Site Classification

Moderator:

DAVID H. VAN LEARClemson University

YELLOW-POPLAR SITE INDEX AND SOIL PROPERTIES INTHE PISGAH NATIONAL FOREST

W. Henry McNab’

Abstract-The relationship of site index (SI) of yellow-poplar (Liriodendron tulipifem L.) to soil series and soil physical andchemical properties was studied in a small area of mountainous terrain in western North Carolina. Site index ranged from26.4 meters (m) to 38.7 m on 39 sample plots. Average SI of stands on Brevard soils (29.6 m) was significantly lower thanthat of stands on Tusquitee (33.2 m) or Haywood soils (33.5 m). Site index was correlated mainly with humic matter andbulk density. Regression models using age and bulk density as independent variables explained 50 percent of the variationin stand height at age 50. Residuals of models in which SI was the dependent variable were highly correlated with standage. Models based on stand height at age 50 were always superior to those based on SI. Soil properties explained only amoderate proportion of variation in total height, and this suggests that other factors, such as topography and climate, mayalso be important determinants of site quality.

INTRODUCTIONSite index (SI-total tree height at a specified age, usually50 years) has been used to express productivity of forestsites for many commercial tree species since the 1920’s.Determination and use of SI are not completelystraightforward. Biased curve construction methods (Beckand Trousdell 1973) influence of stand density (Eigel andothers 1982, Ike and Huppuch 1968) and sample treeselection (Lloyd and Jones 1983) can present problems.Despite these problems, SI is widely used as anexpression of site quality because of its ease of applicationand interpretation. Site index is seemingly well suited foruse in evaluation of sites dominated by yellow-poplarbecause this species is generally shade intolerant, displaysapical dominance during early to middle ages, and occursin even-aged stands.

Site index has long been presented in association withsoil map units to provide information on forest productivityand suitable tree species. Ike and Huppuch (1968) foundthat yellow-poplar height was correlated with groups ofsoil series in north Georgia. Eigel and others (1982)reported that yellow-poplar site quality was correlated withseveral soil series in eastern Kentucky, but Phillips (1966)found little association of SI with soil series in NewJersey. Also, VanLear and Hosner (1967) found thatyellow-poplar SI did not differ from soil mapping unit tosoil mapping unit for five such mapping units in southwestVirginia. Most researchers have found that soil taxonshave not accounted effectively for variation in yellow-poplar SI.

The influence of environmental factors on SI of yellow-poplar has been evaluated in various soil-site studies.Prediction models based on soil, topographic, and climaticvariables have been developed (Auten 1945, Brown andMarquard 1988, Ike and Huppuch 1968, Munn andVimmerstedt 1980, Smalley 1964, Tryon and others 1960).Usually, yellow-poplar growth is correlated mostly with soiland topographic properties that affect moisture availabilityduring the growing season, and somewhat with factors

that influence nutrient availability. In most cases theproportion of SI variation explained is moderate, between50 and 75 percent. A problem with most soil-siterelationships is lack of validation with independent datasets (Broadfoot 1969).

Because the results of studies of soils and SI have oftenbeen inconclusive, and have sometimes beencontradictory, additional study of the factors affecting sitequality is desirable. The purpose of this exploratory studywas to evaluate the relative importance of specific soilvariables that affect site quality for yellow-poplar, and totest for correlations between soil taxonomic units andyellow-poplar SI.

THE STUDY AREAThe study was conducted in the Pisgah Mountains ofwestern North Carolina, near Asheville. Geologicformations of this region consist of highly metamorphosedPrecambrian muscovite and biotite schist and gneisses,and occasional Devonian granitic outcrops. Local reliefranges from gently rolling hills to steep mountain slopes.Surficial geologic deposits consist of Holocene andWisconsin loamy colluvium 2 to 5 meters (m) thick. Soilsare mostly relatively deep [>I00 centimeters (cm)], have aloam surface texture, are well drained, are fine to coarsein texture, and acidic. Dystrochrepts are typically presenton moderate to steep slopes and Ultisols are present inlow elevation intermountain basins where soils haveformed in residuum. Mean annual temperature averages13 “C. and mean monthly temperature ranges from 2 “C.in January to 22 “C. in July. Annual precipitation averages120 cm in low-elevation mountain valleys but increaseswith altitude.

Over 100 arborescent species are indigenous to thisregion. Oaks (Quercus sp.) dominate dry slope and ridgesites at elevations below about 1500 m. Moist lowerslopes and cove forests are dominated by mesophyticspecies such as yellow-poplar. Conifers may be presentalso.

’ Research Forester, USDA Forest Service. Southern Research Station. 1577 Brevard Road. Asheville. NC 28806

57

METHODSPlots used in this study were originally established in theearly 1960’s for study of growth and yield of yellow-poplar(Beck and Della-Bianca 1970). Thirty-nine quarter-acre(0.10 hectare) plots in the Pisgah Mountains were selectedfor use in the present study. These were clustered in twogroups: (1) 19 plots in the Bent Creek Experimental Forest,situated at an elevation of 830 m about 16 kilometers (km)south of Asheville (N35’30’00”, W82’37’30”), and (2) 20plots in Mince Cove (N35’22’30”, W82’40’00”), about 8 kmwest of Bent Creek Experimental Forest, at the sameaverage elevation. Most plots are on cove landforms, havenortheast to southeast aspects, and have slope gradientsbetween 20 and 40 percent. Beck’s (1962) curves wereused to determine SI for each plot when the original studywas installed, and the Sl’s were reported in theestablishment report (On file, Bent Creek ExperimentalForest).

A soil scientist classified the soils on each plot to the seriestaxonomic unit during the summer of 1983. Soils weredescribed in detail by subhorizon to a maximum depth ofabout 140 cm, which on most plots included the entire A-horizon and much of the B-horizon. Depth to each visiblechange of texture and color within each horizon wasmeasured and composite samples were collected fordetermination of physical and chemical soil properties.Analysis was done by the North Carolina Soils TestLaboratory for: (1) humic matter (HM) (percent), (2) bulkdensity (BD) (grams per cm3), (3) P (milligrams per cubicdecimeter), (4) K (millequivalents per 100 cm3), (5) Ca(millequivalents per 100 cm3), (6) Mg (millequivalents per100 cm3), (7) Na (millequivalents per 100 cm3), (8) cationexchange capacity (CEC) (millequivalents per 100 cm3), (9)acidity factors (AF) (millequivalents per 100 cm3), and (10)pH (log H-ion concentration). Three other soil propertieswere derived: (11) total bases (TB) (sum of Ca, Mg, K, Na),(12) total cations (TC) (calculated as TB+AF), and (13)base saturation (BS) (percent) (calculated as (TB/TC)*lOO).Concentrations of soil chemicals and values for physicalproperties were weighted by thickness of each subhorizonto approximate a value for the whole horizon. Values forthe A and B horizons were weighted by the thicknesses ofthe horizons to characterize the solum. Although pH is

expressed on a logarithmic scale, it was treated as anonlogarithmic variable.

Simple relationships between soil properties and sitequality were examined by plotting, correlation, and simpleregression models. Site index was used as the dependentvariable in one model to account for variation in heightassociated with age, thereby allowing maximum emphasison the relative importance of soil properties. However,because SI curves are themselves predictions based onmodels, a more direct accounting of the importance of soilvariables was made by expressing height at age 50 (HT50)as the dependent variable in the second regression model.

Separate regression analyses were made using soilvariables for each horizon and the solum for (1) all seriescombined, and (2) individual series, where adequatereplications were available. A dummy variable was used toaccount for differences in site quality associated with thetwo locations. Simple correlation coefficients (r) were usedto evaluate the relationships between dependent andindependent variables. Analysis of variance was used totest for significant differences in mean SI among series.Scheffe’s test at the 0.05 level was used to separatemeans by soil series. Data were analyzed by stepwiseregression using combination forward and backwardmethods for inclusion of variables in the model thatproduced the largest coefficients of determination (r’). Thelevel of significance for inclusion of variables in the modelwas pcO.10. All analyses were made using SAS (SASInstitute 1985).

RESULTS AND DISCUSSIONAverage stand age was 53.5 (+11.9 standard deviation)years; average height was 32.8 (k3.3) m; and SI averaged29.6 (+2.3) m. Initial analyses indicated no significant effectof group locations on SI.

Soil SeriesThe 39 sample plots were situated on soils representing 6taxonomic series (table 1). All series were well drained,relatively deep (greater than 100 cm), and had loamsurface texture and mesic temperature regime. Brevardand Watauga series have clay accumulation in the B

Table l-Number (N) of plots sampled, site index (SI), site type, and taxonomic class bysix soil series for yellow-poplar stands in western North Carolina

Series N Sia (range) Site Taxonomic class

Ashe 1Brevard 6Haywood 8Porters 1Tusquitee 22Watauga 1

33.7 (-)29.6 (26.7-32.7)33.5 (28.9-38.7)32.3 (-)33.2 (26.4-38.3)34.2 (-)

SlopeBenchSlopeSlopeCoveSlope

Typic DystrochreptsTypic HapludultsCumulic HaplumbreptsUmbric DystrochreptsUmbric DystrochreptsTypic Hapludults

Yn meters, base 50 years

horizon. Only three series (Brevard, Haywood, andTusquitee) were represented by an adequate sample size(three or more plots) for testing mean differences in SI.Scheffe’s test indicated that mean SI of Brevard (29.6k3.6m) was significantly lower than mean SI of Haywood(33.5k3.6 m) or Tusquitee (33.2k2.9 m) and that mean SIof Haywood was not significantly different from mean SI ofTusquitee. The range of SI was least for Brevard (6.0 m)and greatest for Tusquitee (11.9 m).

Site index was correlated with relatively few properties ineach soil series (table 2). Humic matter and Bulk densitywere most consistently correlated with SI. Many soilproperties were significantly correlated with other soilproperties. Correlated properties included humic matterand bulk density (r=-0.62), and CEC and Bulk density(r=-0.48). For Tusquitee, SI was correlated with only onevariable: Humic matter of the solum.

