Printer Friendly

Regional gradient analysis and spatial pattern of woody plant communities of Oregon forests.


Identifying the factors controlling the distribution, abundance, and diversity of species in ecological communities continues to be a central problem in community ecology. Community structure is now considered a product not only of local physical conditions and interactions among species, but also of regional constraints such as climate, and of historical processes such as dispersal and speciation, migration, and extinction (Neilson and Wullstein 1983, Brown 1984, Ricklefs 1987, Menge and Olson 1990, Ricklefs and Schluter 1993). Ecological research in recent decades has focused on intracommunity patterns, especially on the role of competition and other species interactions in community structure. In addition, plant community ecologists have devoted considerable effort to quantifying local- to landscape-scale variation in vegetation. As a result, factors associated with local patterns of community composition often are well known, but differ among localities. Whereas biogeographers and phytosociologists have long recognized the role of environmental and historical factors in regional- to continental-scale patterns of plant community composition, current knowledge remains general and qualitative. Few systematic, quantitative, regional-scale descriptions provide context for landscape differences (but see Denton and Barnes 1987), and the synthesis and collation of plant-community data in regional analysis has been cited as a major research need (Franklin 1988, Franklin and Blinn 1988).

In the Pacific Northwest, much of the research in plant community ecology has involved classification of potential natural vegetation (Wellner 1989), and placement of associations along environmental gradients has only been inferred (e.g., Topik et al. 1988). Most quantitative studies of forest communities of the western United States have been limited to older, natural forests, to incomplete segments of regional environmental gradients, or to certain forest community types, and most were conducted at the landscape scale (Whittaker 1960, Whittaker and Niering 1965, Daubenmire and Daubenmire 1968, Fonda and Bliss 1969, Minore 1972, Dyrness et al. 1974, Zobel et al. 1976, del Moral and Watson 1978, Peet 1981, del Moral and Fleming 1979, Allen and Peet 1990, Allen et al. 1991, Spies 1991, Spies and Franklin 1991, Riegel et al. 1992, Parker 1994). Other regional- and continental-scale studies quantified patterns of species diversity (Glenn-Lewin 1977, Richerson and Lum 1980, Currie and Paquin 1987), but did not evaluate gradients in community composition. In the only regional-scale account of Oregon vegetation, Franklin and Dyrness (1973) synthesized local studies and related vegetation to environmental factors only qualitatively.

We undertook a quantitative, systematic analysis of plant community data spanning a broad region - forest lands in Oregon - to examine contributions of broad- and local-scale factors to regional variation in community composition, and to explore the influences of physical environment, biotic factors, and disturbance processes. Our primary study objectives were to identify and quantify environmental factors associated with regional gradients in the composition of woody plant communities of forests and woodlands throughout Oregon and its subregions [ILLUSTRATION FOR FIGURE 1 OMITTED]; to examine how these factors change with geographic extent and location; and to describe and map geographic patterns of compositional gradients. Our study addresses long-standing questions about factors controlling the distribution, abundance, and diversity of species in ecological communities, but considers a somewhat broader region and a much larger dataset than most previous studies, and uses contemporary multivariate statistical and spatial analytical tools.

Factors associated with regional variation in species composition

We hypothesized that macroclimate, an expression of broad-scale temperature and moisture environment, is the primary associate of regional-scale patterns of community composition, and that substrate (geologic parent material and soils) and local factors (microclimate, topography, and site disturbance) are secondary. It is almost axiomatic that, at a continental scale, patterns of vegetation physiognomy and community composition are associated primarily with large-scale climate (Daubenmire 1978, Woodward 1987). Macroclimate is thought to influence communities directly through physiological effects on organisms and by limiting populations. Macroclimate indirectly influences communities by modifying or regulating the importance of local-scale factors and by favoring certain species and growth forms in interspecific competition. Empirical studies that quantified environmental associations with patterns of forest vegetation in the Pacific Northwest (Whittaker 1960, Whittaker and Niering 1965, Daubenmire and Daubenmire 1968, Fonda and Bliss 1969, Waring 1969, Minore 1972, Dyrness et al. 1974, Zobel et al. 1976, del Moral and Watson 1978, del Moral and Fleming 1979, Spies 1991, Spies and Franklin 1991, Riegel et al. 1992, and others), in other mountainous regions of the western United States (Peet 1978, 1981, Allen and Peet 1990, Allen et al. 1991, Parker 1994), and elsewhere (Denton and Barnes 1987) have consistently demonstrated the primary importance of temperature and moisture. The studies, however, either did not distinguish effects of macroclimate from microclimate and other factors, or considered broadscale climate only indirectly by observing vegetation change along complex gradients of elevation or latitude.

Vegetation patterns have been associated with topography in gradient studies at smaller geographic extents, and in larger areas that were prestratified by elevation or substrate (e.g., Whittaker 1960, Peet 1978, 1981, Allen and Peet 1990, Allen et al. 1991). However, topographic position did not differentiate vegetation types in the west-central Oregon Cascades (Zobel et al. 1976). Studies in several western mountain systems have shown that soil differences usually are important only secondarily (Whittaker and Niering 1965, Daubenmire and Daubenmire 1968, Fonda and Bliss 1969, Minore 1972, Zobel et al. 1976). Only occasionally is soil chemistry associated with vegetation patterns, such as in the Klamath Mountains and redwood regions (Whittaker 1960, Waring and Major 1964, Waring 1969) and in the pumice region of central Oregon (Volland 1985). Interactions among disturbance and successional processes with other environmental factors in determining regional vegetation patterns are poorly understood, and have been difficult to quantify in regional gradient studies.

The influence of geographic extent and location

Within a local area, the range of climatic conditions is small, and most sites fall within species' physiological tolerances. Thus, shifts in species' relative abundances are thought to be associated with local variations in topography, microclimate, and substrate, and with biotic interactions (Neilson 1987, Neilson et al. 1992), as well as with stochastic disturbances that are highly variable over time and space. We therefore hypothesized that species variation explained by regional climate decreases, and variation explained by local factors increases, with decreasing geographic extent.

Different environmental factors probably assume varying degrees of importance among locales within a region, and limiting factors likely do not vary in parallel from place to place (Neilson et al. 1992). Local shifts in community composition have been proposed to be less pronounced in favorable climates, and more striking where climate is more stressful (Neilson et al. 1992), but rarely has this been demonstrated (but see Neilson and Wullstein 1983). The western United States is climatically diverse, and shifts in controls on community composition among landscapes are poorly understood. Within Oregon, we hypothesized that moisture and topographic factors assume greater importance where climate is drier and more erratic (i.e., in eastern Oregon and the interior valleys of western Oregon). Franklin and Dyrness (1973: 50) proposed a geographic shift in major environmental constraints across Oregon's forests, but this hypothesis has not been tested with regional data. In their xerophytic zones of eastern Oregon and the western interior valleys, they hypothesized that moisture is limiting for many species, and that communities respond to small differences in environment affecting moisture stress. Neither moisture nor temperature are limiting within their temperate zones, and species composition does not shift so markedly in response to local differences in site conditions. Temperature is the major factor separating their temperate from their subalpine types. Zobel et al. (1976) found that temperature differentiated major vegetation zones in the central western Cascades, and moisture distinguished communities within the warmer zones. They hypothesized that moisture stress might be more important at elevations below their study area. Waring (1969) found similar relations in the Siskiyou Mountains. Riegel et al. (1992) concluded that precipitation was the key environmental factor governing distribution and composition of community types in the foothill Quercus spp. (oak) woodlands of southwest Oregon.


Study area

The study area, the state of Oregon, is between 41.9 [degrees] and 46.4 [degrees] N latitude and between 116.5 [degrees] and 124.9 [degrees] W longitude. Patterns of physiography, geology, and soils are varied and complex. The region is characterized by the north-south oriented Cascade and Coast ranges, as well as the Siskiyou Mountains in southwest Oregon and the Blue-Ochoco-Strawberry-Wallowa complex of mountains in the Blue Mountains subregion of eastern Oregon [ILLUSTRATION FOR FIGURE 1 OMITTED]. The Cascade Range, which is the major topographic and climatic divide in Oregon, is bisected at the northern Oregon boundary by the Columbia River. Elevations range from sea level to [greater than] 3000 m [ILLUSTRATION FOR FIGURE 2A OMITTED]. Geologic formations date from the Paleozoic (over 400 x [10.sup.6] yr old) to the Holocene (Walker and MacLeod 1991) [ILLUSTRATION FOR FIGURE 3 OMITTED]. Vulcanism has shaped much of the landscape, but sedimentary and metamorphic rocks are plentiful, and deposition of parent materials by alluvial, colluvial, and eolian processes is common (Walker and MacLeod 1991). Soil types are primarily inceptisols, spodosols, and ultisols (Franklin and Dyrness 1973). Volcanic activity during the Pleistocene and Holocene has mantled large tracts at higher elevations in the Cascade Range and in central Oregon with pumice and ash.

Climatic variation in the region is influenced by the interplay among westerly winds of maritime air masses, northeasterly winds of continental air masses, and the blocking effects of mountain ranges (Franklin and Dyrness 1973). Western Oregon has a maritime climate characterized by mild temperatures; a long frost-free season; prolonged cloudy periods; narrow seasonal and diurnal fluctuations in temperature; mild, wet winters and cool, relatively dry summers; and heavy precipitation (Table 1). Most precipitation falls as rain between 1 October and 31 March, resulting from cyclonic storms that approach from the Pacific Ocean on the dominant westerlies. Storm tracks shift to the north during summer, and high-pressure systems bring fair, dry weather for extended periods. To varying degrees, coastal mountains block maritime air masses from the interior valleys of western Oregon, where climates are less muted and precipitation declines markedly. Orographic effects influence local amounts of precipitation [ILLUSTRATION FOR FIGURE 2B OMITTED] and the proportion that falls as snow (Franklin 1988). A general latitudinal increase in precipitation and decrease in temperature occurs from south to north, so the interior valleys of southwest Oregon have the hottest and driest climate in western Oregon (Franklin and Dyrness 1973).

In eastern Oregon, temperatures fluctuate more widely and are more extreme, and frost-free seasons are shorter than in western Oregon (Table 1, [ILLUSTRATION FOR FIGURE 2D OMITTED]). Precipitation is primarily cyclonic in origin but is considerably less than to the west due to the rain shadow of the Cascade Range. Summers are very dry, although convective storms are common in mountainous areas, and a high proportion of annual precipitation falls as snow.

Forests dominate the landscapes west of the Cascade Range and the mountain slopes to the east, covering 11 x [10.sup.6] ha (46%) of the total land area (Powell et al. 1993) and confined to the five subregions of our study [ILLUSTRATION FOR FIGURE 1 OMITTED]. About 60% of the forest land is publicly owned, 23% is owned by the timber industry, and 17% belongs to nonindustrial private landowners (Powell et al. 1993) [ILLUSTRATION FOR FIGURES 4 AND 5 OMITTED].

Coniferous tree species dominate most of Oregon's forest communities (Franklin and Dyrness 1973, Waring and Franklin 1979, Franklin 1988). Outside of the mixed-evergreen zone of southwest Oregon, where several evergreen hardwood trees codominate, hardwoods tend to occupy harsh sites or specialized habitats or serve as pioneers. The present climate of wet, mild winters and relatively dry summers favors evergreen, needle-leaved conifers by permitting extensive photosynthesis outside of the growing season and reducing net photosynthesis during the summer (Waring and Franklin 1979). The mesic temperate coniferous forests of northwestern Oregon contain the greatest biomass accumulation and some of the highest productivity rates of any forests in the world (Franklin 1988). Many of the dominant tree species (e.g., Tsuga heterophylla and Thuja plicata) are endemic to the Pacific Northwest, and others find their center of distribution and attain maximum development here. Southward, where the climate is warmer and drier, Pinus lambertiana, Calocedrus decurrens, and Lithocarpus densiflorus are added to the flora. The forest zones of interior southwest Oregon represent northern extensions of the mixed-conifer forest of the Sierra Nevada and the mixed sclerophyll forest of the California Coast Ranges. In eastern Oregon, Pacific coastal elements mix with Rocky Mountain elements. Pinus ponderosa characterizes forests at lower elevations, and Abies lasiocarpa those at higher elevations. Franklin and Dyrness (1973) provide detailed descriptions of Oregon plant communities.

