sampling evaluation of sampling methods and species ... · evaluation of sampling methods and...

SAMPLING

Evaluation of Sampling Methods and Species Richness Estimators forAnts in Upland Ecosystems in Florida

JOSHUA R. KING1, 2, 3 AND SANFORD D. PORTER2

Environ. Entomol. 34(6): 1566Ð1578 (2005)

ABSTRACT The growing emphasis on including invertebrates in global biodiversity conservationefforts has prompted an increase in the study of invertebrate assemblages. Invertebrate samplingdesigns and the bias of individual methods, nevertheless, remain poorly understood for a variety ofhabitats. We used a structured inventory approach to sampling ants in Þve upland ecosystems inFlorida. We evaluated the efÞciency of quantitative and nonquantitative methods for sampling ants.We also evaluated the performance of four species richness estimators. A total of 3,774 speciesoccurrences were distributed among 1,732 samples that contained 94 species from 31 genera. Twentyunique species and 10 duplicate species were collected. Compared with a comprehensive species listfor Florida, sampling captured �66% of the regional fauna and �70Ð90% of species within theecosystems studied. Combinations of sampling methods were much more effective for assessingspecies richness. Individual methods were complementary and sampled only part of the entireassemblage. Nonparametric estimators (an incidence-based coverage estimator [ICE] and a jackknifeestimator [Jack2]) performed better than lognormal Þtting, and Michaelis-Menten curve extrapola-tion. However, none of the estimators was stable, and their estimates should be viewed with skepticism.The results of this study provide support for the use of the Ants of the Leaf Litter (ALL) protocolfor thoroughly sampling ant assemblages in temperate and subtropical ecosystems. Furthermore, ourresults indicate that even in relatively species-poor (compared with the tropics) temperate andsubtropical regions, a large sampling effort that includes multiple sampling methods is the mosteffective manner of thoroughly sampling an ant assemblage. Therefore, we suggest that structuredinventory should be adopted for a wider variety of terrestrial invertebrate studies.

KEY WORDS Florida, Formicidae, rarefaction, species richness estimation, structured inventory

BIOLOGICAL INVENTORY IS A FUNDAMENTAL component ofthe life sciences. Inventories provide the foundationfor improving the applied pursuits of sustainable re-source management, conservation biology, and pestmanagement (Price and Waldbauer 1994, New 1998).Appreciation of the importance of biological inven-tory and the value of biodiversity has steadily grownin the last two decades as the potential impact of thebiodiversity crisis has been recognized (Wilson 1988,Raven and Wilson 1992). For many terrestrial verte-brates and some plants, intensive local or regionalsampling can be expected to produce a comprehen-sive inventory when integrated with existing informa-tion, such as the taxon-based work of systematists(Eldredge 1992). However, for the vast majority ofterrestrial organisms, particularly hyperdiverse groupssuch as arthropods, a comprehensive inventory is dif-Þcult to achieve except at very small scales or in

isolated regions with depauperate faunas (e.g., smalloceanic islands; Disney 1986).

The goal of most arthropod inventories commonlyfalls into one of two categories: strict inventory orcommunity characterization (Longino and Colwell1997). Strict inventory generates a nearly comprehen-sive species list for a discrete spatiotemporal unit,which requires species-level identiÞcation of samples(Longino and Colwell 1997). Comparisons with otherspatiotemporal units are not necessarily desirable.Traditionally, strict inventories have been carried outby systematists and museum collectors. In contrast,community characterization uses structured sampling(i.e., randomization and repetition) to permit statis-tical separation of different spatiotemporal units. Thisis done for the purposes of ranking units according tothe goals of conservation or pest management (Coch-ran 1963, Longino and Colwell 1997). Unit rankingmay not require sample identiÞcations to the specieslevelbecause theprimaryconcern is the relativeabun-dances of focal taxa and how they change across spaceor time (Colwell and Coddington 1994, Oliver andBeattie 1996). Normally, community characterizationis carried out using one or a few sampling techniquesand comprehensive species lists are neither necessary

1 University of Florida, Entomology and Nematology Department,PO Box 110620, Gainesville, FL 32611.

2 USDAÐARS, Center for Medical, Agricultural and VeterinaryEntomology, PO Box 14565, Gainesville, FL 32604.

3 Corresponding author: Department of Biological Science,Florida State University, Tallahassee, FL 32306Ð4370 (e-mail:[email protected]).

0046-225X/05/1566Ð1578$04.00/0 � 2005 Entomological Society of America

nor feasible (Disney 1986, Longino and Colwell 1997).Community characterization has most often beenused by entomologists, ecologists, and conservationbiologists.

Recently, analytical advances focused on examiningarthropod ecology have converged with a growingemphasison includingarthropods in rapidbiodiversitysurveys for conservation purposes (Wilson 1988, Cod-dington et al. 1991, Kremen 1992, 1994, Kremen et al.1993, Kim 1993, Colwell and Coddington 1994, Silvaand Coddington 1996, Longino and Colwell 1997, New1998, Fisher 1999a, Anderson and Ashe 2000, Gotelliand Colwell 2001, Sørensen et al. 2002). Accordingly,a number of taxon-speciÞc, structured inventory tech-niques have been introduced that use a variety ofsampling methods that combine the “species hunting”techniques of systematists with the more quantitativemethods of ecologists (Coddington et al. 1991,Longino and Colwell 1997, Fisher 1999a). This ap-proach serves as a practical, short-term alternative tolong-term, comprehensive surveys (e.g., Deyrup andTrager 1986, Lawton et al. 1998) for assessing localarthropod diversity. By design, this methodology com-bines the quantitative approach of community char-acterization with the objectives of strict inventory. Itpermits analyses of inventory completeness and anassessment of the costs and beneÞts of methods whenused individually and in combination (Longino andColwell 1997, Fisher 1999a, Delabie et al. 2000).

To assess inventory methods, an account of speciescaptured per sampling method (as well as per unit areaor time) is necessary to evaluate effectiveness. Speciesaccumulation curves are an effective method for eval-uating the efÞciency of various inventory techniquesfor sampling species richness (Clench 1979, Soberonand Llorente 1993, Colwell and Coddington 1994). Aspecies accumulation curve is a plot of the cumulativenumber of species discovered within a deÞned area(and/or time) as a function of some measure of effort.Curves can be constructed to measure either speciesdensity (the number of species per unit area) or spe-cies richness (the number of species per individual;Gotelli and Colwell 2001). Species accumulationcurves are similar to rarefaction curves (Gotelli andColwell 2001), which are produced by randomly andrepeatedly resampling a pool of individuals or samplesand plotting the number of species represented byincreasing numbers of individuals. In fact, rarefactioncurves can be viewed as the statistical expectation ofa corresponding species accumulation curve whensamples are repeatedly reordered (Gotelli and Col-well 2001).

When the actual number of species in an area isunknown (as is typically the case for arthropods) theshape of a rarefaction curve can be used to estimatehow completely an area has been sampled and howefÞciently different methods have captured species inthat area (Colwell and Coddington 1994, Longino andColwell 1997, Fisher 1999a). A curve that approachesan asymptote after a large sampling effort is represen-tative of a decrease in species accrualÑa measure ofsampling completeness (Colwell and Coddington

1994, Longino and Colwell 1997, Gotelli and Colwell2001). Sampling completeness can be further evalu-ated by comparing curve asymptotes with values de-termined by species richness estimators or by ob-served species richness from well-sampled localities.When used in combination with a structured inven-tory, rarefaction curves are intended to permit quan-titative analyses of species richness for comparisonbetween methods or between sites. Additionally, sub-samples can be evaluated within the context of theentire data set to determine the relative efÞciency ofmethod combinations and changes in design (e.g., theeffect of increasing distance between samples; Fisher1999a).

