ABSTRACT: Southeastern United States habitats dominated by longleaf pine (Pinus puhtris P. Miller)have declined precipitously in area and extent. Conservation of diverse ground-layer vegetation in theseendangered habitats depends on prescribed fire. While the need for prescribed fire is now generallyaccepted, there is disagreement concerning the most appropriate fire regime. One of the more importantvariables is frequency of fire. Several hypothetical relationships between fire frequency and vascularplant richness and composition are suggested by the existing literature. Results of two long-termprescribed fire studies support the hypothesis that burning as frequently as fuels permit is optimal formaintaining the largest number of native ground-layer plant species. However, fire frequency effectson species composition differed between the two studies. Increasing fire frequency in South CarolinaUltisol flatwoods and wet savannas was associated with a distinct shift from woody to herbaceous-dominated communities. Herbs, particularly bunchgrasses and perennial forbs, dominated annual- andbiennial-bum treatment plots, whereas triennial- and quadrennial-bum plots were shrub-dominated. Incontrast, annual and biennial fires did not produce herbaceous dominated ground-layer vegetation inNorth Florida Spodosol flatwoods. Reduced dominance of saw palmetto and somewhat increasedimportance of forbs and grasses, particularly rhizomatous grasses, distinguished the annually burnedplots. However, biennial- and quadrennial-bum plots were similar in composition and did not differsignificantly in species richness at the largest spatial scale.

Efectos de la Frecuencia del Fuego en la Vegetacicin de Pino de Hoja Larga(Pinus palustris P. Miller) en Carolina de Sur y el Noreste de Florida, USA

RESUMEN: Los hjbitats de1 Sur de USA, dominados por el pino de hoja larga (Pinuspalustris P. Miller)han declinado drkticamente en Area y extensicin. La conservacidn de capas diversas de vegetaci6n enesos hgbitats amenazados dependen de1 fuego recetado. Mientras la necesidad de1 fuego recetado estdahora generalmente aceptada, hay un desacuerdo acerca de1 rkgimen de fuego m&s apropiado. Una delas variables m&s importantes es la frecuencia de1 fuego. En la literature existente se sugieren muchasrelaciones hipotkticas entre la frecuencia de1 fuego y la riqueza de plantas vasculares y la composicibn.Resultados de dos estudios de large tkrmino de fuego recetado, apoyan la hipcitesis que la quema tanfrecuente coma sea posible es el ciptimo para mantener el mayor nlimero de plantas de especies nativas.No obstante, el efecto de la frecuencia de1 fuego en la composici6n de especies vari6 entre 10s dosestudios. El aumento de la frecuencia de1 fuego en 10s bosques de llanura Ultisol de Carolina de1 Sury en las savanas htimedas estuvo asociado con un cambio distinto desde comunidades dominadas porleiiosas a herbkeas. Las hierbas, particularmente pastos asociados (‘bunchgrasses’), un ‘forb’ perenne,dominci las lotes tratados con fuego anual y bienal, mientras que 10s plots tratados con fuegos trienalesy cuadrienales estuvieron dominados por arbustos. En contrate, 10s fuegos anuales y bienales noprodujeron una vegetacidn dominada por herbkeas en 10s bosques llanos Spodosoles en el Norte deFlorida. La disminuci6n de la dominancia de ‘saw palmetto y el aumento de la importancia de ‘forbs’y pastas, particularmente 10s rizomatosos, distinguieron 10s plots anualmente quemados. No obstante,10s plots bienales y 10s cuadrienales fueron similares en composicicin y no difkieron significativamenteen riqueza de especies a una gran escala espacial.

Index terms: fire frequency, fire regime, longleaf pine, Pinus palustris, prescribed burning

INTRODUCTION way and Lewis 1997). Historically, recur-ring low-intensity fires perpetuated

Longleaf pine (Pinus pulustris P. Miller)dominated woodlands and savannas areamong the most species rich plant com-munities in North America (Bridges andOrzell 1989, Peet and Allard 1993). Un-fortunately, these habitats are also highlyendangered. Of an estimated 36 million haof presettlement old-growth habitat, onlyabout 3% remains in anything close to theoriginal condition (Frost 1993). Fire ex-clusion is the primary factor responsiblefor the loss of longleaf pine habitat (Brock-

longleaf pine savannas. In the absence offire the diverse longleaf ground-layer veg-etation, characterized by numerous grass-es, forbs, and low shrubs, is replaced by adepauperate understory dominated byhardwood trees and large shrubs (Hey-ward 1939, Lemon 1949, Komarek 1974,Abrahamson and Hartnett 1990, Waldropet al. 1992, Brockway and Lewis 1997).

The need for prescribed lire in manage-ment of longleaf pine dominated habitats

22 Natural Areas journal Volume 23 (1 ), 2003

is well established. However, numerousquestions remain concerning the most ap-propriate burn regime and how that mightvary among habitats and along environ-mental gradients. One important questionconcerns frequency of fire. Field observa-tions and some experimental data tend tosuggest that very short fire return times,that is, 1 to 3 y, are necessary to maintainspecies richness of longleaf ground-layervegetation (Lewis and Harshbarger 1976,Komarek 1974. Walker and Peet 1983,Peet et al. 1983, Waldrop et al. 1992; seealso Tester 1989, 1996 for similar resultsobtained in midwestern oak savannas).These studies also documented strong ef-fects of fire frequency on vegetation com-position. Generally, annual or biennialburning resulted in ground-layer commu-nities dominated by grasses, albeit oftenwith small shrubs and some forbs in asubdominant posi t ion. In contrast , less fre-quent or periodic fires tended to favorshrubs and woody sprouts, with reducedimportance of grasses and forbs.

The studies cited above support what wewill henceforth refer to as the “Most Fre-quent Fire Hypothesis” (MFFH) of longleafpine ground-layer community manage-ment . This hypothesis suggests that burn-ing as frequently as fuels will allow is thebest strategy for maintaining species r ich-ness and composi t ion of the nat ive longleafground layer. At present, the MFFH formsthe scientific basis of most current firemanagement in longleaf pine stands. It is,however, at variance with results of sever-al recent publications as well as a popularhypothesis f rom plant community ecologytheory. The “Intermediate DisturbanceHypothesis” (IDH: Connell 1978) postu-lates that highest levels of species r ichnessin plant communities should occur at “in-termediate” disturbance frequencies thatallow persistence of both “early” and “late”successional species. Testing the IDH isdifficult because of the difficulty of defin-ing an “intermediate” fire frequency. Fromthe theoret ical point of view, an intermedi-ate fire return interval would be definedrelative to the extremes of fire return timesnaturally occurring in a particular habitat(Connell 1978). However, investigatorsattempting to test the IDH have generallydefined “intermediate” relative to the range

of fire frequencies that happened to beencompassed by their own studies (Col-lins et al. 1995, Beckage and Stout 2000).

Several recent prescribed fire studies, in-cluding three in southeastern pinelands,have challenged the validity of both theMFFH and IDH. Mehlman (1992), in along-term study of old field loblolly pine(Pirzus tcwclcr L.) stands in north Florida(the Stoddard Fire Plots at Tall TimbersResearch Station), found that increases infire frequency were associated with in-creased species richness but only up tosome threshold level. Additional analysisof Mehlman’s (1992) data suggested thatfire return intervals of about six yearswould be sufficient to maintain maximallevels of species richness in the StoddardPlots and that more frequent burning wouldhave little additional effect (Beckage andStout 2000). We will henceforth refer tothe hypothesis that f ire effects tend to pla-teau or “saturate” as the Saturation Hy-pothesis (SH).

In addition to reanalyzing Mehlman’s(1992) data, Beckage and Stout (2000)presented results of their own study offrequency of burning in central Floridasandhills. They were unable to detect astatistically significant effect of fire histo-ry on either species composition or spe-cies r ichness in these sandhill p lots . Beck-age and Stout (2000) acknowledged thatthis study, which was observational ratherthan experimental, lacked replication andstatistical power, and therefore did notconstitute a strong test of the IDH or theMFFH. The authors concluded that at leastthree treatment replicates would be need-ed to have at least a 50% probability ofdetecting a real fire frequency effect,whereas six replicates would be requiredfor a statistical power of 0.8. As Beckageand Stout (2000) pointed out, few fire ex-periments, or observational studies, haveanywhere close to this level of replicat ion.

Brockway and Lewis (1997), in an exper-iment carried out in south Georgia longleafpine-1lc.x h’l~hrci-Aristicla heyrichiana

Trinius & Ruprecht-Sporoholus urtissii

flatwoods (nomenclature follows Kartesz1994 unless otherwise indicated), demon-strated major differences in composition

and vascular plant species richness be-tween burned and long-term unburnedplots. However, differences among threefire frequency treatments, that is, annual,biennial, and triennial dormant-seasonburning, were relatively minor and, for themost part, not statistically significant (P >0.05). An important point is that the firefrequency treatments in this study werecarried out from 1942 to 1954 and that aperiod of 24 y elapsed between the termi-nation of t reatments and data collect ion in1980. During the interval al l the plots wentunburned for 9 y and, when burning wasresumed, the former annual, biennial, andtriennial plots were al l burned on a bienni-al schedule. There may well have beenclear effects of the different frequency ofburn treatments in 1954 that disappearedduring the long interval prior to collectionof data. This study was further hamperedby low replication (n=2), making it evenmore difficult to detect a treatment effect.

Two recent studies in tal lgrass prair ie alsochallenged the MFFH and IDH. Collins etal. (1995) demonstrated a negative corre-lation between fire frequency and speciesrichness. They suggested that too frequentburning might enhance the compelitive-ness of dominant grasses at the expense offorbs, thereby reducing rather than en-hancing species richness. We will refer tothis suggest ion as the “Frequent Fire Spe-cies Loss Hypothesis” (FFSLH). Engle etal. (2000) did not, however, find this effectin “highly disturbed early successionalprairie communities” in Oklahoma. In theirexperiment, differences in species compo-sition and richness among treatment plotswere related to edaphic factors and timesince the last burn rather than the fire fre-quency treatments.

One problem in comparing resul ts of thesevarious studies is scale. For example,Mehlman (1992) and Beckage and Stout(2000) measured species richness on arelatively large scale, while Walker andPeet (19X3) and Brockway and Lewis(1997) made their observations on a muchsmaller scale. Sampling over a range ofspatial scales might help to resolve someinconsis tencies .