Disturbance regimes vary tremendously across Oregon. Before European settlement, natural disturbance regimes were dominated by infrequent, catastrophic fires (Agee 1993). Periodic, low-intensity underburns were common in some places (Teensma 1987, Morrison and Swanson 1990), as were fires set by American Indians (Agee 1993). Natural fire-return intervals ranged from 15 yr in drier eastside pine forests, to 400 yr in moist, coastal forests, to 800 yr in subalpine forests (Agee 1993). Windstorms, pathogens, and other disturbance agents also have influenced forest development.

In the last 100 yr, natural disturbance regimes have been supplanted by timber management and wildfire suppression. Timber management activities generally are more frequent and intense and less variable in size and intensity than natural disturbances, and simplify stand and landscape structure (Hansen et al. 1991). Intensive management of forest plantations in northwestern Oregon consists of clear-cutting most live trees and snags, site preparation by prescribed fire or herbicides to control competing vegetation, replanting with a single species (usually Pseudotsuga menziesii), periodic thinning to maintain vigorous and evenly spaced crop trees, and harvesting at 40-100 yr intervals (Hansen et al. 1991). In southwestern and eastern Oregon, stands more often are partially harvested. In eastern Oregon, fire suppression has allowed fire-seral and late-successional tree species to overstock forest stands (Johnson et al. 1994). Selective harvests and overstory removals have converted large areas of parklike forests of Pinus ponderosa to greater dominance by Abies spp.

Forest management practices have differed among land ownerships and over time. Logging of old-growth forest from federal lands began after World War II and accelerated in the 1970s. Most remaining old growth is concentrated on federal lands and at higher elevations (Bolsinger and Waddell 1993). Almost all of the vegetation plots in our study on federal lands were established in older, natural stands [ILLUSTRATION FOR FIGURES 4 AND 5 OMITTED]. In contrast, virtually all forest lands in private and state ownership, and almost all vegetation plots in our study on nonfederal lands, have been harvested in some manner [ILLUSTRATION FOR FIGURE 4 OMITTED] and are [less than]120 yr old [ILLUSTRATION FOR FIGURE 5 OMITTED]. Logging of the most accessible and productive lands in private ownership began in the mid-1800s, and some areas have been harvested a second time. By the 1980s, 97% of the nonfederal timberland was in an early- or mid-successional stage, and only 3% was late successional (Pacific Resource Inventory, Monitoring, and Evaluation Program, Pacific Northwest Research Station, Portland, Oregon, unpublished data). Grazing, fuel-wood harvesting, introduction of nonnative plant species, and urbanization also have altered forest composition, especially near population centers in the interior valleys of western Oregon (Franklin and Dyrness 1973, Riegel et al. 1992).

Three hierarchical geographic extents

We divided the major forested areas of the state a priori into five geographic subregions, based primarily on physiography [ILLUSTRATION FOR FIGURE 1 OMITTED]. Boundary placement was somewhat arbitrary but consistent with widely accepted physiographic provinces (e.g., Franklin and Dyrness 1973, Bailey et al. 1994). We aimed to broadly stratify the region into relatively homogenous subregions to examine within-region ecological variation, and to analyze effects of geographic extent on ecological pattern. We conducted most of our analyses independently at each of three nested, hierarchical geographical extents: state (broadest), half-state (western and eastern Oregon), and subregion [ILLUSTRATION FOR FIGURE 1 OMITTED].

We limited our study to forested lands because of availability of field plot data. The data sources defined forest as areas with at least 10% canopy cover of trees, including closed forest and open woodland. We did not include small inclusions of nonforest land, usually associated with soil conditions, within forested areas. These natural openings comprise a considerable portion of the landscape in the Blue Mountains subregion.

Vegetation data

We obtained vegetation data collected on [approximately]10000 field plots installed by the Area Ecology Program (Ecology) of the Pacific Northwest Region, USDA Forest Service, on National Forest lands; the Forest Inventory and Analysis (FIA) inventory conducted by the Pacific Resource Inventory, Monitoring, and Evaluation Program of the Pacific Northwest Research Station, USDA Forest Service, on nonfederal lands; and the Forestry Intensive Research Program (FIR) of Oregon State University on USDI Bureau of Land Management lands in southwest Oregon (Table 2). We obtained digital data files from each source, extracted vegetation and site data elements common to all data sets, and compiled the extracted data into a single database. Vegetation measurements used in our study consisted of crown cover for woody plant species. For [TABULAR DATA FOR TABLE 1 OMITTED] all component data sets, crown cover was visually estimated on fixed-radius plots (Table 2).

The component data sets were collected under different objectives and sampling designs (Table 2). The Ecology and FIR plots were selected subjectively without preconceived bias (Mueller-Dombois and Ellenberg 1974), primarily from older, natural stands. The Ecology plots were established across all National Forest lands statewide [ILLUSTRATION FOR FIGURE 4 OMITTED], and the FIR plots on Bureau of Land Management lands in southwest Oregon. The FIA plots were established systematically across all nonfederal forest lands statewide [ILLUSTRATION FOR FIGURE 4 OMITTED], at intersections of a 5.5 x 5.5 km grid. Plot sizes and configurations varied among the component data sets, and thus also with geographic location, ecological conditions, and ownership (Table 2). We subsampled from the complete vegetation data set to reduce it to a manageable size, to even-out strong differences in sampling intensity among the component data sets, to introduce an element of randomness, and to reduce the possibility of geographic bias. To obtain an even geographic distribution of the subsample, we randomly selected 10 plots from each 635-[km.sup.2] hexagon in the tessellation developed by White et al. (1992). This yielded 2443 plots for analysis, encompassing 30 families, 71 genera, and 181 species. Nomenclature follows Hitchcock and Cronquist (1973) and Hickman (1993).

Explanatory variables

We compiled data on the physical environment and disturbance history of each plot from field-recorded measures and through overlay of plot locations with mapped features in a geographic information system (GIS) (ARC/INFO software, version 7.0; Environmental Systems Research Institute [ESRI] 1991) (Table 3). We used field-recorded data for slope, aspect, elevation, overstory crown cover, stand age, and history of clearcut timber harvesting. We used a cosine transformation of aspect: ASPECT = cosine(45 - DEGREES) + 1 (Beers et al. 1966). This equation transforms a circular variable, where large values are close to small values, into a measure of aspect that is relevant for vegetation. Values range from 0.0 on the most exposed (southwest) slopes to 2.0 on the least exposed (northeast) slopes. For the Ecology and FIR plots, AGE was based on the oldest of several trees on the plot, determined by increment boring. For FIA plots, AGE was based on all over- and mid-story trees on each plot, determined by increment boring. Where a history of clear-cutting was recorded on FIA plots, we coded CLEARCUT = 1. We assumed all other plots never had been harvested (CLEARCUT = 0).

We derived climate data from precipitation map surfaces generated by the Precipitation-elevation Regressions on Independent Slopes Model (PRISM) (Daly et al. 1994) and temperature map surfaces generated by a model developed by Dodson and Marks (1997). Both models use digital elevation models (DEMs) to account for topographic effects in interpolating weather measurements from an irregular network of weather stations to a uniform grid. Thirteen precipitation surfaces (mean annual and 12 mean monthlies) were generated at 10-km resolution from 1961-1990 weather data. We log-transformed all precipitation surfaces, because vegetation does not respond linearly to amount of precipitation. A 1.0-cm difference in precipitation is more important at low than at high levels. We generated 15 temperature surfaces (mean annual, all-time recorded minimum, all-time recorded maximum, and 12 mean monthlies) at 500-m resolution from 1981-1992 weather data. We assigned precipitation and temperature values from each map surface to plot locations, by using bilinear interpolation (LATTICESPOT function, ESRI 1991). From the mean monthly precipitation and temperature values for each plot, we computed indices that approximate conditions during the growing season (SMRPRE, SMRTMP, SMRTSMRP) and the cool season (WTRPRE, WTRTMP), as well as seasonal variability and continentality (Table 3). Continental climates experience greater diurnal and seasonal variability in temperature (reflected in CVTMP), less seasonal variability in precipitation because of increased prevalence of rainfall from summer convective storms (reflected in CVPRE and CONTPRE), and colder minimum temperatures (reflected in MINTMP) than do maritime climates (Table 3).

We estimated each plot's potential solar radiation from latitude, elevation, aspect, and slope using program SOLARPDX (Smith 1993). SOLARPDX computes [TABULAR DATA FOR TABLE 2 OMITTED] total insolation, including direct (direct beam) and indirect (skylight), based on the algorithms of Lowry and Lowry (1989). Calculations assume an average profile for atmospheric transmissivity and absorbtivity for humid, temperate climates, and do not adjust for cloud cover.

We obtained data on lithology and geologic age from a geologic map of Oregon (Walker and MacLeod 1991) obtained as a GIS coverage from the Corvallis Forestry Sciences Laboratory, USDA Forest Service, 3200 SW Jefferson Way, Corvallis, Oregon. We assigned a type to each field plot using the INTERSECT command (ESRI 1991). We grouped the types into 14 generalized geologic types, which we converted to dummy variables. Because direct gradient analysis can only accept K - 1 dummy variables for a categorical variable with K categories, we dropped the uninformative "mixed rocks" (MIXR) category (Table 3, [ILLUSTRATION FOR FIGURE 3 OMITTED]) from our analyses.
TABLE 3. Codes and definitions of explanatory variables used in
canonical correspondence analysis, by variable subset.

Code                         Definition


ASPECT        Cosine transformation of aspect (degrees), 0.0
              (southwest) to 2.0 (northeast)
SLOPE         Slope (percent)
SOLAR         Total potential solar radiation (joules/[m.sup.2] x


AGE           Stand age (years)
CLEARCUT      History of clearcut harvest (1 = clearcut, 0 = not
TREOCOV       Crown cover of all overstory trees (percent)


ELEV          Elevation (m)
ANNPRE        Mean annual precipitation (natural logarithm, mm)
WTRPRE        Mean precipitation in November-March (natural
              logarithm, mm)
SMRPRE        Mean precipitation in May-September (natural
              logarithm, mm)
CVPRE         Coefficient of variation of mean monthly precipitation
              of the wettest and driest months, usually December
              and July

CONTPRE       Percentage of mean annual precipitation falling in
SMRTSMRP      Moisture stress during the growing season,
ANNTMP        Mean annual temperature ([degrees] C)
MAXTMP        All-time maximum recorded temperature ([degrees] C)
MINTMP        All-time minimum recorded temperature ([degrees] C)
WTRTMP        Mean monthly temperature ([degrees] C) from
SMRTMP        Mean monthly temperature ([degrees] C) from
CVTMP         Coefficient of variation of mean monthly temperature
              of the coldest and warmest months, usually January and
              August, calculated after rescaling [degrees] C (-40
              [degrees] to 40 [degrees]) to (0 [degrees] to 255


VOLC          Igneous: volcanic and intrusive rocks
SILR          Igneous: silicic rocks (granite, diorite, rhyolite,
              and dacite)
MAFO          Igneous: mafic rocks (basalt, basaltic andesite,
              andesite, gabbro), Miocene or older
MAFY          Igneous: mafic rocks (basalt, basaltic andesite,
              andesite, gabbro, Pliocene and younger
UMAF          Igneous: ultramafic rocks (serpentine)
PYRO          Igneous: mafic pyroclastic rock, ash, ejecta, vent
              deposits, Miocene and older
PYRY          Igneous: mafic pyroclastic rock, ash, ejecta, vent
              deposits, Pliocene and younger
SILV          Igneous: silicic vent deposits
META          Metamorphic
SEDR          Sedimentary: siltstones, sandstones, mudstones,
TUFO          Sedimentary: tuffaceous rocks and tuffs, pumicites,
              silicic flows, Miocene and older
TUFY          Sedimentary: tuffaceous rocks and tuffs, pumicites,
              silicic flows, Pliocene and younger
DEPO          Depositional: dune sand, alluvial, glacial,
              glaciofluvial, loess, landslide and debris flow,
              playa, lacustrine, fluvial
MIXR          Mixed rocks


X             Longitude (decimal degrees)
Y             Latitude (decimal degrees)
Y2            [Y.sup.2]
X3            [X.sup.3]
X2Y           [X.sup.2]Y

We obtained the latitude (Y) and longitude (x) of each plot. The derivation and accuracy of these coordinates varied greatly among the component data sets. To account for complex spatial trends, we developed quadratic and cubic combinations of X and Y (sensu Borcard et al. 1992). We identified the five spatial variables (Table 3) that explained the most species variation in a preliminary gradient analysis for use in subsequent analyses.