Among arthropods, ants (Hymenoptera: Formici-dae) have been a focal group for development ofstructured inventory protocols and novel techniquesfor analyzing data generated from structured inven-tory. Ants are an appropriate group for testing theeffectiveness of new methods. They are diverse, abun-dant, and nearly ubiquitous. They inßuence the bioticand abiotic processes of the ecosystems where theyoccur. A majority of species nest in Þxed positions,largely ensuring that species dwell where they aresampled. They are among the most studied terrestrialinvertebrate taxa and have been used to monitor en-vironmental impact and ecosystem recovery(Andersen 1990, 1993, Holldobler and Wilson 1990,Folgarait 1998). Numerous, established sampling tech-niques are available, representing a range of costs andyields (Bestelmeyer et al. 2000). Furthermore, struc-tured inventory techniques and biodiversity data anal-yses have been established for inventorying and esti-mating ant species richness in tropical rainforestecosystems (Fisher 1996, 1998, 1999a, b, 202, Longinoand Colwell 1997, Agosti et al. 2000a, b, Longino et al.2002). Much less is known about the performance ofstructured inventory methods to capture ants in sub-tropical and temperate ecosystems. Similarly, the per-formance of species richness estimators remainslargely untested on data drawn from temperate orsubtropical ant communities. The biodiversity crisis isnot conÞned to the tropics (Platnick1992) and neitherare the goals of conservation biology, entomology, andecology. Accordingly, improving structured inventorymethods and species richness estimation for ants andarthropods in a variety of ecosystems is warranted.

In this study, we evaluated the efÞciency of a struc-tured inventory of ants in northern and central Flor-ida. In so doing, we compared the efÞciency of fourindividual sampling methods and the performance ofthree species richness estimation techniques. The antfauna of Florida has been thoroughly surveyed in thepast 50 yr throughout the state (Van Pelt 1956, 1958,Deyrup and Trager 1986, Deyrup et al. 2000, Luber-tazzi and Tschinkel 2003, M. Deyrup, L. Davis, Z.Prusak, S. Cover, and J. R. King, unpublished data). Asa result, there is a clear record of distribution andhabitat association for a majority of species (Deyrup2003). Consequently, this study presented a uniqueopportunity to compare inventory results and speciesrichness estimations with approximate species rich-

December 2005 KING AND PORTER: ANT SAMPLING AND SPECIES RICHNESS ESTIMATION 1567

ness values expected at local and regional scales. Inevaluating sampling methods our objectives were (1)to compare the number of ant species captured bybaits, pitfalls, litter extraction, and hand-collectingmethods (and combinations thereof) in Þve differentecosystems, (2) to compare the complementarity ofsampling methods, and (3) to determine the relativecosts (in time) of sampling methods. These objectivespermit an analysis of the efÞciency of sampling meth-ods in the context of facilitating conservation or land-planning decisions that require comparable estimatesof species richness and endemism (Coddington et al.1991, Platnick 1992).

Materials and Methods

Study Area. A detailed description of the ecosys-tems sampled and the methods used in this study canbe found in King (2004). Brießy, this study was carriedout in 3 sites in each of Þve different ecosystems (atotal of 15 study sites) in north and central Florida.Sampling was performed in temperate hardwood for-ests in the San Felasco Hammock State Park, pineßatwoods in the Osceola National Forest, high pine inthe Katherine Ordway Biological Preserve, and Flor-ida scrub forest in theArchboldBiological Station.Theplant community descriptions of Myers and Ewel(1990) were used as a basis for ecosystem selection.These siteswere selectedaprioribecause theycontainsome of the remaining nearly undisturbed native up-land ecosystems in Florida. We also sampled a non-native ecosystem, consisting of converted (previouslycleared) Þelds, to represent moderately disturbedhabitat. A Þeld was sampled in an area adjacent tonatural areas at San Felasco Hammock State Park, theKatherine Ordway Biological Preserve, and the Arch-bold Biological Station. The ecosystems sampled rep-resent a gradient of upland plant communities thatinclude closed-canopy, hardwood forests, open-can-opy pine and oak woodland, and completely open,herbaceous savannah.Sampling. Ants were sampled from June through

September 2001 using baits, pitfall traps, leaf litterextraction with Berlese funnels, and hand collecting.A total of 36 pitfall and 36 litter samples were takenseparately at 5-m intervals (180 m total) in two parallellines separated by 10 m (Fisher 1996, 1998, 1999b,2002). A transect of 36 baits was placed between thepitfall and litter extraction lines, with each bait cor-responding to pitfall and litter extraction samples. Pit-fall traps were 85-mm-long plastic vials with a 30-mminternal diameter partially Þlled with �15 mm of non-toxic, propylene glycol antifreeze. Traps were buriedwith the open end ßush with the surface of the groundand operated for 3 d. Two 0.25-m2 litter samples weretaken after setting pitfall traps at each litter samplepoint. Litter samples were obtained by collecting allsurface material and the Þrst �1 cm of soil withinquadrats. The two samples were pooled; larger objects(e.g., logs) were macerated with a machete and siftedthrough a sieve with 1-cm grid size. Sifted litter wasplaced in covered metal 32-cm-diameter Berlese fun-

nels under 40-W light bulbs. The funnels were oper-ated until the samples were dry (�48Ð72 h). Baitswere 12 by 75-mm test tubes with a piece (�2 g) of hotdog (Oscar Meyer, NorthÞeld, IL) inserted �2 cminto the tube. At each sampling point, a small spot wascleared of any litter, and the bait tube was placeddirectly on the ground (to speed discovery and ac-cess) and shaded with one half of a Styrofoam plate.The baits were left for 0.5 h and collected, and the endswere plugged with small cotton balls to prevent theants from escaping. Hand collecting consisted of sys-tematically searching vegetation, tree trunks, logs, andsmall twigs for 2 h per site. Hand collecting was per-formed within and immediately adjacent to (�5Ð10m) sites. Time records were kept for each step of thesampling and sorting process to estimate costs. Timecosts included installation, operation, and collecting ofsamples, traps, and baits, and time spent processingand identifying specimens. For the sake of technicalconsistency, and with the exception of one high pinesite, all sampling was performed by J. R. King. Simi-larly, all ants, from all samples, were sorted, counted,and identiÞed to species by J. R. King.Analysis. All records used in this study were based

on the worker caste as their presence provides evi-dence of an established colony (Fisher 1999a, Longinoet al. 2002). The purpose of the sampling design wasto produce a representative, nearly complete specieslist for each ecosystem type. The relative abundanceof individuals is an important measure when consid-ering species richness (Gotelli and Colwell 2001). Forants, however, the sampled abundance of foragingworkers is not comparable with individuals of otheranimals. The sociality of ants can often lead to extremeclumping of individuals within samples (particularlylitter samples, which may include entire colonies),which may skew species richness comparisons andspecies/abundance relationships (Gotelli and Colwell2001). To partially remedy this, species occurrences(incidence data) were used in place of the abundanceof individuals when evaluating species-based abun-dance measures (Longino 2000, Fisher and Robertson2002, Longino et al. 2002). Therefore, data for thethree sites sampled within each ecosystem type werepooled and converted to a species � sample incidence(presence-absence)matrix(Longino2000,Longinoetal. 2002). A regional data set was also generated bypooling all of the data into a single species � sampleincidence matrix.

Sample-based rarefaction curves were used to com-pare total species richness within ecosystems and forthe regional data set. Total species richness capturedby each method in each ecosystem was similarly eval-uated. Rarefaction curves were generated by randomreorderings (50 times) of samples using the programEstimateS (Colwell 2000). We compared these sam-ple-based rarefaction curves with the observed num-ber of species plotted on the y-axis (ordinate) againstspecies occurrences (incidence data) plotted on thex-axis (abscissa) to assess differences in species rich-ness between and among curves representing ecosys-

1568 ENVIRONMENTAL ENTOMOLOGY Vol. 34, no. 6

tems and methods (Colwell and Coddington 1994,Gotelli and Colwell 2001).