We present results from two ongoing long-

Volume 23 (l), 2003 Natural Areas Journal 23

term studies that const i tute a s t rong test ofthe MFFH and the various alternative hy-potheses. The two studies were carried outin longleaf pine-dominated flatwoods (wetsavannas were admixed at one of the loca-tions), albeit in different geographic loca-tions and with very different species com-positions. Both studies were experimentswith random treatment assignment andadequate replication (n=4 in one study,n=6 in the other) to detect any meaningfuleffects. The range of fire frequency treat-ments in these experiments (I- to 4-y firereturn intervals) was not sufficient to elu-cidate the full relationship between firefrequency and species richness. However,it probably was sufticient to allow us totest for a “saturation” effect of the sortfound in Mehlman’s (1992) data or a neg-ative effect of very frequent fire as foundby Collins et al. (1995). Our sampling andanalytical techniques also allowed for atest of scale effects on species richness.

METHODS

Tiger Corner Study

Francis Marion National Forest (FMNF)is located in the Atlant ic Coastal Plain , jus tnortheast of Charleston, South Carolina.In addit ion to the Tiger Corner Study, to bediscussed herein, FMNF was the site ofanother long-term fire study, the well-known Santee Fire Study located at theSantee Experiment Station near Cordes-ville (Waldrop et al. 1992 and earlier pub-lications cited therein). The Santee Study,which spanned the 43-y period between1946 and 1989, was discontinued follow-ing Hurricane Hugo, a category 4 stormthat came ashore in September 1989 andcaused substantial canopy damage through-out much of the national forest.

The Tiger Corner study site is located ap-proximately 4 km southeast of Jamestown,South Carolina, and IS km northeast ofthe old Santee Fire Study plots . The exper-iment, ongoing since 1958, originally con-sis ted of 20 O.&ha plots arranged into fourblocks of five plots each. Five experimen-tal treatments were randomly assigned toplots within blocks. The t reatments includ-ed fire frequencies ranging from annual toquadrennial as well as an unburned “con-

trol.” One control and one triennial plotwere lost to salvage operations followingHurricane Hugo in 1989. One plot assignedan annual burn treatment was so wet thatit rarely burned; it was consequently ex-cluded from the study. Numbers of treat-ment fires applied during the study periodranged from 44 for the annual plots to 11for the quadrennials . Biennial plots burned22 times and triennials were burned 14times. Fires were generally administeredin late winter, usually in February or earlyMarch. Wade et al. (1993) documentedpost-Hugo fire behavior in the plots.

The vegetation in the Tiger Corner plotscan be classified according to Peet andAllard (1993) as Atlantic Longleaf Flat-woods with inclusions of Atlantic MesicLongleaf Woodland, Atlantic LongleafSavanna, and Longleaf Seepage Bog (seePeet and Allard 1993 for lists of speciesassociated with each of these communitytypes). Dominant Atlantic Longleaf Flat-woods species cited by Peet and Allard(1993) include Pinus pulustris, P. elliottii,P semtina, I1e.x &bra, Serenoa repens,Quercus pumila, llex coriucea, Cyrilluruccmc~7oru, Myrica ceriferu, Guylu.s.suciufrondosu, Lyoniu murinnn, Pteridium uq-uilinum, and Aristidu strictu, “althoughnot all the species occur throughout therange.” In central South Carolina whereour plots are located, Pinus elliottii is rareexcept in mari t ime vegetat ion and Aris t idu.strictu/beyrichiunu and Sercnou repens areabsent. Instead, Pinus tuedu, Aristidu vir-gatu, Clethru ulnifoliu, Schizuchyrium sco-puriLm~, Andropogon virginicus var. decip-iens, Andropogon glomerutu.v, Vclcciniumtenellum, Lyonia lucidu, Lyoniu lipstri-nu, and Arundinuricr tectu (Walt.) Muhl.are characteristic, dominant, or subdomi-nant, tlatwoods plants (Komarek 1974;Percher 1995; E. Kjellmark, P. McMillan,R.K. Peet, J.S. Glitzenstein, D.R. Streng,unpubl. data). The combination of the lackof wiregrass and the admixture of bog andcane-break type shrubs within flatwoodsis apparently unique to the outer CoastalPlain of central South Carolina. The Na-ture Conservancy (TNC) recognizes threeglobally rare South Carolina variants (i.e.,associations in TNC terminology) of At-lantic Longleaf Flatwoods that are com-mon in FMNF and comprise the majority

of the vegetat ion in our Tiger Corner studyplots (NatureServe Explorer 2001). Theseinclude ( 1) Pinus I.‘ulu,stri,r--ArL4rzdirlariagigunteu ssp. tectu-Liquidumbur styruci-flua-Andropgon glomerutus-Surrrrcenicrminor Woodland (Gl global ranking), (2)Pinus plustri.s-Clethru Lllni~jliu-Gn~llI.v-saciu ,frondosa-Quercus pumilu-Schizu-chyrium scopurium Woodland (G 1 globalranking), and (3) Pinus palustris-I? sero-tinu-Ilex glabru-Lyorziu lucidu Woodland(G3-G4 global ranking) . Maintaining theseglobally rare associat ions could be consid-ered a management priority despite theirlocal abundance.

In the Tiger Corner study plots, longleafpine is the dominant canopy tree in onlyone of the four blocks (mean canopy coverapproximately 18%, vegetation samplingmethods are discussed below). In the otherblocks loblolly pine was the canopy dom-inant prior to Hurricane Hugo (in fact, thatwas one reason for blocking), but the hur-ricane decimated loblolly pine throughoutmuch of FMNF (Sheffield and Thompson1992), including at Tiger Corner. Present-ly, loblol ly pine s t i l l dominates the canopyin a single block, though mean canopycover is less than 10%. In one block loblollyand longleaf pines now co-dominate, eachwith less than 5% mean cover, and in theremaining block canopy dominance is spl i tequivalently between longleaf, loblolly,and pond pines, a l l with less than 5% can-opy cover. Despite these differences inpine species dominance, now much re-duced due to Hurricane Hugo, similarmixtures of ground-layer communitiesoccur in each of the blocks. This is indicat-ed by the fact that Redundancy Analysis(RDA) ordination ( this analyt ical techniquewill be discussed in detail below) failed todetect any statistically significant differ-ences in species composition related toblocking (F = 1.06, P = 0.46, Monte-Carlon = 99). The conclusion seems to be thatcurrent differences in pine species domi-nance among blocks are most likely relat-ed to logging history and perhaps f ire his-tory during the early phases of post- loggingstand regeneration and do not indicate sig-nificant environmental differences. In allprobability, longleaf pine was historicallythe dominant canopy tree on all the sites(Frost 1993). We therefore feel justified in

24 Natural Areas Journal Volume 23 (I), 2003

classifying the plots as longleaf woodlandswhether or not longleaf pine is presentlythe dominant canopy tree.

Soi ls in the Tiger Corner plots are Ul t isols ,primarily of the Lynchburg series (fineloamy, si l iceous, thermic Aeric Paleaquul t ;see Binkley et al. 1992). Ultisols are soilscharacterized by a sandy surface soil over-lying a loamy or clayey subsoil (i.e., anargil l ic horizon). Aquults are “Ultisols thatoccur in wet places where groundivaterapproaches the soil surface for large partsof most years” (Brown et al. 1990: SO).The clayey subsoil tends to retain mois-ture and nutrients during dry periods to alarger extent than in Spodosols where theargillic horizon is lacking (see discussionof Osceola study site soils below). Thismay account for the greater frequency ofbog-type shrubs in Ultisol Fla twoods (Tag-gart 1990). A detailed study of soil nutri-ents at Tiger Corner found little effect offire frequency treatments except in thecontrol plots. In these plots only the finefraction of the forest floor (Oe+, Oa hori-zons) was significantly enriched in carbonand nitrogen (Binkley et al. 1992).

We collected two types of data in the TigerCorner plots. During 1992-I 993 we sam-pled plant biomass from eight 0.25-m?locations within each of the 14 fire treat-ment plots (the “controls” were not sam-pled). Prior to sampling, we laid out a gridof 10-m x 10-m cells in each plot andmeasured elevation al each grid intersec-tion using a laser-plane. This was done inan effort to control for the effect of eleva-tion, which was assumed to be a surrogatefor hydrology. Biomass was sampled ineach plot from eight grid cell intersectionsrandomly located within the same rela-tively narrow range of elevations (approx-imately 0.2 m). Biomass samples weresorted to species, dried, and weighed. Bio-mass was sampled in each plot at the endof the first growing season after burningso that results would not be confoundedby time-since-burn effects.

During 2000-2001 we used the NorthCarolina Vegetation Survey (NCVS) (Peetet al. 1998) methodology to sample vege-tation in the annual, biennial, and quadri-ennial burn plots (1 1 total plots). All plots

but one were sampled during the first grow-ing season after burning and the other plotwas sampled early in the following grow-ing season. Triennial plots were not sam-pled since they were not due to burn againuntil late winter 2002 (they will be sam-pled during the 2002-growing season).Controls also have not yet been sampledfor either biomass or NCVS data. Thenegative consequence of long-term fireexclusion, that is, almost complete loss ofcharacteristic longleaf ground-layer vege-tat ion, is clearly evident in these plots evenin the absence of data. Furthermore, ef-fects of not burning are already well estab-lished by other studies, and lack of burn-ing is not considered to be a validmanagement strategy in southeastern pine-lands (see Introduction above; also Ko-marek 1974, Bridges and Orzell 1989,FNAI-FDNR 1990, Taggart 1990, Frost1993, Peet and Allard 1993, Platt 1999).Lastly, fire frequency and time since burneffects are unavoidably confounded in thecontrol plots. Thus sampling those plotshas thus far been a low priority.