The insensitivity of SI to soil series in my study is similar tothat reported for other locations in the SouthernAppalachians. Ike and Huppuch (1968) found a range of 35feet (10.7 m) in SI for eight soil series, but differences in SIwere only at the extremes. VanLear and Hosner (1967)reported a lack of significant correlation between SI andsoil series and suggested that classification criteria fortaxonomic units are more important for agricultural cropsthan for trees. Also, individual soil series can be describedrather broadly, and broadness of description could allowexcessive variation in site quality within series. Althoughthis study revealed that yellow-poplar SI for the Brevardseries differed significantly from that for other series, thisfinding is of limited usefulness because areas of Brevardsoils are small and restricted to floodplain terraces. Thefact that SI is about identical for Haywood and Tusquitee is

surprising because these series differ considerably inthickness of A-horizon, which is typically correlated withsite quality for yellow-poplar and other species. Shortlyafter this study was concluded, the Tusquitee taxon wassubdivided into two series (the new series is Tuckasegee),and this subdivision may reduce the broad range of soilproperties associated with Tusquitee.’

Soil PropertiesDescriptive statistics for soil fertility factors summarized intable 2 are presented by individual cations for Tusquiteeand all series in table 3. Ca accounts for about half or moreof all cations in the A and B horizons and in the solum; Nais the least common cation. Acidity factors (mostly H andAl) are relatively constant among all soil layers forTusquitee and all series. Site index was not significantlycorrelated with level of any cations for the Tusquitee seriesor for the pooled series data sets. The range of cationlevels found on the study plots with Tusquitee soils nearlyequaled the range reported by the Natural ResourcesConservation Service that characterizes this series3.

’ Personal communication. 1997. Sara A. Browning, AssistantRanger, Nantahala National Forest, 90 Sloan Road, Franklin, NC28734.

3 Unpublished Natural Resource Conservation Service regionaldata for Tusquitee and Tuckasegee series were combined toapproximate the Tusquitee series recognized in 1983 and resultedin a range of CEC for the A horizon of 3.1-20.0 millequivalents per100 cubic centimeters and 1.2-4.9 millequivalents per 100 cubiccentimeters for the B horizon. In this study, CEC ranges forTusquitee soils were 3.1-14.7 and 2.2-5.2 millequivalents per 100cubic centimeters for the A- and B- horizons, respectively.

Table 2-Mean soil propertiesa and correlation with site indexb by series and horizon for yellow-poplar standssampled in western North Carolina

Series Hzn Thk TB CEC BS HM PH P BD

Brevard A 15 4.1* 6.3 56 1.1* 5.5* 2.3 0.9B 114* 1.4 2.8 45 0.2 5.2 0.9 1.1*Solum 129* 1.7* 3.2 46 0.3 5.3* 1.1 1.1*

Haywood A 55 3.3 6.3 41 2.4* 5.3 2.7 0.8BC 92* 0.9 4.0* 20 1.2* 4.9 2.3 1.0Solum 147* 1.6 4.7* 26 1.8* 5.1 2.6 0.9*

Tusquitee A 23 3.5 6.5 45 1.9 5.4 2.8 0.9B 108 1.0 3.4 29 0.7 5.1 1.3 1.0Solum 131 1.5 3.9 32 0.9* 5.2 1.5 1.0

All A 28 3.7 6.5 48 1.8* 5.4 2.8 0.9B 103 1.1 3.5* 31 0.7* 5.1 1.4 1.0*Solum 131 1.6 4.0* 34 0.9* 5.2 1.7 1.0*

a Soil property abbreviations and units of measure: Hzn=horizon, Thk=thickness (cm), TB=total bases (meq/l 00cm3),CEC=cation exchange capacity (meq/100cm3), BS=base saturation (pet), HM=humic matter (pet), BD=bulk density (g/cm3)b *Indicates significant (pcO.1) correlation with site index.’ No B horizon measured on one plot where A horizon >I40 cm.

59

Table 3-Means and ranges of selected chemical properties of the A and B horizonsof Tusquitee and all soil series for yellow-poplar stands in western North Carolina

Tusquitee (n=22) All series (n=39)

Soilpropertya Mean Range Mean Range

A horizonThicknessCaMgKNaTBAFCECBS

23.22.540.780.200.013.522.946.46

54

15.2-45.70.10-11.40.10-2.180.06-0.440.00-0.03

b

b

b

b

28.02.620.860.210.013.702.796.49

57

10.2-I 39.70.10-11.40.10-2.180.06-0.480.00-0.03

b

b

b

b

B horizonThicknessCaMgKNaTBAFCECBS

108.30.500.390.140.011.042.403.44

30

86.4-124.50.10-2.040.08-I .Ol0.04-0.290.00-0.03

b

b

b

b

102.70.510.440.410.011.112.363.47

32

58.4-124.50.10-2.040.08-I .620.03-0.360.00-0.20

b

b

b

b

a Soil property abbreviations and units of measure: Thickness in cm, Cations (Ca, Mg, K, Na) inmeq/l 00cm3, TB=total bases (meq/l 00cm3), AF=acidity factors (meq/l 00cm3), CEC=cationexchange capacity (meq/l 00cm3), BS=base saturation (pet).b Range not determined.

Regression analysis of SI as a function of soil propertieswas done using all series pooled (39 plots) and forTusquitee, the only series with adequate sample size (22plots). The best regression model for all series combinedpredicted SI as a function of Humic matter alone, butvariation explained was low (r2= 0.28). Examination ofresiduals around the regression revealed a highlysignificant correlation of SI with stand age (r=-0.64). Whenage was included in the regression the total proportion ofvariation explained increased (r’=0.50), but variationexplained by bulk density, the only significant soil variable,was low (r’=0.18). For Tusquitee samples, age explained asignificant proportion of variation in SI (?=0.37), but no soilvariables entered the model. The Tusquitee series appearsto be intermediate to the Brevard and Haywood series inmany soil characteristics, but characteristics of the threeseries overlap considerably. Comparison of data uponwhich the model for all series is based suggests that datafor the Brevard and Haywood series, with respectivelyhigher and lower bulk densities than Tusquitee, account forthe significance of the relationship between SI and bulkdensity (fig. 1).

P 30-3 A EIREVARD0 HAYWOOD 0

28 0 OTHER0 TUSQUITEE a- PREDICTED SI a A26

24 ,0.7 0.8 0.9 1.0 1.1 1.2

BULK DENSITY (G/CM3)

Figure l-Actual and predicted site index in relation tosolum bulk density and soil series for yellow-poplar standsin western North Carolina.

60

Results using HT50 as the dependent variable in multivariatemodels indicated that stand age accounted for mostvariation (table 4). Bulk density was the most important soilvariable for models of the B horizon and solum. Horizonthickness was also related to HT50, but explained littlevariation. For models based on Tusquitee series, stand agewas the only significant variable. Although HT50 is a moreappropriate dependent variable to use for predicting yellow-poplar site quality, the model accounted for about the sameproportion of variation as SI.

Results of this study suggest that height of yellow-poplar ismore a function of soil physical properties than of soil fertility.Many researchers have reached similar conclusions aboutthe importance of variables that influence the soil moistureregime (Auten 1945, Hay and others 1987, Phillips 1966,Schomaker 1958, Smalley 1964). Fertilization studiesconducted on poor sites indicate that yellow-poplar respondsin height and diameter to increased N, P, and K (Finn andWhite 1966, Schomaker and Rudolph 1964) but response isgenerally less on good sites (Hay and others 1987, McAlpine1959). Loftus (1971) reported that development of yellow-poplar root systems may be sensitive to Ca, particularly inthe B horizon where increased acidity reduces potentialavailability of this cation. Ca levels in this study weresomewhat above those reported by Loftus for Hartsells soilsin Tennessee and were not significantly related to SI.However, the level of some soil cations could be influencedby decomposition of yellow-poplar litter because foliage ofthis species is high in Ca and Mg (Cotrufo 1977). Soils in thepresent study, represented mainly by Tusquitee, appeared toprovide minimum requirements for moderate but notexceptional height growth.

The overall lack of association between soil cations andheight of yellow-poplar is consistent with findings ofSchomaker and Rudolph (1964). Schomaker and Rudolph(1964) found no correlation between growth of yellow-poplar and soil cation levels in a 25-year-old plantation inMichigan, but relationships between growth and levels of

nutrients in foliage and litter were highly significant. Finnand White (1966) and Gilmore and others (1968) alsofound high correlations between levels of nutrients infoliage and yellow-poplar growth. Heiberg and White (1956)state that foliar analysis provides a direct assessment ofsoil nutrients available and used by trees for growthcompared to indirect evaluation based on soil analysis.

Other environmental variables in addition to thoseassociated with soils must be considered when predictinggrowth of yellow-poplar in the Southern Appalachians.Topographic variables have an important influence onavailability of moisture (Ike and Huppuch 1968) and mayinfluence transpiration and respiration, length of growingseason, and other factors related to tree growth but notexpressed by soil properties. Climatic variables are alsocorrelated with yellow-poplar growth (Beck 1985, Tryon andMyers 1952). As Smalley (1964) suggests, growthresponse of yellow-poplar to environmental factors iscomplex and a number of compensating and interrelatedsite factors are likely to be important.

In summary, even though wide ranges of SI and soilcharacteristics were included in this study, soil series or soilcharacteristics appear to be of limited usefulness forpredicting site quality for yellow-poplar. The few soilvariables that were significantly related to yellow-poplar SIsuggest that variables affecting soil moisture regime aremore important than those affecting soil fertility. Morestudy-and especially study that includes topographicvariables, samples in soil series other than Tusquitee, andnutrient analysis of foliar samples-is needed to determinethe actual value of soil characteristics as predictors of sitequality for yellow-poplar.

ACKNOWLEDGMENTDonald E. Beck, retired Research Forester, USDA ForestService, Bent Creek Experimental Forest, was responsiblefor the design and field implementation of this study.