Species richness was estimated in three ways foreach ecosystem and for the regional data set: (1) byÞtting a lognormal distribution, (2) by extrapolatingrarefaction curves, and (3) by using nonparametricestimators. For the parametric model Þtting, the sam-ple data from each site were Þtted to the lognormaldistribution using the method of Preston (1948). Inthis method, the sample data were Þtted to the log-normal distribution, S(R) � S0e

(�a2R2), where S(R) isthe number of species in the Rth octave, S0 is thenumber of species in the modal octave, and a is aparameter related to the width of the distribution.Parameters of the lognormal were estimated using amodiÞed version of the method of Ludwig and Reyn-olds (1988). Octaves were assigned to abundanceclasses (observed), and parameters S0 and a wereestimated using nonlinear curve Þtting (proc nlin,NewtonÕs estimation method and least squares Þtting,SAS release 8.02; Longino et al. 2002). Species richnesswas estimated by calculating the total area under theÞtted curve, including the portion of the curve hiddenbehind the “veil line” (Magurran 1988).

Projection of the rarefaction curves for the regionaldata set was accomplished using sample-based rar-efaction plots (Soberon and Llorente 1993, Gotelli andColwell 2001). Patterns observed in the rarefactioncurves generated for individual ecosystems were sim-ilar to those of the regional data set and were notdisplayed. The smoothed curves were created by av-eraging repeated, random reorderings (50 times) ofthe samples with the mean number of species occur-rences from each sample computed in succession(Colwell and Coddington 1994, Gotelli and Colwell2001). The smoothed curves were projected by Þttingthe asymptotic Michaelis-Menten equation,

S(n) �Smaxn

B � n,

where S(n) is the number of species, n is the numberof samples, and Smax and B are Þtted constants (Col-well and Coddington 1994). Following Raaijmakers(1987),maximumlikelihoodestimators forparameters(Smax and B) were determined for the Eadie-Hofsteetransformation of the equation (Colwell and Codding-ton 1994, Longino et al. 2002).

We also used two nonparametric methods to esti-mate species richness for comparison with the rar-efaction curves. A jacknife (henceforth Jack2) esti-mator,

S � Sobs � �L(2n � 3)

n�

M(n � 2)2

n(n � 1) � ,

was used, where L � the number of species in onesample, M � the number of species that occur in twosamples, and n � the number of samples (Burnhamand Overton 1978, 1979). An incidence-based cover-age estimator (ICE),

Sice � Sfreq �Sinfr

Cice

�Q1

Cice

�ice2 ,

was also used, where Sfreq � number of frequentspecies (found in �10 samples), Sinfr � number in-frequent species (found in �10 samples), Cice � sam-ple incidence coverage estimator, Q1 � frequency ofuniques (the number of species known from only onesample), and �ice

2 � estimated coefÞcient of variationof the Q1 for infrequent species (Lee and Chao 1994).These estimators were chosen because they have beenshown to reliably provide intermediate (ICE) andupper (Jack2) level species richness estimates relativeto observed species richness and other nonparametricestimators in biological inventories (Colwell and Cod-dington 1994, Chazdon et al. 1998, Sørensen et al.2002).

All species richness estimates were compared withtotal species richness values generated for each eco-system and the region from previous collectionrecords. These records included published (Van Pelt1956, 1958, Deyrup and Trager 1986, Lubertazzi andTschinkel 2003) and unpublished inventories (VanPelt 1947, Prusack 1997) and personal collection datasets (J. R. King, L. Davis, M. Deyrup). Amassed over�50 yr, these collections represent a comprehensivespecies occurrence data set for several localities andthe region as a whole. Using these data, a potentialupper limit of species richness was determined for theregional data set by including all of the species withcollection records coinciding with the study area(north Florida in the region stretching from Columbiaand Baker Counties southward along the central ridgeof the peninsula to the end of the Lake Wales Ridgein Highlands County).

At the scale of the ecosystem, the potential speciesrichness value was also established using collectionrecords. The potential species richness values weredetermined using the highest number of species pre-viously captured by an approximately similar samplingeffort (many samples taken using multiple methods)within a study area of similar size to those sampled inour study (e.g., the hardwood hammock within SanFelasco Hammock State Park). The potential speciesrichness values are approximations and represent agenerous upper limit estimate of the potential of spe-cies that may occur at the regional scale and withinecosystems. Similarly, collection records were used toassess the observed rarity of species collected in thisstudy. Species that were unique to either the study asa whole (uniques: the number of species known fromonly one sample) or to individual sampling methodswere compared with collection records to determinewhether the rarity of these species was a product ofinsufÞcient (or inappropriate) sampling or if theywere truly rare.

To assess complementarity among methods, weused a measure that describes the proportion of allspecies in two sites that occurs in only one or the othersite (Colwell and Coddington 1994). Pairwise comple-mentarity or distinctness values were calculated usingthe Marczewski-Steinhaus (M-S) distance:


CMS �a � b � 2j

a � b � j,

where a � the number of species at site A, b � thenumber of species at site B, and j � the number ofspecies common to both. Methods were comparedwithin ecosystems and for the regional data set. Tofurther analyze the sampling efÞciency of quantitativemethods, litter and pitfall data were examined withinsites to determine the impact of spatial separationamong sample points on the similarity of species sam-pled. Within sites, faunal similarity among samplesalong each transect was determined using the Jaccardindex:

Sj �j

a � b � j

(the complement of the M-S distance). The similarityof samples for all pairwise combinations of samplepoints (5Ð175 m apart) along each transect was plottedagainst increasingly larger distances along each sitetransect to determine the homogeneity of samples(Fisher 1999a). These were averaged among siteswithin each ecosystem. Comparisons beyond 150 mwere excluded from the analysis because there wereincreasingly fewer replicates to generate means (e.g.,there is only one pairwise comparison of samples175 m apart).

We examined the time costs of individual methods.Time costs included time spent in the Þeld collectingsamples and in the laboratory sorting and identifyingspecimens. Laboratory hours were spent on tasks thatrequired previous, rigorous scientiÞc training (e.g.,sorting and identifying specimens to the specieslevel). Field hours were an account of time spent ontasks that did not require specialized training (e.g.,laying out pitfalls, sifting litter). Although the timerequired to accomplish these tasks would undoubt-edly vary among different workers, a comprehensiveaccount of time costs per method provides an estimateof the amount of time required to reproduce similarresults under similar conditions. The time cost permethod would remain consistent relative to the timecosts for the other methods even if, for example, mul-tiple workers were using the same methods.

Finally, the effectiveness of quantitative methodsfor predicting total site species richness was deter-

mined. Ordinary least-squared regressions were usedto determine the relationship between total speciesrichness and species richness per method for baits,pitfalls, and litter samples within sites.

Results

The inventory captured 37,961 individual ants rep-resenting 94 species from 31 genera. A total of 3,774species occurrences were distributed among 1,732samples. Seventy-four ground-dwelling species, 12 ar-boreal species, and 8 subterranean species were cap-tured. A species list and detailed discussion of therelative abundance, body size, and ecology of indi-vidual species captured in this study can be found inKing (2004).Observed and Estimated Species Richness. Ob-

served species richness varied among ecosystems. Themost species were captured in high pine and Floridascrub ecosystems, whereas fewer species were cap-tured in pine ßatwoods, Þeld, and hardwood hammockecosystems, respectively (Table 1, observed species).At the regional scale, the 94 species sampled included20 uniques (the number of species known from onlyone sample) and 10 duplicates (the number of speciesknown from two samples; Fig. 1). Two-thirds of thespecies were captured within the Þrst �750 speciesoccurrences. Beyond 750 species occurrences, the ad-dition of species was considerably diminished.

