Assessing spider community structure in a beech forest: Effects of sampling method

The spider community of a beech forest on limestone was studied for one year using four sampling techniques: emergence traps, pitfall traps, soil samples, and arboreal eclectors. 87 spider species were recorded. Emergence traps and arboreal eclectors were particularly efficient in detecting spider species. Dominance identity (percentage similarity) was highest for catches from emer­ gence traps and pitfall traps. Species recorded were assigned to various ecological groups. In terms of proportional abundance, repre­ sentation of the ecological groups varied and appeared related to the sampling method used. Stratum type and type of prey capture strategy accounted for >60% of the variance in the catch results (canonical correspondence analysis). Proportional abundance of funnel-web spiders was much higher in pitfall trap catches (31.7%) than in any other method (1.0-11.6%).


INTRODUCTION
Various methods are available for sampling soil ani mals in terrestrial habitats (Dunger & Fiedler, 1989).Many of these techniques, such as pitfall traps, heat extraction of soil samples and emergence traps also col lect spiders, with pitfall traps apparently being the method most widely employed in studies of spider communities.
Some previous studies of beech forests on acid soils (Luzulo-Fagetuwi) have used more than one sampling method to describe spider communities (e.g., Albert, 1976Albert, , 1979;;Dumpert & Platen, 1985;Platen, 1985), and some of these authors have commented on differences in the performance of different methods.The present study was conducted in a beech forest on limestone.First, we evaluate the performance of four methods, i.e., emergence traps, pitfall traps, Kempson extraction of soil samples, and arboreal eclectors (see Funke, 1971), which were concurrently employed during one study year.Our approach in analysing the data, however, differs from ear lier comparisons in that we employ rarefaction methods to achieve better comparability of catch results.
Following Tretzel (1952), many arachnologists have attempted to define ecological groups of spiders with respect to moisture and light preferences, periods of maturity, vegetation strata occupied, prey capture strate gies, and other characteristics.Different sampling tech niques are likely to collect such ecological groups of spi ders in different proportions.The second part of this paper will consider the extent to which different sampling methods differ in their estimation of the proportional importance of these ecological groups.

Study site
The study site, a beech forest on limestone (age: 120 years; 420 m asl.), is located some 10 km east of Gottingen (Lower Saxony; Germany) (see Schaefer & Schauermann, 1990).The plant association has been classified as a Melico-Fagetum (see Dierschke & Song, 1982;Dierschke, 1989).Two sub-areas were studied, the first of which has a herb layer dominated by Allium ursinum L. in spring, while the second one has Mercurialis perennis L. as the dominant herb species.The study was conducted in a year characterized by below-average mean monthly tem peratures and high monthly precipitation rates (Stippich, 1986).

Sampling methods
Two identical sampling programmes were performed in each of the sub-areas from 10 March 1981 until 8 December 1981.The sampling programmes included: Emergence traps [abbreviated: E]: Twelve emergence traps (ground area = 1 m2; Funke, 1971) were set up in each sub-area.They were moved to new places every four weeks in spring and then every eight weeks later in the year.
Pitfall traps [P]: Six pitfall traps (0 = 5 cm; preservative: aqueous picrinic acid) arranged in a row at 10 m intervals were operated in each sub-area.Catches were collected at weekly intervals.

General results
Our analyses are based on a total catch of 4,257 adult spiders.Overall, 87 species were identified, with 69 and 72 species recorded from the Allium and the Mercurialis sub-areas, respectively.The linyphiid sub-families Linyphiinae and Erigoninae were particularly species-rich and represented by 28 and 26 species, respectively, while spe cies numbers of the remaining families were much lower and varied between 1 and 7 species per family.
The species recorded are listed in Appendix 1.A spe cies list containing the full quantitative data and assigna tions of species to ecological groups is available from the authors on request.

Similarity of the catches
Only eight out of 87 species were detected by all sam pling methods.The proportion of species detected exclu sively by a single method was high in the arboreal eclectors (43.8%), lower in emergence traps (20.3%) and pit fall traps (14.3%), and lower still in the soil samples (7.4%) (Fig. 1).
In order to illustrate similarity patterns, cluster analyses were performed.Catch results from the two sub-areas studied were treated separately.First, species identity was calculated using the Sorensen index and an UPGMAdendrogram generated (Fig. 2a).Catches from the two sub-areas paired according to sampling technique.Spider assemblages as determined by arboreal eclectors (A) dif fered considerably in species composition from those obtained from the three other methods.Fig. 1 suggests that A shares only few species with P and K, while there is considerable overlap with E.
If the proportional abundances of species are taken into account and percentage similarity used as a similarity measure (Fig. 2b), the composition of the catches differed again considerably between the arboreal eclectors and the other three methods, but catches from emergence and pit fall traps were more similar to each other than to Kempson samples.Since assemblages from the two sub areas paired again according to trapping method, i.e., catches from the two sub-areas were very similar, all sub sequent analyses were done using the pooled data.