Table 4-Partial and total correlation of significant regression variablesa withtotal height at age 50 years for yellow-poplar stands on Tusquitee and allsoil series in western North Carolina

Soil series Horizon Variable 1 Variable 2 Variable 3 Total

____________ Variable (rz) _ _ _ _ _ _ _ _ _ _ _ _ _

T u s q u i t e e A Age(0.47) - 0.47B Age(0.47) - 0.47Solum Age(0.47) - 0.47

All senes A Age(0.34) TH(0.08) PH(0.07) 0.49B Age(0.32) BD(0.15) TH(0.03) 0.50Solum Age(0.32) BD(0.18) - 0.50

a Age=stand age (years), TH=thickness of soil layer (cm), BD=bulk density (g/cm3),PH=pH

61

LITERATURE CITEDAuten, J.T. 1945. Prediction of site index for yellow-poplar

from soil and topography. Journal of Forestry. 43: 662-668.

Beck, D.E. 1962. Yellow-poplar site index curves. Res.Note 180. Asheville, NC: USDA Forest Service,Southeastern Forest Experiment Station. 2 p.

Beck, D.E. 1985. Is precipitation a useful variable inmodeling diameter growth of yellow-poplar. In:Shoulders, Eugene, ed. Proceedings of the third biennialsouthern silvicultural research conference; 1984November 7-8; Atlanta, GA. Gen. Tech. Rep. SO-54.New Orleans, LA: U.S. Department of Agriculture,Forest Service, Southeastern Forest Experiment Station:555-557.

Beck, D.E.; Della-Bianca, L. 1970. Yield of unthinnedyellow-poplar. Res. Pap. SE-58. Asheville, NC: U.S.Department of Agriculture, Forest Service, SoutheasternForest Experiment Station. 20 p.

Beck, D.E.; Trousdell, K.B. 1973. Site index: accuracy ofprediction. Res. Pap. SE-108. Asheville, NC: U.S.Department of Agriculture, Forest Service, SoutheasternForest Experiment Station. 7 p.

Broadfoot, W.M. 1969. Problems in relating soil to siteindex for southern hardwoods. Forest Science. 15: 354-364.

Brown, J.H.; Marquard, R.D. 1988. Site index of yellow-poplar in relation to soils and topography in theAllegheny Plateau of Ohio. Northern Journal of AppliedForestry. 5: 34-38.

Cotrufo, C. 1977. Nutrient content in litterfall of anAppalachian hardwood stand. Journal of the ElishaMitchell Scientific Society. 93: 27- 33.

Eigel, R.A.; Blevins, R.L.; Wittwer, R.F.; Liu, C.J. 1982.Site index of yellow-poplar in relation to soil andtopography in eastern Kentucky. In: Muller, R.N., ed.Fourth central hardwood forest conference proceedings.1982 November 8-l 0; [Proceedings city and state notavailable][Publication city and state not available]: 207-219.

Finn, R.F.; White, D.P. 1966. Commercial fertilizersincrease growth in a yellow-poplar plantation. Journal ofForestry. 64: 809-810.

Gilmore, A.R.; Geyer, W.A.; Boggess, W.R. 1968.Microsite and height growth of yellow-poplar. ForestScience. 14: 420-426.

Hay, R.L.; Hammer, R.D.; Conn, J.P. 1987. Soil propertiesdominate yellow-poplar seedling growth. SouthernJournal of Applied Forestry. 11: 119-123.

Heiberg, S.O.; White, D.P. 1956. A site evaluationconcept. Journal of Forestry. 54: 7-10.

Ike, A.F.; Huppuch, C.D. 1968. Predicting tree heightgrowth from soils and topographic site factors in theGeorgia Blue Ridge Mountains. Pap. 54. Macon, GA:Georgia Forestry Commission, Georgia Forest ResearchCouncil. 11 p.

Lloyd, F.T.; Jones, E.P., Jr. 1983. Density effects onheight growth and its implications for site indexprediction and growth projection. In: Jones, Earle P., Jr.,ed. Proceedings of the second biennial southernsilvicultural research conference; 1982 November 4-5;Atlanta, GA. Gen. Tech. Rep. SE-24. Asheville, NC:U.S. Department of Agriculture, Forest Service,Southeastern Forest Experiment Station: 329-333.

Loftus, N.S. 1971. Yellow-poplar root development onHartsells subsoils. Res. Note SO-131. New Orleans, LA:U.S. Department of Agriculture, Forest Service,Southern Forest Experiment Station. 5 p.

McAlpine, R.G. 1959. Response of planted yellow-poplarto diammonium phosphate fertilizer. Res. Note 132.Asheville, NC: U.S. Department of Agriculture, ForestService, Southeastern Forest Experiment Station. 2 p.

Munn, L.C.; Vimmerstedt, J.P. 1980. Predicting heightgrowth of yellow-poplar from soils and topography insoutheastern Ohio. Soil Science Society of AmericaJournal. 44: 384-387.

Phillips, J.J. 1966. Site index of yellow-poplar related tosoil and topography in southern New Jersey. Res. Pap.NE-52. Upper Darby, PA: U.S. Department ofAgriculture, Forest Service, Northeastern ForestExperiment Station. 10 p.

SAS Institute Inc. 1985. SAS User’s Guide, Version 5Edition. Cary, NC: SAS Institute Inc. 1290 p.

Schomaker, C.E. 1958. Two-year results of plantingyellow-poplar in north Alabama. Journal of Forestry. 56:37-38.

Schomaker, C.E.; Rudolph, VJ. 1964. Nutritionalrelationships affecting height growth of planted yellow-poplar in southwestern Michigan. Forest Science. 10:66-76.

Smalley, G.W. 1964. Topography, soil, and the height ofplanted yellow-poplar. Journal of the Alabama Academyof Science. 35: 39-44.

Tryon, E.H.; Myers, C.A. 1952. Periodic precipitationaffects growth of yellow-poplar on a West Virginiahillside. Castanea. 17: 97-102.

Tryon, C.P.; Beers, T.W.; Merritt, C. 1960. Themeasurement of site quality for yellow-poplar. Journal ofForestry. 58: 968-969.

VanLear, D.H.; Hosner, J.F. 1967. Correlation of SI andsoil mapping units poor for yellow-poplar in southwestVirginia. Journal of Forestry. 65: 22-24.

BOTTOMLAND HARDWOOD CLASSIFICATIONFOR USE IN WILDLIFE MANAGEMENT PLANNING

D. David Pritchard, John D. Hodges, and Bruce D. Leopold’

Abstract-A bottomland hardwood classification system for use in wildlife management planning in the Lower MississippiAlluvial Valley was developed and has specific implications as a practical and flexible means of identifying and delineatinghabitats based on characteristics meaningful to a broad range of native wildlife species. Classification methods weredeveloped and evaluated on Twin Oaks and Mahannah Wildlife Management Areas north of Vicksburg, MS. Subsequently,maps were created using a geographical information system and have been used for management planning. A review ofthis classification and mapping process is provided and offers insight into opportunities and limitations with use of anysimilar system in this or other forest types or geographical regions.

INTRODUCTIONIncreasingly, land managers are obligated to include, andoften to prioritize, wildlife management objectives whenmanaging forested land. Successful incorporation of wildlifeobjectives into forest management planning requires thatmanagers know the types and distribution of existingwildlife habitat as it relates to wildlife species of interest.Historically, classification of forest characteristics has beenlimited to identifying stand types as they relate to timberproduction objectives. Recent efforts to identify forestcharacteristics for use in wildlife management planninghave emphasized the development of more intensive, andoften species-specific, habitat evaluation systems such ashabitat suitability indices. Though clearly valuable to somemanagers, these often laborious and costly systems maynot be practical or economical for managers withoutprimary wildlife management objectives or with limitedfinancial and professional resources.

Furthermore, no system has previously been developed toclassify bottomland hardwood forest habitats for use inmanaging a broad range of native wildlife species. Thoughwork has been done to develop a more comprehensivesystem to evaluate wildlife communities and habitats forbottomland hardwood forests, at the initiation of this projectno other system was completed or published and itremained unclear whether any of the ongoing workaddressed the need for a system that is practical for use bymanagers with the aforementioned limited resources.Therefore, a need existed to develop a system that, whilelimiting the need for labor and professional resources,would allow managers to identify bottomland hardwoodforest habitats meaningful to a broad range of nativewildlife species.

Subsequently, a project was initiated by the College ofForest Resources at Mississippi State University incooperation with the U.S. Army Corps of Engineers(COE) and the Mississippi Department of Wildlife,Fisheries and Parks to develop a forest managementplan for Twin Oaks and Mahannah Wildlife Management

Areas. These two areas, located north of Vicksburg, MS,in the Lower Mississippi Alluvial Valley, were purchasedby the COE for mitigation of habitat losses during theconstruction of the Tennessee-Tombigbee Waterway andwere to be managed for “maximum sustained publicrecreational opportunities for consumptive and non-consumptive users” (U.S. Army Corps of Engineers1994). For managers to create and implement a forestmanagement plan to meet these objectives, existingforest conditions had to be identified, described, andmapped in detail.

This need offered an opportunity to evaluate a bottomlandhardwood classification system for use in wildlifemanagement planning. Thus, a classification system wasdeveloped to provide a practical means of obtaining andmapping habitat information meaningful to a wide range ofnative wildlife species. This system was then implementedand evaluated for use in wildlife management planning onTwin Oaks and Mahannah

METHODSThough the system developed for this project is designedfor use in the bottomland hardwood forests on Twin Oaksand Mahannah, the process of classification offers insightinto using similar systems in other forest types orgeographical regions. The process is as follows: (1) designthe classification system by determining variables to beclassified and establishing categories for each variable; (2)define stands on aerial photographs by delineatingdifferences in classification; (3) verify and describe standsby classifying variables in the field; (4) create stand mapsusing final classifications; and (5) incorporate into ageographical information system (GIS).

Classification SystemThe classification system is divided into two sections (fig.1). The first section includes “definitive variables” and thesecond section “auxiliary variables.” Definitive variablesdefine stand boundaries and stand type, whereas auxiliaryvariables further describe the stand.