Species richness estimates performed poorly at theregional scale (Fig. 1). At the local scale, nonpara-metric estimators performed better than projectingthe Michaelis-Mentin curve or lognormal curve Þtting(Table 1). The method of lognormal curve Þtting toestimate species richness consistently predicted spe-cies richness levels at or only slightly above the ob-served richness (Fig. 1; Table 1). All other richnessestimators generally did not stabilize with increasingsampling effort. As sample size increased, so too didthe estimates (Fig. 1). Estimators also consistentlypredicted widely separated results relative to eachother. Projecting the Michaelis-Menten curve pro-duced estimates at or, most frequently, below ob-served species richness. The MMMeans estimate alsoincreased steadily with sample size with no evidenceof leveling off (Fig. 1). The ICE consistently predictedspecies richness greater than the observed species

Table 1. Species richness estimates and measures of inventory completeness for each ecosystem and for the regional data set

Hardwoodhammock

Pineßatwoods

Floridascrub

Highpine

Field Regional

Potential species richness 41 46 48 58 45 142Observed species richness (% of potential) 29 (71%) 39 (85%) 43 (90%) 48 (83%) 35 (78%) 94 (66%)Richness estimators

Lognormal 29 39 43 48 35 96MMMean 26 38 42 50 37 87ICE 35 49 50 53 41 113Jack2 41 52 56 57 47 124

Singletons 7 9 9 8 7 20

Potential species richness values represent known species richness values determined from previous inventories. Richness estimator(MMMean, ICE, and Jack2) values represent the mean of 50 randomizations of sample order. Singleton values represent the no. of specieswith only one individual in samples. All values were rounded to the nearest whole no.


richness but less than Jack2 estimates. Both ICE andJack2 showed some evidence of leveling off between1,500 and 2,200 species occurrences; however, theyboth became unstable beyond 2,200 occurrences, ris-ing steadily with increasing sample size. Neitheruniques (the number of species known from only onesample) nor duplicates (the number of species knownfrom two samples) showed any sign of decreasing.Rather, they remained at nearly the same levelthroughout the majority of the sampling effort.

The estimates for the completeness of the inventoryranged from 71 to 90% of the potential species richnessestimates within ecosystems (Table 1). Similarly, anestimated 66% of the total regional fauna was sampledduring this study. Within ecosystems, the ICE andJack2 estimators were very near the potential siteestimates predicted from previous survey work. Incontrast, MMMeans and lognormal estimates werelittle different from observed species richness values.For the regional data set, Jack2 was nearest to thepotential site estimate while all other estimators pro-duced considerably lower species richness estimates.Rarity.Uniques (the number of species known from

only one sample) accounted for �21% of the totalspecies richness. The 20 uniques captured in this studyincluded Þve arboreal species. All of the arboreal spe-cies captured are common, although not abundantthroughout the range of the study and were poorlysampled by the methods used. Similarly, two of theuniques were subterranean species that are rarely cap-tured, although an alternative sampling method (sub-

terranean baiting; J. R. King, unpublished data) hasrevealed that one of these species is common. Fiveground-dwelling species among the uniques are fre-quently encountered within peninsular Florida, al-though their low abundance here suggests that theyare uncommon in the upland habitats sampled. Theremaining eight ground-dwelling species have notbeen commonly collected previously, suggesting thatthese species are truly uncommon or rare. For a list ofthe unique species, see King (2004).Complementarity of Ecosystems. Comparing the

fauna of different sites revealed that, on average, spe-cies composition was much more similar among sitesin the same ecosystem type than among differentecosystems (Table 2). Between different ecosystems,the complementarity of species lists was similar. A

Fig. 1. Observed species richness, uniques (the number of species known from only one sample), duplicates (the numberof species known from two samples), and estimated species richness for the jacknife (Jack2), incidence-based coverage (ICE),lognormal, and Michaelis-Mentin curve (MMMean) estimators for the regional (all samples pooled) data set (see text forestimator formulas). Curves are sample based rarefaction curves generated from 50 randomizations of sample order. Thelognormal estimate is a single value determined by Þtting a lognormal curve to the full species abundance distribution.

Table 2. Mean percent faunal complementarity � SD betweenecosystems

Hardwoodhammock

Pineßatwoods

Highpine

Floridascrub

Field

Hardwood hammock 40� 5 76 � 4 78 � 2 69 � 8 79 � 5Pine ßatwoods 47� 3 73 � 8 65 � 5 81 � 8High pine 43� 9 71 � 5 72 � 9Florida scrub 45� 15 76 � 5Field 64� 2

Complementarity values are M-S distances between species lists foreach ecosystem. Bold values represent comparisons among siteswithin the same ecosystem. Higher values indicate that sites haveincreasingly dissimilar fauna.


similar pattern was also seen among sites of the sameecosystem with the exception of Þeld sites. Field siteswere less similar to one another than other ecosys-tems, probably as a result of their geographic separa-tion. The greatest difference in species compositionoccurred between pine ßatwoods and Þeld sites,whereas the least difference was seen between Floridascrub and pine ßatwoods sites.Effectiveness of Sampling Methods. Sample-based

rarefaction curves for individual methods (Fig. 2) re-vealed that pitfall and litter samples were nearly iden-tical in the number of species each captured and bothmethods captured more species than hand collectingand baiting. This pattern was consistent within indi-vidual ecosystems as well, except for hardwood ham-mocks (data not shown). In hardwood hammocks,litter sampling captured a much greater number ofspecies (26) than pitfalls (17). Baiting was the leastproductive method for capturing species richness. Thecombination of quantitative methods (pitfall, litter,and bait samples) was the most thorough approach forcapturing species richness. The hand collecting curveshows little evidence for an asymptote at all, indicatinga very rapid increase in species per individual col-lected and no evidence of an asymptote. The rapidaccrual of species by hand collecting and the lack ofan asymptote suggest that collecting in this manner isthe most efÞcient method for maximizing species rich-ness. The lack of an asymptote suggests that too littletime was spent hand-collecting per site. However, thisis a methodological artifact, because duplicate collec-tions were generally avoided. Consequently, time,rather than species, occurrences are a better ordinatefor assessing the effectiveness of this method.

Litter and pitfall samples were the most similar inspecies composition among all sampling methods (Ta-ble 3). Baits and litter samples were the least similarin species composition. Overall, the fauna captured byhand collecting was the least similar in species com-position to any of the other collecting methods. Litterextraction sampling was particularly useful for cap-turing cryptic species living in rotting wood and in leaflitter. Pitfall samples were most effective at capturinghighly active, surface foraging species. Hand collect-ing captured a number of arboreal species that werenot captured by other methods.

Within ecosystems, the similarity of species in sam-ples along 180-m linear transects generally varied onlyslightly between nearby and distant samples (Fig. 3Aand B). Field sites, however, exhibited the most dis-similarity between nearby and distant sites with asteady decrease in the similarity of species composi-tion per sample as a function of increasing distance(Fig. 3A and B). Sample autocorrelation differedamong ecosystems and methods. Hardwood hammocksamples exhibited greater similarity among samples,

Fig. 2. Effectiveness of individual and combined sample methods. Curves are sample-based rarefaction curves based on50 randomizations of sample order. Curves correspond to individual methods (H, hand collected; P, pitfall; L, litter extraction;B, bait) and a combination of methods (P � L � B, combined pitfall, litter extraction, and baits).

Table 3. Percent faunal complementarity of sample methodsfor the regional data set

Pitfall Litter Bait Hand

Pitfall Ñ 38 47 55Litter Ñ 60 57Bait Ñ 56Hand Ñ

Values are M-S distances generated from the regional (all sitespooled) species list. Higher values indicate that methods are moredissimilar in the species captured.


on average, than other ecosystemsÑparticularlyamong litter samples (Fig. 3A). Pine ßatwoods andFlorida scrub ecosystems were most similar in thedegree of sample autocorrelation, whereas hardwoodhammock and high pine were the least similar (Fig. 3Aand B). With the possible exception of Þelds, theimpact of ecosystem type seems to be a greater de-terminant of sample autocorrelation than distance be-tween samples.