Gradient analyses

Direct gradient analyses. - Our primary analytical tool was canonical correspondence analysis (CCA) (ter Braak 1986, 1987a, b, ter Braak and Prentice 1988), a direct gradient analysis method used widely in community ecology (Palmer 1993). We chose CCA, because our goal was to better understand environmental factors associated with vegetation pattern. In CCA, sites and species are arranged in a multidimensional space, with the restriction that the ordination axes must be linear combinations of the specified environmental variables. We used program CANOCO, version 3.12 (ter Braak 1987a), recompiled with increased array sizes to accommodate our large data set. In all CCAs, we log-transformed species relative abundance values to dampen the influence of dominant species. All other CANOCO defaults were used. All CCA plot scores in this paper are linear combinations. In CCA and in this paper, the fraction of species variation explained by a set of explanatory variables (total variation explained, TVE) is the sum of all constrained eigenvalues divided by the total variation (TV) in the species data (or "total inertia," sensu ter Braak 1987a), which is the sum of all unconstrained eigenvalues. The TV is the ratio of the dispersion of the species scores to the dispersion of the plot scores (ter Braak 1987a), a property of the species-by-plot data matrix. Data matrices with greater TV contain many species and little overlap of species occurrence among plots.

CCA with variance partitioning. - We performed CCA with variance partitioning (ter Braak 1988, Borcard et al. 1992, Okland and Eilertsen 1994), using partial CCA, to quantify the relative contributions of variable subsets to explained variation. In partial CCA, species variation associated with explanatory variables that are not of direct interest (i.e., covariables) is partialled out, in order to examine a selected set of explanatory variables of interest. The usual explanatory variables are replaced by the residuals obtained by regressing each of the variables of interest on the covariables. We analyzed several combinations of two sets of explanatory variables, denoted {A} and {B} in the general sense (sensu Okland and Eilertsen 1994). We denoted variation explained by {A} as A and by {B} as B. We found the fraction of variation explained by {A} and not shared with {B}, denoted A[where]B, by partial CCA (ter Braak 1988), using the variables in {B} as covariables and {A} as the constraining variables. We found the fraction of variation explained by {B} and not shared with {A }, denoted B [where] A, by partial CCA using the variables in {A} as covariables and {B} as the constraining variables. We found the fraction of variation explained that is shared by {A) and {B}, denoted A [intersection] B, by (A - A [where] B).

We grouped factors thought to operate at a local scale into the topography ({T}) and disturbance ({D}) subsets, regional-scale factors into the macroclimate ({C}) and geologic ({G}) subsets, and measures and transformations of spatial position into the spatial ({S}) subset (Table 3). We performed CCA with variance partitioning to quantify relative contributions to explained variation of environmental vs. spatial factors ({T [union] D [union] C [union] G} vs. {S}), of local vs. regional factors ({T [union] D} vs. {C [union] G}), and of each variable subset rs. the other three subsets ({T} vs. {D [union] C [union] G}, {D} vs. {T [union] C [union] G}, {C} vs. {T [union] D [union] G}, and {G} vs. {T [union] D [union] C}).

Stepwise CCA. - We performed stepwise CCA on all woody species, tree species only, and shrub species only, for each geographic area. We excluded spatial variables because they explained very little of the species variation in CCA with variance partitioning. We added explanatory variables to the model in the order of greatest additional contribution to TVE, but only if they were significant (P [greater than or equal to] 0.01), where significance was determined by a Monte Carlo permutation test using 99 permutations ([H.sub.o]: additional influence of variable on vegetation is not significantly different from random), and if adding the variable did not cause any variance inflation factors to exceed 20. Variables with large inflation factors are strongly multicollinear with other variables and contribute little unique information to the model (ter Braak 1987a). We excluded them to improve model interpretability and parsimony. Although several of the explanatory variables included in the stepwise models still were intercorrelated, CCA is robust to this multicollinearity (Palmer 1993).

We graphed results as biplots [ILLUSTRATION FOR FIGURE 6 OMITTED], in which arrow length and position of the arrowhead indicate the correlation between the explanatory variable and the CCA axes, arrow direction indicates how the variable is correlated with the CCA axes, and smaller angles between arrows indicate stronger correlations between variables (ter Braak 1986, 1987a, b). We evaluated CCA model fit for plots using residual distances, an ordination diagnostic provided by CANOCO (ter Braak 1987a). Residuals are the squared chi-square distances between each plot and the centroid of all plots in m-dimensional species space for CCA axes 1-4 (ter Braak 1987a).

Indirect gradient analyses. - In order to more fully explore how successfully the environmental variables used in the constrained ordination explained species variation, we analyzed the vegetation data using detrended correspondence analysis (DCA), an indirect gradient analysis method (Gauch 1982), and with detrended canonical correspondence analysis (DCCA) using the same explanatory variables identified in the stepwise CCAs. Species relative abundance was log-transformed cover; all other defaults were used. We used detrending by segments, with 26 segments for both DCA and DCCA, so we could compare results from the two methods (Palmer 1993). We compared eigenvalues and gradient lengths from the two methods. By rescaling in DCA and DCCA, gradient length is set equal to species turnover (standard deviations, SD) so as to make gradient length interpretable and readily comparable between different datasets (Peet et al. 1988). We also computed Spearman's rank correlations (PROC CORR) (SAS 1990) of plot scores from DCA and DCCA.

Mapping of dominant gradients. - We kriged (Isaacs and Srivastava 1990) the plot scores from stepwise CCA in a GIS (command KRIGING, ESRI 1991) to help visualize the ordination results through interpolation. Kriging is a linear, weighted-average interpolation method that considers spatial autocorrelation in the data and does not require that the data be normally distributed or uncorrelated. We selected the Gaussian model as having the best fit between actual and predicted semivariograms for CCA axes 1-4 at the state scale. We interpolated the plot scores to a lattice with 8000-m spacing, equal to the average distance between adjacent field plots, and then contoured the lattice using command LATTICECONTOUR (ESRI 1991). We kriged the Blue Mountains subregion independently from the other four subregions because it is not geographically contiguous. We selected sizes of the sampling windows and contour intervals subjectively to achieve comparable appearance among contour maps.

Species diversity measures

We computed measures of the three levels of species diversity defined by Whittaker (1960) for each geographic area and vegetation layer. We estimated alpha diversity ([Mathematical Expression Omitted]), or intracommunity diversity, as the mean number of species sampled (species richness) on a field plot. We estimated gamma diversity (S), the diversity of an entire landscape or region, from the total number of species sampled on all plots within each geographic area. We computed two measures of beta diversity, which is defined as the change in species composition along environmental gradients and often is measured as the degree to which the species compositions of samples differ. We used Whittaker's (1960) measure of beta diversity, [Mathematical Expression Omitted], and total variation (TV) from stepwise CCA. The TV is the ratio of the dispersion of the species scores to the dispersion of the plot scores (ter Braak 1987a). Data matrices with greater TV contain many species with little overlap of species occurrence among plots, and thus higher beta diversities.


Comparison of direct and indirect gradients

Eigenvalues and gradient lengths were moderately higher for detrended correspondence analysis (DCA) than for detrended canonical correspondence analysis (DCCA) for the first two axes in most of the geographic areas (Table 4), indicating that a portion of the species variation was not accounted for by the explanatory variables in DCCA. However, the strong correlations between the DCA axis 1 and the explanatory variables (Table 5) suggested that much of the variation in vegetation is related to the measured explanatory variables. Furthermore, the striking similarity between correlations of DCA scores and canonical correspondence [TABULAR DATA FOR TABLE 4 OMITTED] analysis (CCA) scores with the explanatory variables indicated that the dominant compositional gradients related to the same explanatory variables (Table 5). Geographic patterns of DCA scores at the state level also were quite similar to CCA scores for the first two axes.
TABLE 5. Spearman's rank correlations of explanatory factors with
axis scores from detrended correspondence analysis (DCA) and
intraset correlation coefficients from canonical correspondence
analysis (CCA) that included all explanatory variables, all woody
species in Oregon.

                    Axis 1                   Axis 2
               DCA       CCA         DCA              CCA

ASPECT        0.03       0.08        0.00                  0.03
SLOPE        -0.38      -0.35       -0.03                 -0.01
SOLAR         0.36       0.27        0.06                 -0.02
AGE           0.16       0.07       -0.14                  0.27
CLEARCUT     -0.36      -0.33        0.00                 -0.14
TREOCOV      -0.58      -0.51       -0.33                  0.21
ELEV          0.77       0.82       -0.01                  0.32
ANNPRE       -0.82      -0.85       -0.49                  0.30
WTRPRE       -0.82      -0.80       -0.47                  0.25
SMRPRE       -0.67      -0.58       -0.52                  0.48
CVPRE        -0.68      -0.67       -0.22                 -0.03
CONTPRE       0.72       0.68        0.22                  0.04
SMRTSMRP      0.06      -0.06        0.49                 -0.68
ANNTMP       -0.59      -0.63        0.13                 -0.48
MAXTMP       -0.75      -0.80        0.03                 -0.34
MINTMP       -0.79      -0.83       -0.15                 -0.16
WTRTMP       -0.72      -0.76        0.02                  0.02
SMRTMP       -0.29      -0.33        0.31                 -0.58
CVTMP         0.81      -0.84        0.21                 -0.19
VOLC         -0.05      -0.05       -0.05                 -0.01
SILR          0.04      -0.03        0.01                 -0.04
MAFO          0.27       0.33        0.14                  0.05
MAFY          0.10       0.11       -0.11                  0.18
UMAF         -0.02      -0.05        0.01                 -0.01
PYRO          0.12       0.13        0.03                 -0.01
PYRY          0.14       0.14       -0.05                  0.02
SILV          0.08       0.08        0.03                 -0.02
META          0.00       0.01        0.01                 -0.01
SEDR         -0.43      -0.42       -0.07                 -0.10
TUFO         -0.14      -0.14       -0.02                 -0.01
TUFY          0.16       0.17        0.11                 -0.08
DEPO          0.00       0.01       -0.07       [less than]0.01

DEPO          0.00       0.01       -0.07       [less than]0.01

Note: Sign reflects arbitrary selection of gradient direction by

Plot scores from DCA and DCCA for axis 1 were highly or moderately correlated in all geographic areas (Table 4). Correlations declined - sometimes markedly-with the second and subsequent axes. Differences in the ordering of plots by DCA and DCCA beyond the first or second axis appeared to have been caused by poor performance of DCA on our very long gradients. We therefore present DCA results for axes 1 and 2 only. Because of similar geographic patterns and ordering of plots between DCA and DCCA, the similar correlations of DCA and CCA axes with explanatory variables, and the questionable performance of DCA beyond axis 1, we emphasize results from direct gradient analysis in this paper.

Dominant gradients at state, half-state, and subregion scales

Results from stepwise CCAs varied with geographic extent, geographic location, and sample size (numbers of plots and species) (Table 6). Total variation (TV) and sample size increased, and total variation explained (TVE) decreased, with increasing geographic extent (Table 6). In all geographic areas, TV was higher for shrub species than for tree species (Table 6). TVE was highest for all woody species (15%) and for shrubs (12%) in the Klamath subregion, and for trees in the eastern Cascades (19%). Since TVE decreased with increasing sample size, we could not discern whether the explanatory power of the models changed with geographic [TABULAR DATA FOR TABLE 6 OMITTED] extent or location. Because sample size did not appear to affect the proportions of TVE attributable to different subsets of explanatory variables, we present our findings as proportions.

We present results of stepwise CCAs at state, half-state, and subregion scales below. The half-state and subregion analyses revealed much variation in associations between environmental and compositional gradients. The within-region differences underscored the importance of ecological modeling approaches that consider a location's particular biota, physical environment, and history. For each geographic area, species with highest and lowest scores on stepwise CCA axes are shown in Table 7.

Oregon. - At the state scale, the dominant compositional gradient (CCA axis 1) reflected a gradient from the maritime climate of the Coast to the more continental climate of eastern Oregon [ILLUSTRATION FOR FIGURES 6A AND 7A OMITTED]. Plots with the lowest scores on axis 1 were at lower elevations and experienced lower seasonal variability in temperature, warmer maximum and minimum temperatures, higher rainfall, and a higher incidence of clear-cutting [ILLUSTRATION FOR FIGURE 6A OMITTED]. These plots were concentrated along the length of the coast [ILLUSTRATION FOR FIGURE 7A OMITTED] within the Picea sitchensis zone of Franklin and Dyrness (1973). Plots with high scores on axis 1 were at higher elevations and were characterized by greater seasonal variability in temperature, cooler maximum and minimum temperatures, lower precipitation, and less clear-cutting [ILLUSTRATION FOR FIGURE 6A OMITTED]. The highest plot scores were concentrated along the eastern boundary of the eastern Cascades subregion south of Bend and the southern boundary of the Blue Mountains subregion, at the transition to the shrubsteppe [ILLUSTRATION FOR FIGURE 7A OMITTED]. High-scoring plots fell largely within the Pinus ponderosa and Juniperus occidentalis zones of Franklin and Dyrness (1973).