’ Executive Assistant to the CEO, National Hardwood Lumber Association, P.O. Box 34518, Memphis, TN 38148-0518; Vice President, LandManagement, Anderson-Tully Company, P.O. Box 28, Memphis, TN 38101; and Professor, Wildlife Ecology, College of Forest Resources,Mississippi State University, MS 39762 (respectively).

63

L

I. Definitive variables1. Cover class

(1) Forested land(2) Nonforested land

2. inundation(1) Uncontrolled(2) Controlled

ii. Auxiliary variables7. Vertical stratification

(1) Main canopy only(2) With understory(3) With midstory(4) With mid- and understory

8. Topographical position(1) Ridge

3. Main canopy species association(1) Red oaks, sweetgum, (sweet pecan)(2) Elm, ash, sugar-berry

(2) Fiat(3) Slough

(3) Red oaks, overcup, (elm/ash/sugarberry) 9. Presence of ground flora(4) Overcup, ash, bitter pecan (1) Sparse(5) Pure red oak (2) Moderate(6) Pure sweetgum (3) Heavy

4. Average size class of main canopy stems 10. Woody debris coverage(1) Regeneration (1) NoneI;; Spating (2) 1 to 5 percent

(4) SOmali sawtimber(3) 6 to 10 percent(4) Greater than 10 percent

(5) Large sawtimber(6) Veterans

5. Main canopy density(1) Sparse(2) Normal(3) Dense

6. Residuals(1) Without residuals(2) With residuals

Figure l-Forest classification variables for Twin Oaks Wildlife Management Area, SharkeyCounty, MS.

Any variable can be included in the second section of thesystem but, as will be discussed later, any definitivevariable must be readily identifiable on aerial photographs.In fact, any variable that can be directly or indirectlyclassified on aerial photographs should first be included asa definitive variable even if it is likely to be changed to anauxiliary variable later. By so doing, initial stands aredivided as finely as possible and can always be mergedlater if needed.

Practical implementation of this system requires that allvariables, whether definitive or auxiliary, be easilyrecognizable and classifiable. Physical characteristics ofthe forest, as is pointed out by Smith (1990), commonlymeet these criteria. However, easily recognizedcharacteristics are not necessarily easily classified and,therefore, care must be taken to define, for each variable,classes (or categories) that are meaningful, but practical.

Because any one species of wildlife may perceive aspecific forest characteristic with finer or coarser resolutionthan another species, it is not practical to define categoriesbased on its meaning to every species. It is practical,however, to define categories that can be readily identified

by managers and yet still provide some indication of habitatsuitability for most native wildlife. Also, starting with toomany classes may require changes later but, again, it iseasier to merge two separate stands than it is to divide oneexisting stand.

Though it may seem more logical to order variables bytheir importance within the system, any one characteristiccould be the most important, or not important at all, to anygiven wildlife species of concern. For this reason, it isactually more practical to order variables by their relativeease in delineating and classifying. For example, it iseasier to determine whether a stand is forested ornonforested than to determine its species association (fig.1).

Aerial ImageryDelineating stands on aerial imagery is a commonly usedstratification technique to reduce sample size (Avery andBerlin 1992) and allows for more accurate boundarydefinition than could be accomplished in the field. Colorinfrared photographs taken during spring greenup or duringfall drawdown provide excellent species distinction. At anapproximate scale of 1:12,000, good quality photographs

64

contain enough detail to classify all defining variables whilestill providing an aerial view wide enough to delineateentire stands on a single frame. Images are viewed instereo and a change in any one of the definitive variablesconstitutes the delineation of a separate stand.

Field ClassificationUsing the stands defined and classified on aerialphotography, the next step of this process is to verifyclassification and further describe stands with auxiliaryvariables. Sampling and measuring classified variablesmay assure accurate classification, or simple observationmay provide sufficient verification when time or funding islimited (assuming variables can be easily identified andclassified).

Auxiliary variables are not a part of stand delineation onaerial imagery, but rather are classified in the field. Aspreviously discussed, auxiliary variables can be added atany time without changing stand boundaries or type.Classification of these variables simply allows a previouslydefined stand to be described in any amount of detaildeemed important for management planning. Again,classification of auxiliary variables may be achieved bysampling and measuring or by observation only.

MappingOptions for the form of the final map range from blueprintstyle, hand-drafted maps to comprehensive GIS’s. For thisproject, geographically corrected stand maps were digitizedwith Arc/Info (ESRI 1992a) to establish a GIS for bothstudy areas. The resulting data were then transferred toArc/View (ESRI 199213) and merged with database filescontaining stand classifications. Finally, a color coded mapwas created for each variable or combination of variablesneeded.

RESULTS AND DISCUSSIONThough development of a classification process was theobjective of this project, this process was implemented ontwo study areas and, therefore, has been evaluated insome respects. Results and discussion of this evaluationwill be presented in the same order as methods.

Classification SystemFigure 1 represents the classification system used fordescribing habitat conditions on Twin Oaks. This was thefirst system developed and, with minor modifications, itallowed for what appears to be a relativelycomprehensive, broad-based description of existing foresthabitat conditions. However, when this system and itsvariables were applied on Mahannah, moderatemodifications were necessary to achieve a similardescription of habitat conditions. Though thesemodifications required additional resources, the processwas still valid and the system was still practical toimplement. It appears that with appropriate modificationsto certain variables and their respective classes, and theaddition of any pertinent auxiliary variables, thisclassification and mapping process can be used in otherforest types and geographical regions.

Aerial ImageryThough practical and accurate classification would not bepossible without the information gained from aerialphotographs, there appear to be three major limiting factorsassociated with this step of the classification process.These factors are summarized as follows: (1) acquisition ofusable photographs, (2) opportunities to work with skilledphotographic interpreters in a cooperative environment,and (3) practical limitations as to what variables can beused to define and delineate stands.

Acquisition of quality aerial photographs was limited by theavailability of skilled aerial photographers and weatherconditions during ideal greenup or drawdown periods.Though acceptable 1:24,000 spring color infrared photosalready existed for Twin Oaks, no such images were foundfor Mahannah. Contracting a capable aerial photographerto photograph Mahannah during ideal periods with clearweather proved to be difficult.

Once all photographs were finally obtained, skilledphotographic interpreters were readily available through theMississippi Remote Sensing Lab at Mississippi StateUniversity. These resources, however, may not be asreadily available elsewhere and, thus, may limitopportunities to successfully implement this process.

Assuming acquisition of good-quality photography andavailability of skilled interpreters pose no problem, alimitation to this step of the process still exists in that eventhe most experienced photographic interpreter cannotclassify on aerial photographs every variable that may beimportant to managers. It therefore stands to reason that adefinitive variable must be directly or indirectly classifiableon aerial photography. Nonetheless, any variable deemedessential can be included as an auxiliary variable todescribe the defined stand.

As for photographic requirements for this project, early-to-mid-April or early-to-mid-October appear to be best forideal color signatures. Though a scale of 1:6,000 may haveprovided the best detail for interpretation, the increasedchance of delineating entire stands on one frame and thelower cost of power frames made 1:12,000 an effectivescale for this study. It was also noted that even with1:12,000 scale images, some frames did not containenough visible landmarks to allow for stand location on theground.

Field ClassificationAs mentioned previously, verifying classifications in thefield could potentially include sampling and measuringvariables, or observation alone may offer sufficientassurance of accurate classifications if all variables can bereadily identified and classified. While it does notnecessarily require more professional knowledge toaccurately classify variables by observation than to sampleand measure those same variables, classifying byobservation may, in fact, require a better understanding ofthe ecology common to the forest type being classified.Without quantitative measurements of certain forest

65

characteristics or an in depth understanding of forestecology, confidence in the accuracy of stand classificationsis likely to be low.

MappingBecause the time and economical requirements of thisclassification process are crucial to its usefulness, andbecause the choice of final map form can be relativelyinexpensive or expensive, it is helpful that the choice offinal map form is last in this process. Given that the GIScreated for this project ultimately provides an efficientmeans for evaluating potential habitat alterations, addingother geographically referenced information, and identifyingstands that contain unique habitat characteristics, creationof a GIS appears to be a useful and practical choice forfinal maps if sufficient time and funding remain afterclassification and verification are complete.

CONCLUSIONSThis classification system and process appear to provideinformation meaningful in managing bottomland hardwoodforests for a wide range of native wildlife species and wasrelatively practical to implement. However, because thiswas an initial evaluation of the process, a need still existsto quantify the meaningfulness of the classified variables tonative wildlife species and to determine whether this

process has any real economical and practical advantageover more intense and quantitative habitat evaluationsystems. It may be that the professional resources requiredfor acquisition and interpretation of aerial photography, andthe cost of creating final maps, negate any advantage tousing this process instead of existing habitat evaluationsystems.

LITERATURE CITEDAvery, T.E.; Berlin, G.L. 1992. Fundamentals of remote

sensing and airphoto interpretation. New York:Macmillan Publishing Company. 472 p.

ESRI, Inc. 1992a. ARC/INFO computer software.Redlands, CA: Environmental Systems ResearchInstitute, Inc.

ESRI, Inc. 199213. ARC/VIEW computer software.Redlands, CA: Environmental Systems ResearchInstitute, Inc.

Smith, R.L. 1990. Ecology and field biology. New York:Harper and Row. 922 p.

U.S. Army Corps of Engineers. 1994. Tennessee-Tombigbee Waterway wildlife mitigation project:mitigation implementation plan for the Mississippi Deltalands: Mahannah and Twin Oaks Wildlife ManagementAreas, MS. Mobile, AL: U.S. Army Engineer District,Mobile.

LANDSCAPE ECOSYSTEM CLASSIFICATION FORTHE MOUNTAIN HIGHLANDS OF THE CHEROKEE NATIONAL FOREST

Benjamin E. West and John C. Rennie’

Abstract-A landscape ecosystem classification (LED) was developed for the 26,100- hectare (ha) Mountain Highlandssection [610 to 1660 meters (m) in elevation] of the Tellico Ranger District, Cherokee National Forest, Tennessee.Vegetation, soils, and landform data were obtained from 90 0.04-ha plots located in stands representing late successionalstages. Indirect ordination was used to distinguish site units (plots with similar vegetation), followed by stepwise discriminantanalysis to determine those environmental variables that contributed most to the classification of the vegetation. Atechnique of direct ordination, detrended canonical correspondence analysis, was also used to identify site types fromvegetation, soils, and landform.

