Numerous plant species, with population centres scattered along environmental gradients, each with binomial distributions broadly overlapping those of other species, freely and variously combine into communities which predominantly intergrade with one another, forming a complex and potentially continuous but variously interrupted population pattern.

Robert H. Whittaker (1967)

As the quotation from Whittaker emphasizes, the vegetation of an area seldom consists of a mosaic of discrete types having unique species assemblages distinct from other types.  For the most part, plant species are distributed individualistically, each according to its own requirements, characteristics, and interactions with other species in a particular locale (Gleason 1926).  As a consequence, vegetation is a continuously varying phenomenon that depends on the distributions and proportional abundances of individual species.  No two patches of vegetation are identical in the combinations and proportions of species present (Miles 1979).  Even replicate samples from a small, relatively homogeneous patch of vegetation typically have average percent similarity values of only 50 to 90% (Gauch 1982).  Such “noise” among samples, which results from “chance distribution and establishment of individuals, animal activity, local disturbances, and environmental heterogeneity at scales below that of the sample area” (ibid.), limits the precision to which one can estimate the abundance of species present (see Floyd and Anderson [1987] in relation to INEEL vegetation).  It is also unlikely that any two patches will follow highly similar trajectories through time.  Despite this variability, different patches existing under similar environmental conditions in a region tend to have similar assemblages of species, making it possible to recognize “types.”  Whittaker (1975) likened the recognition of community types to our recognition of colors within the continuous spectrum of wavelengths of light.  Some rather distinct community types will be readily apparent, whereas recognition of other types necessarily will be quite arbitrary.  It is important to bear in mind that community type designations are ultimately arbitrary, abstract, ad hoc divisions.  They reduce the inherent complexity of vegetation to something manageable;  they are necessary to facilitate communication and management.   

Community types typically are distinguished by the dominant growth forms and the visual aspect created by the dominant species.  We assume that such types reflect interpretable differences in environment, but disturbance history of a site may be equally important, as is clearly shown by the numerous fire scars that are readily identified in aerial photographs or satellite images of the INEEL (see Fire History).

Development of the Vegetation Map.    The use of satellite imagery to map vegetation is based on the assumption that there will be a close correspondence between the properties of the vegetation and the spectral properties of the site.  In arid regions where vegetation is sparse, however, the spectral signature of an area may depend largely on spectral characteristics of the soil surface and/or the shadows cast by individual trees or shrubs (Tueller 1987, Smith et al. 1990).  To the extent that soil spectral properties and vegetation are not related, we can expect limits on the ability to accurately map vegetation from satellite imagery.  Furthermore, as we have explained above, the continuously varying nature of vegetation places constraints on the precision and accuracy of any classification scheme.  Because of the inherent variability in vegetation, precision (prediction of species composition) and accuracy (correct identification of a “type” at a particular location) of a vegetation map tend to be inversely related.  Broad vegetation classes (e.g., sagebrush/grassland) may be very accurate but not provide sufficient precision to be useful for environmental assessment or management.  On the other hand, precise predictions may be possible with narrow classes, but if accuracy is low, those predictions will be erroneous and misleading. 

The vegetation map of the INEEL was developed from Landsat satellite images (Kramber, et al. 1992).  Two Landsat scenes were selected to provide contrasting vegetation conditions;  one was from May 8, 1987, during the spring growth period, and the other was from August 17, 1989, when most herbaceous plants were senescent.  Spectral data from the two scenes were combined, and a preliminary classification was developed consisting of 27 cover classes that potentially represented vegetation types.  This classification was accomplished by a remote sensing analyst (William Kramber, Idaho Department of Water Resources) working with three of the authors (Rope, Anderson, and Glennon) and was based on known or inferred vegetation patterns from our field experience.  Aerial photography from 1976 was used to help identify patterns in some areas.  Thirty-two 1:24,000 scale maps, corresponding to USGS 7.5’ quads, were produced to facilitate field sampling for refining this initial classification.

We sampled vegetation, soils, and other parameters at plots representing all 27 classes of the preliminary classification in July and August, 1990.  Sixty-six plots were sampled.  At each plot, abundance of each vascular plant species occurring on the site was ranked using a four-point scale.  Slope, aspect, and soil surface characteristics also were recorded.  Comparable data were collected from 35 plots on the permanent vegetation transects.

The vegetation data were used to develop an error matrix highlighting discrepancies between the draft land cover classes and actual vegetation.  This provided a framework for reclassifying the 200 spectral classes.  The process used was one of successive refinements based on the plot data, field notes, and our field experience at the INEEL.  Ordinations and cluster analyses were used to classify the vegetation samples and identify assemblages of species that corresponded to the cover classes on the map (Anderson 1991).  These results were used to make further refinements of the map cover classes.  Several spectral classes were associated with areas dominated by perennial grasses, but the individual classes did not consistently have the same species composition.  For example, it was clear that we would be unable to distinguish between areas that had been seeded to crested wheatgrasses (Agropyron desertorum or Agropyron cristatum) from native grasslands (the crested wheatgrasses are perennial bunchgrasses that are native to the steppes of Asia).  We therefore combined several spectral classes into one “grasslands” class.  Spectral classes associated with various disturbances and bare soil were also combined.  Eleven vegetation classes are recognized on the INEEL Vegetation Map.  Some of these classes are quite distinct in species composition, whereas others are much more heterogeneous.  A description of each of the vegetation classes is given in the next section.

Vegetation integrates climate, soils, aspect, evolution, site history, species interactions, and chance into a single expression.  (JEA)