With minor exceptions, the axis 1 gradient was longitudinal from the Coast to the foothills of the eastern Cascades, reflecting the strong climatic influence of the Pacific Ocean and the north-south orientation of the Coast and Cascade ranges. The dominant gradient in the Blue Mountain subregion was more latitudinal. Forests in the northern part of the Blue Mountain subregion receive a stronger maritime influence, and were most similar in composition to forests of the Cascades [ILLUSTRATION FOR FIGURE 7A OMITTED]. Positions of selected species in ordination space are shown in Fig. 8. Species with highest and lowest scores on the CCA axes are shown in Table 7.
TABLE 7. Species with -[greater than]1% constancy that scored
highest and lowest in stepwise canonical correspondence analyses
of all woody species, by geographic area and axis.

Axis 1                             Axis 2



Picea sitchensis                   Quercus garryana
Umbellularia californica           Quercus kelloggii
Alnus rubra                        Fraxinus latifolia
Rhamnus purshiana                  Juniperus occidentalis
Menziesia ferruginea               Arctostaphylos viscida
Vaccinium ovatum                   Rhus diversiloba
Vaccinium parvifiorum              Lonicera hispidula
Rubus spectabilis                  Ceanothus integerrimus
Oplopanax horridum                 Rubus discolor
Sambucus racemosa                  Symphoricarpos spp.


Juniperus occidentalis             Tsuga mertensiana
Cercocarpus ledifolius             Abies amabilis
Pinus contorta                     Abies lasiocarpa
Artemisia tridentata               Abies procera
Chrysothamnus viscidiflorus        Abies magnifica var. shastensis
Chrysothamnus nauseosus            Pinus monticola
Haplopappus bloomeri               Gaultheria ovatifolia
Purshia tridentata                 Vaccinium scoparium
Symphoricarpos oreophilus          Vaccinium alaskaense
Ribes cereum                       Rubus lasiococcus

Axis 3                             Axis 4


Larix occidentalis                 Quercus chrysolepis
Salix scouleriana                  Quercus kelloggii
Physocarpus malvaceus              Pinus lambertiana
Lonicera utahensis                 Cornus nuttallii
Spiraea betulifolia                Arbutus menziesii
Philadelphus lewisii               Lonicera ciliosa
Acer glabrum                       Lonicera hispidula
Ribes lacustre                     Arctostaphylos viscida
Oplopanax horridum                 Berberis piperiana
Symphoricarpos albus               Whipplea modesta


Chamaecyparis lawsoniana           Picea sitchensis
Abies magnifica var. shastensis    Alnus rubra
Lithocarpus densiflorus            Umbellularia californica
Umbellularia californica           Vaccinium ovatum
Quercus chrysolepis                Rubus spectabilis
Pinus monticola                    Sambucus racemosa
Arctostaphylos viscida             Menziesia ferruginea
Berberis piperiana                 Arctostaphylos uva-ursi
Arctostaphylos patula              Chrysothamnus viscidiflorus
Ceanothus protratus                Purshia tridentata

Axis 1                             Axis 2

Western Oregon


Picea sitchensis                   Picea sitchensis
Fraxinus latifolia                 Tsuga mertensiana
Rhamnus purshiana                  Abies amabilis
Rubus discolor                     Abies procera
Rubus spectabilis                  Pinus contorta
Menziesia ferruginea               Menziesia ferruginea
Sambucus racemosa                  Oplopanax horridum
Myrica californica                 Vaccinium scoparium
Rosa spp.                          Vaccinium alaskaense
Symphoricarpos spp.                Rubus spectabilis


Pinus contorta                     Quercus kelloggii
Pinus monticola                    Quercus garryana
Tsuga mertensiana                  Pinus ponderosa
Abies magnifica var. shastensis    Ceanothus cuneatus
Picea engelmannii                  Ceanothus integerrimus
Abies amabilis                     Arctostaphylos viscida
Vaccinium scoparium                Lonicera hispidula
Ribes viscosissimum                Rhus diversiloba
Arctostaphylos nevadensis          Amelanchier pallida
Gaultheria ovatifolia              Berberis piperiana

Eastern Oregon


Juniperus occidentalis             Pinus albicaulis
Quercus garryana                   Pinus contorta
Artemisia arbuscula                Pinus monticola
Artemisia tridentata               Tsuga mertensiana
Chrysothamnus nauseosus            Abies lasiocarpa
Chrysothamnus viscidiflorus        Abies magnifica var. shastensis
Haplopappus bloomeri               Vaccinium scoparium
Purshia tridentata                 Arctostaphylos patula
Arctostaphylos patula              Haplopappus bloomeri
Ceanothus prostratus               Ceanothus velutinus


Tsuga mertensiana                  Quercus garryana
Abies lasiocarpa                   Philadelphus lewisii
Pinus albicaulis                   Physocarpus malvaceus
Taxus brevifolia                   Holodiscus discolor
Picea engelmannii                  Acer glabrum
Vaccinium scoparium                Prunus virginiana
Vaccinium membranaceum             Ribes spp.
Ribes lacustre                     Symphoricarpos albus
Sorbus scopulina                   Spiraea betulifolia
Berberis nervosa                   Rubus parviflorus



Quercus kelloggii                  Abies procera
Quercus garryana                   Cornus nuttallii
Fraxinus latifolia                 Castanopsis chrysophylla
Arbutus menziesii                  Prunus emarginata
Calocedrus decurrens               Vaccinium membranaceum
Rosa eglanteria                    Oplopanax horridum
Rosa spp.                          Berberis aquifolium
Amelanchier alnifolia              Berberis nervosa
Symphoricarpos spp.                Ribes spp.
Rhus diversiloba                   Acer circinatum


Picea sitchensis                   Pinus contorta var. contorta
Chamaecyparis lawsoniana           Chamaecyparis lawsoniana
Tsuga heterophylla                 Fraxinus latifolia
Abies procera                      Salix hookeriana
Menziesia ferruginea               Arctostaphylos columbiana
Vaccinium alaskaense               Myrica californica
Vaccinium membranaceum             Baccharis pilularis
Oplopanax horridum                 Cytisus scoparius
Myrica californica                 Amelanchier alnifolia
Rubus spectabilis                  Ceanothus velutinus

Western Cascades


Fraxinus latifolia                 Quercus kelloggii
Quercus garryana                   Quercus garryana
Quercus kelloggii                  Pinus ponderosa
Rhamnus purshiana                  Abies magnifica var. shastensis
Acer macrophyllum                  Arctostaphylos nevadensis
Prunus emarginata                  Arctostaphylos patula
Rubus discolor                     Berberis piperiana
Rubus laciniatus                   Ribes binominatum
Symphoricarpos spp.                Ribes viscosissimum
Rhus diversiloba                   Lonicera hispidula


Abies lasiocarpa                   Thuja plicata
Abies magnifica var. shastensis    Tsuga heterophylla
Pinus contorta                     Oplopanax horridum
Tsuga mertensiana                  Rubus pedatus
Picea engelmannii                  Rubus spectabilis
Vaccinium scoparium                Vaccinium ovalifolium
Arctostaphylos nevadensis          Vaccinium alaskaense
Ribes viscosissimum                Vaccinium parvifolium
Ribes binominatum                  Sambucus racemosa
Sorbus sitchensis                  Oemleria cerasiformis



Cercocarpus montanus               Fraxinus latifolia
Cercocarpus ledifolius             Cercocarpus montanus
Quercus garryana                   Cercocarpus ledifolius
Quercus kelloggii                  Quercus garryana
Fraxinus latifolia                 Alnus rubra
Ceanothus cuneatus                 Rubus discolor
Ceanothus sanguineus               Rubus spectabilis
Symphoricarpos albus               Baccharis pilularis
Ceanothus integerrimus             Ceanothus thyrsiflorus


Sequoia sempervirens               Abies magnifica var. shastensis
Pinus contorta var. contorta       Pinus monticola
Alnus rubra                        Ribes binominatum
Berberis pumila                    Acer glabrum
Juniperus communis                 Vaccinium membranaceum
Rhododendron occidentale           Quercus sadleriana
Baccharis pulularis                Arctostaphylos nevadensis
Rhamnus californica                Arctostaphylos patula
Rubus spectabilis                  Ceanothus pumilus

Eastern Cascades


Juniperus occidentalis             Abies lasiocarpa
Cercocarpus ledifolius             Abies procera
Artemisia arbuscula                Tsuga mertensiana
Artemisia tridentata               Pinus albicaulis
Chrysothamnus viscidiflorus        Pinus monticola
Chrysothamnus nauseosus            Pinus contorta
Salix geyeriana                    Vaccinium scoparium
Purshia tridentata                 Arctostaphylos nevadensis
Haplopappus bloomeri               Ribes viscosissimum
Spiraea douglasii                  Spiraea douglasii


Abies lasiocarpa                   Quercus garryana
Abies procera                      Pseudotsuga menziesii
Tsuga mertensiana                  Larix occidentalis
Vaccinium scoparium                Lonicera ciliosa
Vaccinium membranaceum             Corylus cornuta var. californica
Berberis nervosa                   Ceanothus integerrimus
Acer circinatum                    Spiraea betulifolia
Spiraea betulifolia                Holodiscus discolor
Rubus ursinus                      Symphoricarpos spp.
Corylus cornuta var. californica   Berberis nervosa

Blue Mountains


Abies lasiocarpa                   Betula occidentalis
Pinus albicaulis                   Taxus brevifolia
Pinus contorta                     Philadelphus lewisii
Taxus brevifolia                   Berberis nervosa
Picea engelmannii                  Holodiscus discolor
Chimaphila menziesii               Acer glabrum
Chimaphila umbellata               Physocarpus malvaceus
Vaccinium scoparium                Rubus parviflorus
Sorbus scopulina                   Amelanchier alnifolia
Lonicera involucrata               Rosa spp.


Juniperus occidentalis             Pinus albicaulis
Cercocarpus ledifolius             Abies lasiocarpa
Haplopappus bloomeri               Juniperus occidentalis
Artemisia arbuscula                Artemisia tridentata
Artemisia tridentata               Arternisia arbuscula
Artemisia rigida                   Artemisia rigida
Chrysothamnus nauseosus            Chrysothamnus viscidiflorus
Chrysothamnus viscidiflorus        Vaccinium scoparium
Purshia tridentata                 Haploppapus bloomeri
Prunus spp.                        Purshia tridentata

The second CCA axis was a gradient in growing season moisture stress, from areas of warm, dry growing seasons at lower elevations to areas of cool, wet growing seasons at higher elevations [ILLUSTRATION FOR FIGURES 6A AND 7B OMITTED]. Areas of low summer precipitation and high summer temperature included the interior valleys of western Oregon, especially the Rogue and Umpqua Valleys, the foothills of the eastern Cascades from Bend northward to the Columbia River Gorge, and lower elevation areas of the Blue Mountains subregion. Lowest plot scores were concentrated in these areas as well as in the foothills of the Coast Range bordering the Willamette Valley and in scattered locations throughout the Ochoco Mountains [ILLUSTRATION FOR FIGURE 7B OMITTED], and tended to be in younger stands. In addition to Rubus discolor (Table 7), several introduced species found primarily in the interior valleys of western Oregon and along the coast were among the lowest scoring species on axis 2, but had low constancy: Crataegus monogyna, Rosa eglanteria, and Cytisus scoparius. Highest plot scores on axis 2 were in older, subalpine forests of the Cascade Range and in high-elevation areas of the northern Coast Range and the Siskiyou, Blue, and Wallowa mountains [ILLUSTRATION FOR FIGURE 7B OMITTED].