Fire treatment plots were subdivided into20-m x 50-m (the size of an NCVS plot)sections and one section was randomlyselected for sampling. NCVS data werecollected following the procedures of Peetet al. (1998). The NCVS plot was itselfsubdivided into 10 10-m x 10-m “mod-ules,” and 4 conliguous modules in prede-termined locations (the so-called “inten-sive modules”) were sampled for coverand “level” data. Cover was estimated ac-cording to a semi-quantitative scale rang-ing from I (trace) to 10 (95%100% cov-er). “Level” referred to the scale at whicha species was first encountered. A series ofnested plots was searched, beginning withthe smallest (10 cm x 10 cm) and endingwith the largest (10 m x 10 m). Specieswere assigned scores as follows depend-ing on the scale at which they were firstencountered: (5) 10 cm x IO cm, (4) 32 cmx 32 cm, (3) 1 m x 1 m, (2) 3.16 m x 3.16,( 1) IO m x IO m. Level data were collectedat two of the corners of each intensivemodule. We summed the level data to pro-vide an overall measure of abundance.Species not present in the intensive mod-ules but present elsewhere in the 20-m x50-m sample plot were assigned a total

score of 0.5. For these “residual” speciesonly, cover was estimated across the entire20-m x 50-m plot.

A list of all the vascular plant species en-countered in the Tiger Corner NCVS studyplots is provided in Appendix A, avai lableonline at <http://www.talltimbers.org/research.html>. All plant species encoun-tered in both long-term studies discussedherein were natives, with the exception ofLespedeza cmeato, which occurred at tracelevels in a single Tiger Corner plot.

Osceola Study

Osceola National Forest (ONF) is in theCoastal Plain of eastern Florida, approxi-mately 65 km west of Jacksonville. Thelong-term study plots are located alongCounty Road 250A, about 6 km northeastof Olustee, Florida. This experiment con-sisted of 24 O.&ha plots arranged in sixblocks of four plots each. Treatments werethe same as in the Tiger Corner study,except that there was no triennial burntreatment. The Osceola experiment wasinitiated simultaneously with the TigerCorner experiment, and numbers of treat-ment f ires were equivalent at the two si tes.As at Tiger Corner, burns were typicallycarried out in late winter.

Vegetation in the Osceola plots can beclassified primarily as Southern LongleaiFlatwoods (Peet and Allard 1993) withsmall inclusions of Southern MesicLongleaf Woodland (see Peet and Allard1993 for lists of indicator species charac-teristic of these communities). Using theFNAI-FDNR (1990) system, the prevail-ing vegetat ion would be classif ied as MesicFlatwoods. Seretmt repens was the domi-nant understory species in all plots, withcover values generally exceeding 50%. Thetree stratum was composed almost entire-ly of longleaf pine (mean canopy cover =35%). Other common species in the plotsconsidered characteristic of SouthernLongleaf Flatwoods (Peet and Allard 1993)included Myrim cer(fet*u, Ilex glnhnr,

Kalmia hirsu~a, Vrrcciniurtz tnyrsitdes, andAri~tidu heyriclt ianu. Quercus tttinimt andSporolx~1u.s curtissii, two other commonspecies in our Osceola plots, are importantspecies of longleaf flatwoods communi-

Volume 23 (I), 2003 Natural Areas Journal 25

ties in the north Florida-south Georgiaregion (FNAI-FDNR 1990; Streng et al.1993; Brockway and Lewis 1997; E. Kjell-mark, P. McMillan, and R.K. Peet, unpubl.data).

Soi ls in the Osceola s tudy plots are Spodo-sols, pr imari ly Leon sands (sandy si l iceousthermic Aeric Haplaquod; see McKee1982). Spodosols are sandy soils distin-guished by a “spodic horizon”, that is, “asubsurface zone in which organic matterin combination with aluminum and/or ironhas accumulated due to downward leach-ing.” Spodosols lack the clayey subsoilthat is characterist ic of Ultisols and conse-quently have somewhat poorer moistureand nutrient retention during dry periods.“Aquods are spodosols that are wet forextended periods in most years” (Brown etal. 1990). Perhaps due to the difference insoil orders, soil nutrient responses to fireappeared to differ between the two studyareas. McKee (1982) reported generallyhigher soil nutrient concentrations inburned as compared to unburned plots at avariety of Coastal Plain locations, includ-ing our ONF study plots.

We did not collect biomass data from ONEHowever, NCVS data were collected us-ing the same methods as in the Tiger Cor-ner Study. Once again, all burn treatmentplots were sampled in the growing seasonfollowing burns the previous winter. Con-trols were not sampled. A vascular plantspecies list for the ONF experiment is pro-vided in Appendix B, available online at<http://www.talltimbers.org/research.html>.

Analyses

Effects of fire frequency on abundance ofindividual species (cover, level data) wereanalyzed using model I, two-way ANOVAas appropriate for a randomized blocksdesign (see Sokal and Rohlf 1969: 325).As i t turned out, the two measures of abun-dance were highly correlated (Tiger Cor-ner r = 0.X 1, P= .OOO, II = 1230; Osceolar = 0.79, P = .OOO, n= 787) and results ofanalyses were similar. Consequently, onlythe cover results will be presented herein.

Block and fire effects on species composi-

tion data were analyzed using redundancyanalysis (RDA), available through theCANOCO software package (Ter Braak1987-1992). RDA is closely related toPCA, a commonly used multivariate tech-nique (Ter Braak 1995). Like PCA, RDAassumes l inear relat ionships between spe-cies abundances and environmental gradi-ents. Unlike PCA and other indirect ordi-nation techniques, the ordination axes inRDA and other types of direct ordinationsare “constrained” (this term will be ex-plained below) to be linear functions ofthe independent variable(s). When com-bined with the Monte-Carlo randomiza-tion test included with CANOCO, RDApermits a direct statist ical test of the effectof the independent variables on speciescomposition. RDA is particularly usefulfor analyses of experimental data becauseexperimental treatments typically produceunidirectional response patterns that meetthe required assumptions of linearity. It istherefore recommended for analysis ofvegetation experiments with large num-bers of species (Ter Braak 1987-1992,1995). MANOVA, the multivariate exten-sion of ANOVA that would generally beused for such analyses, is useless when thenumber of dependent variables (species inan ordination analysis) exceeds the num-ber of experimental units. Thus RDA canbe viewed as a replacement for MANOVAwhen the limitations of the latter are ex-ceeded (Ter Braak 1987-1992).

Since readers may be unfamiliar with PCAand RDA we will attempt a brief summaryof the mathematical concepts underpin-ning these techniques. Readers wishing afuller understanding of this topic are urgedto consult Ter Braak ( 1995) or some othertext in quantitative plant ecology.

PCA is easiest to conceptualize using asimple two-species system. Imagine a num-ber of sites that are inhabited by two spe-cies that differ in abundance among thesites. If the x-axis represents the abun-dance of species-A and the y-axis repre-sents the abundance of species-B then onecan plot the location of each si te using theabundance values for the two species. Onecan then draw a line through the cloud ofdata points in the direct ion of the greatestscatter or variance among the points (see

Figure 5.13 in Ter Braak 1995). This lineis the first principal component, or first“PCA axis” in ordinat ion terminology. Thefirst axis site scores are determined byextending a perpendicular from each pointto the line. The two-dimensional variabil-i ty among si tes has now been reduced to asingle dimension, albeit with some loss ofinformation. If instead of two species wehave many species, the concept is the samealthough impossible to visualize graphi-cally. A single ordination axis will gener-ally be inadequate to describe the varia-tion in a multidimensional system and itwill consequently be necessary to extractadditional axes. The second and higheraxes are located analogously to the first,except that they are oriented in the direc-tion of the highest amount of residual var i-ance that is orthogonal to (i.e., not corre-lated with) the axes already derived.

A PCA ordination axis, like any indirectordination axis, is a purely mathematicalentity that must be interpreted with refer-ence to external data. For those of us wholike to think in terms of real environmentaleffects , i t is convenient to think of an indi-rect ordination axis as an underlying, butunknown, environmental gradient that in-fluences species composition in some im-portant way (Ter Braak 1995). Once theordinat ion is completed one is s t i l l , howev-er, left with the task of determining which,if any, actual environmental gradients arerepresented by a part icular ordination axis.RDA is an extension of PCA that attempts tocircumvent this problem. It does so by in-sert ing a regression step into the weightedsummation iterative algorithm by which PCAattempts to extract the ordination axes (wewill not attempt to explain how this algo-rithm works; interested readers are referredto Ter Braak 1995). In each cycle of theiteration, the site scores are regressed on theenvironmental (i .e. , independent) variablesand the predicted values from the regressionare taken as the new site scores for the nextiteration. Instead of searching for an axisassociated with the maximal amount of re-maining unexplained variance, this additionalstep forces the ordinat ion solut ion to con-verge on an axis that is already related to thepre-selected independent variables.

In the case of our study, the only indepen-

26 Natural Areas journal Volume 23 (11, 2003

dent variable was fire frequency, that is,time between fires, so we were certain thatthe first ordination axis extracted by RDAwould be related to this variable. Since thesecond and higher axes were not con-strained by any independent variables,these axes were the same as would bederived through PCA with the qualifica-tion that they must be orthogonal to thefirst constrained axis. At this point therestill remained the issue of statistical sig-nificance. As noted above, this is solved inCANOCO through a Monte-Carlo random-ization procedure. The observed eigenval-ue, a measure of the amount of explainedvariance for the first (i .e. , constrained) axiswas compared to eigenvalues obtained byordinating artificially generated datawherein species and associated abundancedata were randomly assigned to experi-mental treatments. If eigenvalues from 5or fewer in 100 random permutations ex-ceed the actual observed eigenvalue, theexperimental treatment &ect is consid-ered to be significant at the 0.05 level.

Blocking in experiments can be accommo-dated in the CANOCO Monte-Carlo test byrestricting permutations to plots withinblocks (Ter Braak 1987- 1992 refers to thissort of restr icted permutation test as part ialRDA). Alter ing the permutat ion test in thisfashion is only justified if the block effecthas been previously demonstrated to be sta-t is t ical ly s ignif icant in i ts own r ight . As wenoted above, this was not the case for theTiger Corner Study, so the full or unrestrict-ed Monte-Carlo test was employed for thats tudy. In contrast , the block effect was sig-nificant for the Osceola plots (F = 2.15, P =0.04, number of permutations = 99); thusthe partial or restricted permutation test wasjustified for that analysis. Incidentally, theanalysis of block effects at Osceola identi-fied two groups of plots corresponding todifferent sections within the overall studyarea. Basically, the three southern blocksappeared to be compositionally distinct fromthe three northern blocks, for reasons thatwere not apparent but may have been relat-ed to some aspect of management history.