Five ecological site units were identified and found to be recurring across the landscape. These were: (1) yellow pine/mixedoak forest, (2) mixed oak/red maple forest, (3) cove hardwoods/mixed mesophytic forest, (4) eastern hemlock/yellow birchforest, and (5) beech forest. Each site unit represents an array of site types which are identifiable combinations of landformand soil features, associated with discrete vegetation communities. A four-variable model consisting of elevation, aspect,slope position, and A horizon depth accurately predicted site unit membership more than 60 percent of the time. For thepurposes of developing a management strategy, elevation and aspect appear to be sufficient for predicting the distributionof site units across the landscape of the Mountain Highlands. Using a geographic information system (GIS), an ecosystemmap was developed for the Mountain Highlands based on elevation and aspect.

INTRODUCTIONNatural resource managers are being asked to manage awider range of natural resources, to meet broader andmore diverse objectives, and to accomplish this with fewerfinancial resources. To achieve this, they will need newtools and techniques. Landscape ecosystem classification(LED) offers one approach to management that meetsthese criteria.

Development of an LED is generally completed in fourphases (Jones and Lloyd 1993): Phase l-identifyingvegetation types (site units) from late successional stages;Phase 2-determining seral vegetation stages associatedwith each site unit; Phase 3-mapping site units; andPhase 4-developing management strategies. The actualclassification of ecosystems across the landscape is basedon specific vegetation, soils, and landform characteristics.The advantage of LED models is that once a model for aparticular landscape has been developed, it is possible topredict vegetation types from the more permanent featuresof its landform and soil characteristics.

This study is part of similar LED research being conductedthroughout the Blue Ridge province of the SouthernAppalachians by the USDA Forest Service and coordinatedby Henry McNab of the Bent Creek Experimental andDemonstration Forest. The overall research effort includes11 sites in Virginia, North and South Carolina, Georgia, andTennessee.

STUDY SITEThis study represents Phase 1 of an LED for the MountainHighlands section of the Tellico Ranger District in theCherokee National Forest, Tennessee (fig. 1). Previous

work (Yoke 1994, Yoke and Rennie 1996) defined twogeographic sections of the district based on homogenousgeological groups. These were then related to elevation forfield use. The Foothill section coincides with the WaldenCreek geologic group and ranges in elevation from 305 to610 meters (m) (Yoke 1994). The Mountain Highlandssection coincides with the Great Smoky group and occursabove an elevation of 610 m.

The Mountain Highlands section consists of approximately26,100 hectares (ha) in Monroe County, TN. Temperaturesrange from 2 to 8 degrees cooler than the mean annualtemperature for the rest of the county (13 “C.). Annualrainfall is about 191 centimeters (USDA 1981). All of thesoils are classified as lnceptisols with the main subgroupsincluding Typic and Umbric Dystrochrepts, and TypicHaplumbrepts (Scott ND).

Human activity in the Mountain Highlands section has beenlong and varied. Early use by Native Americans (about10,000 years ago) was primarily for hunting and impactsfrom human occupation were likely minimal (Bass 1977).With the more permanent settlement patterns associatedwith the introduction of agriculture, clearing of the forestoccurred and use of fire became more extensive (Chapman1985). Logging in the higher elevations and on the steeperslopes of the Mountain Highlands was not widespread untilafter the turn of the 20th century (Weals 1991).

METHODS

Field MethodsWith the assistance of personnel from the Tellico RangerDistrict, mature stands more than 75 years old were

’ Graduate Research Assistant and Professor, respectively, Department of Forestry, Wildlife and Fisheries, The University of Tennessee,Knoxville 37901-I 071.

67

Figure l-Location of Tellico Ranger District, Cherokee National Forest, Tennessee. Dark area represents theMountain Highlands section (areas above 610 meters in elevation).

identified. A matrix of aspect (north-facing and south-facingslopes) and slope type (ridges, upper, middle, and lowerside slopes, and coves) was used to distribute plots overas much of the vegetation and environmental conditions aspossible. Data were collected only in stands with few signsof disturbance, based on ground observations.

Vegetation, soils, and landform data were obtained from 90stands representing late successional stages. Vegetation

was sampled in four strata: trees, saplings and shrubs,regeneration, and herbs and vines. In each stand, a set ofnested plots was established (fig. 2) with measurementsbeing taken to estimate coverage and density by species foreach stratum. Soil variables, determined from a soil pitwithin the 0.04-ha plot, included: humus thickness, Ahorizon thickness, and solum thickness. Topographicvariables were: aspect, slope angle, slope position,landform index (McNab 1993) terrain shape index (McNab

68

X

1 -------=6.32 m t r a n s e c t

X 4Herbs 8z Vines

Figure 2-Arrangement and sizes of plots used to sample vegetation

1989) and elevation. Aspect was transformed from degreesto a linear scale ranging from 0 for southwest aspects to 2for northeast aspects (Beers and others. 1966).

Data AnalysisVegetation data were summarized by species and stratumin terms of percent cover and density of the plot area.Importance Value (IV) 200 (relative coverage + relativedensity) was calculated for all strata except the herbaceousstratum, where IV 100 (relative coverage) was used. A datamatrix based on these importance values was composed of294 species and 90 plots.

Two techniques of indirect ordination were used todistinguish site units, i.e., plots with similar vegetation:polythetic divisive classification using TWINSPAN (Hill1979a) and detrended correspondence analysis (DCA)using DECORANA (Hill 197913). Stepwise discriminantanalysis (SWDA) was then used to determine thoseenvironmental variables that contributed most to theclassification of the vegetation. Combinations oftopographic and soil variables were used to predict site unitmembership for each plot with cross-validation of thediscriminant analysis.

Direct ordination with detrended canonical correspondenceanalysis (DCCA) using CANOCO (ter Braak 1987) wasalso used to identify site types from vegetation, soils, andlandform. Results of the direct and indirect gradientanalyses were compared and evaluated for prediction ofsite units. A model to predict site units from soil andtopographic variables was developed with particularconcern for its use on the ground by natural resourcemanagement personnel.

RESULTS AND DISCUSSIONDCA ordination using DECORANA and TWINSPANclassification of the vegetation data resulted in theidentification of five recurring ecological site units. Thesewere: (1) yellow pine/mixed oak forest, (2) mixed oak/redmaple forest, (3) cove hardwoods/mixed mesophytic forest,(4) eastern hemlock/yellow birch forest, and (5) beechforest. Figure 3 shows the clustering of plots fromDECORANA in ordination space. In this figure, site unitsare represented by numbers and were derived using theTWINSPAN classification. Axis 1 and Axis 2 representhypothetical environmental gradients. Because indirectordination of the plots and species is not constrained byenvironmental variables, Axis 1 and Axis 2 in figure 3

69

3.5 --.4i.*:

3 .~ : .I._ .,:*2 ‘...,

.. .4. . . : . . . . . .

..‘. .5’*.i‘4

i’ . . 3 :

.tz,ii..;*...f. .404 . . . . . . ..--‘;‘r l 5

. i:..*._.-. l rI ;......”

*.*.5__ . ..i’i l b, l ,’ .2:

l = :l 3:* 5

l 5 /... ...‘.. . . .

c1: :*2 l 2

.I2l 4 : .5*.,!....”

..a *

3 ._...... -‘$-J

c .$I

.1*’.I ::

l 1:012l 2.2

‘0

. . . . * l 2* 02l 2 .*

..f.’ .+I’ :..._l 1.2 ** : 43 t:

. . . .1; *.:

._ ., . . . . . . . . . . .‘.. . ..$!$..’:

1.5 ~ - a._ *.*.*

l 2.2 -s...--” l 3 l 3

.2.2:. .3.3

.*._ tj .3 03*...*2 . ..”I l 3 ::

. ..*I -- *.

. 3 ‘:.

**.. fs ’ l a’ l ’

*._ * l 3. . . . l ‘* 3 l 3 ::

0.5 ~ - . .._ .*.a..

. . . . :. . :

0 ,.._ . :

. . . .0 1 2 3 4 5. . 6

A x i s 1

1 - Yellow pine/mixed oak forest2 - Mixed oak/red maple forest3 - Cove hardwoods/mixed

mesophytic forest4 - Eastern hemlock/yellow birch

forest5 - Beech forest

7

Figure 3-DCA ordination of ninety plots using DECORANA. Numbers represent TWINSPAN derived site units. Units of axesare standard deviations. Lines are subjectively drawn to show site unit membership.

remain unlabeled. Assignment of a particular environmentalvariable(s) to those axes occurs after stepwise discriminantanalysis.

To determine the relationship between vegetation, soils,and landform, a SWDA was performed using the nineenvironmental variables and the site units designated byDECORANA and TWINSPAN. Six soil and topographicvariables were determined to be significant at the 0.15 level(table 1). Elevation appears to be the most importantvariable in discriminating between site units, accounting formore than 54 percent of species variation, followed byslope position and aspect based on R2 values.

Table I-Significant variables identified through stepwisediscriminant analysis from nine environmental variablesand five site units (significance level for entry and retention= 0.15)

Variable R2 Prob.>F

Elevation 0.547 4.0 x IO-13Slope position 0.359 3.0 x IO-7Aspect 0.340 9.0 x IO-7A horizon depth 0.280 2.0 x IO-5Solum depth 0.193 1.7 x IO-3Humus depth 0.167 5.2 x 1O-3

70

Cross-validation of the environmental variables selected bySWDA was performed to determine how successfully thevariables could be used in combination to predictmembership in the five site units. Results are shown in thetabulation below:

VariablesCorrect

classification

- - Percent - -

Elevation and slope positionElevation and aspectElevation, slope position,

5655

and aspectElevation, slope position,

55

aspect, and A horizon depthElevation, slope position,

60

aspect, A horizon depth,solum depth, and humus depth 41

Little improvement in classification resulted from usingmore than two variables. However, a four-variable modelconsisting of elevation, slope position, aspect, and Ahorizon depth, predicted site unit membership the best (60percent).