Axis 3 was most strongly correlated with geologic variables. Low-scoring plots were on mafic rocks dating to the Miocene and older. High-scoring plots were on ultramafic parent materials, primarily in the Klamath subregion [ILLUSTRATION FOR FIGURES 6B AND 7C OMITTED]. Axis 3 also reflected a climatic gradient from low to high potential solar radiation, from high to low seasonal variability in temperature, from high to low summer precipitation, and from low to high minimum temperatures [ILLUSTRATION FOR FIGURE 6B OMITTED]. The gradient was latitudinal, with low scores in the north and high scores in the south [ILLUSTRATION FOR FIGURE 7C OMITTED]. Low-scoring species on axis 3 were those with affinities for moist or disturbed sites at forest edges or openings, and were predominantly species distributed east of the Cascade crest (Table 7). Highest scoring species included several serpentine associates: Chamaecyparis lawsoniana (Zobel 1990) and Arctostaphylos viscida with [greater than]1% constancy (Table 7) and others with [less than]1% constancy (Pinus attenuata, P. jeffreyi, Ceanothus pumilus, C. cuneatus, Garrya buxifolia, Berberis pumila, Rhamnus californica, Quercus vaccinifolia, and Juniperus communis).

Axis 4 was less interpretable than the first three axes. Local site and disturbance factors were the strongest environmental correlates. Low-scoring plots were on steep slopes and in older stands with dense canopies, and concentrated in the Klamath Mountains [ILLUSTRATION FOR FIGURE 7D OMITTED]. High-scoring plots were on gentle terrain and in younger stands with sparse canopies, primarily along the southern coast and in the Paulina Mountains of central Oregon [ILLUSTRATION FOR FIGURE 7D OMITTED].

Western Oregon. - In western Oregon, the first axis was strongly correlated with elevation and temperature and moderately correlated with stand age and clearcutting [ILLUSTRATION FOR FIGURE 6C OMITTED]. Low plot scores were on warm, low-elevation sites along the coast and in the interior valleys in stands previously clear-cut, and high scores were on colder, high-elevation sites in older, uncut stands. The second axis was correlated with growing season precipitation and temperature [ILLUSTRATION FOR FIGURE 6C OMITTED]. Low scores were in areas with cool, wet summers along the coast and on Mount Hood, and high scores were in areas with hot, dry summers in the Rogue and Umpqua Valleys and in the eastern portion of the Columbia River Gorge.

Eastern Oregon. - In eastern Oregon, axis 1 followed a gradient from younger stands in areas with warm, dry summers, sparse canopies, and high moisture stress to older stands in areas with cool, wet summers, dense canopies, and low moisture stress [ILLUSTRATION FOR FIGURE 6D OMITTED]. Axis 2 followed a gradient from areas of cool temperatures at higher elevations to warmer temperatures at lower elevations. Low-scoring plots also tended to be on flat sites with high solar radiation, and high-scoring plots on steep slopes with low solar radiation. The axis 2 gradient was latitudinal, with low plot scores on mountain peaks and in the southern portion of the eastern Cascades and high scores in the northern portion of the eastern Cascades and Blue Mountains subregions.

Coast subregion. - In the Coast subregion, axis 1 reflected a gradient in summer moisture stress, annual precipitation, and seasonal temperature variability [ILLUSTRATION FOR FIGURE 6E OMITTED]. Lowest plot scores were concentrated in the Umpqua and Willamette valleys, and highest scores were along the northwestern coast. Axis 2 was strongly associated with geology, with high-scoring plots in the coastal dunes, the Willamette Valley floor, coastal forests south of Coos Bay, and wet hardwood sites in the extreme northwest corner of the state. Remaining plots were largely undifferentiated along axis 2. The strongest environmental correlates were depositional soils, elevation, and seasonal temperature variability [ILLUSTRATION FOR FIGURE 6E OMITTED].

Western Cascades subregion. - The dominant compositional gradient in the western Cascades was most strongly associated with temperature [ILLUSTRATION FOR FIGURE 6F OMITTED]: elevation, summer temperature, and seasonal temperature variability contributed 45% of TVE (Table 8). Lowest plot scores were in the foothills of the Willamette and Umpqua valleys and on previously clear-cut sites. The lowest scoring tree species on axis 1 were hardwoods (Table 7). Highest plot scores were on the peaks of the Cascade Range in stands without cutting. The second axis was strongly correlated with summer precipitation and less so with overstory cover, solar radiation, and seasonal temperature variability [ILLUSTRATION FOR FIGURE 6F OMITTED]. Axis 2 was latitudinal, from low plot scores in the south to high scores in the north.

Klamath subregion. - In the Klamath subregion, axis 1 followed a gradient from a continental climate in the eastern portion of the subregion to a maritime climate with low seasonal temperature variability, high winter precipitation, and high seasonal variability in precipitation along the coast [ILLUSTRATION FOR FIGURE 6G OMITTED]. Lowest plot scores on axis 1 were in the Rogue Valley and eastward toward the Cascade crest, and highest scores were along the coast. Axis 2 followed a gradient from low elevations and warm temperatures to high elevations and cool temperatures [ILLUSTRATION FOR FIGURE 6G OMITTED]. Plots with low scores also tended to be in younger stands with a history of clearcutting. Low scores were clustered in the Rogue Valley, the southern end of the Umpqua Valley, and along the coast. High plot scores were in the mountainous areas of southwest and southern Oregon. Despite contributing 10% of TVE (Table 8) in the Klamath subregion, ultramafic parent materials were not important until the third CCA axis.

Eastern Cascades subregion. - In the eastern Cascades, axis 1 was most strongly correlated with overstory tree cover and precipitation [ILLUSTRATION FOR FIGURE 6H OMITTED]. The axis was strongly dominated by high-scoring plots along the crest of the Cascade Range, especially near the Three Sisters and Mount Hood, which had dense canopies and high summer rainfall. Remaining, lower scoring plots were not well differentiated, but had sparser canopies, warmer and drier summers, and a higher proportion of total precipitation falling in summer [ILLUSTRATION FOR FIGURE 6H OMITTED], and were located throughout the mid- to low elevations of the subregion. Axis 2 correlated most strongly with elevation, high and low temperature extremes, and summer temperature. Lowest scores were in the Three Sisters and Paulina mountains, and the highest scores were in the extreme northern and southern portions of the subregion.

Blue Mountains subregion. - The dominant gradient in the Blue Mountains subregion reflected growing season conditions and overstory canopy cover [ILLUSTRATION FOR FIGURE 6I OMITTED]. Plots with low scores on axis 1 had low summer moisture stress and dense canopies, primarily found at high elevations in the Wallowa and Blue Mountains. High plot scores were in low-elevation areas with low summer precipitation, high summer temperatures, and sparse canopies. Axis 2 was associated with elevation, summer climate, and topography. Low plot scores were in areas of low elevation, hot and dry summers, steep slopes, and low solar radiation, and were concentrated in canyons along the Snake River and in the Blue Mountains along the northern Oregon boundary. High scores were concentrated near the peaks of the Wallowa and Strawberry mountains, and in the southern portions of the Ochoco Mountains.

Associations of local and regional factors with species gradients

The relative contributions of explanatory variables to TVE in stepwise and partial CCA were influenced by geographic extent and location, and varied with the scale of the explanatory factor (Tables 8 and 9). Unfortunately, the statistical significance of TVE differences among CCA models cannot be tested, and thus interpretations of differences are somewhat subjective. In partial CCA, regional factors (climate and geology) accounted for more of the TVE (70-79%) than local factors (topography and disturbance) (10-20%) at all geographic locations and extents (Table 9). Local factors contributed less to TVE at the state (10%) than at the subregional scale (14-20%), whereas contributions of regional measures were comparatively strong at all geographic extents (Table 9). Contributions of local [TABULAR DATA FOR TABLE 8 OMITTED] factors to TVE were slightly greater (17%) in eastern Oregon, where climate is more stressful, than in western Oregon (10%). Although subregional differences were slight, local factors were least important in the Coast (14% of TVE) where climate is most maritime and benign, and most important in the Blue Mountains (20% of TVE) where climate is most continental and variable.

At all extents and locations, climate contributed far more to TVE (46-60%) in partial CCA than any other variable subset (Table 9). Climate was particularly important (60% of TVE) in the Coast subregion. Geologic variables were second to climate in total contributions to TVE (11-19%) in all geographic areas. Disturbance (6-12%) and topography (4-8%) contributed least to TVE in our sample. The contributions of climate, geology, and disturbance variables were unrelated to geographic extent. Contributions of topographic variables to TVE increased very slightly with decreasing geographic extent, from 4% at the state scale to 5-8% at the subregion scale.

Geographic patterns of CCA model fit

Plot-score residuals from stepwise CCA illustrate geographic patterns of model fit [ILLUSTRATION FOR FIGURE 9 OMITTED]. High residuals were concentrated in locations that represented ends of the gradients of the first two CCA axes, where environmental features and community composition diverged most from regional norms. These areas included the foothills of the Willamette Valley west of Salem, the interior valleys of southwest Oregon, the coastal dunes north of Coos Bay, and subalpine areas of the Three Sisters and Wallowa mountains. Extensive areas of the state had large residual scores, as well as high among-plot variability in residual scores. The southern half of the Klamath subregion and most of eastern Oregon exhibited this "salt-and-pepper" pattern of residuals.

Spatial structure in regional gradients

Spatial position alone accounted for 7-15% of TVE (Table 10). Patterns in the relative contributions of spatial position and environment to TVE were more strongly associated with geographic location than with geographic extent. At broader geographic extents, spatial position explained only slightly more species variation, and environmental measures uncorrelated with spatial position explained slightly less, than at smaller extents (Table 10). Spatial position explained slightly more of the TVE in eastern Oregon (14%) and its subregions (11-14%) than in western Oregon (9%) and its subregions (7-9%) (Table 10). Environmental factors in the Coast and western Cascades subregions were more spatially structured (30-34% of TVE) than in the Klamath, eastern Cascades, and Blue Mountains subregions (17-24% of TVE). The subregions also differed somewhat in terms of proportions of TVE attributed to environmental measures uncorrelated with spatial position (E|S), with the Coast having the least and the Klamath the most.

Species diversity patterns

Alpha and gamma diversity were greater in western than in eastern Oregon (Table 11). Alpha diversity was greatest in the Klamath and western Cascades, lowest in the eastern Cascades and Blue Mountains, and intermediate in the Coast for both trees and shrubs. Gamma diversity was greatest in the Klamath subregion for both vegetation layers. Alpha, gamma, and beta diversity of the shrub layer were greater than for trees in all geographic areas. Beta diversity was greater in eastern than in western Oregon for both tree and shrub layers.


Environmental correlates of regional species gradients

Elevation. - Elevation contributed more to total variation explained (TVE) than any other variable (Table 8). [TABULAR DATA FOR TABLE 9 OMITTED] However, elevation is a complex-gradient (sensu Whittaker 1960, 1965) that covaries with a host of historical and environmental factors such as soil chemistry, amount and persistence of winter snowpack, and climate. In addition, in much of our dataset the elevation gradient was confounded with land ownership patterns and thus disturbance gradients. ELEV and AGE were moderately correlated in all geographic areas except for the Coast subregion (Table 12), but only in the western Oregon canonical correspondence analysis (CCA) model were these gradients collinear [ILLUSTRATION FOR FIGURE 6 OMITTED]. Elevation and clear-cutting were confounded in parts of the western half of the state, where even-aged forest management predominates. ELEV and CLEARCUT were moderately correlated (Table 12) and CCA gradients were collinear [ILLUSTRATION FOR FIGURE 6 OMITTED] at the state scale, in western Oregon, and in the western Cascades and Klamath subregions. Our sample thus did not allow us to distinguish the effects of elevation from stand age in western Oregon, nor elevation from clear-cutting in Oregon, western Oregon, and in the western Cascades and Klamath subregions. ELEV contributed most to TVE in these geographic areas (Table 8), and some of this explained variation could have been associated with the confounded disturbance variables. Nevertheless, the strong association between elevation and species composition we observed in our sample is consistent with other vegetation studies.

We debated excluding elevation from our analysis because it measures spatial position and only indirectly reflects physical environment. Excluding ELEV from stepwise CCAs, however, did not appreciably affect TVE, probably because of multicollinearity with other climate measures, nor the relations among species, plots, and explanatory variables. This robustness to multicollinearity among explanatory variables, as well as to omission of important explanatory variables, is a strength of CCA (Palmer 1993). We retained ELEV in our analyses because of its value in interpreting results.

Macroclimate. - Results supported our hypothesis of the primary importance of macroclimate and the secondary role of substrate (geology), topography, and disturbance in controlling regional compositional gradients in the predominantly mid- to late-successional forest communities we sampled (Tables 8 and 9). The minor degree of confounding of disturbance measures with climate, primarily with MAXTMP and MINTMP in the western half of the state, was insufficient to raise questions about the primary importance of climate overall.
TABLE 10. Proportion of total variation explained (TVE) by subsets
of explanatory variables in canonical correspondence analyses with
variance partitioning, all woody species, by geographic area. S =
spatial variables; E = topography, disturbance, climate, and geology
variables (see Table 3).