In summary, ordination output from Species richness data from NCVS plotsCANOCO includes two sets of scores, s i te were analyzed with model I two-wayscores and species scores. Site scores were ANOVA appropriate for a randomizeddiscussed above. Species scores are relat- blocks experiment (see Sokal and Rohlf

ed to site scores and essentially indicatethe contribution of each species to theplacement of the ordination axis. Thusspecies with high scores on a particularaxis are those that are strongly positivelycorrelated with that axis. Treatment means,referred to as centroids, can be displayedin ordination space by averaging acrosssites receiving the same experimental t reat-ments .

When interpreting ordinations, one ordi-narily graphs the site and species scoresonto the coordinate space defined by themajor axes (Ter Braak 1995). Because ofthe large numbers of species at our twosites (Tiger Corner n = 278, Osceola n =1 lo), d isplaying ordinat ion resul ts of indi-vidual species was confusing and uninfor-mative. Consequently, we divided speciesinto 32 groups (see f igure legends for Fig-ures 1 and 4 and the appendices <http://www.talltimbers.org/research.html>) andaveraged the species scores. Groups werebased mostly on life form, life history(Godfrey and Wooten 1979, 1981), andcomparative ecology (Tag&art 1990, Peetand Allard 1993), but two families, Fa-baceae and Orchidaceae, were recognizedas separate groups due to their unique formsof nutrient acquisition and reproductivebiology. Some groups consisted of onlysingle species or genera-for example, PV= parastic vine (Cuscutcr compacta), SP =shrubby p a l m (Srrmort relwn.y). S G =shrubby grass (Arundinarict tecta [Walt. ]Muhl.), TG = shade-tolerant grass (Claus-tnnnthiurn Icr.rum), SF = shrubby forb (By-tisia tinctoria), and BS = biennial shrub(Ruhus spp.). Some species were includedin more than one group (e.g., HerbaceousVine + Fabaceae). Decisions about group-ings and, to some extent, assignment 01species to groups inevitably involved acertain amount of subjectivity. Readerswishing to display the results using theirown groupings may do so by consultingthe RDA species scores provided in Ap-pendices A (Tiger Corner) and B (OsceolaNV, available online at <http://www.talltimbers.org/research.html>.

1969: 325). Single degree of freedom poly-nomial contrasts were used to test for lin-ear and second order trends, thereby eval-uating the hypotheses discussed in theintroduction. A significant increasing lin-ear trend coupled with a nonsignificantsecond-order term would be consistent withthe MFFH but not wi th any other hypoth-esis. In contrast, a significant quadraticcontrast coupled with an insignificant lin-ear contrast would be consistent with theIDH. Support for the FFSLH would comefrom a negative linear contrast. A com-plete lack of significance (i.e., no signifi-cant contrasts) might indicate the absenceof any meaningful effect of frequency ofburn on species richness. However, suchan outcome might also be consistent with“saturation” at a longer fire return intervalthan the range encompassed by our stud-ies. A final possibi l i ty, “saturat ion” at veryshort fire return times, would show up inour analyses as the combination of s ignif-icant linear and quadratic contrasts.

RESULTS AND DISCUSSION

Species Composition

Tiger Corner Experiment

Significanl (P 5 0.05) effects of fire fre-quency on vegetation composition wereevident at both study sites. However, thedominant mode of species compositionvariation in the cover data (2nd RDA axis,1 st unconstrained axis, 28.4% explainedvariance in the Tiger Corner RDA) wasnot a function of fire frequency. Instead,this axis (henceforth referred to as the“moisture axis”) was clearly associatedwith soil moisture/hydrology. Speciesgroup FA (Fabaceae) was a good indica-tor. Almost without exception, ground-lay-er species in this family occur in mesic todry habitats and not in wetter longleafhabitats (Taggart 1990, Peet and Allard1993). In the Tiger Corner ordination, FAwas located close to the lower end of themoisture axis (i.e.. the left side of Figure1 C, henceforth referred to as the “dry end”).Other groups with a predominance of le-gumes-for example, HV (herbaceousvine), PF (prostrate forb), and SF (Bcqti-,sicr tinctorict j-were also found toward thedry end of this axis. In contrast, SG (Arun-

Volume 23 (I), 2003 Natural Areas Journal 27

dincrria tectu, switchcane) a dominant spe-cies of wet flatwoods and swamp ecotonesin the outer South Carolina Coastal Plain(Komarek 1974), was a good indicator forthe wet end of the moisture axis (r ight s ideof Figure IC). Group GM (non-Poaceaegraminoids, particularly wet savanna Cur-cx and Rh~~r~~ho.sl,orn j was another goodwet end indicator.

The first RDA cover axis (y-axis in Figurel), constrained to be a function of firefrequency, was the second most importantin terms of percent of explained covervariance (17.7%). This axis clearly sepa-rated the Tiger Corner plots according to

frequency of burn treatments. Quadrenni-al-burn plots were located toward the topof this axis and annual-burn plots were atthe bottom. Biennial-burn plots occupiedan intermediate posit ion between the othertwo treatments. but tended to be more close-ly associated with the annual-burns thanthe quadrennial-burns (Figure 1 A, note thelocations of treatment centroids). Accord-ing to the CANOCO Monte-Carlo test , thefrequency of burn effect was marginallysignificant (F = 1.96, P = 0.06, n = 99permutations).

The most obvious change in vegetationcomposition across the Tiger Corner tire

frequency gradient was in the relat ive dom-inance of woody and herbaceous plants(Figure IB). Woody species were clus-tered toward the low fire end of the gradi-ent while herbaceous species predominat-ed at the high end. This trend was evidentin the biomass data as well (Figure 2B).Ordination trends may sometimes reflectrelative rather than absolute changes forcertain groups, but that was not the casefor Tiger Corner. Absolute cover and bio-mass of woody plants declined substan-tial ly with increases in f ire frequency whileherbaceous species showed the oppositetrend (Figures 2A, 3). Within the woodyand herbaceous categories, most dominant

TIGER CORNER EXPERIMENT

-0.8 -

-1.2 -0.8 -0.4 0 0.4 0.8 1.2

MESIC R D A A X I S 2 WE7

C LOFIRE 0.8 -

pv TG

SF

8s0.4 -

L S

r;REI I I I t I I I

-0.8 -0.4 0 0.4 0.8

R D A A X I S 2MESIC WE1

Figure 1. Results of the RDA ordination of the Tiger Corner cover data.(A) Plot scores. (B) Species scores classified according to three major lifeforms: Woody [WJ, Grass [G] and Forb [F]. (C) Species scores averagedinto “functional groups”: abbreviations and number of species pergroup are as follows: AF = Annual Forb, 14; AG = Annual Grass, 2; BF= Biennial Forb, 3; BG = Buneh Grass, 20; BS = Biennial Shrub, 2; CG= Climax Grass, 8; CT = Coniferous Tree, 4; FA = Fahaceae, 30; FE =Fern, 5; GM = Graminoid Monocot, excluding Poaceae, 20; HV =Herbaceous Vine, 8; IF = Insectivorous Forb, 4; LF = Leafy Forb, 64;LS = Large Shrub, 15; OR = Orchidaceae, 9; PF = Prostrate Forb, 7; PV= Parasitic Vine, 1; RF = Rosette Forb, 37; RG = Rhizomatous Grass,6; SB = Small Bunchgrass, 19; SF = “Shrubby” Forb, 1; SG = ShrubbyGrass, 1; SR = Small Rosette, 14; SS = Small Shrub, 11; TG =Shade-Tolerant Grass, 1; TS = Hardwood Tree-Sprout, 16; WF = Weedy Forb,13; WG = Weedy Grass, 14; WV = Woody Vine, 7. Species included ineach group are listed in Appendix A ihttp://www.talltimbers.org/research.htmI>.

28 Natural Areas Journal Volume 23 (I), 2003

species showed similar, and statistically woody plants and herbs, grass and forbsignificant, trends (Figure 3). species were distributed similarly with

respect to fire frequency (Figure IB). InIn contrast to the different responses of fact, the centroids of these groups were in

TIGER CORNER EXPERIMENT

WOODYPLANTS

-100 ’ _LIYR 2 Y R 3 Y R 4 Y R

750 GRASSESANDFORBS

-50 -IYR 2 Y R 3 Y R 4 Y R

FIRE RETURN INTERVAL

B BIOMASS CCA (P < .02)

Figure 2. Effects of fire frequency on biomass composition in the Tiger Corner Study plots. (A) Absolutedifferences in woody and herbaceous biomass. ANOVAs include a block and treatment effect. Errordegrees of freedom were reduced by two to account for two missing observations (Sokal and Rohlf 1969:338): the annual plot that was too wet to burn and a triennial plot excluded due to salvage damage (seetext). (B) WA ordination scatterplot. Each symbol represents a single species. First axis is constrainedto be a function of tire frequency. Moisture eftects are reduced in this ordination because elevation,presumed to be a surrogate for hydrology, was explicitly controlled for when collecting the data (seetext). CCA ordination is similar to RDA ordination discussed in the text but assumes unimodal ratherthan linear species responses to treatments. Similar results for the two types of ordinations and inputdata (i.e. biomass and cover) provide confidence that the fire treatment effect is indeed meaningful.

almost the same location in the ordinationscat terplot .

Centroid plots of finer groupings were abit more interesting (Figure I C). As wouldbe expected, most of the woody plantgroups occurred near the low fire frequen-cy end of the fire axis (upper half of FigureIC). Ruhus spp. (group BS, biennial shrub)responded as typical for woody plants. Incontrast, Arundinuria tecta (SG, shrubbygrass) did not. This species occurred closeto the high fire end of the fire frequencyaxis (bottom right of Figure IC), suggest-ing considerable tolerance for even veryfrequent f i res . This resul t is consistent withthe observation that Arundinaria is, andwas even in presett lement t imes, an impor-tant dominant of wetter fire-maintainedpinelands in the Carolinas (Lawson 1709,Hughes 1966, Komarek 1974). Anotherinteresting observation was the tendencyfor small shrubs (group SS), such as Gcrq-lussucia spp., Hypericum crux-undreae,H. @ides, Quercus pumila, Vacciniumtenellum, to occur closer than the otherwoody plant groups (TS, LS, WV) to thehigh fire frequency end of the gradient.This finding is consistent with plant com-munity surveys (Peet and Allard 1993)and results of other fire studies (e.g. , Abra-hamson 1984, Waldrop et al. 1992, Strenget al. 1993) in demonstrating that thesesmall, mostly rhizomatous shrubs can beimportant components of frequently burnedpinelands .