DCCA combines indirect ordination (DCA) and multipleregression. The resulting ordination (fig. 4) represents thevariation in species composition over the plots as can beexplained by the environmental variables. Similar patternsof variation in plot distribution resulted from DCCA (fig. 4)as from DCA (fig. 3). However, there was greater overlap

LEGEND1 - Yellow pine/mixed oak forest2 - Mixed oak/red maple forest3 - Cove hardwoods/mixed

mesophytic forest4 - Eastern hemlock/yellow birch

forest5 - Beech forest

0 05 I 15 2 25 3 35 4 45

Axis 1

Figure 4-DCCA ordination of 90 plots using CANOCO. Numbers represent TWINSPAN derived site units. Units of axes arestandard deviations. Lines are subjectively drawn to show site unit membership.

of site units with DCCA, especially for Site Unit 4. Thespread of Site Unit 4 agreed with the cross-validationresults from the discriminant analysis, which showed lowpredictive ability for Site Unit 4.

The correlation coefficients of environmental variables withthe two axes (table 2) indicate that aspect (Axis 1) andelevation (Axis 2) are the most important predictors ofvegetation distribution. An important part of the output fromCANOCO is the creation of biplots of the vegetationordination and the environmental variables. Figure 5displays graphically the correlation results from table 2-

Table 2-Correlation coefficients between environmentalvariables and species ordination axes from directordination (DCCA)

Variable Axis 1 Axis 2

Aspect 0.48 -0.54Elevation 0.39 0.59A horizon depth 0.32 -0.15Landform index 0.22 -0.37Slope position -0.18 0.40Solum depth 0.17 0.13Humus depth 0.14 0.18Slope -0.09 0.17Terrain shape index 0.08 -0.04

DCCA ordination of plots and environmental variables. Theenvironmental variables are represented by arrows orvectors. The directions and relative lengths of the arrowsconvey the most meaningful information. The anglebetween an arrow and each axis is a representation of itsdegree of correlation with the axis, and the vector lengthindicates the strength of the correlation (ter Braak 1987). Inaddition, it can be inferred from figure 5 that Site Unit 5 islocated at the highest elevations, and Site Unit 3 is locatedin areas with the highest transformed aspect (northeast-facing slopes). Similar trends can be interpreted for theother environmental variables.

Similar studies in the Southern Appalachians have foundvegetation distribution to be related to differentcombinations of elevation and landform features.Whittaker’s (1956) monograph on the vegetation of theGreat Smoky Mountains concluded that moisture, asdefined by the interaction of elevation, topographicshape, and slope aspect, was the primary environmentalvariable that determined the vegetation present at agiven site. Golden (1974) concluded that vegetationpattern in the Great Smoky Mountains appeared to bemost directly related to elevation and topography.Callaway and others (1987), Gattis (1992), and Yoke(1994) all reached similar conclusions. Each studyidentified slightly different combinations of topographicvariables that were the best predictors of vegetationoccurrence and productivity differences across theSouthern Appalachian landscape. Overall, however,elevation was consistently the most importantenvironmental variable. This research obtained similar

71

4s I------- --/

‘i l 5 05l 5

3.5 1

ELEVATION. 5

SLOPE POSITION

.2.4 .= 04 4IJORIZON DEPTH

ASPECT

OJ 1

0 0.5 1 1.5 2 2.5 3 3.5 4

Axis 1

LEGEND1 - Yellow pine/mixed oak forest2 - Mixed oak/red maple forest3 - Cove hardwoods/mixed

mesophytic forest4 - Eastern hemlock/yellow birch

forest5 - Beech forest

LFI = Landform IndexTSI = Terrain Shape Index

s

Figure 5-DCCA ordination of 90 plots and significant environmental variables (arrows) using CANOCO. Units of axes arestandard deviations.

results. In addition, the five site units identified in this Based on the results of the DCA, SWDA, and DCCA, astudy were consistent with plant community types two-variable model (fig. 6) was developed for use bydescribed by these other studies. natural resource managers working in the Mountain

MOUNTAIN HIGHLANDS610 m to 1550 m elevation

ELEVATION> 1lOOm

0’ < Aspect < 114”

Site Unit 1Yellow Pine/Mixed Oak Forest

Site Unit 2Mixed Oak/Red Maple Forest

Site Unit 3Cove Hardwood/Mixed MesophyticForest

Site Unit 4E. Hemlock/Yellow Birch Forest

Site Unit 5Beech Forest

MESICFigure 6-Landscape ecosystem classification model for the Mountain Highlands of the Tellico Ranger District, CherokeeNational Forest, Tennessee.

72

Highlands section. Elevation and aspect were selectedbecause they appear to be sufficient for predicting thedistribution of site units across the landscape of theMountain Highlands, and because of their ease of use bothin the field and in creating ecosystem maps with a GIS. Anecosystem map of the Mountain Highlands section wascreated using the model shown in figure 6 and a GIS;however, it is not shown due to graphical limitations.

ACKNOWLEDGMENTThis research was supported by the USDA Forest Service,Southern Research Station and Cherokee National Forest;and the Tennessee Agricultural Experiment Station.

LITERATURE CITEDBass, Q.R., II. 1977. Prehistoric settlement and

subsistence patterns in the Great Smoky Mountains.Knoxville, TN: The University of Tennessee. Knoxville163 p. thesis.

Beers, T.W.; Dress, P.E.; Wensel, L.C. 1966. Aspecttransformation in site productivity research. Journal ofForestry. 64: 691-692.

Callaway, R.M.; Clebsch, E.E.C.; White, P.S. 1987. Amultivariate analysis of forest communities in thewestern Great Smoky Mountains National Park.American Midlands Naturalist. 118: 107-120.

Chapman, J. 1985. Tellico archaeology: 12,000 years ofNative American history. Knoxville: The University ofTennessee Press. 136 p.

Gattis, J.T. 1992. Landscape ecosystem classification onthe Highlands Ranger District, Nantahala NationalForest in North Carolina. Clemson, SC: ClemsonUniversity. 55 p. thesis.

Golden, M.S. 1974. Forest vegetation and siterelationships in the central portion of the Great SmokyMountains National Park. Knoxville, TN: The Universityof Tennessee. Knoxville. 275 p. dissertation.

Hill, M.O. 1979a. TWINSPAN: A FORTRAN program forarranging multivariate data in an ordered two-way table

by classification of the individuals and attributes. Ithaca,NY: Cornell University. 90 p.

Hill, M.O. 197913. DECORANA: A FORTRAN program fordetrended correspondence analysis and reciprocalaveraging. Ithaca, NY: Cornell University. 36 p.

Jones, S.M.; Lloyd, F.T. 1993. Landscape ecosystemclassification: the first step toward ecosystemmanagement in the Southeastern United States. In:Aplace, G.H. [and others]., eds. Defining sustainableforestry. Washington, DC: Island Press: 181-201.

McNab, W.H. 1989. Terrain shape index: quantifying effectof minor landforms on tree height. Forest Science. 35:91-104.

McNab, W.H. 1993. A topographical index to quantify theeffect of mesoscale landform on site productivity.Canadian Journal of Forest Resources. 23: 1100-1107.

Scott, B.L. [n.d.] Soil resource inventory report: Ocoee,Hiwassee, and Tellico Ranger Districts, National Forestsin Tennessee. USDA Forest Service In-Service Report.233 p.

ter Braak, C.J.F. 1987. The analysis of vegetation-environment relationships by canonical correspondenceanalysis. Vegetatio. 79: 69- 77.

USDA. 1981. Soil survey of Monroe County, Tennessee.U.S. Department of Agriculture, Soil ConservationService, and Forest Service. 145 p.

Whittaker, R.H. 1956. Vegetation of the Great SmokyMountains. Ecological Monographs 26: I-80.

Weals. I/. 1991. Last train to Elkmont: A look back at lifeon Little River in the Great Smoky Mountains. Knoxville,TN: Olden Press. 150 p.

Yoke, K.A. 1994. Landscape ecosystem classification inthe Cherokee National Forest. Knoxville: The Universityof Tennessee, Knoxville. 100 p. thesis.

Yoke, K.A.; Rennie, J.C. 1996. Landscape ecosystemclassification in the Cherokee National Forest, EastTennessee. Environmental Monitoring and Assessment.39: 323-338.

PRELIMINARY ECOLOGICAL LAND CLASSIFICATION OFTHE CHAUGA RIDGES REGION OF SOUTH CAROLINA

Curtis J. Hutto, Victor B. Shelburne, and Steven M. Jones’

Abstract-An ecological method of integrated multifactor land classification was applied in the Chauga Ridges region, asubregion of the Blue Ridge Physiographic province, in northwest South Carolina. Phase I sampling involved measuring 61reference plots in undisturbed late-successional upland hardwood stands. Four distinct site units were identified usingordination, cluster analysis, discriminant function analysis, and canonical correspondence analysis. Using cross-validation ofa discriminant function, a combination of five environmental features (landform index, weighted terrain shape index, slopeposition, root mat thickness, and distance-to-bottom) identified site units with an accuracy rate of 78 percent. A predictiveGeographic Information System (GIS) model using spatial variables only (landform index, terrain shape index, slopeposition, distance-to-bottom, transformed aspect, and slope gradient) was also developed which identified site units with asuccess rate of 76 percent.

INTRODUCTIONLandscape Ecosystem Classification (LEC) is an integrated,multifactor, ecological classification system that is relativelynew to the Southeastern United States. Although manyapproaches to classification are in use today (Bailey andothers 1978), not all are designed to quantitatively assessthe interrelationships of ecosystem components. Single-component classifications of soil only, vegetation only, orlandform only ignore the complex relationship evidentbetween plants and their environment. The LEC approachprovides a taxonomic separation of sites within a hierarchicalframework, and allows the model user to identify ecologicallysimilar areas without the use of vegetation.