                                   Proportion of TVE(**)
Geographic area       S [where] E   S [intersection] E   E [where] S

Oregon                    0.15             0.30              0.55

Western Oregon            0.09             0.35              0.56
Eastern Oregon            0.14             0.22              0.64

Coast                     0.09             0.34              0.57
Western Cascades          0.07             0.30              0.63
Klamath                   0.07             0.18              0.75
Eastern Cascades          0.11             0.24              0.65
Blue Mountains            0.14             0.17              0.69

** S [where] E is variation explained by S not shared with E, S
[intersection] E is variation explained by S that is correlated with
E, and E [where] S is variation explained by E not shared with S.

Our findings cannot be compared directly with previous studies because of differences in methods. Most investigators considered broad-scale climate only indirectly, by studying vegetation change along complex-gradients of elevation or latitude, and none explicitly examined relative contributions of regional (macroclimatic) and local (microclimatic) measures of temperature and moisture. Most studies emphasized local topography as an indirect measure of moisture stress through solar heating and desiccation from wind exposure. Plant evapotranspiration stress (Waring 1969, Zobel et al. 1976) is the measure of direct interest, but it is impractical to measure in regional studies.

Seasonal variability and extremes in climate were more important in explaining species gradients than were mean annual climatic conditions. In addition, the two dominant species gradients at the state scale were associated with continentality and growing season moisture stress, climatic gradients that integrate elements of both temperature and moisture. The importance of continentality in our study was consistent with dei Moral and Watson's (1978) study that spanned an east - west gradient across the Cascade Range, although our study spanned a much wider latitudinal range. Species gradients across Oregon were much more strongly associated with moisture stress during the characteristic warm, dry summers than with winter conditions. This association was consistent with observed correspondence between growing season precipitation and temperature gradients and latitudinal vegetation gradients in the Appalachians (Cogbill and White 1991), the southern Rockies (Allen et al. 1991), and the northern California uplands (Parker 1994).

Topographic factors. - Topographic measures contributed slightly more to TVE in areas of greater climatic stress (Tables 8 and 9), as hypothesized, although subregional differences were most apparent in the stepwise CCAs (Table 8) and rather insubstantial in the partial CCAs (Table 9). Local and topographic factors explained the least variation in partial CCA in the Coast (Table 9), where climate is most mild, maritime, and wet. The Coast was the only geographic area where neither SOLAR nor ASPECT were significant in a stepwise model (Table 8). In the Eastern Cascades and Blue Mountains, SLOPE and especially TREOCOV explained more species variation than in the western subregions (Table 8). TREOCOV probably reflects a complex-gradient of disturbance, microsite environment, large-scale climate, net primary productivity, and other factors. However, we think the variable most strongly [TABULAR DATA FOR TABLE 11 OMITTED] indicates local topographic moisture, and our grouping of TREOCOV with the disturbance variables in partial CCA may have obscured subregional differences in topographic effects. Water availability strongly influences leaf area index (LAI) in northwestern coniferous forests (Grier and Running 1977, Waring et el. 1978, Gholz 1982), and LAI probably is strongly correlated with our measure of canopy cover. In fact, TREOCOV may have better integrated microtopography, soils, and large-scale climatic factors that influence actual site moisture availability to plants than did SOLAR or SLOPE.
TABLE 12. Spearman's rank correlations of disturbance variables with
elevation for 2443 field plots, by geographic area.

                               Spearman's rank correlation
                                         with ELEV

Geographic area                   AGE             CLEARCUT

Oregon                          0.40(*)          -0.42(**)

Western Oregon                  0.53(**)         -0.38(**)
Eastern Oregon                  0.32(*)          -0.09(***)

Coast                          -0.06(***)         0.05(*)
Western Cascades                0.49(***)        -0.41(**)
Klamath                         0.31(*)          -0.38(**)
Eastern Cascades                0.41(*)          -0.06(***)
Blue Mountains                  0.26(*)          -0.13(***)

* Variable included in stepwise model, not collinear with ELEV on

** Variable included in stepwise model, collinear with ELEV on axes

*** Variable not included in stepwise model, not collinear with ELEV
on axes 1 and 2.

Our findings were consistent with Del Moral and Watson (1978), who found that topography more profoundly differentiated community composition on the eastern than on the western slopes of the Washington Cascades. They also attributed major contrasts in community structure between eastern and western Washington to variation in overstory dominance, an indirect expression of climatic and topographic moisture gradients. In eastern Oregon, moisture is more limiting and temperatures are colder and more variable, topography is often deeply dissected, and topographic effects would be expected to be more pronounced (del Moral and Watson 1978, Neilson et al. 1992).

In general, other gradient studies appear to emphasize greater topographic control over vegetation pattern than we observed in our study, probably because most addressed smaller geographic extents or prestratified their study areas by elevation or substrate before examining topographic effects (e.g., Whittaker 1960, Peet 1978, 1981, Allen and Peet 1990, Allen et al. 1991). Whereas aspect and topographic position have been emphasized in other studies, slope accounted for more TVE than aspect or solar radiation in our study (Table 8). Inferring mechanisms is difficult, however, because slope integrates many hydrological and geomorphic processes such as soil formation and movement (Swanson et al. 1988), seepage, shading, and snow movement. Flat sites are more likely to have deep, moist soils, although low slopes also can occur on dry, rocky ridge-tops. Steep sites are more likely to be dry with shallow, rocky soils.

Disturbance. - Our findings on the influence of disturbance on regional species gradients were inconclusive for most geographic areas due to confounding of our explanatory variables, especially CLEARCUT with ELEV (Table 12). In western Oregon, where even-aged management predominates, the clear-cutting and elevation gradients were strongly collinear at the half-state scale and in the western Cascades and Klamath subregions (Table 12, [ILLUSTRATION FOR FIGURE 6 OMITTED]). This confounding influenced the state-level analysis as well. CLEARCUT was not significant in eastern Oregon (Table 8), where selective harvesting is most commonly practiced and very few of our plots had been clear-cut [ILLUSTRATION FOR FIGURE 4 OMITTED]. The confounding of explanatory variables was due primarily to the nonrandom geographic distribution of federal and nonfederal land ownerships across Oregon's major environmental gradients, which was reflected in our sample of field plots. The systematic grid of FIA plots provided an unbiased sample of nonfederal lands, but these forests occur at lower elevations and are younger [ILLUSTRATION FOR FIGURE 5 OMITTED] and more heavily disturbed by logging [ILLUSTRATION FOR FIGURE 4 OMITTED] than federal lands. National Forest lands occur at higher elevations and the Ecology plots were almost all from older, natural stands [ILLUSTRATION FOR FIGURES 4 AND 5 OMITTED]. The BLM lands span a range of ages and elevations, but the FIR plots were restricted to older stands in southwest Oregon.

Disturbance and elevation were not confounded in the Coast subregion: AGE and CLEARCUT were uncorrelated with ELEV (Table 12) and gradients were not collinear [ILLUSTRATION FOR FIGURE 6E OMITTED]. The Coast was the only subregion where private lands predominated (70% of the plots vs. 23-28% elsewhere), and where the mean elevations of federal and nonfederal plots did not differ. Clear-cutting was not strongly correlated with any of the first four CCA axes in the Coast. In partial CCA, only 2% of TVE was associated with CLEARCUT and uncorrelated with the other variables. Stand age was not significant in stepwise CCA (Table 8), and accounted for only 1% of TVE in partial CCA.

We attribute the small contribution of disturbance measures to explained species variation in the Coast to several factors. First, species presence-absence strongly influences regional-scale ordinations, where gradients are long and species constancy is low. Existing studies suggest that disturbance influences the relative abundances of species within a local area, but that few species are totally eliminated from a site by disturbance in general or by logging in particular (Ohmann and Bolsinger 1991, Ramey-Gassert and Runkle 1992, Halpern and Spies 1995). Furthermore, most early invaders on disturbed sites are herbaceous (Halpern and Spies 1995) and were not included in our study. In addition, the ubiquity of several long-lived, early-seral, shade-intolerant tree species in Oregon (e.g., Pseudotsuga menziesii) probably obscured some successional patterns in our analyses. Nevertheless, our findings in the Coast are consistent with Spies (1991) and Spies and Franklin (1991), who found that ecological differences among physiographic provinces were more important than stand age in explaining regional patterns of community composition in Pseudotsuga menziesii forests. Our finding in the Coast of the secondary association of disturbance factors with regional species gradients should not be confused with demonstrated effects of disturbance on other measures of biological diversity in forest ecosystems, such as vegetation structure and wildlife diversity (Hansen et al. 1991).

Across most of our Oregon study area, the confounding of explanatory variables prevented us from determining whether species gradients were influenced by elevation or by clear-cutting, or by both. In an ideal sample, field plots would be balanced across the multivariate and geographic space defined by stand age, disturbance history, and physical environment. However, very little of today's forest landscape at any elevation is in early-successional, natural forest (Hansen et al. 1991), and late-successional forests at low elevations are virtually nonexistent. Quantification of the role of disturbance relative to other environmental factors in controlling compositional gradients may be intractable using gradient analytical or other methods in regions where land ownership patterns and associated disturbance histories are confounded with environmental gradients. Chronosequence and experimental studies would be similarly challenged by the lack of available stands across the various gradients for sampling or treatment. Ultimately, research on disturbance and successional patterns needs to be based on long-term remeasurement of permanent plots from a design that is balanced across major disturbance and environmental gradients.

Geology. - More geologic variables were significant in stepwise CCA at broader geographic extents (Table 8), but overall contributions of geology to TVE were unrelated to geographic extent (Table 9). Four of the 14 geologic types were most strongly correlated with species gradients in stepwise CCA: ultramafic rock; mafic rock of the Pliocene and later; mafic pyroclastic rock, ash, ejecta, and vent deposits of the Pliocene and later; and depositional materials (Table 8, [ILLUSTRATION FOR FIGURE 6 OMITTED]). However, multicollinearity between geologic and climatic variables precluded firm conclusions about associations of specific geologic types with species gradients in much of Oregon. Additional study is needed to separate effects of climate and geology, and to determine which properties of the parent materials influence community composition.

Geographic variation in environmental correlates of species gradients

The contrast in species-environment associations among geographic subregions was an important finding of our study. Despite the strong contrasts among subregions, we were able to synthesize patterns of vegetation and environment at the regional scale. Apparent contradictions in findings among landscape-scale vegetation studies thus can be attributed to real differences in species and ecological relations among places. Nevertheless, within-region variation in the environmental associations of species gradients in our study must be interpreted carefully, due to confounding of some of the explanatory variables in some geographic areas in our sample. In particular, disturbance gradients were confounded with elevation at the state level, in western Oregon, and in the western Cascades and Klamath subregions.

The greatest contrast in Oregon's forest vegetation was between the maritime, moist climate of the Coast and western Cascades and the drier, continental climate of the eastern Cascades and Blue Mountains. The Klamath subregion was intermediate in character, reflecting the subregion's central nature to forest flora of the western United States from both historical and climate perspectives (Whittaker 1960). Chief among the within-region differences was the greater importance of moisture in explaining species variation in eastern Oregon. Continentality measures (CVPRE, CONTPRE, MINTMP, CVTMP) were more important in explaining compositional gradients in western Oregon, especially in the Coast subregion, than in eastern Oregon (Table 8, [ILLUSTRATION FOR FIGURE 6 OMITTED]). Western Oregon encompasses the influence of the Pacific Ocean, and the orographic effects of the Coast Range with the Willamette Valley in its rain shadow. Indeed, the influence of climate in general was especially strong in the Coast subregion, where elevation, temperature, and precipitation variables contributed 60% of TVE in CCA with variance partitioning (Table 9), and 78% of TVE in stepwise CCA (Table 8). The observed shifts in environmental correlates with geographic location also corroborated relations between major forest zones and primary environmental constraints hypothesized by Franklin and Dyrness (1973:50).