Most herbaceous groups were closely as-sociated with the high end of the fire fre-quency gradient (Figure IC). Group CGwas particularly favored by very frequentfire. This group included “climax” or ma-trix grasses Schizachyrium scopurium,Ctenium aromaticurn, Muhlenbergia ex-prznsm (Poir.) Trin., Andropogon gerurdii,Sorghastrum nutuns, Andropogon virgini-C L L S var. decipiens, and Punicum virgutum.These grasses are collectively characteris-tic of pristine soils in the outer CoastalPlain region of central South Carolina (Peetand Allard 1993; Percher 199.5; P. Mc-Millan, J.S. Glitzenstein, D.R. Streng, andR.K. Peet, unpubl. data). Other groups ap-pearing to be exceptionally favored by, ordependent on, frequent fire included ro-sette forbs (group RF), small rosettes

Volume 23 (11, 2003 Natural Areas Journal 29

TIGER CORNER EXPERIMENT

7F; -COVER DATA

2 2 0

y 15 PERENNIAL FORB

syj 1 0

=ii 5

0 I0 1 2 3 4 5

20PERENNIAL GRASS

CL& 16

Y 12

8

88

5 4ul

0

40

2 30

Y

8 20

8

510

VI0

0Flfk RET”FiN INTERtAL (YR:)

5

Figure 3. Effects fire frequency on individual and cumulative species cover scores in the Tiger Cornerstudy plots. All species occurring in at least ten plots were included. Cover scores and associated coverranges are as follows: 1 = trace, 2 = O-l%, 3 = l-2%, 4 = 2-S%, 5 = S-10%, 6 = 10-25%, 7 = 25-50%,8 = S&75%, 9 = 75-S%, 10 = Y5-100%. Note that the sum of cover scores for a particular plantcategory is not equivalent to plot cover for that category (e.g., sum of cover scores for woody plants isnot equivalent to woody plant cover). Nevertheless, the sum of scores does represent an alternativeunbiased indicator of the importance of a particular group in a plot. Species abbreviations: Woody: SA= Sorhus urbul~fi~lia (I>.) Heynh., RA = Rubus argutus, KC= Rhus copallinum, MC = Myrica cerifera, MH= Myrica heterophylla, MV = Magnolia virginiana, IG = 1Ie.x gIabra, NC = IIypcricum crux-andrcue, GF= Gayhssucia froridosn, CA = Clethra aln(folia, AR = Acer rubrum; Grasses: SS = Schizachyriumscoparium, SG = Saccharum giguuteum, CA = Ctcnium aromaticum, DS = Dichanthclium strigosum, UT= Dichanthclium dichotomum var. tenue, AV = Andropogm virginicus var. decipicns; @‘orbs: PG =Pityopsk graminifXa, HA = Nelianthus angustifXius, EP = Eupatorium pilosum, El, = Euputorium

leucolcpis, CP = Carphephorus paniculatus, CL = Coreopsis linifolia, RN = Bigelowia nudata, AD = Asterdumosus, RA = Rhexia nlifnnus. P-values from ANOVA tests are listed next to the species codes.

(group SR; e.g., Lachnocaulon uncqs,Rhynchosporu chuptnanii), insectivorousforbs (group IF, including S~trrucmia “pp.,Drosem “pp., Pinguiculu lutea), and Or-chidaceae (group OR).

All herbaceous groups were not associat-ed with frequent burning, however. Onecurious example was the parasitic forbCusculu cotnpac~c~ (PV = parasitic vine inthe RDA ordination scatterplot). This ge-nus lacks roots as mature plants and de-rives nutr i t ion through haustorial connec-tions with host plants. According toGodfrey and Wooten (19% l), C. compacmhas been shown to parasitize a fairly largenumber of host plants, many of which arecommon hardwood trees and shrubs offlatwoods and swamp ecotones (e.g., Mag-n o l i a virginiuna, Cyrilla racem$ora,Clethra uln$~lia, Rubus spp., Myrica spp.).We have observed it to be particularlyabundant on sprouts l-2 y after fire. Ap-parently i t was favored in the quadrennial-burn plots by the greater densi ty and vigorof hardwood sprouting.

Another herb associated with lower firefrequencies was the shade-tolerant grassChasmunthium luxum (TG). This grass isfound in a wide range of habitats frompine savannas and flatwoods through var-ious types of closed woodlands and evenhardwood bottomlands (Godfrey andWooten 1979, Weakley 1999). Comparedto most other Poaceae i t appeared to toler-ate and perhaps even prefer longer firereturn intervals. On the drier end of themoisture gradient, the same could perhapsbe said for (SF) Bnptisia tiwtoria. Basedon its position in ordination space, thisrobust forb appeared to be more tolerantof less frequent fire than most other le-gumes. Ferns (group FE, including Pterid-ium oquilittum, Osmun&t spp., and Wood-cr,rtrdia spp.), another shade-tolerantherbaceous group (Grime et al. 1988), alsoappeared to tolerate a somewhat reducedfrequency of fire.

Last, and perhaps most unexpectedly, thegroups WG (weedy perennial grasses) andWF (weedy forbs) were also associated, atleast to a greater extent than most otherherb groups, with the low end of the firefrequency axis. We hypothesize that these

30 Natural Areas journal Volume 23 (l), 2003

ruderal species may benefit from tempo-rary windows of reduced competition inthe less frequently burned plots. Such“windows” may open briefly when shrubsare temporarily damaged following higherintensity f ires result ing from excessive fuelaccumulations and less weedy herbs havebeen reduced or eliminated during severalfire-free years.

A final interesting pattern in the TigerCorner ordination data concerned a possi-ble interaction between soil moisture andfire frequency. Moving from wet to dryacross the moisture axis, there was somesuggestion of increasing similarity in spe-

cies composit ion between quadrennial andmore frequently burned plots (Figure 1 A).This may reflect the fact that most of thelarger shrubs (e.g., Lykcr lucidu, Ikx gla-bra, Clethrn uln(fdioliu, Myrica certferajwere more prevalent toward the wetter endof the moisture gradient. With even minorreductions in burn frequency-for exam-ple, from biennial to quadrennial burn-ing-these wet flatwoods shrubs increasegreatly in cover (Figure 3) and biomass(Figure 2), in the process competitivelyexcluding grasses and forbs. Toward thedrier end of the moisture gradient, wherelarge shrubs are less important due toedaphic restr ict ions, the ground-layer com-

munity may be able to tolerate longer in-tervals between fires without loss of spe-cies. Plant community differences relatedto fire history did appear to be more dis-tinct in the wetter Tiger Corner plots. An-nually and biennially burned plots on the“wet” end of the soil moisture gradientencompass some of the finest wet savan-nas in the FMNF while the less frequentlyburned triennial and quadrennial plots onsimilarly moist soils are shrub-dominatedflatwoods.

Osceola Experiment

The dominant mode of variation in the

OSCEOLA EXPERIMENT

COVER DATA

A L OFIRE *

-0.8 - FG$

FOR8F

i

G R A S S

+ W O O D Y

!RE

-1.2 , , t , I , I , I

-0.8 -0.4 0 0.4 0.8 1.2

M O I S T RDA AXIS 2 SUBXERIC

LOC FIRE 0.8 --,

kE -0.8

-0.4 -0.2 0 0.2 0.4 0.6

MOIST RDA AXIS 2 SUBXERIC

Figure 4. RDA results from the Osceola study site. (A) Plotscores. (I%) Species scores classified according to three majorlife-forms: Woody [WI, Grass [G], Forb [F]. (C) Species scoresaveraged into “functional groups”; abbreviations and numberof species per group are as follows: AF = Annual Forb, 4; BG =Bunch Grass, 9; BS = Biennial Shrub, 1; CG = Climax Grass, 5;CT = Coniferous Tree, 2; FA = Fabaceae, 9; GM = GraminoidMouocot, excluding Poaceae, 6; HV = Herbaceous Vine, 3; LF= Leafy Forb, 23; LS = Large Shrub, dicots only, 7; PF =Prostrate Forb, 4; RF = Rosette Forb, 13; RG = RbizomatousGrass, 4; SB = Small Bunchgrass, 11; SP = Shrubby Palm, 1; SR= Small Rosette, 3; SS = Small Shrub, 12; TS = Hardwood TreeSprout, 3; WF = Weedy Forb, 3; WG = Weedy Grass, 8; WV =Woody Vine, 3. Species included in each group are listed inAppendix B <http://www.talltimbers.org/researeh.html>.

Volume 23 (I), 2003 Natural Areas Journal 31

Osceola cover data (RDA axis 2, 1st un-constrained axis, 30.9% explained vari-ance, x-axis in Figure 4) was again relatedto soil moisture rather than fire frequency.Again, groups FA (Fabaceae), HV (herba-ceous vine), and PF (prostrate forb) weregood indicators for the dry end of themoisture gradient (right side of Figure 4),while SR (small rosettte) and GM (non-Poaceae graminoid monocot) helped toidentify the wetter end. Group LS (largeshrub), including moist flatwoods domi-nants llex gluhru and L.yoniu fruticosu, aswell as shrub bog species Ilex coviuceaand Ilex myrtijolia, also occurred close tothe wet end of the Osceola moisture gradi-ent. In addition to the group means, theidentities of individual species found atthe opposi te ends of the gradient suggest-ed moisture differences as the explana-tion. The six species with the lowest RDAsecond axis scores included moist/mesicflatwoods indicators Ilex glahm, Rhpcho-spru plurnosu, Aristidu .spic~f~mnis, andAndropogon glaucopsis Elliott (Godfreyand Wooten 1979, 1981; FNAI-FDNR1990; Peet and Allard 1993; Streng et al.1993). At the other extreme, the six spe-cies with the highest scores on this axisincluded dry flatwoods/sandhill speciesQurrcus minima, Cnidoscolus stimulosus,Elephantopus elatus, Aster walteri, Seric-ocurpus tort@iu,s (Michx.) Nees, andTephraria hispidulu (FNAI-FDNR 1990,Peet and Allard 1993, Streng et al. 1993).