Litter samples required the most time to collect,sort, and process, followed by pitfalls, baits, and handcollected samples, respectively (Fig. 4A). Baits com-monly produced very large numbers of individuals ofthe same species, particularly mass recruiting speciessuch as Pheidole dentata Mayr and Solenopsis invictaBuren (Fig. 4B). Litter sampling often captured entirecolonies or colony fragments of multiple species, alsoresulting in large numbers of individuals. Pitfall sam-ples and hand collecting tended to produce morespecies per individual sampled. There were 8 speciescaptured only in pitfall samples, 13 species capturedonly in litter samples, 8 species captured only by handcollecting, and 1 species captured only by baiting (Fig.4A).

Pitfall species richness predicted 69% of total sitespecies richness (Fig. 5; total � 3.58 � 1.29pitfall, R2

� 0.69,P� 0.01). Litter species richness accounted forslightly less (61%) of the variability in total speciesrichness (total � 7.47 � 1.04litter,R2� 0.61,P� 0.01),whereas bait species richness predicted 46% of totalsite species richness (total � 15.98 � 1.29bait, R2 �0.46, P � 0.01) across all sites.

Discussion

Inventory Completeness. There are 218 ant speciescurrently known from the entire state of Florida, 142of which are known to occur within the region of thisstudy (Deyrup 2003). It is less clear how many speciesmight dwell within a given upland ecosystem in thestate, although previous inventories provide reason-able estimates (Table 1). At the regional scale, �66%of the ant fauna was captured. Within the ecosystemssampled, �70Ð90% of the ant fauna was captured.Because our estimates of the upper limit of speciesrichness were conservative (see Materials and Meth-ods), it is quite possible that we actually captured ahigher percentage of the ant species present, partic-ularly at the local scale. At the local (ecosystem) scale,this level of inventory completeness is equal to orbetter than similar studies conducted in tropical rain-forests (Longino and Colwell 1997, Fisher 1999a, Ag-osti et al. 2000a, b). There are no comparable studiesof structured inventory for capturing the ant speciesrichness of a regional fauna, although intensive sam-pling in temperate and subtropical localities has cap-tured similar levels of species richness (Talbot 1975,Deyrup and Trager 1986). Among similar, thoroughinventories of ant species richness, regardless of scale,the sampling effort put forth in this study was com-parable in the total number of samples taken (Talbot1975, Deyrup and Trager 1986, Fisher 1996, 1998,1999a, b, 2002, Longino and Colwell 1997).

Although our sampling design captured a large por-tion of the species present, it still cannot be described

Fig. 3. Similarity of ant species in (A) litter and (B) pitfall samples as a function of distance. Curves represent Jaccardindex values of pairwise comparisons of pitfall and litter data at increasing distances between samples. Jaccard index valueswere averaged across (3) replicate sites for each ecosystem, where higher values indicate greater similarity among samples.


as comprehensiveÑundoubtedly species were missedat local and regional scales. At both scales, the short-comings of the sampling methods and the spatial lim-itations of the sample design were probably largelyresponsible for the underestimates, rather than thesampling effort (the number of samples; Longino andColwell 1997). Sampling additional sites or using dif-ferent methods would capture more species muchmore readily than additional sampling within sites.Nevertheless, the structured inventory protocol usedhere provided a sufÞciently thorough sample of localand regional ant species to permit an accurate com-parison of the complementarity and total species rich-ness of different areas. Traditionally, invertebrateshave been excluded from natural area inventories be-cause of the difÞculty of obtaining accurate estimatesof local and regional species richness. This approachhas been adopted despite their importance in ecosys-tem functioning and their contribution to total local

and regional biodiversity. The results presented hereand for tropical ant inventories (Fisher 1996, 1998,1999a, b, 2002, Longino and Colwell 1997) suggest thatstructured inventory can provide sufÞciently accuratemeasures of ant diversity to include them in naturalarea inventories in temperate, subtropical, and trop-ical regions. Similar results have been obtained usingstructured inventory to capture spiders (Coddingtonet al. 1991, Silva and Coddington 1996, Dobyns 1997,Toti et al. 2000, Sørensen et al. 2002). Accordingly,structured inventory of arthropods should be adoptedas a primary component of future natural area inven-tories.Performance of Species Richness Estimators.

Whether evaluated for individual ecosystems or forthe regional (combined) data set, the four speciesrichness estimators differed considerably among eachother. They produced a wide range of estimates andgenerally did not stabilize as sample numbers in-creased (Fig. 1). Instability of coverage-based estima-tors, curve extrapolation, and the lognormal modelgenerally seems to be the rule rather than the excep-tion, regardless of the origin of the data set (Chazdonet al. 1998, Toti et al. 2000, Longino et al. 2002, Sø-rensen et al. 2002). This trend suggests that species

Fig. 4. The efÞciency of methods as measured by (A) thetotal time required and the number of unique species (num-ber of species captured only by that method; represented bythe values above the bars) sampled by different collectingmethods and (B) the abundance/richness index. Sorting (inhours) represents the total amount of time required to sortand identify specimens per method. Collecting (in hours)represents the total amount of time required to collect sam-ples per method. It does not include the time required tooperate pitfall traps (3 d). The abundance/richness index isthe ratio of total number of ants captured divided by the totalnumber of species captured per method. Higher numbersrepresent fewer species per individual captured.

Fig. 5. The relationship between total site species rich-ness and total method species richness for bait, pitfall, andlitter samples. All regressions were signiÞcant (P � 0.01).Pitfall site species richness accounted for the most variabilityin total site species richness (R2 � 0.69) followed by litter(R2 � 0.61) and baits (R2 � 0.46).


richness estimates calculated with these estimatorsshould be very cautiously interpreted as approxima-tions of potential species richness. If used at largerscales, they are probably even less reliable. Lognormaland MMMeans estimates were at or slightly belowobserved species richness. The ICE estimator pre-dicted higher numbers of species than either lognor-mal or MMMeans, and Jack2 consistently predictedthe highest number of species. Only ICE estimatesshowed any evidence of stability for the regional dataset (Fig. 1). However, the curve stabilized only be-tween �1,000 and 2,000 species occurrences andshowed no evidence of stabilization above or belowthese values (Fig. 1). Within ecosystems, both ICEand Jack2 estimates were very near to potential spe-cies estimates (Table 1). However, because the esti-mators were unstable and were proportional to samplesize, the values of species richness they predict shouldalso be viewed with skepticism. It has been suggestedthat using nonparametric estimators in combinationmay predict upper and lower boundaries of local spe-cies richness (Toti et al. 2000). We would suggest thatit has not yet been clearly shown that such an ap-proach is valid.

The poor performance of the lognormal andMMMeans estimators for our data set contrasts withtheir performance in other intensive inventories (Totiet al. 2000, Longino et al. 2002). Our data were poorlyÞt by lognormal and asymptotic functions (King2004), and as such, the estimators were inappropriatefor the data set. The use of these estimators should beclosely evaluated based on their Þt to data and thenumber of rare species in the data set (Longino et al.2002). The lognormal in particular may be problem-atical because of the necessity of Þtting a continuousdistribution to discrete data, the variability of esti-mates resulting from the selection of different inter-vals of abundance categories, and the lack of Þt tosome data sets (Colwell and Coddington 1994). Thelognormal model should not be assumed to Þt theabundance distribution of species (Lambshead andPlatt 1985, Hughes 1986), particularly for temperateand subtropical invertebrate assemblages (Siemann etal. 1999).Rarity. In contrast to many tropical insect (e.g.,

