comm.). Map excludes Crawford County collections, which may represent a distinct species. Specimens mapped are from (or were at the time the information about them was captured in the Michigan Flora database) herbaria at the following institutions: Butler University, Central Michigan University, Cornell University, Cranbrook Institute of Science, Harvard University, Michigan State University, Ohio State University, University of Michigan, University of Michigan Biological Station, Wayne State University, and Western Michigan University. One reason for the difference is that the goal of the latter project was to pinpoint the location as accurately as possible and to provide, with full documentation, a measure of the maximum inaccuracy. An example of a type of simple map that the data now support is shown in Figure 2, where locality data for Centaurea maculosa are classified by time period of collection. For an introduced species such as this, depiction of the spatiotemporal pattern of collections can provide insight into the history of invasion. Figure 3 is an example of how in some cases, even at a small scale a map

Page  10 ï~~10 THE MICHIGAN BOTANIST Vol. 46 10 THE MICHIGAN BOTANIST Vol. 46................... I W Ei'' i'. SENT,.... t\ w 0 100 -----r----- I Miles.. Species density (number of species recorded per sq. m. and area) in 30 km x 30 km grid ceils 0.08901-0 158v 0.1580o1 - o.347 0.347001- 13,264 Herbarium,\ FIGURE 4. Species density in 30 km by 30 km grid cells, as measured by the number of species recorded per square km of land area circumscribed by the grid cell. Superimposed are the locations in Michigan of major collections of the state's flora. Results for specimen density are similar and are therefore not shown. showing point locations rather than county-level distributions can convey some information about the species's requirements. While not obvious from the county-level distribution for Solidago houghtonii, its association with Great Lakes shoreline habitat is apparent from the distribution of localities. Taken as a whole, the data suggest that knowledge of the species occurring in many areas of Michigan could benefit from further collecting/documentation. Figure 4 suggests visually that certain areas, such as islands and certain coastal areas, have been better sampled by collectors than others. In addition the data suggest that botanical knowledge of areas further from major herbaria might especially benefit from further collecting. This apparent pattern could simply be an artifact of the use of just those Michigan Flora records that were available electronically. However, data for the number of species documented in each county,

Page  11 ï~~2007 THE MICHIGAN BOTANIST 11 2007 THE MICHIGAN BOTANIST 11 L I 2 r.,...... j... VXI L) f f t.:.: f.3y.,t...? W a N 0 100..... -- - M iles Number of spec ies recordedi n county in proportion to size of county in square kilometers 0.10 - 0.327 * 0327001 - 0448 4 0448001-w0>652 k + 4 f41,. ip K4 \ti. 0652001 - 54,635 nHerbarium. ountyboundary FIGURE 5. Number of species recorded by the complete Michigan Flora database as of March 2004 per county unit in proportion to the size of the unit in square kilometers. Superimposed are the locations of major collections in Michigan of the state's flora. Species richness data are from Reznicek et al. 2004 and are used with permission of the coauthors. County unit designations follow Voss (1972, 1985, 1996). using the complete Michigan Flora data set as of March 2004, are available, and a pattern of a higher number of species documented in areas near major herbaria (as well as on islands) is suggested visually by these data as well (Fig. 5). Finally, Figure 6 shows the relationship of the number of specimens documented in these 30 km by 30 km cells to the number of species recorded. Many of the cells show few specimen and species records. Based on the species richness and size of better documented counties, areas of this size would be expected to have well over 500 species (e.g., Kalamazoo and Washtenaw counties, with 1604 and 1603 species respectively, have 963 and 771 per 900 sq. km respectively). One limitation of the mapping method as applied here is that it uses only co