The RDA ordination identified frequencyof burn as the second most important in-fluence on vegetation composition in theOsceola study plots. Axis 1 (y-axis in Fig-ure 4), constrained to be a function of firefrequency, accounted for 2 1.8% of thespecies variance and was statistically sig-nificant (Monte-Carlo F = 3.07, P = 0.02).As at Tiger Corner, quadrennial-burn andannual-burn plots were at opposite ends ofthe lire frequency axis. Biennial-burn plotsagain occupied an intermediate position,but here, unlike the Tiger Corner ordina-tion, they were more closely associatedwith, and tended to overlap, the quadren-nial-burn plots (Figure 4A).

Distr ibutions of species groups across theOsceola fire frequency axis were perhapsdeceptively similar to Tiger Corner. Woody

OSCEOLA EXPERIMENT

PERENNIAL FORB

PERENNIAL GRASS

WOODY

1 2 3 4 5

FIRE RETURN INTERVAL (YEARS)

Figure 5. Effects of frequency of fire on individual and cumulative species cover scores in the Osceolastudy plots. All species occurring in at least I6 plots were included. Cover scores and associated coverranges are as follows: 1 = trace, 2 = O-I %, 3 = l-2%, 4 = 2-S%, 5 = 5-IO%, 6 = IO-25%, 7 = 25-SO%,8 = 50-75%, 9 = 75-95%, 10 = 95-100%. See note in the legend for Figure 3 concerning interpretationof sum of cover score measurements. Species abbreviations: Woody: AA = Asimina angusf{f&z, GD =Gaylussacia dumosa, G T = Gaylussacia tomentosa (Gray) Pursh ex Small, IG = Ilex glahru, LF = Lyonia

fruticosa, SR = Serenoa repens, VM = Vaccinium myrsinites, QM = Quercus minima; Grasses: AG =Andropogon glaucopsis Elliott, AV = Andropogon virginicus var. decipiens, AB = Aristidu heyrichianaTrinius & Ruprecht, AS = Aristidu spiciformis, DC = Dichanthelium chamaelonche (=Pnnicumchamaelonche Trin.), III, = Dichanthelium portoricense (Desv. ex Ham.) B. F. Hansen & Wunderlin, PA= Panicum anceps var. rhizomatum (A. S. Hitchc. & Chase) Fern., SS = Schizachyrium stolon(fernm, SC= Sporobolus curtissii; Forbs: EM = Eupatorium mohrii, PG = Pityopsis graminifolia, PP = Pterocaulonpycnostuchyum (Michx.) Eli., XC = Xyris carolinianu. P-values from ANOVA tests are listed next to thespecies codes.

32 Natural Areas Journal Volume 23 (l), 2003

plants and herbs were again clustered inopposite halves of the ordination graph,woody plants associated with less frequentfire, herbs the reverse (Figure 4B). How-ever, an examination of absolute covervalues revealed an important differencebetween the two sites. At Tiger Corner,changes in fire frequency were associatedwith absolute changes in cover and biom-ass of both woody and herbaceous spe-cies. At Osceola, in contrast, total woodycover and mean cover values for dominantwoody species were relatively unaffectedby frequency of fire (Figure 5). The oneexception was Serenoa repens, which hadsignificantly lower cover scores in theannually burned plots. Overall, however,differences in relative abundance of woodyand herbaceous species suggested by theordination results were due primarily toeffects of fire frequency treatments onherbaceous cover. Furthermore, increasesin herbaceous cover were most evident inthe annually burned plots. The obviousinference is that decreased saw palmettocover associated with annual burningopened up space for herbs.

Differing responses of plant groups to f irefrequency are indicated by the graph ofgroup centroids in RDA ordination space(Figure 4C). Herbaceous groups that ap-peared to benefit especially from high firefrequencies included rhizomatous grasses(group RG), such as Panicun~ unceps var.r#zizomururn (A. S. Hitchc. & Chase) Fern.,Schizachyrium stolonijerum, and Cteniumjloridanum; small rosettes (group SR, thisgroup included some sedge species thatwere technically small bunches rather thantrue rosettes), Hypoxis juncca, Rhynchos-pow plumosa, and Lcrchnocaulon anceps;and, in contrast to Tiger Corner, weedyforbs (group WF), such as Diodia teres,Eutha?~irr minor (Michaux) Greene. Cli-max grasses (group CC, defined in thiscontext as matrix grasses of high qualityEast Gulf Coastal Plain ilatwoods; Peetand Allard 1993) were also favored byfrequent fire, but not to the same extent asat Tiger Corner. Wiregrass (Aristida heyri-chkm), expected to dominate the herba-ceous layer in eastern Florida longleaf pinewoodlands (Peet and Allard 1993), showedlittle response to frequency of burning(Figure 5). A possible explanation may be

that sexual reproduction and hence popu-lation growth of this important grass mayhave been limited by the lack of growing-season tires (see Streng et al. 1993).

As expected from mean cover scores (Fig-ure 5), Serenoa repens (group SP) wasclosely associated with the low fire end ofthe Osceola burn frequency axis. Otherwoody plant groups were also associatedwith this end of the f i re gradient , though toa lesser extent than Serenoa. Smal l shrubs(group SS, e.g., Quercus minima, Q. pumi-la, Vuccinium myrsinites, G: stamineum,Hypericum microsepalum, Gaylussacinspp.) once again appeared to be more tol-erant of closely spaced fires than did othergroups of woody plants, but the differencebetween LS and SS was narrower than atTiger Corner. A similarity with Tiger Cor-ner was the relative proximity of WG(weedy bunchgrasses) to the low fire fre-quency end of the gradient, a result thatwas discussed earlier.

Finally, there was some indication of thesame sort of interaction between fire fre-quency and soil moisture noted earlier inthe Tiger Corner ordination results. Onceagain, species composit ion of plots burnedat different frequencies appeared to con-verge to some extent at the dry end of thesoil moisture gradient (Figure 4A).

Rare Species

Effects of fire frequency on globally rareplants or endangered species are of specialinterest to conservation. Two such speciesoccurred in our study plots. Ptero~~los.scr.s-pis ec’ristatrx (spiked-medusa orchid, Na-ture Conservancy G-Rank = 2) was con-fined to a single biennial burn plot in theTiger Corner Study. During the survey ofthis plot in autumn 2000 we observed sev-en flowering stems of this plant in or aroundthe NCVS plot located within burn treat-ment plot 2A. Though small. this is one ofthe largest concentrations of this speciesin the FMNF (J. Glitzenstein, pers. ohs.;also J. Townsend, former curator of Clem-son Herbarium, pers. corn.). Presence of aspecies in a single plot is not an adequatesample with which to speculate about firefrequency effects. However, most obser-vations of this species range-wide are con-

sistent with its occurrence in frequentlyburned flatwoods or mesic savannas(Weakley 1999).

Asclepins pedicelktu (stalked milkweed),another globally rare plant (Nature Con-servancy G-Rank = 3) occurred in all an-nual and quadrennial NCVS sample plotsat the Osceola study site. However, thespecies was present in only three of sixbiennial burn plots. Given the sparse dis-tribution of this plant (overall mean cover< 1 %), lower occurrence rates in the bien-nial-burn treatment plots are probably oflittle consequence. However, i ts occurrencein al l of the quadrennial plots suggests thatit may be more tolerant of longer intervalfires than many forb species (Appendix B<http://www.talltimbers.org/research.html>). Careful demographicmonitoring is needed to better documentresponses to variable t i re regimes of theseand many other rare plants (e.g., Kirkmanet al. 1998).

Species Richness

Plots of species richness versus fire fre-quency were, for the most part , consistentwith the MFFH (Figures 6A, B). At bothsites, species richness tended to increaselinearly with decreasing fire return times,and this pattern was observed across a vari-ety of spatial scales. Predictions of the otherhypotheses were not supported. Species rich-ness did not peak at intermediate fire fre-quencies as predicted by the IDH, there wasno evidence of a “saturation effect,” andspecies richness was not highest at the long-est tire return interval as predicted by theFFSLH. Our results were, however, consis-tent with the observat ions of Beckage andStout (2000) in one important respect. Atboth Tiger Corner and Osceola the P-valueindicating the strength of the statistical rela-tionship between fire frequency and speciesrichness decreased noticeably with spatialscale (Figures 6A, B). In fact , at both si tes,the linear contrast for the largest plot size(1000 m’) fell short of significance at the0.05 level (Tiger Corner P = 0. I 1, OsceolaP = 0.29). Thus, our results at this largestscale of measurement were similar to Beck-age and Stout’s (2000) findings for theirsandhill s tudy, which were based on a sim-ilarly large plot size (500 m”).

Volume 23 (11, 2003 Natural Areas Journal 33

TIGER CORNER EXPERIMENTA 160

1SCALE P-VALUE

cn 120-W looo�o l @-li!% AA .11

b 8 0 -

5 ‘0°‘0!!i -2

*\.07

40 -

10.0

1;:a Iit

0 .Ol * l .03w . .04

I I I I I

0 1 2 3 4 5

FIRE RETURN INTERVAL (YRS)

B60

1 SCALE

40

20

0

OSCEOLA EXPERIMENT

P-VALUE

--__-2..

--__l

1 �O��O l \ o,05I 1

0 1 2 3 4 5FIRE RETURN INTERVAL (YRS)

Figure 6. Species richness curves for the (A) Tiger Comer and (R) Osceola studies. P-values of singledegree of freedom linear contrasts are shown to the right of each fitted line. Each point represents anaverage of all data collected at the scale in square meters indicated to the left of the line.

We tentatively conclude, therefore, thatdecreases in longleaf ground-layer speciesrichness with less frequent burning aremore evident at smaller spatial scales. Acrit ical issue is therefore whether this buff-ering capacity, as we might refer to it, at alarge scale is l ikely to be a stable feature ofthese habitats or represents instead anephemeral condit ion. We suspect that i t isephemeral and is due to patchiness in therate at which large woody species becomeestablished and displace other ground-layerplants. Thus, we hypothesize that reduc-tions in species richness related to reduc-tions in fire frequency appear first at smallscales and are then translated to increas-ingly larger scales, leading ultimately tocommunity scale and even landscape scaledeclines in biodiversity. An alternativehypothesis is that patches of habitat suit-able for herbs and small shrubs may beconsistently generated at longer fire returnintervals, though these patches may makeup a smaller proportion of the total areathen would be the case with more frequentburning. Repeated sampling over time inthe same plots will be necessary to dis-criminate between these two hypotheses.