Hemiptera, Lepidoptera) inventories, rare species didnot account for a majority of the species captured(King 2004). Among unique species, a majority wererare because of methodological edge effects (Longinoet al. 2002), range limitations, or small populationswithin the sampled area (i.e., they are known to becommon elsewhere within the region). Arboreal spe-cies and some of the subterranean species were un-dersampled because of the limitations of the samplingdesign. Of 20 unique species, only 8 species (�8% ofthe total number of species) could be considered trulyrare compared with their known distributions. Theseresults are similar to other comprehensive temperate,subtropical, and tropical ant inventories, where trulyrare species comprise a minority of species (Van Pelt1956, 1958, Talbot 1975, Deyrup and Trager 1986,Longino et al. 2002). The occurrence of only a small

number of truly rare species (apparently range re-stricted and numerically uncommon) within a givenregion seems to be a persistent pattern among ants asrevealed by intensive inventory efforts (Longino et al.2002). This pattern of rarity in ants contributes to thestability and accuracy of species richness estimators,particularly nonparametric estimators, because theircalculation is dependent on the number of rare species(Colwell and Coddington 1994, Chazdon et al. 1998,Longino et al. 2002). It is unclear, however, howchanges in the relative proportion of rare species fromdata set to data set are ultimately expressed in thestability and success of the estimators.Effectiveness of Sampling Methods. The complete-

ness of the structured inventory shows the effective-ness of the combined methods for accurately samplingthe species richness and relative abundance of ants inupland ecosystems in Florida. Individual methodswere much less effective than combined methods atcapturing total species richness (Fig. 2). The com-bined quantitative methods (pitfalls, litter samples,baits) also provided an accurate measure of the rela-tive abundance of ground-dwelling species (King2004). Both measures are crucial components for ac-curatelymeasuringandcomparingbiodiversity amongdifferent areas or times (Gotelli and Colwell 2001).Relative to other thorough sampling projects, such aslong-term, strict inventories (e.g., Talbot 1975,Deyrup and Trager 1986), a structured inventory is afaster approach to capturing a large majority of localspecies richness. However, we do not suggest thatstructured inventory is a replacement for long-termtaxonomic inventories, if time allows. These resultsand those of other structured inventories reveal that,regardless of where the sampling occurs, the fastestway to thoroughly sample a given area is to use avariety of sampling methods and take a large numberof samples (and in so doing, capture a large number ofindividuals; Longino and Colwell 1997, Majer 1997,Fisher 1999a, Longino et al. 2002). This is not newinformation, but the increasing number of rapidly gen-erated, thorough inventories from a variety of regionsand habitats serve as an important reminder of a sim-ple truth of invertebrate sampling that bears repeat-ing: a large sampling effort is required to accuratelymeasure the biodiversity of invertebrates. Some ca-veats are that the application of individual methodsmay be adequate in certain habitats, for use in sam-pling projects that do not require thorough invento-ries, or perhaps if an extremely large number of sam-ples are applied.

Different ecosystems and individual methods gen-erated complementary species lists (Tables 2 and 3).Additionally, the spacing of samples at or above 5 mhad little or no effect on the similarity of samples (Fig.3A and B) within ecosystems. These results suggestthat samples spaced 5 m or further apart may beconsidered independent. Similarity among sampleswas more dependent on the ecosystem where sam-pling occurred. These results emphasize the differ-ences among ecosystems and the sampling bias ofindividual sampling methods. Hardwood hammock


and Florida scrub ecosystems have closed canopiesand a deep layer of litter. Pine ßatwoods, high pine,and Þeld ecosystems are increasingly open with adecreasing amount of litter. Accordingly, the ant faunaof hardwood hammocks is, for example, rich in crypticspecies that nest in rotten wood and leaf litter relativeto high pine ecosystems that have much less litter.Pitfalls and litter samples produced species lists thatwere the most similar among sampling methods (Ta-ble 3). The underlying reason for the difference seenbetween the two methods was the number of uniquespecies captured by litter sampling in hardwood ham-mock. In all of the other ecosystems, pitfalls wereequal to or better than litter sampling. In combinationwith studies conducted in tropical forests where littersampling was superior to pitfall sampling (Fisher1999a), these results provide further support for theapplication of the Ants of the Leaf Litter (ALL) pro-tocol in a wide variety of ecosystems for generatingestimates of species richness (Agosti et al. 2000a). Thestandardized ALL protocol emphasizes pitfall and lit-ter sampling in addition to hand collecting to generateaccurate estimates of local ant biodiversity. The pro-tocol is quantitative and ßexible enough to permit itsapplication in a variety of habitats. The deÞciencies ofindividual methods in a given habitat are compensatedfor by the performance of other methods. For exam-ple, in habitats with a well-developed leaf litter, theinadequacy of pitfalls will be overcome by the supe-riority of litter sampling techniques and the ßexibilityof hand collecting (Melbourne 1999).

Despite the obvious advantages of applying multi-ple sampling methods to generate a structured inven-tory, the expertise and time required to execute aninventory often precludes this approach to samplinginvertebrates (Burbidge 1991). Individual samplingmethods are, therefore, often applied to generate acommunity characterization (Longino and Colwell1997). The inadequacies of individual sampling meth-ods suggest that their unaccompanied application re-quires additional consideration. Among the methodsapplied in this study, baits were the least effective atproducing estimates of biodiversity and unique spe-cies (Figs. 4A and 5). Despite the small amount of timerequired to sort and identify species collected in baits,the limited amount of information they produced di-minished their value as a biodiversity sampling tech-nique (Fig. 4B). In contrast, hand collecting was themost efÞcient method for capturing species richnessand was an effective method for producing uniquespecies (Fig. 4A and B). The rarefaction curve sug-gested a much greater potential for capturing speciesrichness if more time were spent hand collecting ineach site (Fig. 2). However, the success of hand col-lecting is dependent primarily on the experience ofthe collector (Longino et al. 2002, Sørensen et al.2002)Ñexperienced collectors will collect more spe-cies more quickly than inexperienced collectors. Handcollecting is also not quantitative, so relative abun-dance cannot be reliably estimated.

Both pitfall and litter extraction sampling weremost closely correlated with total site species richness

(Fig. 5) and provided relative abundance data. Therewere also a number of species unique to each method(Fig. 4A). However, pitfall sampling and leaf litterextraction were costly, requiring the most time to sortand identify specimens and the use of specializedequipment (i.e., Berlese funnels). Pitfall samples re-quired less time (�18%) to sort and identify speci-mens and, with the exception of hardwood hammocksites, were equal or better than litter sampling forupland ecosystems in Florida. Although pitfall samplestend to undersample cryptic species and arboreal spe-cies (Majer 1997) and do not perform as well as litterextraction sampling in ecosystems with a well-devel-oped litter layer (Fisher 1999a, Melbourne 1999), theirgreater efÞciency suggests that pitfalls are the bestchoice for an individual sampling method in a majorityof warm temperate and subtropical upland ecosystems(i.e., those without a deep layer of litter). Pitfall trapscan be made even more effective if they are operatedfor longer periods, and more traps of larger diameterare used (Abensperg-Traun and Steven 1995, Majer1997).

Although it is a central tenet of ecological method-ology (Southwood and Henderson 2000), the impactof methodological bias on experimental and samplingdesign in the study of invertebrate assemblages isoften a neglected point of discussion. Before a designthat uses a single sampling method is initiated, itshould be considered that the effect of treatments maybe either obscured or falsiÞed by sampling bias ifspecies richness and relative abundance are the vari-ables of interest (Melbourne 1999). Sampling biasshould be of particular concern in studies where com-munity-level effects are assessed by sampling assem-blages of invertebrates at local scales. More speciÞ-cally, hypothesis-testing studies designed to explorecomplex ecological processes, such as the factors de-termining assembly rules, require accurate measure-ment of species richness and relative abundance andmust account for sampling bias. Ideally, this account-ing should be done a posteriori and should considerthe results of previous inventories whenever possible.

A structured inventory approach should also beadopted for larger-scale invertebrate studies, de-spite the cost savings of using sampling methods in-dividually. Structured inventories are often contrastedwith single-species time-series population studies,land management programs, or regional and geo-graphic scale biodiversity assessment. In these types ofstudies, costly sampling protocols are considered un-necessary, intractable, and undesirable (Andersen1997, Andersen et al. 2002). Nevertheless, even thesestudies would likely beneÞt from incorporating someof the methods of structured inventory, such as usingmultiple (more than one or two) sampling techniques.The shown effectiveness of using a structured inven-tory approach to sampling ants (Longino and Colwell1997, Fisher 1999a, Agosti et al. 2000a, Longino et al.2002) and spiders (Coddington et al. 1991, Silva andCoddington 1996, Dobyns 1997, Toti et al. 2000, Sø-rensen et al. 2002) also supports its broader use intaxonomic studies. For these studies, accurate assess-


ments of species richness and the time savings relativeto traditional taxonomic studies will be particularlyvaluable. Finally, an increase in the use of structuredinventory to study invertebrates in a variety of habitats(regardless of the type of study) will make a signiÞcantcontribution toward cataloguing, conserving, and un-derstanding biodiversity.