Managing forests on an ecosystem basis is a necessity forthe development of a supporting framework used to sustaindesirable landscape attributes. An ecological classificationis one that expresses the interrelationships between plantsand soil, plants and landform, and soil and landform. Thesereciprocal interactions between biota and the physicalenvironment serve to define the concept of the ecosystemas a unit of management. Landscape ecosystemclassification uses undisturbed late-successional referencestands to assess the plant community as a type ofphytometer that expresses the combined forces ofphysiography and soils. Using an array of multivariatetechniques, a model is constructed that will identify complexenvironmental gradients, and break out ecologically similarareas which will recur across the landscape.

The objectives of this study were to complete Phases I toIII of an LEC for the Chauga Ridges region of the BlueRidge Physiographic povince (Myers and others 1986).This paper will focus on Phase I, which expresses spatialvariability in the landscape, and involved locating andsampling undisturbed reference stands to identifydiscriminating soil and landform variables. These were thenused to develop the discriminant functions unique to eachecological site unit.

STUDY AREAThe study was conducted in northwest South Carolina,within the southern half of the Andrew Pickens RangerDistrict, Sumter National Forest, Oconee County. This area(approximately 38,000 acres) is bounded on the north bythe higher elevations and steeper, broken topography ofthe Blue Ridge Mountain province (Myers and others1986). To the south and east lies the Blue Ridgeescarpment, which rises abruptly more than 650 feet abovethe Piedmont Foothills Region near the towns of Walhallaand Westminster, SC. The area is drained by the ChaugaRiver and the Chattooga River, which also forms thewestern boundary.

Representative topography includes short, steep slopesformed by the dissection through biotite schist, phyllite, andhorneblende gneiss parent materials from first- andsecond-order tributaries of the Chauga and ChattoogaRivers. Other geologic substrates consist of granitic andmylonitic gneiss, amphibolites, and scattered carbonateareas (Hatcher 1969). The study area is roughly bisectedby the Brevard Fault Zone, a narrow belt of low-grademetamorphic rocks that marks the southeastern edge ofthe Blue Ridge Belt from northern North Carolina tonorthwestern Alabama (Reed and others 1970). Elevationranges from about 900 feet to slightly over 1800 feet abovemean sea level. Roughly 75 percent of the study area ismapped as the Evard soil series, and classified as a fine-loamy, oxidic, mesic Typic Hapludult. These reddish uplandsoils occupy most slope categories, have a solumthickness of 21 to 40 inches, and an argillic horizon that is12 to 28 inches thick (USDA Soil Conservation Service1985).

The climate of the area is mild, and is one of transitionbetween the cooler, moist conditions of the Blue Ridge andthe warmer, drier climate of the Piedmont. Mean July hightemperature ranges from 75 to 77 “F., with a mean Januarylow temperature of about 30 “F. Mean annual rainfall is

’ Graduate Research Assistant, Clemson University, Department of Forest Resources, Clemson, SC 29634-1003; Associate Professor, ClemsonUniversity, Department of Forest Resources; and Senior Environmental Scientist, Environmental Services, Inc., 2169 West Park Court, Suite A,Stone Mountain. GA 30087 (respectively).

74

about 62 inches and is fairly evenly distributed throughoutthe year, with the exception of a slightly drier period in thefall. Forest vegetation is dominated by mesic mixedhardwood forest and oak-hickory forest, with the manyspecies of oaks most abundant on dry to intermediatesites. Shortleaf pine (Pinus echinata) is the most abundantconiferous species, although eastern white pine (Pstrobus) and eastern hemlock (Tsuga canadensis) are alsocommon.

METHODS

SamplingCompletion of Phase I of this study involved locatingrelatively undisturbed late-successional stands forsampling. Selected stands were those that showed littlesigns of vegetation or soil disturbance, and showedcharacteristic signs of mature forest or old-growth structure,such as large flat crowns or high levels of coarse woodydebris. Stand age was also a determining factor; standsgreater than 100 years old were searched out andevaluated for the above conditions.

Sixty-one late-successional stands were sampled usingthe Braun-Blanquet subjective selection system describedabove, with an attempt to distribute the plots equallythroughout the different landscape positions. To insureecological uniformity within each plot, site factors such asaspect, slope gradient, topographic position, andexposure were required to be consistent. Plot centerswere located as close as possible to the center of auniform stand area, and circular nested plots of 0.10 acreand 0.25 acre were marked on the ground. Diameter atbreast height (d.b.h.) in l-inch classes, and frequency byspecies were recorded for saplings (1.0 inch d.b.h. to 4.9inches d.b.h.) within the smaller plot. Diameter at breastheight in 2-inch classes, and frequency by species wererecorded for trees (5.0 inches d.b.h. and greater) withinthe larger plot. Vines, shrubs, herbs, and seedlingsgreater than 1 foot tall were also tallied and recorded byspecies within the larger plot, using Braun-Blanquetcoverage classes.

Soil and physical variables were also collected at each plot.Soil variables included thickness of the B horizon (BTH),clay content of the B horizon (BCL), root mat thickness(RM), solum depth, and particle size distribution of the Aand B horizons. Soil samples and data were collectedsystematically by spacing three auger holes along thecontour, with the center auger hole at plot center and theouter two near the plot perimeter. These samples werethen analyzed in the laboratory for particle size distribution.Soil chemical properties of previous LEC studies in theSouthern Appalachians have been shown to be weakdiscriminators (Moffat 1993, Gattis and others 1993) andthus were not assessed.

Landform variables were also recorded from plot center.These measurements consisted of landform index (LFI)and terrain shape index (TSI), as developed by McNab(1993 and 1989, respectively), slope gradient (SG), slope

position (SP), aspect (TASP), and slope distance to thenearest drain (DTB). Terrain shape index was multiplied byslope position to form a variable called weighted terrainshape index (WTSI). Aspect was transformed by the sinemethod of Beers and others (1966). Large LFI valuesindicate a higher degree of radiant energy protection fromsurrounding slopes. Large positive TSI readings indicateconcave landform conditions for the plot, while negativereadings are associated with drier, convex conditions.Slope position was recorded as a percentage on a relativesliding scale, with ridges classed as 0 percent and streambottoms representing 100 percent. Importance value 200(relative basal area + relative frequency/2) was used as ameasure of tree and sapling abundance. These values,along with the coverage class midpoints for the herbs,vines, shrubs, and seedlings, were converted to apresence/absence structure and used for the vegetationanalysis.

Data AnalysisFor each plot, vegetation data were summarized for eachspecies by canopy class. Members of the same species indifferent canopy layers were treated as separate“pseudospecies.” Data were analyzed using a series ofmultivariate techniques. Using indirect gradient analysis,species abundance and plot data were initially organizedand displayed in multidimensional space using theordination technique of Detrended CorrespondenceAnalysis (DECORANA, Hill 1979a). The distances betweenplot locations on the graph relate the relative degree ofsimilarity or difference (fig. 1). A numerical method ofpolythetic divisive classification was then performed on themain matrix for vegetation data using two-way indicatorspecies analysis (TWINSPAN, Hill 197913). This is asubjective classification, and allows the investigator to drawa separation between the groups identified in the initialordination of plots.

Significant separation between groups was then testedusing discriminant function analysis of the 16environmental variables collected (SAS Institute 1985). Theinitial goal was to develop a single model using physicalvariables that were cost-effective and relatively easy tomeasure in the field, and which could be integrated into aGIS predictive algorithm. We felt that soil variables couldnot be accurately assessed from existing digitalinformation, and thus developed two models, eachconsisting of a unique suite of variables.

3

2.5T

. Submeslc :Mesic

Axis -2 1.5

(s.d.11

0 0 5 1 1.5 2 2.5 3 3.5

Axis 1 (s.d.1

Figure I-Initial ordination of late-successional plots for theChauga Ridges region.

75

Canonical correspondence analysis (CCA) was thenapplied to the main matrix of vegetation data, along withthe secondary matrix of environmental data. We used theCANOCO program developed by ter Braak (1988). CCA isa method of direct gradient analysis which integrates linearregression and ordination to reveal relationships betweenplant communities and physical variables.

RESULTSAfter deleting outlier plots and those thought to besuccessional, the primary data matrix consisted of 49 late-successional plots with 245 species and pseudospecies.The ordination procedure arranged the plots on a moisturegradient (axis 1) that showed a beta diversity of 3.5standard deviations. Plots with low LFI and TSI scores(high exposure and convex microsites, respectively)expressed xeric soil moisture conditions, and tended to bedistributed near the graph origin. Initial cluster analysisshowed three groups that could be delineated (fig. 1). Afterdeleting the two groups on the moist end of the gradient,ordination and cluster analyses were again performed; betadiversity decreased to 2.5 standard deviations and twomore groups were separated (fig. 2). The speciesassociations in each group were reviewed and labeled asmesic, submesic, intermediate, and xeric. Nearly 75percent of the sites were classified as intermediate in soilmoisture.

The primary and secondary matrices were then subjectedto CCA, providing a validation source from physicalvariables for the ordination and cluster analysis performedpreviously. The first two axes were found to account for 74percent of the variation in the species-environmentrelationship. Landform index and SP had the strongestcorrelation with Axis 1 (r=0.86 and 0.62 respectively), whileSG had the highest correlation with Axis 2 (r=0.65).Another important output from CCA is the species-environment biplot (fig. 3). The arrow for eachenvironmental variable points in the direction of maximumchange for that variable across the diagram, and its lengthis proportional to the rate of change in this direction.Environmental variables with long arrows are more stronglycorrelated with the nearest ordination axis than are thosewith shorter arrows, and have a greater influence on thepatterns of community variation as shown in the ordinationdiagram. Thus, landform index can be seen to have a highassociation with the moisture gradient of Axis 1.Transformed aspect, BCL, and BTH show relatively short

2.5 .Intermediate

2 : XericAxis .

2 1.5 l * .

(s.d.) l * . l .*1 l '. :* l

.0.5 l ..: . .