Performance of sampling methods
Since the sampling methods differed with respect to both sampling effort (i.e., number of replicates) and sample size (i.e., numbers of individuals caught), we used the appropriate rarefaction methods to achieve compara bility.
Sampling effort: In this analysis, one replicate equals the annual catch of one pitfall trap, one emergence trap, and the arboreal eclector of one beech stem, respectively.Catches from soil samples taken on different dates throughout the season were combined according to the original numbering of the samples; this gives a total of 18 replicates.Fig. 3a shows "Shinozaki curves" (see Achtziger et al., 1992), which represent the number of species expected to  be detected by a given number of replicates of the respec tive sampling method.Obviously, emergence traps will yield more species than the same number of both pitfall traps and soil samples.On the other hand, species num bers expected to be returned for a given sampling effort were high for arboreal eclectors compared with emer gence traps.It is impossible to tell, however, how quickly the curve for arboreal eclectors will level off.Given the mean numbers of species per replicate (Table 1), one can conclude that species turnover between replicates is high in A and E but low in P and K.
Sample size: Mean sample sizes per replicate differed between methods (Table 1).For E, P, K and A, rarefac tion (see Hurlbert, 1971;Magurran, 1988) was used to calculate species numbers expected in samples of stan dardized size (Fig. 3b).Again, given identical sample sizes, expected species numbers for both emergence traps and arboreal eclectors were much higher than the respec tive values for pitfall traps and soil samples.

Representation of ecological groups
If sampling techniques differ in species composition (see Fig. 2), it does not follow that catches also differ in the representation of ecological groups.
Species numbers: Litter-dwelling species (Str1) formed the most species-rich stratum type in the catches of each sampling method but their relative importance varied: it was particularly high (74.1%) in soil samples and low (29.2%) in arboreal eclectors, the assemblage of which was characterized by stratum types associated with higher vegetation strata.In the arboreal eclectors, stenochronous summer-breeders represented 35.4% of the species, while the respective values varied between 11.1 and 15.3% for the three other methods.Apart from these examples, rela tive importance of ecological types in terms of species numbers was fairly similar for all sampling methods.
Proportional abundance: Differences between sampling methods become more pronounced when we look at the proportional abundance of ecological groups.In order to summarize our results, we applied canonical correspon dence analysis (CCA) using ecological groups as explana tory (nominal) variables.First, we did separate CCAs for each of the six ecological categories (i.e., moisture types, size classes, etc.) to identify the ones with the highest explanatory potential.We found that prey capture strategy and stratum type performed best.So we con ducted a CCA with the ecological types represented by these two categories and also included three maturity types.Species represented by six or fewer individuals were removed from the analysis.The sum of all uncon strained Eigenvalues was 0.749, of which 63.2% were explained by the first two axes.
The resulting ordination graph (Fig. 4) shows a trian gular arrangement of sampling methods: catches from arboreal eclectors (A) and soil samples (K) are more or less isolated, while those from emergence (E) and pitfall traps (P) are fairly similar.Some species are strongly associated with individual sampling methods but there are also groups of species which are intermediate between P/E and K, and P/E/K and A: Axis 1 separates A from P, E and K. Abundance of spider species constructing webs on bark (WB), inhabitants of higher vegetation (Str4), and sit-and-wait predators (SW) were positively corre lated with this axis, while the correlation was negative for litter-dwelling species (Str1).Axis 2 separates sampling methods (P and E) with high abundance of funnel-web spiders (WF) from those where such species are under represented (A and K).
Catches from arboreal eclectors differ from the overall pattern in that stenochronous autumn-breeders (stau) are very abundant.This is largely due to the dominant spe cies D. socialis, and, hence, collinear with variable WB (and also variable Str5, which is not shown in Fig. 4).Soil sample assemblages are characterized by high num- bers of stenochronous spring-breeders (stsp), while steno chronous summer-breeding species (stsu) are virtually lacking.