In general, vegetation was more complex and spatial pattern was finer grained in eastern than in western Oregon. Beta diversity was greater in eastern Oregon for both tree and shrub layers (Table 11), as was similarly observed by del Moral and Watson (1978), and more species were rare [ILLUSTRATION FOR FIGURE 10 OMITTED]. This finer scale pattern of variation in vegetation in eastern Oregon was corroborated in the geographic pattern of residuals [ILLUSTRATION FOR FIGURE 9 OMITTED] and by the greater contribution of local measures of slope and canopy cover to explained variation (Table 8). In contrast to other studies (e.g., Whittaker 1960, del Moral and Watson 1978), however, our results showed that alpha and gamma diversity were greater in western than in eastern Oregon (Table 11), probably because we excluded the species-rich herbaceous layer. In addition, we sampled a much longer latitudinal gradient than did del Moral and Watson, and western Oregon is hotter and drier than western Washington.

Collectively, our results supported the notion that where regional climate is more stressful for plant life - overly cold, hot, or dry - species respond to smaller scale variations in substrate, topography, and biotic interactions (Neilson et al. 1992). However, these fine-scale factors were poorly captured by our explanatory variables. Vast portions of the mid-elevation Coast and Cascade ranges, where the CCA model performed well more consistently [ILLUSTRATION FOR FIGURE 9 OMITTED], are within the core of the biome where climate is benign and most fine-scale resource variation is within a given species' range of tolerance (Neilson et al. 1992). In these areas the CCA models, weighted heavily toward regional climate variables, performed better than at the periphery. Gradient methods yield lower TVEs (Table 6), and presumably larger residuals, under conditions of high beta diversity such as we encountered in eastern Oregon.

Sampling effects and unexplained variation in CCA

Several characteristics of CCA and our data affected our levels of explained variation and interpretation of results. Chief among these was the confounding of TVE with sample size (numbers of plots and species) (Table 6). We propose the following mechanism behind the sample-size effect. Greater numbers of plots sample increasing numbers of species as more rare species are encountered. Rare species are those with small geographic ranges, narrow habitat specificity, or small, nondominant population sizes (Rabinowitz 1981). The rare species increase total variation in the species-by-plot data matrix, but their presence is unlikely to be explained by environmental variables, and TVE thus decreases. The species-area relation and the positive relation between species distribution and abundance account for this sampling effect (Hanski et al. 1993). Current models of regional species distribution (Levins 1969, Brown 1984, Collins and Glenn 1991, Brown 1995) predict that most species are locally distributed and regionally rare. Studies of vegetation (Whittaker 1960, del Moral and Watson 1978) and other taxa have shown that species with more extensive distributions tend to be more locally abundant than species with more restricted distributions (Hanski et al. 1993). Because locally rare species are more difficult to detect than locally abundant species, the number of sites at which a species is found with some fixed scheme of sampling is a monotonically increasing function of the average abundance of the species (Hanski et al. 1993). The shapes of our species frequency distributions were consistent with these models [ILLUSTRATION FOR FIGURE 10 OMITTED]. Species rarity also increased with geographic extent in our study: at the subregion scale, 28-44% of species had constancies of [less than]1%, but 50% of species were this rare at the state scale.

The sample-size effect limited our ability to compare CCAs among geographic extents and locations with different sample sizes, led us to present TVEs as proportions rather than actual amounts, and led us to emphasize the direction rather than magnitude of TVE differences. The sample-size effect also calls into question the comparison of TVEs among studies. Our TVEs fell on the low end of the 10-50% range that is typical of gradient analyses (Gauch 1982, Palmer 1993), but this may be attributable to our substantially longer gradients and sample sizes. Sampling issues as they pertain to CCA merit further investigation, especially in regional studies of long gradients.

Several aspects of our sample design and data contributed to our unexplained variation and limited our conclusions. Of primary importance was the confounding of land ownership patterns, and associated disturbance histories, with major environmental gradients in Oregon. Also, our estimates of species diversity were influenced by sample size (number of plots) and plot size (Magurran 1988) to an unknown degree (Table 11). Diversity comparisons also were complicated by the different areal extents of the subregions. Other sampling effects and sources of error can be attributed to the scale of the field plots (plot size) relative to the spatial resolution of the explanatory variables, misregistration of the plot locations to the mapped explanatory variables, errors in the climate models and other mapped data, and errors in the field-recorded data. Although our purpose was to infer regional-scale pattern, observations of community composition were from plots of predetermined size, shape, and distribution. These traits varied among the component data sets (Table 2), but we had no reason to suspect the differences would bias our results. Plot size was not associated with CCA axes 1-4 in any geographic area.

We doubt that sampling or data quality had any substantive effects on our overall findings, other than limiting conclusions about clear-cutting associations with regional species gradients to the Coast subregion. The indirect gradient analyses corroborated conclusions from the direct gradient analyses on the important environmental controls of regional species gradients. The species variation explained by spatial position that was uncorrelated with environmental measures (Table 10) can be considered a synthetic descriptor of unmeasured underlying processes (Borcard et al. 1992). Our results suggested that omitted explanatory factors were potentially most important in eastern Oregon and to a lesser degree at broader geographic extents (Table 10). In the drier habitats of eastern Oregon, measures of fine-scale environment such as understory light and moisture conditions probably would explain additional variation. At the state scale, unmeasured historical factors such as evolutionary processes and species dispersal and migration probably contributed to spatial pattern in the species data. Future advances in the spatial representation of environmental measures thought to control regional species gradients, and advances in the understanding of sampling issues as they pertain to CCA, will improve regional gradient analyses.


Our study was the first systematic quantification, synthesis, and mapping of vegetation-environment gradients across a large, contiguous region of the western United States based on field plot data. The study examined associations of both local- and regional-scale environmental factors with regional species gradients across forest land of all land ownerships. Regional patterns of variation in species composition were complex and multidimensional. Environmental correlates of species gradients, species diversity patterns, and the spatial patterning of plant communities varied with geographic extent and location.

Our findings supported a conceptual model of multiscaled controls on vegetation distribution, and the related notion that local community structure is the result of both regional- and local-scale processes. Broad-scale climate was the primary control on regional species gradients: climatic factors were the strongest associates of community gradients across the range of geographic extents we examined. At the state scale, the two dominant gradients reflected climatic regimes that integrate temperature and moisture factors: continentality and moisture stress during the growing season. Geology, topography, and disturbance factors were secondary in explaining species variation across all of our geographic extents and locations. Local topographic factors (slope and tree canopy cover, an indirect measure of site moisture) explained slightly more variation at smaller geographic extents and in areas of less equable climate. Spatial structure in our species data that was uncorrelated with explanatory variables suggested that fine-scale understory environment and unmeasured historical factors may have influenced present-day community patterns.

Our gradient study was the first to examine associations between human disturbance and species gradients using plot data at the regional scale. However, the confounding of land ownership patterns and their associated disturbance histories with elevation gradients across much of Oregon, and in our sample, rendered our findings inconclusive in all but the Coast subregion. In the Coast, clear-cutting accounted for only 2% of explained variation in partial canonical correspondence analysis (CCA). Stand age was not significant in stepwise CCA, and accounted for only 1% of explained variation in partial CCA. Ordinations of long, regional gradients are influenced more by species presence than by abundance, and few woody species are totally eliminated or introduced to sites by clear-cutting. Quantifying the role of disturbance in regional species gradients may be intractable in regions where geographic patterns of ownership and environment are confounded. Further research on successional patterns should be based on long-term remeasurement of permanent plots from a design that is balanced across regional disturbance and environmental gradients.

Within Oregon and for the range of geographic extents we examined, variation in environmental correlates of species gradients was more strongly associated with geographic location than with geographic extent. The most striking contrast in vegetation within our region was between the wet, maritime climate of the Coast and western Cascades subregions in northwest Oregon, and the dry, variable, continental climate of eastern Oregon. The Klamath subregion was intermediate in character. Our environmental measures, which were heavily weighted to climate, more successfully explained patterns of community composition in western than in eastern Oregon. In the benign climate of western Oregon, topographic effects were minimal and geographic patterns of variation in gradients were coarse grained. In the drier and less equable climate of eastern Oregon, both climatic and topographic moisture were substantially more important in explaining compositional gradients, local measures of slope and tree canopy cover assumed greater importance, beta diversity was greater, more species were rare, and geographic patterns of variation in community composition and environment were more fine grained.

Despite strong ecological contrasts within the region, we were able to synthesize species-environment relations at the regional scale. These findings suggested that apparent conflicts among local vegetation studies can be explained by real ecological differences among places. Indeed, our regional study provides a broader context for considering gradient and classification studies conducted at smaller, landscape scales within the region.

We demonstrated the tremendous potential for quantifying and mapping regional ecological patterns through multivariate and geographic analysis of large plot-scale data sets and geographic data. Regional gradient studies can provide the quantitative, ecological basis for stratifying regions into relatively homogeneous subregions for finer scale modeling and analysis, as well as for developing regional strategies for conserving biological diversity, for predicting and planning for the effects of global climate change, and for the inventory and monitoring of forest vegetation at broad geographic scales. Although there are several important limitations to direct gradient analyses based on regional, plot-scale databases, such findings provide a strong quantitative and empirical basis for generating hypotheses about regional ecological pattern and process for further study with complementary approaches, such as regional experiments, retrospective studies, and paleoecological studies.


We thank Lisa Lackey and Paul Hardwick for their assistance in geographic information systems. Brad Smith shared several of his software tools as well as plot data from southwest Oregon. We are indebted to numerous individuals of the Area Ecology Program and the Pacific Resource Inventory, Monitoring, and Evaluation Program for providing plot data. Review comments from William Baker, Fred Hall, Mike Palmer, Mark Roberts, Don Zobel, and two anonymous reviewers substantially improved our manuscript. The study was funded by the Pacific Northwest Research Station, USDA Forest Service.


Agee, J. K. 1993. Fire ecology of Pacific Northwest forests. Island Press, Covelo, California, USA.

Allen, R. B., and R. K. Peet. 1990. Gradient analysis of forests of the Sangre de Cristo Range, Colorado. Canadian Journal of Botany 68:193-201.

Allen, R. B., R. K. Peet, and W. L. Baker. 1991. Gradient analysis of latitudinal variation in Southern Rocky Mountain forests. Journal of Biogeography 18:123-139.

Bailey, R. G., P. E. Avers, T. King, and W. H. McNab, editors. 1994. Ecoregions and subregions of the United States (map), scale 1:7 500 000. U.S. Department of Agriculture, Forest Service, Washington, D.C., USA.

Beers, T. W., P. E. Dress, and L. C. Wensel. 1966. Aspect transformation in site productivity research. Journal of Forestry 64:691-692.

Bolsinger, C. L., and K. L. Waddell. 1993. Area of old-growth forests in California, Oregon, and Washington. U.S. Forest Service Resource Bulletin PNW-RB-197.

Borcard, D., P. Legendre, and P. Drapeau. 1992. Partialling out the spatial component of ecological variation. Ecology 73:1045-1055.

Brown, J. H. 1984. On the relationship between abundance and distribution of species. American Naturalist 124:255-279.

-----. 1995. Macroecology. University of Chicago Press, Chicago, Illinois, USA.

Cogbill, C. V., and P. S. White. 1991. The latitude-elevation relationships for spruce-fir forests and treeline along the Appalachian mountain chain. Vegetatio 94:153-175.

Collins, S. L., and S. M. Glenn. 1991. Importance of spatial and temporal dynamics in species regional abundance and distribution. Ecology 72:654-664.

Currie, D. J., and V. Paquin. 1987. Large-scale biogeographical patterns of species richness of trees. Nature 329:326-327.

Daly, C., R. P. Neilson, and D. L. Phillips. 1994. A statistical-topographic model for mapping climatological precipitation over mountainous terrain. Journal of Applied Meteorology 33:140-158.

Daubenmire, R. 1978. Plant geography with special reference to North America. Academic Press, New York, New York, USA.

Daubenmire, R., and J. B. Daubenmire. 1968. Forest vegetation of eastern Washington and northern Idaho. Washington Agricultural Experiment Station Technical Bulletin 60.

del Moral, R., and R. S. Fleming. 1979. Structure of coniferous forest communities in western Washington: diversity and ecotope properties. Vegetatio 41:143-154.

del Moral, R., and A. F. Watson. 1978. Gradient structure of forest vegetation in the central Washington Cascades. Vegetatio 38:29-48.

Denton, S. R., and B. V. Barnes. 1987. Tree species distributions related to climatic patterns in Michigan. Canadian Journal of Forest Research 17:613-629.

Dodson, R., and D. Marks. 1997. Daily air temperature interpolation at high spatial resolution over a large mountainous region. Climate Research 8:2-20.

Dyrness, C. T., J. F. Franklin, and W. H. Moir. 1974. A preliminary classification of forest communities in the central portion of the western Cascades in Oregon. Coniferous Forest Biome Bulletin 4.