DISCUSSION AND CONCLUSIONS

Our resul ts overal l s t rongly supported theMFFH and will perhaps serve to furtheremphasize to ecologically oriented landmanagers the need for short interval burnsin southern pinelands. This conclusion isconsistent with the findings of the SanteeStudy (Waldrop et al. 1992) which, to-gether with our own results, would appearto const i tute the best avai lable informationon this topic. Annual and biennial burnsalso produced the f inest quali ty, most spe-cies-rich wet savanna communities in therenowned Green Swamp area of NorthCarolina (Walker and Peet 1983) and innumerous si tes sampled by Taggart (1990).Our ONF results suggest that long-termannual burning, particularly during resto-ration, may be necessary in systems dom-inated by highly fire-tolerant shrubs suchas Serenou rtywns.

Before concluding, however, we wouldlike to advance a few caveats and sugges-tions for future research. Perhaps the mostimportant caveat is that before prescribing

34 Natural Areas Journal Volume 23 (l), 2003

any particular fjre regime, managers needto consider the conservation priorities altheir own site. For example, if one is man-aging a rare shrub (e.g., Elliottiu rucemo-sa, Stmhydeonza gruveolens, IWhergillagardenii, Lindem melissifolia), quadren-nial burning in late winter may be a per-fectly appropriate management strategy.The same may apply to rare shrub-domi-nated communities. For example, two lo-cally abundant but globally rare FMNFflatwoods communities (Pinus palustris-Clethra Lllrz~~~lia-Gaylussaciu ,fror&sa-Quercus pumila-Schizachyrium scopari-urn Woodland Association and Pinuspalustris-P. semtina-llex glubra-Lyoninlucidu Woodland Associat ion) in the TigerCorner plots may be threatened over thelong term by long-term annual and bienni-al fires. If maintaining these shrub-domi-nated associations is considered to be animportant management priority, it mightbe helpful to set aside certain sections ofFMNF for slightly longer interval burns.

A second important caveat is that we needto recognize that most fire frequency ex-periments in longleaf pine communities,including our own, have been carried outin f latwoods. Flatwoods, by defini t ion, arecommunities with a strong shrub compo-nent or , at least , environmental condit ionsconducive to shrub invasion (Abrahamsonand Hartnett 1990). Reducing fire frequen-cy, even s l ight ly, s t imulates sprout ing andvegetative proliferation of shrubs, reduc-ing space available for herbaceous plantsand decreasing species richness (Waldropet al. 1992, Brockway and Lewis 1997). Itis therefore not surprising that data col-lected in f latwoods are consistent with theMFFH.

In contrast to the various studies of flat-woods, Beckage and Stout (2000) is theonly fire frequency study that we know ofthat focused on drier longleaf pine habi-tats. Despite the limitations of that study,straightforwardly acknowledged by theauthors themselves, the results are inter-

1 esting. In contrast to flatwoods, sandhilllongleaf pine forests typically lack an un-derstory of dense shrubs, although an un-derstory-midcanopy layer of “scrub” oaks(e.g., Quercus Levis, Q. incam) is presenteither as sprouts or small trees. With re-

ductions in fire frequency the scrub oaksget larger but, being more widely spaced,they may not be as effective as flatwoodsshrubs in competitively excluding herbs.Based on our own field experience, andresults from the dry end Tiger Corner andOsceola plots, we tend to agree with Beck-age and Stout (2000) that sandhil ls may beless sensitive to variations in fire returnintervals than are flatwoods.

In addition to drier longleaf pine sites,some studies have suggested that wet sa-vanna sites and the species therein mayalso be less sensitive than flatwoods tolonger fire return intervals (Streng andHarcombe 1982, Brewer 1999). The datafrom this study, however, do not supportthis hypothesis . Most groups of wet savan-na indicator species, such as small rosetteforbs and sedges, Orchidaceae, and insec-tivorous plants, were amongst the mostsensitive to reductions in fire frequency.Brewer (1999) criticized previous studiesas not providing truly conclusive evidenceof the need for annual or biennial fires formaintenance of Surruceniu populations.Our results, involving long-term random-ly applied burn treatments and observa-tions standardized for time-since-burn ef-fects, appear to avoid most of Brewer’s(1999) criticisms. Perhaps our disagree-ment can be resolved when we considerthat Brewer’s (1998, 2002) own studiesindicate that dominant flatwoods shrubssuch as Ilex glabru can invade pine savan-na habitats under certain circumstances.Furthermore, Brewer (2002) acknowledgedthat reduced tire frequency is one of sev-eral factors that may facilitate invasion offlatwoods shrubs into bogs and wet savan-nas. It therefore seems reasonable to hy-pothesize that lower fire frequencies main-tained over several decades, as in our study,might gradually lead to conversion of wetgrass-dominated savannas into wet shrub-dominated flatwoods. This is a likely sce-nario in our Tiger Corner s tudy plots s incemost of the wet savanna patches occuralong ecotones or in small inclusions withinthe flatwoods. Hurricanes such as Hugomight accelerate the conversion of savan-nas to flatwoods by increasing the areaand rate of natural soil disturbances in-cluding tip up mounds and high intensityfires associated with fuel accumulations

(Wade et al. 1993). Brewer (2002) exper-imentally demonstrated increased rates ofIlex glubru seedling appearance in smallartif icial ly generated soil disturbances, andthe same phenomenon presumably occursin natural disturbances. Thus hurricanesand reductions in fire frequency may in-teract to promote establishment and growthof shrubs in wet savanna habitats. Increas-ing competition from shrubs might thenlead to substantial reductions in abundanceand diversity of wet savanna herbs, suchas we observed at Tiger Corner.

One conclusion from the above discussionis that cross-habitat comparisons of firefrequency treatments would appear to be aprofitable area for future research (see alsoLiu et al. 1997). Even within flatwoodscommunities, comparisons across geogra-phy and soi l formations may be of interest .Comparisons of Tiger Corner and Osceoladata, along with results of other studies,s t rongly suggest tha t Spodosol Fla twoodsand Ultisol Flatwoods respond differentlyto the same fire frequency treatments. Firefrequencies (i.e., annual, biennial burns)that produced herbaceous dominated com-munities in Ultisol flatwoods (SanteeStudy, Tiger Corner Study, Brockway-Lewis Study) had little effect on shrubcover in Spodosol flatwoods (OsceolaStudy). Biennial burning, regardless ofburn season, has also not reduced biomassof dominant flatwoods shrubs in the long-term St. Marks season-of-burn study (J.S.Glitzenstein, D.R. Streng, unpubl. data).Flatwoods plots in the St . Marks Study arealso located on Spodosols. It would ap-pear that, on the whole, shrubs on Spodo-sols are more tolerant of closely spacedfires (see also Eggart 1990). This hypoth-esis is consistent with other observationsthat woody plants on coarser textured soi lsare generally more resistant to stress. Forexample, as climate becomes limiting inOklahoma and central Texas, eastern treesand shrubs become restricted almost en-tirely to sandy soils (Costello 1969).

Given that we accept the MFFH, a relatedissue concerns fire season. Prior to thearrival of humans in North America it isprobable that most fires in southeasternlongleaf pine woodlands, ignited by light-ning associated with thunderstorms, oc-

Volume 23 (I), 2003 Natural Areas Journal 35

curred during the growing season (May-September; Komarek 1968). Thus, thegrowing season is the “natural” fire seasonto which, presumably, most native plantshave adapted. It is also true that humanshave been burning during the fall and win-ter for hundreds, if not thousands of years,and that these human-started fires coveredlarge areas of the landscape (Lawson 1709,Elliott 18 16- 1824). Plants not also adapt-ed to this anthropogenic f ire season wouldlong since have been selected out of exist-ence. It is not within the scope of thispaper to take sides in the issue of growing-versus dormant-season fire (see Streng etal. 1993). We merely note, as has beennoted previously by others (Waldrop et al .1992, Brockway and Lewis 1997), thatfrequent dormant-season fires can be usedto maintain species-rich and apparentlyhigh quality longleaf pine ground-layer ina variety of habitat types. Of this there isno doubt. It may be desirable to switchfrom dormant-season burning to growing-season burning while maintaining an equal-ly high fire frequency (e.g., growing-sea-son burns may have more effectivelypromoted establishment of wiregrass andother dominant bunchgrasses in our Os-ceola plots). However, managers shouldbe aware that reducing fire frequency orarea burned in order to burn at a more“natural” season is likely to be a riskystrategy.

ACKNOWLEDGMENTS

The long-term studies described hereinwould have been impossible without thesustained help and cooperation of the Fran-cis Marion National Forest and the Osceo-la National Forest. We are deeply gratefulto numerous individuals in both nationalforests. Ted Ash, David Combs, and An-drew Hulin deserve special recognition forlong-term help with data collection andplot maintenance. Jimmy Rickards, SusanCarr, and members of South Carolina Na-tive Plant Society, especial ly John Brubak-er, provided additional field assistance. Wethank Randy Heidorn for organizing theFire Forum at the 2001 Natural AreasConference, and Bill Platt for solicitingour contr ibut ion. Steve Orwell, Steve Brew-er, and two anonymous reviewers provid-ed comments on a previous draft of the

manuscript. Lastly, we are indebted toSteve Brewer, Bill Platt, Brian Beckageand other conference participants for aseries of particularly stimulating e-mailexchanges. Financial assistance was pro-vided by the US Forest Service, SouthernResearch Station, Grant # 29-634.

JeflGlitzenstein is a Research Associate utTull Timbers Reseurch Stution. His inter-ests include @ect.s of environmentul gru-clients und nutural and managed distur-bance regimes on plant communities andpopulations. Recent research he has con-ducted with his wtfe, Donna Streng, in-volves experiments with prescribed burn-ing rure plunt population establid~ment,and other restorution techniques in longle@pine ,savunna.s.

Donnu Streng, u Reseurch Associate atTull Timbers Research Stution, is particu-lurly interested in the conservution ofplanthiodiver.sity in jire-mmintuined communi-t ies of the southeastern United States. Herother interests include rare and commonplant dernogrq7hy us well us floristics ofsoutherrstern Coustul Plain plant commu-ni t i es .

Dule Wcrde is u Reseurch Forester with theUSDA Forest Service, Southern ResearchStation. His interests include,fire behuviorund vegetat ion clynamics in southern pine.stund.s, with particulur emphasis on link-ages between these two sets of variubles.He maintains long-term .rtudies of pre-*scribed burning in several southeasternecosystems.