DISCUSSION
Our 1-year sampling programme involving four sam pling methods yielded 87 species for the Gottingen beech forest.As suggested by the shapes of rarefaction curves (Figs 3 a, b), the sampling programme did not detect all the spider species occurring in the Gottingen beech forest, and 31 additional species (see Appendix 1) have been recorded during later studies (mostly from the Allium sub area) giving a total of 118 species (Stippich 1981(Stippich , 1986).Further species have been added to the list recently by the pitfall trap studies of Sührig (1997) and Rothlânder (1998), which were, however, conducted in a larger area covering differently aged beech stands.Some earlier studies using various combinations of sampling methods were conducted in beech forests on acid soils and reported roughly similar species numbers: S = 83 (Platen, 1985), S = 95 (Dumpert & Platen, 1985), and S = 109 (Albert, 1979).
Sampling methods differed in species numbers detected: Emergence traps and arboreal eclectors proved to be more efficient than pitfall traps and soil samples.This was not only true considering absolute species num bers (Fig. 1) but also when expected species numbers had been calculated for standardized sample sizes using rare faction (Figs 3a, b).These patterns agree with the find ings from two earlier studies: Dumpert & Platen (1985) recorded higher species numbers from emergence traps than pitfall traps and soil samples in spite of much larger sample sizes obtained by the latter two techniques, and in Albert's (1979) study, species numbers were much higher for arboreal eclectors than for pitfall traps and soil sam ples (see Table 2).While in the present study more spe cies were recorded from emergence traps than from arboreal eclectors, Platen's (1985) 2-year study resulted in the opposite pattern with 69 species trapped by a single arboreal eclector and 43 species recorded from six emer gence traps.So, the relative efficiencies of arboreal eclectors and emergence traps remain uncertain.Based on our results, we would think that arboreal eclectors offering open access to spiders are more likely to collect high spe-Table 2. Species numbers (S) and sample sizes (N) recorded during studies on spider communities of beech forests using four dif ferent sampling methods.See Fig. 1 for abbreviations of sampling methods; a -duration of the study (years).Sampling effort and sample size may not be the only factors influencing the number of species recorded.In methods strongly dependent on the activity of adult spi ders (e.g., pitfall traps and arboreal eclectors), an extra increase may be achieved by extending the study period to more than one year.Thereby, the chances of the maturity periods of individual species coinciding with favourable weather conditions will increase and so will the probability of detecting additional species.
Representation of ecological groups can obviously be related to specific properties of the sampling method.For example, arboreal eclectors are efficient in sampling spe cies associated with higher strata, and not surprisingly, such stratum types are over-represented.Since Drapetisca socialis is the dominant species in arboreal eclectors, pro portional abundances of ecological types is strongly biased towards the ecological characteristics of this spe cies.Assemblages of soil samples are characterized by high proportional abundance of litter-dwelling species, whereas funnel web spiders such as Coelotes terrestris and C. inermis are over-represented in pitfall trap catches.This is not a special feature of our study but a general pat tern (Table 2; bottom).
Our study was characterised by an exceptionally high sampling effort in terms of replicates (e.g., 24 emergence traps).This enabled us to generate fairly reliable rarefac tion curves, and we feel it might be helpful to summarize our results in the form of a few recommendations, which should also be applicable to studies with lower input of labour.

CONCLUSIONS
Which sampling methods should be applied depends on the type of study.(i) If the major goal is to record as many species as possible from a habitat, the best choice is probably a combination of emergence traps and arboreal eclectors, since both these methods collect most species detectable by pitfall traps and soil sample extraction and high numbers of additional species, (ii) If the major goal is to analyse community patterns, quantitative sampling is required, Soil samples yield highly quantitative results but are restricted to a single habitat stratum, whereas catches from pitfall traps and arboreal eclectors depend com pletely on spider activity and cannot be related to a defined sampling area, While emergence trap catches, too, will depend on the inclination of spiders to move, they sample a defined area of ground and provide at least a semi-quantitative abundance estimate of spiders ranging from litter-dwellers to inhabitants of the herb layer, Whenever catches from emergence traps are available, we would, therefore, recommend to use them in studies of spider communities.(iii) While we only considered results from beech forests, we feel our conclusions should apply to many other woodland habitats.

Fig. 1 .
Fig. 1.Species numbers of spiders recorded using four dif ferent sampling methods in a beech forest.Sampling methods: E -emergence traps; P -pitfall traps; K -soil samples; A -arboreal eclectors.Proportions of species recorded by one, two, three or four methods (= occurrence classes) are indicated by different hatchings, and sub-divisions of occurrence classes indicate the pattern of overlap between sampling methods.

Fig. 2 .
Fig. 2. Species identity (a) and dominance identity (b) of spider assemblages as recorded by four sampling methods in two sub-areas of a beech forest.Sub-areas are indicated by AL (= Allium sub-area) and ME (= Mercurialis sub-area); attached characters indicate sampling method (see Fig. 1).

Fig. 3 .
Fig. 3. Effect of sampling effort (a) and sample size (b) on expected species numbers obtained by different sampling meth ods.(a) Curves were generated by Shinozaki rarefaction; (b) curves were generated by Hurlbert rarefaction; solid symbolsactual values; light symbols -rarefied values.

Table 1 .
Sampling statistics for the four methods used for catching spiders.S -number of species; N -number of indi viduals; SE -standard error; n -number of replicates; see Fig.1for abbreviations of sampling methods.