Page  12 ï~~12 THE MICHIGAN BOTANIST Vol. 46 900 800 - a)" = 700 E o 600 0 S 500 a) S400 -o 300- ** E 200. ' * 100 0 I 0 500 1000 1500 2000 2500 Number of specimens FIGURE 6. Relationship of the number of species documented over 30 km by 30 km grid cells to collection effort as measured by number of specimens collected. Displayed are data for all cells in which the land area circumscribed is within 0.2% of 900 sq. km. ordinates associated with places named in a federal database. However, one could add known botanical collecting localities to the GNIS data or use another gazetteer. Another limitation is that the GNIS currently supports the representation of places only as points, or at best a series of points. However, a component to the GNIS incorporating the data necessary to display geographic entities as multi-dimensional (e.g., a line feature representing a stream, or a polygon representing a populated place, lake or island) is reportedly in development (Payne and MacIntosh 2004). Such data would facilitate creation of a "footprint" over which a specimen is likely to have been collected. This would be especially valuable for the 15% of gazetteer-based records for which the specimens had no locality information more specific than county and place name. They would also enable automated calculations of accuracy values in multiple directions. Because at a later point it could be desirable to utilize the matched feature name and type, those data, rather than just its x and y coordinates, were retained in the final data set. ACKNOWLEDGEMENTS The author thanks Ed Voss for information about the Michigan Flora project, for his willingness to make the electronic records available, and for helpful comments on a draft of this manuscript. The author thanks Tony Reznicek for facilitating use of the Michigan Flora data, for helpful discussion, and for comments on a draft of this manuscript. Any errors in the description of the Michigan Flora

Page  13 ï~~2007 THE MICHIGAN BOTANIST 13 project are those of the author. Jeremy Young generously built the 4D database structure and scripts. The author thanks Katherine Bridges and Robert Fogel for helpful discussion. The manuscript benefited from the comments of Mike Penskar and those of an anonymous reviewer. The author was supported during the execution of the research by a joint postdoctoral fellowship from the University of Michigan Library and University of Michigan Rackham Graduate School. LITERATURE CITED Anonymous. (2004). Invasive plant mapping in Wisconsin: a workshop held on 25 February 2004. Plants out of place: The newsletter of the Invasive Plants Association of Wisconsin 7:2. Hanes, C. R. and F. N. Hanes. (1947). Flora of Kalamazoo County, Michigan. [Authors]. Schoolcraft, Michigan: xii + 295 pp. Meades, S. J., D. Schnare, K. Lawrence and C. Faulkner. (2007). Northern Ontario Plant Database Website. Version 1, January 2004. Algoma University College and Great Lakes Forestry Centre, Sault Ste. Marie, Ontario, Canada. Available online at URL: Payne, R. L. and B. MacIntosh. (2004). 22nd Session of the United Nations Group of Experts on Geographical Names: Report of the United States/Canada Division. 5 pp. Reznicek, A. A., E. G. Voss, and R. A. Simpson. Online Atlas of Michigan Plants. Edition 1. April 2004. University of Michigan. Ann Arbor, Michigan. Available online at URL: http://herbarium. Romig, W. (1972). Michigan Place Names. Walter Romig Publ., Grosse Pointe. [Facsimile reprint, 1986, by Wayne State Univ. Press, Detroit]. 673 pp. Simpson, R. A. (2006). Technical documentation for the approximate point locations and year of collection for over 50,000 specimens in the University of Michigan Herbarium's Michigan Flora database. Available online at URL: Stein, B. R. and J. Wieczorek. (2004). Mammals of the world: MaNIS as an example of data integration in a distributed network environment. Biodiversity Informatics 1:14-22. Stewart, L. O. (1935). Public Land Surveys: History, Instructions, Methods. Collegiate Press, Inc. Ames, Iowa. U.S. Geological Survey. May 1, 1981. Geographic Names Information System (GNIS). Reston, Virginia. Available online at URL: Voss, E. G. (1972). Michigan Flora. Part I. Gymnosperms and Monocots. Cranbrook Institute of Science Bulletin 55 and University of Michigan Herbarium. xv + 488 pp. (1978). Botanical beachcombers and explorers: pioneers of the 19th Century in the upper Great Lakes. Contributions from the University of Michigan Herbarium 13. viii + 100 pp. (1985). Michigan Flora. Part II. Cranbrook Institute of Science Bulletin 59 and University of Michigan Herbarium. xix + 724 pp. S(1996). Michigan Flora. Part III. Dicots (Pyrolaceae-Compositae). Cranbrook Institute of Science Bulletin 61 and University of Michigan Herbarium. xix + 622 pp. (1999). Labeling of herbarium specimens. Michigan Botanist 38(4): 57-63. S(2005). Gazetteer of some possibly puzzling collecting localities for Michigan plants. Contributions from the University of Michigan Herbarium 24:189-225.