Acknowledgments

We thank L. Davis and M. Deyrup for assisting with andverifying species identiÞcations. We thank the Archbold Bi-ological Station and M. Deyrup for laboratory space andaccommodation during part of this work. We thank the Uni-versity of Florida, the Florida Department of EnvironmentalProtectionÕs State Parks Division, and the U.S. National For-est Service for permission to perform sampling in the Kather-ine Ordway Biological Preserve, San Felasco Hammock StatePark, and Osceola National Forest, respectively. Voucherspecimens from this project have been donated to HarvardÕsMuseum of Comparative Zoology and the Archbold Biolog-ical Station. Comments from J. Capinera, M. Deyrup,J. Longino, R. McSorley, and K. Portier greatly improved anearlier version of this manuscript. J.R.K. was supported by aUniversity of Florida Alumni Fellowship for part of this work.This manuscript was written while J.R.K. was supported byUSDA-NRI Grant 2003-01453 to W. R. Tschinkel.

References Cited

Abensperg-Traun, M., and D. Steven. 1995. The effects ofpitfall trap diameter on ant species richness (Hymenop-tera: Formicidae) and species composition of the catch ina semi-arid eucalypt woodland. Aust. J. Ecol. 20: 282Ð287.

Agosti, D., J. D.Majer, L. E. Alonso, and T. R. Schultz (eds.).2000a. Ants: standard methods for measuring and mon-itoring biodiversity. Smithsonian Institution Press, Wash-ington DC.

Agosti, D., J. D.Majer, L. E. Alonso, and T. R. Schultz (eds.).2000b. Sampling ground-dwelling ants: case studies fromthe worldÕs rain forests. Curtin University School of En-vironmental Biology, Perth, Australia.

Andersen, A. N. 1990. The use of ant communities to eval-uate change in Australian terrestrial ecosystems: a reviewand a recipe. Proc. Ecol. Soc. Aust. 16: 347Ð357.

Andersen, A. N. 1993. Ants as indicators of restoration suc-cess at a uranium mine in tropical Australia. Restor. Ecol.1: 156Ð167.

Andersen, A.N. 1997. Using ants as bioindicators: multiscaleissues in ant community ecology. Conserv. Ecol. 1: 1Ð8.

Andersen, A. N., B. D. Hoffmann, W. J. Muller, and A. D.Griffiths. 2002. Using ants as bioindicators in land man-agement: simplifying assessment of ant community re-sponses. J. App. Ecol. 39: 8Ð17.

Anderson, R. S., and J. S. Ashe. 2000. Leaf litter inhabitingbeetles as surrogates for establishing priorities for con-servation of selected tropical montane cloud forests inHonduras, Central America (Coleoptera; Staphylinidae,Curculionidae). Biodiv. Conserv. 9: 617Ð653.

Bestelmeyer, B. T., D. Agosti, L. E. Alonso, C.R.F. Brandao,W. L. Brown Jr., J.H.C. Delabie, and R. Silvestre. 2000.Field techniques for the study of ground-dwelling ants,pp. 122Ð144. In D. Agosti, J. D. Majer, L. E. Alonso, andT. R. Shultz (eds.), Ants: standard methods for measuringand monitoring biodiversity. Smithsonian InstitutionPress, Washington DC.

Burbidge, A. A. 1991. Cost constraints on surveys for natureconservation, pp. 3Ð6. InC. R. Margules and M. P. Austin(eds.), Nature conservation: cost effective biological sur-veys and data analysis. CSIRO, Canberra, Australia.

Burnham, K. P., andW. S. Overton. 1978. Estimation of thesize of a closed population when capture probabilitiesvary among animals. Biometrika. 65: 623Ð633.

Burnham,K.P., andW.S.Overton. 1979. Robust estimationof population size when capture probabilities vary amonganimals. Ecology. 60: 927Ð936.

Chazdon, R. L., R. K. Colwell, J. S. Denslow, and M. R.Guariguanta. 1998. Statistical methods for estimatingspecies richness of woody regeneration in primary andsecondary rain forests of northeastern Costa Rica, pp.285Ð309. InF. Dallmeier and J. A. Comiskey (eds.), Forestbiodiversity research, monitoring, and modeling. Parthe-non Publishing Group, Paris, France.

Clench,H.K. 1979. How to make regional lists of butterßies:some thoughts. J. Lepidop. Soc. 33: 216Ð231.

Cochran, W. G. 1963. Sampling techniques. 2nd ed. Wiley,New York.

Coddington, J. A., C. E.Griswold,D. Silva, E. Penaranda, andS. F. Larcher. 1991. Designing and testing sampling pro-tocols to estimate biodiversity in tropical ecosystems,pp. 44Ð60. In E. C. Dudley (ed.), The unity of evolution-ary biology: proceedings of the fourth international con-gress of systematic and evolutionary biology. DioscoridesPress, Portland, OR.

Colwell, R. K. 2000. EstimateS: statistical estimation of spe-cies richness and shared species from samples. Version6.0. http://viceroy.eeb.uconn.edu/estimates.

Colwell, R. K., and J. A. Coddington. 1994. Estimating ter-restrial biodiversity through extrapolation. Phil. Trans. R.Soc. Lond. Ser. B. 345: 101Ð118.

Delabie, J.H.C., B. L. Fisher, J. D. Majer, and I. W. Wright.2000. Sampling effort and choice of methods, pp. 145Ð154. In D. Agosti, J. D. Majer, L. E. Alonso, andT. R. Schultz (eds.), Ants: standard methods for measur-ing and monitoring biodiversity. Smithsonian InstitutionPress, Washington DC.

Deyrup, M. 2003. An updated list of Florida ants (Hyme-noptera: Formicidae). Fla. Entomol. 86: 43Ð48.

Deyrup, M., and J. Trager. 1986. Ants of the ArchboldBiological Station, Highlands County, Florida (Hyme-noptera: Formicidae). Fla. Entomol. 69: 206Ð228.

Deyrup, M., L. Davis, and S. Cover. 2000. Exotic ants inFlorida. Trans. Am. Entomol. Soc. 126: 293Ð326.

Disney, R.H.L. 1986. Assessments using invertebrates: pos-ing the problem, pp. 271Ð293. InM. B. Usher (ed.), Wild-life conservation evaluation. Chapman & Hall, London,UK.

Dobyns, J. R. 1997. Effects of sampling intensity on thecollection of spider (Araneae) species and the estimationof spider richness. Environ. Entomol. 26: 150Ð162.

Eldredge, N. (ed.). 1992. Systematics, ecology, and thebiodiversity crisis. Columbia University Press, New York.

Fisher, B. L. 1996. Ant diversity patterns along an eleva-tional gradient in the Reserve Naturelle IntegraledÕAndringitra, Madagascar. Fieldiana Zool. 85: 93Ð108.

Fisher, B. L. 1998. Ant diversity patterns along an eleva-tional gradient in the Reserve Speciale dÕAnjanaharibe-Sud and on the western Masoala Peninsula, Madagascar.Fieldiana Zool. 90: 39Ð67.

Fisher, B. L. 1999a. Improving inventory efÞciency: a casestudy of leaf-litter ant diversity in Madagascar. Ecol. Appl.9: 714Ð731.


Fisher, B. L. 1999b. Ant diversity patterns along an eleva-tional gradient in the Reserve Naturelle IntegraledÕAndohahela, Madagascar. Fieldiana Zool. 94: 129Ð147.