+* l ..

I

0 0.5 1 1.5 2 2.5

Axis 1 (s.d.)

Figure 2-Second ordination after removal of mesic andsubmesic groups.

76

-2 -1 0 1 2 3 4

Axis 1 (s.d.1

Figure 3-Environmental biplot from canonicalcorrespondence analysis. M=mesic plot, X=xeric plotTIAM=Tilia americana, FAGR=Fagus grandifolia,TICO=Tiare//a cordifolia, FRAM=Fraxinus americana,LIBE=Lindera benzoin, BODI=Botrychium dissecturn,ADPE=Adiantum pedatum, PTAQ=Pteridium aquilinum,QUST=Quercus stellata.

arrows, and as such are relegated to a minor discriminatingrole in influencing plant community variation. Ordinalspecies locations can also be projected onto eachenvironmental arrow, and those that are near to or beyondthe tip of the arrow will be most positively correlated withand influenced by that physical factor. Pteridium aquihumand Quercus stellafa are most strongly influenced by DTBand RM, respectively, while distributions of Tiarellacordifolia and Fagus grandifolia are controlled mostly byLFI.

Discriminant function analysis was then used to identifydiscriminating soil and landform variables that could beused to accurately predict site unit membership at the 0.20significance level (table 1). The field model (LFI, WTSI, SP,RM, and DTB) classified accurately at 78 percent, while themodel intended for GIS use had a classification successrate of 76 percent. As root mat type and thickness issensitive to site disturbance (Pritchett and Fisher 1987)the field model is intended for use in undisturbed areas.The GIS model consists of spatial variables only (LFI, TSI,SP, DTB, TASP, SG) that can be built into GIS algorithmsfor predicting site unit membership. These models have yetto be validated by field measurements.

DESCRIPTION OF ECOLOGICAL UNITS

XericWith only four of the reference plots classified as xeric,these sites represent a small fraction of the land area in theChauga Ridges region. The frequency of disturbance isgreater on dry ridges, and in this area of steep, shortslopes, topographic uniformity is rare on xeric sites. Theseunits are characterized by low LFI readings (< 0.24),indicating high exposure. Restricted rates of forest floordecomposition led to the development of a more humustype. Late-successional stands were dominated byQuercus coccinea and Q. alba in the overstory, along with

Table I-Mean values of environmental variables by ecologicalsite unit for the Chauga Ridges region

Site unit

Environmentalvariable Xeric Intermediate Submesic Mesic

Landform index 0.17 0.21 0.42 0.49Slope position (pet) 45 38 97 89Terrain shape index -0.05 0.002 0.22 0.08B horizon thickness (in) 32 22 44 25B horizon clay (pet) 21 24 20 13Root mat thickness (in) 1.40 0.92 0.20 0.31Dist. to bottom (ft) 325 139 27 72Slope gradient (pet) 39 37 25 61Transformed aspect 1.06 0.92 0.93 1.60

scattered residuals of Pinus echinata, and Vacciniumvacillans and Pteridium aquilinum in the understory. Caryaglabra and Carya tomentosa were also components of theoverstory, with Oxydendrum arboreum common in thesapling layer. The herbaceous layer was very depauperate,with VI vacillans often comprising 70 to 80 percent of theground cover. Soil properties were variable, withsubsurface clay content ranging from 20 to 35 percent.

IntermediateWe estimated that this site unit occupied about three-quarters of the study area. Landscape positions typicallywere mid- lower slopes of southerly aspect or mid-slopepositions with northerly or easterly aspects. Exclusivespecies in the overstory included Nyssa sylvatica, Q.prinus, and Q. velufina., in association with Q. alba, Pechinafa, and C. glabra. Flowering dogwood (Cornusflorida) dominated the sapling strata and appeared to reachmaximum development on these sites, often in associationwith Sfyrax americana. Although there were no exclusivespecies in the understory, common herbaceous associatesincluded Polysfichum achrosfichiodes, Collinsoniaverficillafa, Dioscorea villosa, Smilacena racemosa,Goodyera pubescens, and Chimaphila maculafa.

SubmesicThese lower slope sites were most strongly influenced byconcave microsite conditions, indicating soil moistureaccumulation. The influence of aspect is overshadowed bydegree of protection/exposure and TSI. Colluvialaccumulation of coarse-textured materials produced 8- to16-inch sandy loam surface horizons with sandy loam toclay loam subsurface horizons. There were no exclusiveupper canopy dominants on submesic sites, but Q. rubra,Fraxinus americana, C. fomenfosa, and Q. alba were themost common associates, with Liriodendron fulipifera foundin scattered canopy openings. Hydrangea arborescens andCalycanfhus floridus were the preferential species in theshrub layer. These communities had the highest level of

herbaceous species richness of all ecological units, with anaverage of 20 species per plot. Bofrychium dissecturn wasthe only exclusive herbaceous species, and it was oftenfound with Sanguinaria canadensis, Sanicula canadensis,Thalicfrum fhalicfroides, Lysimachia quadrifolia, andBohemaria cylindrica.

MesicFagus grandifolia and Tilia americana dominated theselower slope sites, with five of six plots occupying northeastaspects. These cool and moist sites were found in highlyprotected coves (LFI > 0.37) on steep slopes of 43 to 94percent. Soil epipedons were usually dark (< 3 value andchroma), but were not thick enough to be classified asumbric. Humus layer was very thin to nonexistent,indicating rapid breakdown and incorporation of organicmaterials by soil microbes. Soils were relatively young andlargely of colluvial origin, with low subsurface clay contentsof 7 to 19 percent. Occasional stems of F. americana, Q.alba, T canadensis, L. fulipifera, and Magnolia fraseri werecommon associates in the overstory. Within the sapling andshrub layers, Lindera benzoin and Leucofhoe axillaris wereexclusive species restricted to these sites. A lushherbaceous layer was largely composed of Cimicifugaracemosa, Tiarella cordifolia, Caulophyllum fhalicfroides,Geranium maculafum, Polysfichum acrosfichoides,Thelypferis noveboracensis and Hepafica acufiloba.

SUMMARY AND CONCLUSIONSThe landscape ecosystem approach was used in theChauga Ridges region to classify areas of the recurringlandscape that express similar ecological attributes. Byusing a multifactor approach that integrates soil andlandform variables as the driving force behind plantcommunity variation, we were able to assess the nature ofthe study area as transitional between the Piedmont andBlue Ridge Physiographic provinces. Four multivariateanalytical techniques were used to quantify the separationof four distinct ecological site units at the lowest level of

77

classification heirarchy. A field model has been developedin addition to the GIS model, and is based on landformmeasurements that take 3 to 5 minutes per sample tomeasure.

Landform was found to have the strongest influence onplant community variation, with soil attributes playing aminor role. In this respect the Chauga Ridges region issimilar to its northeastern neighbor, the Blue RidgeMountains region. However, much of the vegetation ismore closely aligned with species associations of thePiedmont province. This may indicate the true transitionalnature of the Chauga Ridges region.

LITERATURE CITEDBailey, R.G.; Pfister, R.D.; Henderson, J.A. 1978. Nature

of land and resource classification-a review. Journal ofForestry. 76( 10): 650-655.

Beers, T.W.; Dress, P.E.; Wensel, L.C. 1966. Aspecttransformation in site productivity research. Journal ofForestry. 64: 691-692.

Gattis, J.T.; Jones, S.M.; McNab, W.H. 1993. Landscapeecosystem classification on the Highlands RangerDistrict, Nantahala National Forest, North Carolina. In:Brisette, J.C. ed., Proceedings, seventh biennialsouthern silviculture research conference; 1992November 17-19; Mobile, AL. Gen. Tech. Rep. SO-93.New Orleans, LA: U.S. Department of Agriculture, ForestService, Southern Forest Experiment Station: 223-228.

Hatcher, R.D., Jr. 1969. Stratigraphy, petrology, andstructure of the low rank belt and part of the Blue Ridgein northwesternmost South Carolina. Geologic Notes,Division of Geology, State Dev. Board, Columbia, SC.13(4): 105-141.

Hill, M.O. 1979a. DECORANA: A FORTRAN program fordetrended correspondence analysis and reciprocalaveraging. Department of Ecology and Systematics,Cornell University, Ithaca, NY. 52 p.

Hill, M.O. 197913. TWINSPAN: A FORTRAN program forarranging multivariate data in an ordered two-way tableby classification of the indivuduals and attributes.Department of Ecology and Systematics, CornellUniversity, Itaca, NY. 52 p.

McNab, W.H. 1989. Terrain shape index; quantifying theeffect of minor landforms on tree height. Forest Science35(l): 91-104.

McNab, W.H. 1993. A topographic index to quantify theeffect of mesoscale landform on site productivity.Canadian Journal of Forest Research. 223: 1100-1107

Moffat, S.O. 1993. Landscape ecosystem classification forthe Chattooga Ranger District, Chattahoochee NationalForest in Georgia. Clemson, SC: Clemson University.M.S. thesis.

Myers, R.K.; Zahner, R.; Jones, S.M. 1986. Forest habitatregions of South Carolina. Department of ForestryForest Research Series 42. Clemson University. 31 p.Map supplement.

Pritchett, W.L.; Fisher, R.F. 1987. Properties andmanagement of forest soils, 2d ed. New York: JohnWiley. 494 p.

Reed, C. Jr.; Bryant, B.; Myers, W.B. 1970. The BrevardZone: a reinterpretation. In: Fisher, J.W.; Pettijohn, F.J.;Reed, J.C. Jr.; Weaver, K.W., eds. Studies ofAppalachian geology: Central and Southern. New York:Interscience Publishers: 261-269.

SAS Institute. 1985. SAS user’s guide: statistics. Version 5ed. SAS Institute, Cary, NC.

ter Braak, C.J.F. 1988. CANOCO - An extension ofDECORANA to analyze species-environmentrelationships. Vegetatio. 75: 159-160.

USDA Soil Conservation Service. 1985. Soil survey ofthe Sumter National Forest Area, Oconee County, SouthCarolina. 76 p. Map supplements.