Considerations for ecological reconstruction of historic vegetation: Analysis of the spatial uncertainties in the California Vegetation Type Map datasetAbstractIntroductionCase study--The Wieslander vegetation type map datasetMethodsDatasetDigitization and georeferencing of plot mapsQuantification of total errorAnalysis of errorResultsDiscussionPlot relocation considerationsOther uses for VTM dataConclusionsAcknowledgmentsReferencesORIGINAL PAPERConsiderations for ecological reconstruction of historicvegetation: Analysis of the spatial uncertainties in theCalifornia Vegetation Type Map datasetMaggi Kelly Æ Ken-ichi Ueda Æ Barbara Allen-DiazReceived: 3 November 2006 / Accepted: 10 February 2007Ó Springer Science+Business Media B.V. 2007Abstract Historical ecological data are valuablefor reconstructing early environmental and vege-tation community conditions and examiningchange to vegetation communities and distur-bance regimes over decadal and longer temporalscales, but these data are not free from error. Weexamine the spatial uncertainties associated with18,000 vegetation plots in the decades-old Cali-fornia Vegetation Type Mapping (VTM) datasetthat has been digitized for use in modern ecolog-ical analysis. We examine the relationship be-tween plot location error and basemap year,basemap scale, plot elevation, plot slope, andgeneral plot habitat type. Bivariate plots andclassification and regression tree analysis (CART)confirm that basemap scale and age are thestrongest explanation of total error. Total errorin spatial location for all plots ranged from126.9 m to 462.3 m; plots drawn on 15-min(1:62,500-scale) basemaps had total error rangingfrom 126 m to 199.7 m, and plots drawn oncoarser-scale basemaps (1:125,000-scale) had totalerrors ranging from 241 m to 461.2 m. Relocationof individual VTM plots is considerably easier forplots originally marked on 1:62,500-scale mapsproduced after 1904, and more difficult for plotsoriginally marked on 1:125,000-scale maps pro-duced before 1898. Biogeographical analyses thatrely less on relocating individual plots, such asenvironmental niche modeling or multivariateanalyses can alleviate some of these concerns, butall researchers using these kinds of data need toconsider errors in spatial location of plots. Thepaper also discusses ways in which the differingspatial error might be reported and visualized bythose using the dataset, and how the data mightbe used in modern environmental niche models.Keywords CART  Historical vegetation data Spatial error  VTMIntroductionDatabases of historical ecological data haveproved useful in modern ecological research forreconstructing early environmental and vegeta-tion community conditions, studying disturbanceregimes over long temporal scales, and examiningchange to vegetation communities over decadaland longer temporal scales. In many cases, dataoriginally captured for taxation or land-surveyingpurposes often included some information onvegetation; this information has become useful tous now for reconstructing historic vegetation. Oneof the most often used historical databases is theM. Kelly (&)  K.-i. Ueda  B. Allen-DiazDepartment of Environmental Science, University ofCalifornia, 137 Mulford Hall #3114, Berkeley, CA94720-3114, USAe-mail: [email protected] EcolDOI 10.1007/s11258-007-9273-1Public Land Survey System (PLS), which wasinstituted by the US General Land Office in 1785and partitioned US land into Townships, Rangesand Sections. At each section corner a surveyorblazed two to four trees, known as ‘‘witnesstrees’’. The species, diameter, compass bearing,and distance to the corner of each tree wererecorded, and these data have become one of themost important sources for reconstructing pre-European settlement vegetation in parts of theUS (Bourdo 1956; White and Mladenoff 1994;Manies and Mladenoff 2000; Manies et al. 2001;Schulte and Mladenoff 2001; Mladenoff et al.2002; Schulte et al. 2002; Wang 2005). Wang(2005) and Mladenoff et al. (2002) provide goodreviews of the data. These witness tree point datahave been used in interpolations and statisticalclustering routines to reconstruct pre-Europeansettlement forest species composition and struc-ture across the Midwest, the Southeast andEastern US, including Wisconsin (He, Mladenoffet al. 2000; Radeloff et al. 2000; Schulte et al.2002; Bollinger et al. 2004); in Alabama (Blacket al. 2002; Rathbun and Black 2006); Minnesota(Friedman and Reich 2005); Michigan (Leahy andPregitzer 2003); Pennsylvania (Black and Abrams2001); West Virginia (Abrams and McCay 1996),and Southern Illinois (Anderson et al. 2006).Other researchers have used the PLS data witharcheological sites to examine Native Americaninfluence on pre-European settlement vegetationpatterns (Black et al. 2002; Foster et al. 2004).Where PLS data is not available, someresearchers have used less systematic historicalrecords to reconstruct past vegetation conditions,or land use patterns (Russell 1981; Jackson et al.2000; Ryavec 2001; Wilson 2005). For example,Wilson (2005) used 19th century lumber surveysto estimate potential forest in Maine; and Jacksonet al. (2000) used land survey notes and forestresource inventories to reconstruct forest abun-dance in Ontario, Canada.While valuable, historical ecological data arenot free from error. Numerous researchers havediscussed caveats associated with use of histor-ical ecological data, and recommended cautionin using older collections. Uncertainties canexist in both aspatial or attribute, and spatialcontent of the data in these collections (Schulteand Mladenoff 2001; Mladenoff et al. 2002;Plewe 2002). For example, reports cite surveyorvariability in reporting vegetation characteriza-tion and species descriptions (i.e. attributeerrors) (Bourdo 1956; Galatowitsch 1990; Man-ies et al. 2001; Schulte and Mladenoff 2001;Schulte et al. 2002) and in the reporting of thelocations of both point samples, linear features,and polygonal features like habitat polygons orland parcels (i.e. positional errors) (Ryavec2001; Gregory 2002). These uncertainties needto be considered and evaluated before the useof historic ecological data begins, and this paperdeals specifically with the positional error foundin a historic database.Case study––The Wieslander vegetation typemap datasetThe Wieslander Vegetation Type Map (VTM)dataset, collected originally in the 1920s and1930s in California is a geographically broad andfloristically