Fisher, B. L. 2002. Ant diversity patterns along an eleva-tional gradient in the Reserve Speciale de Manongarivo,Madagascar. Boissiera. 59: 311Ð328.

Fisher, B. L., and H. G. Robertson. 2002. Comparison andorigin of forest and grassland ant assemblages in the highplateau of Madagascar (Hymenoptera: Formicidae).Biotropica. 34: 155Ð167.

Folgarait, P. J. 1998. Ant biodiversity and its relationship toecosystem functioning: a review. Biodiv. Conserv. 7:1221Ð1244.

Gotelli, N. J., and R. K. Colwell. 2001. Quantifying biodi-versity: procedures and pitfalls in the measurement andcomparison of species richness. Ecol. Lett. 4: 379Ð391.

Holldobler, B., and E. O. Wilson. 1990. The ants. BelknapPress, Cambridge, MA.

Hughes, R. G. 1986. Theories and models of species abun-dance. Am. Nat. 128: 879Ð899.

King, J. R. 2004. Ant communities of FloridaÕs upland eco-systems: ecology and sampling. PhD dissertation, Univer-sity of Florida, Gainesville, FL.

Kim, K. C. 1993. Biodiversity, conservation and inventory:why insects matter. Biodiv. Conserv. 2: 191Ð214.

Kremen, K. 1992. Assessing the indicator properties of spe-cies assemblages for natural areas monitoring. Ecol. Appl.2: 203Ð217.

Kremen, K. 1994. Biological inventory using target taxa: acase study of the butterßies of Madagascar. Ecol. Appl. 4:407Ð422.

Kremen, C., R. K. Colwell, T. L. Erwin, D. D. Murphy,R. F. Noss, and M. A. Sanjayan. 1993. Terrestrial arthro-pod assemblages: their use in conservation planning. Con-serv. Biol. 7: 796Ð808.

Lambshead, J., and H. M. Platt. 1985. Structural patterns ofmarine benthic assemblages and their relationships withempirical statistical models, pp. 371Ð380. In P. E. Gibbs(ed.), Proceedings of the 19th European Marine BiologySymposium, Plymouth UK, 1984. Cambridge UniversityPress, Cambridge, UK.

Lawton, J. H., D. E. Bignell, B. Bolton, G. F. Bloemers,P. Eggleton, P. M. Hammond, M. Hodda, R. D. Holt,T. B. Larsen, N. A.Mawdsley, N. E. Stork,D. S. Srivastava,andA.D.Watt. 1998. Biodiversity inventories, indicatortaxa and effects of habitat modiÞcation in tropical forest.Nature (Lond.). 391: 72Ð76.

Lee, S.M., andA.Chao. 1994. Estimating population size viasample coverage for closed capture-recapture models.Biometrics. 50: 88Ð97.

Longino, J. T. 2000. What to do with the data, pp. 186Ð203.In D. Agosti, J. D. Majer, L. E. Alonso, and T. R. Schultz(eds.), Ants: standard methods for measuring and mon-itoring biodiversity. Smithsonian Institution Press, Wash-ington, DC.

Longino, J. T., and R. K. Colwell. 1997. Biodiversity assess-ment using structured inventory: capturing the ant faunaof a tropical rain forest. Ecol. Appl. 7: 1263Ð1277.

Longino, J. T., J. Coddington, and R. K. Colwell. 2002. Theant fauna of a tropical rainforest: estimating species rich-ness three different ways. Ecology. 83: 689Ð702.

Lubertazzi D., and W. R. Tschinkel. 2003. Ant communitychange across a ground vegetation gradient in north Flor-idaÕs longleaf pine ßatwoods. J Insect Sci. 3: 21.

Ludwig, J. A., and J. F. Reynolds. 1988. Statistical ecology: aprimer on methods and computing. Wiley, New York.

Magurran, A. E. 1988. Ecological diversity and its measure-ment. Princeton University Press, Princeton, NJ.

Majer, J. D. 1997. The use of pitfall traps for samplingantsÑa critique. Mem. Mus. Victoria. 56: 323Ð329.

Melbourne,B.A. 1999. Bias in the effect of habitat structureon pitfall traps: an experimental evaluation. Aust. J. Ecol.24: 228Ð239.

Myers, R. L., and J. J. Ewel (eds.). 1990. Ecosystems ofFlorida. University of Central Florida Press, Orlando, FL.

New, T. R. 1998. Invertebrate surveys for conservation.Oxford University Press, New York.

Oliver, I., and A. J. Beattie. 1996. Designing a cost-effectiveinvertebrate survey: a test of methods for rapid assess-ment of biodiversity. Ecol. Appl. 6: 594Ð607.

Platnick, N. I. 1992. Patterns of biodiversity, pp. 15Ð24. InN. Eldredge (ed.), Systematics, ecology and the biodi-versity crisis. Columbia University Press, New York.

Preston, F.W. 1948. The commonness and rarity of species.Ecology. 29: 254Ð283.

Price, P. W., and G. P.Waldbauer. 1994. Ecological aspectsof pest management, pp. 35Ð72. In R. L. Metcalf andW. H. Luckmann (eds.), Introduction to insect pest man-agement. Wiley, New York.

Prusack, Z. A. 1997. Ant fauna in three plant communitieswithin Wekiwa Springs State Park, Florida: assessmentof diversity and comparison of collecting methods. MSthesis, University of Central Florida, Orlando, FL.

Raaijmakers, J.G.W. 1987. Statistical analysis of theMichaelis-Menten equation. Biometrics. 43: 793Ð803.

Raven, P. H., and E. O. Wilson. 1992. A Þfty-year plan forbiodiversity surveys. Science. 258: 1099Ð1100.

Siemann, E., D. Tilman, and J. Haarstad. 1999. Abundance,diversity and body size: patterns from a grassland arthro-pod community. J. Anim. Ecol. 68: 824Ð835.

Silva, D., and J. A. Coddington. 1996. Spiders of Pakitza(Madre de Dios) Peru: species richness and notes oncommunity structure, pp. 241Ð299. In D. E. Wilson andA. Sandoval (eds.), The biodiversity of Pakitza and itsenvirons. Smithsonian Institution, Washington, DC.

Soberon, J. M., and J. Llorente. 1993. The use of speciesaccumulation functions for the prediction of species rich-ness. Conserv. Biol. 7: 480Ð488.

Sørensen, L. L., J. A. Coddington, and N. Scharff. 2002. In-ventorying and estimating subcanopy spider diversityusing semiquantitative sampling methods in an Afro-montane forest. Environ. Entomol. 31: 319Ð330.

Southwood, T.R.E., and P. A. Henderson. 2000. Ecologicalmethods. Blackwell, Malden, MA.

Talbot, M. 1975. A list of the ants (Hymenoptera: Formici-dae) of the Edwin S. George Reserve, Livingston County,Michigan. Great Lakes Entomol. 8: 245Ð246.

Toti, D. S., F. A. Coyle, and J. A. Miller. 2000. A structuredinventory of Appalachian grass bald and heath bald spiderassemblages and a test of species richness estimator per-formance. J. Arach. 28: 329Ð345.

Van Pelt, A. F. 1947. Ants of the Gainesville region: withspecial reference to ecology and taxonomy. PhD disser-tation, University of Florida, Gainesville, FL.

Van Pelt, A. F. 1956. The ecology of the ants of the WelakaReserve, Florida. Am. Midl. Nat. 56: 358Ð387.

Van Pelt, A. F. 1958. The ecology of the ants of the WelakaReserve, Florida (Hymenoptera: Formicidae). Part II.Annotated list. Am. Midl. Nat. 59: 1Ð57.

Wilson, E. O. (ed.). 1988. Biodiversity. National AcademyPress, Washington, DC.

Received for publication 22 March 2005; accepted 9 Septem-ber 2005.


sampling evaluation of sampling methods and species ... · evaluation of sampling methods and...

Documents