LITERATURE CITED

Abrahamson, W.G. 1984. Species responses tofire on the Florida Lake Wales Ridge.American Journal of Botany 71:35-42.

Abrahamson, W.G., and D.C. Hartnett. 1990.Flatwoods and dry prairies. 1990. Pp. 103-149 in R.L. Myers and J.J. Ewe], eds., Ec-osystems of Florida. University of CentralFlorida Press, Orlando.

Beckage, B., and I.J. Stout 2000. Effects ofrepeated burning in a Florida pine savanna:a test of the intermediate disturbance hy-pothesis. Journal of Vegetation Science1 I:1 13-122.

Binkley, D., D. Richter, M.B. David, and B.

Caldwell. 1992. Soil chemistry in a loblol-ly/longleaf pine forest with interval burn-ing. Ecological Applications 2: 1 S7- 164.

Brewer, J.S. 1998. Patterns of plant speciesrichness in a wet slash-pine (Pin14.y c~lliortii)savanna. Journal of the Torrey BotanicalSociety 125:216-224.

Brewer, J.S. 1999. Short-term effects of fireand competition on growth and plasticity ofthe yellow pitcher plant, Srrrrzrcenirr trltrrcr(Sarraceniaceae). American Journal of Bot-any 86: 1264- 127 1.

Brewer, J.S. 2002. Disturbances increase seed-ling emergence of an invasive native shrubin pitcher plant bogs. Natural Areas Journal22:4-10.

Bridges, E.L., and S.L. Orzell. 1989. Longleafpine communities of the West Gulf CoastalPlain. Natural Areas Journal 9:246-263.

Brockway, D.G., and C.E.Lewis. 1997. Long-term effects of dormant-season prescribedfire on plant community diversity, structureand productivity in a longleaf pine-wire-grass ecosystem. Forest Ecology and Man-agement 96: 167- 183.

Brown, R.B., E.L. Stone, and V.W. Carlisle.1990. Soils. Pp. 35-69 in R.L. Myers andJ.J. Ewel, eds., Ecosystems of Florida. Uni-versity of Central Florida Press, Orlando.765 pp.

Collins, S.L., S.M. Glenn, and D.J. Gibson.1995. Experimental analysis of intermedi-ate disturbance and initial floristic compo-sition: decoupling cause and effect. Ecolo-gy 76:486-492.

Connell, J.H. 1978. Diversity in tropical rain-forests and coral reefs. Science 199:1302-1310.

Costello, D.F. 1969. The Prairie World. Crow-ell Co., New York. 242 pp.

Elliott, S. 1816-1824. A Sketch of the Botanyof South Carolina and Georgia. Vols. 1 and2. Reprint 1971, Hafner Publishing, NewYork. 606 pp. and 743 pp.

Eagle, D. M., M. W. Palmer, J. S. Crockett, R.L. Mitchell, and R. Stevens. 2000. Influ-ence of late season fire on early succession-al vegetation of an Oklahoma prairie. Jour-nal of Vegetation Science 1 I : 135- 144.

FNAI-FDNR (Florida Natural Areas Inventoryand Florida Department of Natural Resourc-es). 1990. Guide to the Natural Communi-ties of Florida. Available online <http://www.fnai.org/NCGuide2p65.pdf>.

Frost, C.C. 1993, Four centuries of changinglandscape patterns in the longleaf pine eco-system. Proceedings Tall Timbers Fire Ecol-ogy Conference 18: 17-44.

Godfrey, R.K., and J.W. Wooten. 1979. Aquaticand Wetland Plants of Southeastern United

36 Natural Areas Journal Volume 23 (I) , 2003

States: Monocotyledons. University ofGeorgia Press, Athens. 712 pp.

Godfrey, R.K., and J.W. Wooten. 198 I. Aquaticand Wetland Plants of Southeastern UnitedStates: Dicotyledons. University of Geor-gia Press. Athens. 933 pp.

Grime. J.P, J.G. Hodgson and R. Hunt. 198X.Comparative Plant Ecology: A FunctionalApproach to Common British Species.Unwin Hyman. London. 742 pp.

Heyward, F. 1939. The relation of fire to standcomposition of longleaf pine forests. Ecol-ogy 20: 287-304.

Hughes, R.H. 1966. Fire ecology ofcanebreaks.Proceedings of the Tall Timbers Fire Ecol-ogy Conference 5: l48- 1%.

Kartesl, J.T. 1994. A Synonymized Checklistof the Vascular Flora of the United States,Canada, and Greenland. Vol. 1.: Checklist.Timber Press, Portland, Ore. 622 pp.

Kirkman, L.K., M.B. Drew, and D. Edwards.lY98. Effects of experimental fire regimeson the population dynamics of Schwalhvcl~rinericarzcr. Plant Ecology 137: 1 15-37.

Komarek, E.V. 1968. Lightning and lightningfires as ecological forces. Proceedings TallTimbers Fire Ecology Conference 8: 169.197.

Komarek, E.V. 1974. Effects of fire on temper-ate forests and related ecosystems: south-eastern United States. Pp. 25 l-277 in C.E.Ahlgren and T.T. Kozlowski, eds., Fire andEcosystems. Academic Press, New York.

Lawson, J. 1709. A New Voyage to Carolina.Reprint 1967, edited by H.T. Lefler, Uni-versity of North Carolina Press, Chapel Hill.305 pp.

Lemon, P.C. 1949. Successional responses ofherbs in the longleaf-slash pine forest afterfire. Ecology 30:35-145.

Lewis, C.E., and T.J. Harshbarger. 1976. Shruband herbaceous vegetation after 20 years ofprescribed burning in the South CarolinaCoastal Plain. Journal of Range Manage-ment 29:13-18.

Liu, C., P.A. Harcombe, and R.G. Knox. 1997.Effects of prescribed fire on the composi-tion of woody plant communities in south-eastern Texas. Journal of Vegetation Sci-ence 8495-504.

McKee, W.H. 1982. Changes in soil fertilityfollowing prescribed burning on CoastalPlain pine sites. Research Paper SE-234,U.S. Department of Agriculture, ForestService, Southeastern Forest ExperimentStation, Asheville, N.C.

Mehlman, D.W. 1992. Effects of fire on plantcommunity composition of North Floridasecond growth pineland. Bulletin of theTorrey Botanical Club 119:376-383.

NatureServe Explorer: an online encyclopediaof life [web application]. 2001. Version I .6.NatureServe, Arlington, Va., USA. <http: /Iwww. natureserve.org/ explorer>

Peet, R.K., and D.J. Allard. 1993. Longleafpine vegetation of the Southern Atlanticand Eastern Gulf Coast regions: a prelimi-nary classification. Proceedings Tall Tim-bers Fire Ecology Conference 18:45-82.

Peet, R.K., D.C. Glenn-Lewin, and J. Walker-Wolf. 1983. Prediction of man’s impact onplant species diversity: a challenge for veg-etation science. Pp. 41-54 in W. Holzner.M.J.A. Werger, and 1. Ikusima, eds., Man’sImpact on Vegetation. Dr. W. Junk Pub-lishers, The Hague, The Netherlands.

Peet. R.K., T.R. Wentworth, and P.S. White.1998. A flexible multipurpose method forrecording vegetation composition and struc-ture. Castanea 63:262-274.

Platt, W.J. 1999. Southeastern pine savannas.Pp. 23-51 in R.C. Anderson, J.S. Fralish,and J.M. Baskin, eds., Savannas, Barrens,and Rock Outcrop Plant Communities ofNorth America. Cambridge UniversityPress, Cambridge, UK.

Percher, R.D. 1995. Wildflowers of the Caro-lina Lowcountry and Lower Pee Dee. Uni-versity of South Carolina Press, Columbia.302 pp.

Sheffield, R.M., and M.T. Thompson. 1992.Hurricane Hugo effects on South Caroli-na’s forest resources. Research Paper SE-284, U.S. Department of Agriculture, For-est Service, Southeastern Forest ExperimentStation, Asheville, N.C.

Sokal, R.R., and F.J. Rohlf. 1969. Biometry.W.H. Freemat? and Company, San Fran-cisco. 776 pp.

Streng, D.R., and PA. Harcombe. 1982. Whydon’t east Texas savannas grow up to for-est? American Midland Naturalist 108:278-294.

Streng, D.R., J.S. Glitzenstein. and W.J. Platt.1993. Evaluating effects of season of bumin longleaf pine forests: a critical literaturereview and some results from an ongoinglong-term study. Proceedings of the TallTimbers Fire Ecology Conference 18:227-263.

Taggart, J.B. 1990. Inventory, classification,and preservation of Coastal Plain savannasin the Carolinas. Ph.D. diss., University ofNorth Carolina, Chapel Hill. 229 pp.

Ter Braak, C.J.F. 1987-1992. CANOCO: afortran program for canonical communityordination by [partial] [detrended] [canon-ical] correlation analysis, principal compo-nents analysis and redundancy analysis (ver-sion 2.1).

Ter Braak, C.J.F. 1995. Ordination. Pp 91-173in R.H.G. Jongman, C.J.F. Ter Braak, andO.F.R. van Tongeren, eds., Data Analysisin Community and Landscape Ecology.Cambridge University Press, Cambridge,UK.

Tester, J.R. 1989. Effect of fire frequency onoak savanna in east-central Minnesota. Bul-letin of the Torrey Botanical Club I 16: 134-144.

Tester, J.R. 1996. Effect of fire frequency onplant species in oak savanna in east-centralMinnesota. Journal of the Torrey BotanicalSociety 123:304-308.

Wade, D.D., J.K. Forbus, and J.M. Saveland.1993. Photo-series for estimating post-hur-ricane residues and fire behavior in south-em pine. General Technical Report SE-82,U.S. Department of Agriculture, ForestService, Southeast Forest Experiment Sta-tion, Asheville, N.C.

Waldrop, T.A., D.L. White, and S.M. Jones.1992. Fire regimes for pine-grassland com-munities in the southeastern United States.Forest Ecology and Management 47:195-210.

Walker, J., and R.K. Peet. 1983. Compositionand species diversity of pine-wiregrass sa-vannas of the Green Swamp, North Caroli-na. Vegetatio 55:163-179.

Weakley, AS. 1999. Flora of the Carolinasand Virginia. Working draft of January19,1999. The Nature Conservancy, South-ern Conservation Science Department,Chapel Hill, NC.

Volume 23 (I), 2003 Natural Areas Journal 37