ESRI. 1991. ARC command references. Volume 6.1. Environmental Systems Research Institute, Redlands, California, USA.

Fonda, R. W., and L. W. Bliss. 1969. Forest vegetation of the montane and subalpine zones, Olympic Mountains, Washington. Ecological Monographs 39:271-301.

Franklin, J. F. 1988. Pacific Northwest forests. Pages 104130 in M. G. Barbour and W. D. Billings, editors. North American terrestrial vegetation. Cambridge University Press, New York, New York, USA.

Franklin, J. F., and T. Blinn. 1988. Natural vegetation of Oregon and Washington - commentary and bibliographic supplement. Supplement to J. F. Franklin and C. T. Dyrness. 1973. Natural vegetation of Oregon and Washington. United States Department of Agriculture Forest Service General Technical Report PNW-8.

Franklin, J. F., and C. T. Dyrness. 1973. Natural vegetation of Oregon and Washington. United States Department of Agriculture Forest Service General Technical Report PNW-8.

Gauch, H. G., Jr. 1982. Multivariate analysis in community ecology. Cambridge University Press, New York, New York, USA.

Gholz, H. L. 1982. Environmental limits on aboveground net primary production, leaf area, and biomass in vegetation zones of the Pacific Northwest. Ecology 63:469-481.

Glenn-Lewin, D.C. 1977. Species diversity in North American temperate forests. Vegetatio 33:153-162.

Grier, C. C., and S. W. Running. 1977. Leaf area of mature northwestern coniferous forests: relation to site water balance. Ecology 58:893-899.

Halpern, C. B., and T. A. Spies. 1995. Plant species diversity in natural and managed forests of the Pacific Northwest. Ecological Applications 5:913-934.

Hansen, A. J., T. A. Spies, F. J. Swanson, and J. L. Ohmann. 1991. Conserving biodiversity in managed forests. BioScience 41:382-392.

Hanski, I., J. Kouki, and A. Halkka. 1993. Three explanations of the positive relationship between distribution and abundance of species. Pages 108-116 in R. E. Ricklefs and D. Schluter, editors. Species diversity in ecological communities-historical and geographical perspectives. University of Chicago Press, Chicago, Illinois, USA.

Hickman, J. C., editor. 1993. The Jepson manual - higher plants of California. University of California Press, Berkeley, California, USA.

Hitchcock, C. L., and A. Cronquist. 1973. Flora of the Pacific Northwest. University of Washington Press, Seattle, Washington, USA.

Isaacs, E. H., and R. M. Srivastava. 1990. Applied geostatistics. Oxford University Press, Oxford, UK.

Johnson, C. G., Jr., R. R. Clausnitzer, P. J. Mehringer, and C. D. Oliver. 1994. Biotic and abiotic processes of eastside ecosystems: the effects of management on plant community ecology, and on stand and landscape vegetation dynamics. United States Forest Service General Technical Report PNW-GTR-322.

Levins, R. A. 1969. Some demographic and genetic consequences of environmental heterogeneity for biological control. Bulletin of the Entomological Society of America 15: 237-240.

Lowry, W. P., and P. P. Lowry. 1989. Fundamentals of biometeorology: interactions of organisms and the atmosphere. Peavine, McMinnville, Oregon, USA.

Magurran, A. E. 1988. Ecological diversity and its measurement. Princeton University Press, Princeton, New Jersey, USA.

Menge, B. A., and A. M. Olson. 1990. Role of scale and environmental factors in regulation of community structure. Trends in Ecology and Evolution 5:52-57.

Minore, D. 1972. A classification of forest environments in the South Umpqua Basin. United States Department of Agriculture Forest Service Research Paper PNW-129.

Morrison, P. H., and F. J. Swanson. 1990. Fire history and pattern in a Cascade Range landscape. United States Forest Service General Technical Report PNW-GTR-254.

Mueller-Dombois, D., and H. Ellenberg. 1974. Aims and methods of vegetation ecology. John Wiley and Sons, New York, New York, USA.

Neilson, R. 1987. Biotic regionalization and climatic controls in western North America. Vegetatio 70:135-147.

Neilson, R. P., G. A. King, R. L. DeVelice, and J. M. Lenihan. 1992. Regional and local vegetation patterns: the responses of vegetation diversity to subcontinental air masses. Pages 129-149 in A. J. Hansen and F. di Castri, editors. Landscape boundaries: consequences for biotic diversity and ecological flows. Springer-Verlag, New York, New York, USA.

Neilson, R. P., and L. H. Wullstein. 1983. Biogeography of two southwest American oaks in relation to atmospheric dynamics. Journal of Biogeography 10:275-297.

Ohmann, J. L., and C. L. Bolsinger. 1991. Monitoring biodiversity with permanent plots - landscape, stand structure, and understory species. Pages 525-526 in L. F. Ruggiero, K. B. Aubry, A. B. Carey, and M. H. Huff, technical coordinators. Wildlife and vegetation of unmanaged Douglas-fir forests. United States Forest Service General Technical Report PNW-GTR-285.

Okland, R. H., and O. Eilertsen. 1994. Canonical correspondence analysis with variation partitioning: some comments and an application. Journal of Vegetation Science 5: 117-126.

Palmer, M. 1993. Putting things in even better order: the advantages of canonical correspondence analysis. Ecology 74:2215-2230.

Parker, A. J. 1994. Latitudinal gradients of coniferous tree species, vegetation, and climate in the Sierran-Cascade axis of Northern California. Vegetatio 115:145-155.

Peet, R. K. 1978. Latitudinal variation in southern Rocky Mountain forests. Journal of Biogeography 5:275-289.

-----. 1981. Forest vegetation of the Colorado Front Range - composition and dynamics. Vegetatio 45:3-75.

Peet, R. K., R. G. Knox, J. S. Case, and R. B. Allen. 1988. Putting things in order: the advantages of detrended correspondence analysis. American Naturalist 131:924-934.

Powell, D. S., J. L. Faulkner, D. R. Darr, Z. Zhu, and D. W. MacCleery. 1993. Forest resources of the United States. 1992. U.S. Forest Service General Technical Report RM-234.

Rabinowitz, D. 1981. Seven forms of rarity. Pages 205-217 in H. Synge, editor. The biological aspects of rare plant conservation. John Wiley and Sons, Chichester, West Sussex, UK.

Ramey-Gassert, L. K., and J. R. Runkle. 1992. Effect of land use practices on composition of woodlot vegetation in Green County, Ohio. Ohio Journal of Science 92:25-32.

Richerson, P. J., and K. Lum. 1980. Patterns of plant species diversity in California: relation to weather and topography. American Naturalist 116:504-536.

Ricklefs, R. E. 1987. Community diversity: relative roles of local and regional processes. Science 235:167-171.

Ricklefs, R. E., and D. Schluter, editors. 1993. Species diversity in ecological communities - historical and geographical perspectives. University of Chicago Press, Chicago, Illinois, USA.

Riegel, G. M., B. G. Smith, and J. F. Franklin. 1992. Foothill oak woodlands of the interior valleys of southwestern Oregon. Northwest Science 66:66-76.

SAS. 1990. SAS procedures guide, version 6, third edition. SAS Institute, Cary, North Carolina, USA.

Smith, B. G. 1993. SOLARPDX - a program to estimate solar insolation. Version 1.0, 25 July 1993. Available from Bradley Smith, Deschutes National Forest, 1645 Highway 20 E, Bend, Oregon 97701 USA.

Spies, T. A. 1991. Plant species diversity and occurrence in young, mature, and old-growth Douglas-fir stands in western Oregon and Washington. Pages 111-121 in L. F. Ruggiero, K. B. Aubry, A. B. Carey, and M. H. Huff, technical coordinators. Wildlife and vegetation of unmanaged Douglas-fir forests, U.S. Forest Service General Technical Report PNW-GTR-285.

Spies, T. A., and J. F. Franklin. 1991. The structure of natural young, mature, and old-growth Douglas-fir forests in Oregon and Washington. Pages 90-109 in L. F. Ruggiero, K. B. Aubry, A. B. Carey, and M. H. Huff, technical coordinators. Wildlife and vegetation of unmanaged Douglas-fir forests. U.S. Forest Service General Technical Report PNW-GTR-285.

Swanson, F. J., N. C. Kratz, and R. G. Woodmansee. 1988. Landform effects on ecosystem patterns and processes. BioScience 38:92-98.

ter Braak, C. J. F. 1986. Canonical correspondence analysis: a new eigenvector technique for multivariate direct gradient analysis. Ecology 67:1167-1179.

-----. 1987a. CANOCO - a FORTRAN program for canonical community analysis by correspondence analysis, principal components analysis, and redundancy analysis (Version 2.1). Technical Report LWA-85-02. Agricultural Mathematics Group, Wageningen, Netherlands.

-----. 1987b. Ordination. Pages 91-173 in R. H. Jongman, C. J. F. ter Braak, and O. F. R. van Tongeren, editors. Data analysis in community ecology. Pudoc, Wageningen, The Netherlands.

-----. 1988. Partial canonical correspondence analysis. Pages 551-558 in H. H. Bock, editor. Classification and related methods of data analysis. Elsevier Science, North-Holland, the Netherlands.

ter Braak, C. J. F., and I. C. Prentice. 1988. A theory of gradient analysis. Advances in Ecological Research 18: 271-313.

Teensma, P. D. A. 1987. Fire history and fire regimes of the central western Cascades of Oregon. Dissertation. University of Oregon, Eugene, Oregon, USA.

Topik, C., N. M. Halverson, and T. High. 1988. Plant associations and management guide for the ponderosa pine, Douglas-fir, and grand fir zones, Mt. Hood NF. U.S. Forest Service Publication R6-ECOL-TP-004-88.

Volland, L. A. 1985. Plant associations of the central Oregon pumice zone. U.S. Forest Service Publication R6-ECOL-104-1985.

Walker, G. W., and N. S. MacLeod. 1991. Geologic map of Oregon, scale 1:500 000, two sheets. United States Geological Survey, Washington, D.C., USA.

Waring, R. H. 1969. Forest plants of the eastern Siskiyous: their environmental and vegetational distribution. Northwest Science 43:1-17.

Waring, R. H., W. H. Emmingham, H. L. Gholz, and C. C. Grief. 1978. Variation in maximum leaf area of coniferous forests in Oregon and its ecological significance. Forest Science 24:131-140.

Warlng, R. H., and J. F. Franklin. 1979. Evergreen coniferous forests of the Pacific Northwest. Science 204:1380-1386.

Waring, R. H., and J. Major. 1964. Some vegetation of the California coastal redwood region in relation to gradients of moisture, nutrients, light and temperature. Ecological Monographs 34:167-215.

Wellner, C. A. 1989. Classification of habitat types in the western United States. Pages 7-21 in D. E. Ferguson, P. Morgan, and ED. Johnson, compilers. Proceedings - land classifications based on vegetation: applications for resource management. U.S. Forest Service General Technical Report INT-257.

White, D., A. J. Kimerling, and W. S. Overton. 1992. Cartographic and geometric components of a global sampling design for environmental monitoring. Cartography and Geographic Information Systems 19:5-22.

Whittaker, R. H. 1960. Vegetation of the Siskiyou Mountains, Oregon and California. Ecological Monographs 30: 279-338.

-----. 1965. Dominance and diversity in land plant communities. Science 147:250-260.

Whittaker, R. H., and W. A. Niering. 1965. Vegetation of the Santa Catalina Mountains, Arizona: a gradient analysis of the south slope. Ecology 46:429-452.

Woodward, F. I. 1987. Climate and plant distribution. Cambridge University Press, New York, New York, USA.

Zobel, D. B. 1990. Chamaecyparis lawsoniana (A. Murr.) Parl.: Port-Orford-cedar. Pages 88-96 in R. M. Burns and B. H. Honkala, editors. Silvics of North America. United States Department of Agriculture Agricultural Handbook 654.

Zobel, D. B., A. McKee, G. M. Hawk, and C. T. Dyrness. 1976. Relationships of environment to composition, structure, and diversity of forest communities of the central western Cascades of Oregon. Ecological Monographs 46: 135-156.
COPYRIGHT 1998 Ecological Society of America
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1998 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Ohmann, Janet L.; Spies, Thomas A.
Publication:Ecological Monographs
Date:May 1, 1998
Previous Article:Effects of plant composition and diversity on nutrient cycling.
Next Article:Responses of early successional northern hardwood forests to changes in nutrient availability.

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters