ESTIMATING SPECIES RICHNESS OF ARTHROPODS IN AZOREAN PASTURES: THE ADEQUACY OF SUCTION SAMPLING AND PITFALL TRAPPING

The advantages of the complementary use of suction sampling (Vortis) and pitfall trapping in estimating the species richness of arthropods in sown and semi-natural pastures were investigated on three recent isolated oceanic islands in the Azores. In addition, the ability of pitfall trapping and suction sampling to describe species composition and richness were tested in terms of taxonomic and ecological groups. Both suction sampling and pitfall trapping techniques were reliable in terms of sampling efficiency, getting a good estimate of the arthropod species richness as judged by percentage sampling completeness, the ratio of observed to estimated species richness. Using the two methods together values of almost 90% completeness were obtained in some guilds. Pitfall and suction alone were reliable in estimating species richness of predatory arthropods (all species) and spiders. Joint data is needed for getting reliable estimates for herbivores guilds. In terms of presence-absence data, sap-feeding (suckers) and chewing herbivores showed very different patterns, with a higher proportion of the former sampled exclusively by suction sampling, whereas more chewing species were sampled only by pitfall traps. Similar patterns were obtained for two guilds of spiders, where more web-building species were sampled only by suction sampling, whereas most other species were sampled only by pitfall trapping. We conclude that pitfall trapping and suction sampling should be considered as complementary methods, not as alternatives. We therefore strongly recommend that they be used simultaneously in grassland and agroecosystem community studies, in order to derive reliable estimates of arthropod species composition and richness.


Introduction
The comprehensive and critical sampling of arthropods is a key requisite in many areas of ecological science.Despite this, there is still little consensus as to the relative merits of different methods.This is mainly because few quantitative comparisons have actually been undertaken (but see papers in Toft & Riedel, 1995;and particularly Standen, 2000).Furthermore, most are either site specific or focus on particular taxa.In this paper, we aim to compare quantitatively two of the most widely employed methods, in respect of a wide range of insect taxa and spiders.It is also compared the efficacy of methods in different pasture types, known to have different arthropod guild structure and spider species richness (Borges, 1999;Borges & Brown, 1999).
Undoubtedly, two of the most commonly used sampling methods are pitfall trapping and suction sampling.Pitfall traps are frequently used in ecological studies, mainly because they are inexpensive, easy to set in the field and can be used continuously over several years.However, their use carries some methodological problems (reviewed in Adis, 1979).In addition, the distribution of specimens in traps may not reflect relative frequencies of species in the field (see Jarosík, 1992) and the method does not provide absolute measures of species density (Southwood, 1978;Topping & Sunderland, 1992;Dinter, 1995;Sunderland & Topping, 1995;Powell et al., 1996;Southwood & Henderson, 2000), as the number of individuals caught depends partly on their activity as well as habitat characteristics.However, Baars (1979) considers that, by continuous trapping, a reliable relative measure of the size of ground-beetle populations can be obtained.Moreover, Downie et al. (1996) concluded that, providing the habitats under study are not too different in terms of vegetation structure, the relative abundance of a single species in different places, during a comparable time period, could be assessed.Habitat structure may, however, have serious implications for the interpretation of relative abundance data (Halsall & Wratten, 1988;Melbourne et al., 1997;Melbourne, 1999) and there are studies that suggest that pitfall trapping is not a suitable method for comparing different species within a habitat (Topping & Sunderland, 1992;Dinter, 1995;Hawthorne, 1995).The use of different killing-preserving agents may also cause some distortion of results (Luff, 1968;Southwood, 1978;Adis, 1979).To achieve a more accurate inventory of a community, various killing-preserving agents can be used simultaneously (Adis, 1979;Borges, 1992).
The smaller suction sampling machines have been developed to replace the heavy, not easily portable, D-vac in ecological and entomological work.Examples include: 1) the ECHO (Summers et al., 1984), a backpack unit similar to the Univac, in which the arthropods are collected in a cloth net bag; 2) the Ryobi RSV3100E sweeper-vac, described in Macleod et al. (1994Macleod et al. ( , 1995)), and the "Blow & Vac" (Stewart & Wright, 1995) are similar light weight suction sampling devices, consisting mainly of modifications of petrol driven machines normally used to collect leaf litter; 3) the "Vortis" suction machine is a light, portable suction sampling system, developed by Arnold (1994), with many advantages over other suction sampling machines.Here, arthropods are collected directly into containers (several sizes and shapes are available) and not in bags or filters.This enables the operator to take a large number of samples in a short period of time, and prevents the loss of specimens while transferring from the machine to containers.The arthropods collected are also sampled without damage and may be returned to their habitat if required.The apparatus is light (7.8 kg) and fully portable (overall height 930 mm) and can be easily operated by one person.The collected material does not impede air flow and the nozzle wind speed is constant (Southwood & Henderson, 2000).
The growing need for land managers and conservationists to characterise sites, highlights the need for practical ways of sampling arthropod fauna (Buffington & Redak, 1998).Less labourintensive, though accurate, sampling techniques are required.The present study investigates the level of sampling completeness and the potential for the complementary use of suction sampling (Vortis) and pitfall trapping to estimate the species richness and composition of herbivorous (sap-feeding and chewing species) and predacious arthropods, particularly spiders, in sown and semi-natural pastures from three oceanic islands, in the Azores.In addition, we assess the predictive value of suction sampling and pitfall trapping used on their own as surrogates for describing the species richness of assemblages in twelve sites.

STUDY AREA AND EXPERIMENTAL DESIGN
This study was undertaken in the Azores, an oceanic volcanic archipelago located in the North Atlantic, roughly between coordinates 37°to 40°N latitude and 25°to 31°W longitude.Two replicates of recently sown (3 years old) and old semi-natural pasture (more than 30 years old) were sampled on three Azorean islands (S.Maria, Terceira and Pico).The replicates were standardised in terms of management history and were located distanced such that they could be considered as separate sites for statistical purposes (see Borges, 1997Borges, , 1999;;Borges & Brown, 1999, 2001 for a detailed description of sites).
In all 12 sites (3 islands x 2 pasture types x 2 replicates), an area of at least 900 m 2 was fenced for livestock early in 1994, with additional fencing to preclude differential rabbit grazing.At each site, 20 3 x 3 (9 m 2 ) plots were marked with coloured short wooden posts.The plots were regularly spaced, with a corridor of 5m between each.All sites were grazed regularly by dairy and beef cattle, thereby maintaining their traditional management.Cattle grazing events were carefully managed in liaison with the farmers.Sampling took place at least three weeks after a grazing event, thereby allowing the sward structure to develop.The study was conducted over two consecutive years and four complete samples were obtained (spring, summer and autumn 1994 and summer 1995 to provide a between year comparison).

SAMPLING METHODS
Pitfall traps.Pitfall trapping comprised periods of seven consecutive days of sampling.A set of twenty pitfall traps (a plastic cup with a trap diameter of 110 mm and a depth of 70 mm) was used, and positioned according to a grid configuration, the minimum distance between traps being 5.30 m.Each trap was dug into the ground flush with the substrate in the corner of each plot (30 cm inside), enabling one of the four traps to be used for each sampling occasion.For the present work, it was desirable to have an inexpensive preservative, but with neutral qualities.Ethylene glycol (anti-freeze) (c) Sociedad de Amigos del Museo Nacional de Ciencias Naturales y Consejo Superior de Investigaciones Científicas Licencia Creative Commons 3.0 España (by-nc) http://graellsia.revistas.csic.eswas therefore chosen from potential candidates (e.g.formaldehyde, alcohol, acetic acid) and has been widely used in ecological work.Due precautions in handling this liquid in the field and laboratory, gloves and a nose mask were taken (Hall, 1991).Each trap was half-filled with the preservative with a little detergent (Teepol) added to lower the surface tension of the solution.Each trap was protected against the rain with a white plastic dish cover, fixed to the ground by three pieces of wire, with a consistent space of 2.5-3.0 cm between the ground and pitfall cover.After collection, the samples were stored in tubes with 70% ethanol with some drops of glycerol prior to sorting.
Suction.Two new and identical "Vortis" suction sampling machines (Burkhard Scientific -Sales -Ltd., Rickmansworth, Hertfordshire, England) (see Arnold, 1994;Southwood & Henderson, 2000) were used to facilitate between-island sampling.The collection nozzle samples an area of 0.2 m 2 .On each occasion sampling was carried out between 11.00 and 18.00, and only when the vegetation was completely dry and the wind conditions moderate.At each site, four subsamples were taken in each of the 20 plots, one in each corner (total area: 0.8 m 2 ).The collection nozzle was held in position for 30 seconds on each occasion.To simplify the sorting process, the four subsamples from a single plot were taken successively without changing the collection vial.Samples were frozen to kill the arthropods and stored as for pitfall trap samples.
Since the pastures were grazed regularly, and sampling took place consistently three weeks after a grazing event, the height of the vegetation did not impair the efficiency of the suction sampler.It was also possible for the suction nozzle to be positioned on the ground, which is optimal.
Direct searching.In order to assess the efficacy of each sampling method and the two combined, direct field observation in Spring and Summer 1995 were made, with equal additional sampling effort applied to each plot.This enabled the assessment of Orthoptera and adult Lepidoptera, groups that are not readily collected by either of the main methods used in the study.

DATA ANALYSIS SAMPLING EFFICIENCY, RICHNESS ESTIMATORS AND ACCUMULATION CURVES
It is very difficult to judge the completeness of the sampling without information on abundance of species.For the herbivores (all species), chewers, suckers, predators (all species) and also Araneae we created three species abundance matrices, one for each method and also the two methods together (5 guilds x 3 methods = 15 matrices) using the 48 available samples (4 sampling periods x 12 pastures), and compared observed species richness with predicted (true) species richness using Bootstrap, Michaelis Menten means (MMMean), Chao 1 and first order Jackknife (Jackknife 1) statistical estimators calculated with the software program EstimateS version 6.0b1 (Colwell, 2000).These estimators were selected in a range of several possible because represented a lower (Bootstrap; MMMean) and upper (Jackknife 1, Chao 1) bound of true species richness and were useful in determining the extent to which we underestimated actual species richness for the three methods (pitfall, suction, both methods).Curves obtained when estimating the true species richness with the four estimators were the average of 100 randomizations and were tested for heterogeneity using the Coleman test, that compares the mean randomized species accumulation curve with the curve expected if all the individuals caught over all the samples were randomly assigned to the samples (   Table 1.-Summary table of herbivores (all species), chewers, suckers, predators (all species) and spider species richness estimates for total, pitfall and suction sampling data.In bold the results of three indices of sampling efficiency (see text for further explanations).Table 1.-Número de muestras, individuos y especies observadas y estimadas según los distintos estimadores utilizados, desglosando los valores para cada tipo de técnica de colecta (técnica de succión y trampas de caída) y el tipo de alimentación de las especies: todas las especies de herbívoros, especies masticadoras, chupadoras, predadoras (todas las especies) y arañas.En negrita se indican los resultados de los tres índices utilizados para comprobar la eficiencia de muestreo (ver texto para una explicación más detallada).
Coddington , 1995).Based on that index, it was confirmed for all 15 matrices that the species accumulation curve was derived from a homogeneous (stable) community, that is, the expected curve did not rise more steeply from the origin than the accumulation curve, implying that heterogeneity was not greater than could be explained by random sampling error alone (Colwell & Coddington, 1995).Therefore, the use of a parametric estimator (MMMean) is adequate and it is possible to apply the randomization of samples.
To evaluate the performance of pitfall, suction and both methods together we computed the following indices: i) inventory completeness index that is computed as 1 -percentage of singletons (species represented by only one specimen).This index should be near 1 in completely sampled communities (Toti et al., 2000); ii) sampling intensity was calculated as the ratio of specimens to species (see Sorensen et al., 2002); iii) percentage sampling completeness as the ratio of observed to estimated species richness using the four estimators and the average.

COMMUNITY COMPOSITION
The presence/absence pooled data matrix covering three seasons, in two different years and three sampling methods is used to provide the most accurate approximation to the total arthropod community, from which we can compare the efficacy of individual methods.For contingency table analysis, the G-test was chosen since it is the most reliable means of analysing frequency data (Crawley, 1993).To compare the two sampling methods we used Sorensen's index of similarity, C s , a simple but reliable measure of the extent by which two sites share species in common (Southwood & Henderson, 2000): C s = 2J/(2J+A+B), where J is the number of species common to the two sampling methods and A and B are the number of species exclusively sampled by suction and pitfall respectively.A Dissimilarity index was obtained calculating 1 -C s .Cluster analysis (Ward´s agglomerative method with Euclidean distance) was performed using the software Community Analysis Programme (CAP, 1999).
For the estimation of the value of pitfall and suction samples as surrogates of community species richness, correlation rather than regression analysis was performed.For instance, where most groundbeetles were caught by one type of sampling tech-nique, it is not surprising that there was a strong relationship between species richness based on that technique and that based upon the pooled data.Therefore, the potential explanatory surrogate variables (pitfall and suction species richness) are not independent from the potential response variable (pooled species richness), which violates a basic assumption of regression analysis.The value of r 2 is used as an indicator of the significance of the correlation, as this value shows the proportion of variation in one variable explained by the other (Dytham, 1999).For several ecological groups and taxa, species richness estimates were log 10 transformed to control heterogeneity of variance (Zar, 1984).When zero values were present, data were log 10 (x+0.5)transformed (see Yamamura, 1999).Multiple comparisons can inflate Type I error rate, so the Bonferroni correction was used to set the Pvalue at 0.0028 (= 0.05/ 18).A correlation (Pearson r) analysis was also performed between pitfall and suction species richness estimates using log 10 or log 10 (x+0.5)transformation of data.

SAMPLING EFFICIENCY, RICHNESS ESTIMATORS AND ACCUMULATION CURVES
A total of 47851 arthropod specimens were collected in pitfall samples (7628 herbivores and 40223 predators) belonging to 182 species, whereas 42023 arthropod specimens were collected in suction samples (31846 herbivores and 10177 predators) belonging to 173 species.Total sampling intensity, the ratio of specimens to species, was similar in both methods (pitfall = 262.92;suction = 242.91),but higher for predators than herbivores in pitfall and herbivores than predators in suction (see Table 1).The overall inventory completeness index (the percentage of species that is not singletons) was 0.81 for pitfall and 0.77 for suction, but the values slightly vary between investigated ecological groups in both methods (see Table 1): in predatory groups, inventory completeness is always slightly higher in pitfall than suction or both methods taken together, whereas in herbivores groups total data gets better results than each individual sampling method.
In terms of percentage completeness, in general this index was high with the Bootstrap and Michaelis-Menten means and reached lowest values with the Jackknife 1 or Chao 1 estimators (Table 1).Using the mean value of percentage completeness, in all three herbivore guilds both The estimators used are Chao 1, Jackknife 1 (Jack 1), Bootstrap and Michaelis-Menten Mean (MMMean) and were generated by EstimateS, version 6.0b1 (Colwell, 2000).The two curves at the bottom of the graph plot mean numbers of singletons and doubletons as a function of cumulative number of samples.For all curves, each point is the mean of 100 values based on 100 randomizations of sample accumulation order.
For none of the 15 sample sets (5 guilds x 3 sampling strategies) does the randomized species accumulation curve reach an asymptote (Figs.1-3).The species accumulation curve for the four estimators has three different general patterns (Figs.1-3): i) a non-asymptote pattern for Jackknife 1 and Bootstrap; ii) a general asymptote pattern for Michaelis-Menten means; iii) a more interesting variation in curvature between the five guilds for Chao 1.In fact, for predators (all species), chewing insects and spiders pitfall data (Fig. 1b, c, d) an asymptote was reached by the Chao 1 curve and the singleton and doubleton curves are also converging.For suction data, an asymptote was reached only for predatory guilds with the singleton and doubleton curves converging (Fig. 2b, d).When both methods are taken together only for sucking insects and spiders there is an obvious asymptote in Chao 1 curve (Fig. 3d, e).

SPECIES COMPOSITION
A total of 227 species of arthropods were sampled by suction and pitfall trapping in the twelve sites, during the four sampling occasions over two years.One hundred and twenty species were herbivores (sap-feeders= 61; chewers= 59) and 115 predators.Table 2 shows the number of species of the different taxonomic or ecological groups that were exclusively caught by pitfall trapping or suction sampling or sampled by both methods.As expected both sampling methods contributed unique species (Table 2): about 22% of herbivore species and 31% of predatory arthropod species were only sampled by pitfall trapping, whereas by suction sampling proportions were 28% and 18% respectively; In the Coleoptera, a higher proportion of species was sampled only by pitfall traps than in other taxonomic groups (G = 29.44,d.f.= 1, p < 0.0001); Slightly more species of herbivores were sampled exclusively by suction than by pitfall traps, while a few more species of predatory arthropods were found only in pitfall traps (G = 4.28, d.f.= 1, p = 0.04); Sucking and chewing herbivores showed very different patterns, with a hig-her proportion of the former sampled exclusively by suction, whereas more chewing species were sampled only by pitfall traps (G = 28.02,d.f.= 1, p < 0.0001); Similar patterns were obtained for two guilds of spiders, with more web-building species sampled exclusively by suction, whereas most other species were sampled only by pitfall (G = 7.41, d.f.= 1, p = 0.0091).
Values of the dissimilarity index (Table 2) shows that both methods differ greatly mainly for Hemiptera (Heteroptera), Curculionidae, phytophagous Coleoptera in general and spiders.
The analysis shown in Figure 4, illustrates that pitfall and suction samples of the same islands or habitats are clustered together.For Santa Maria and Pico, suction and pitfall samples form two separate clusters, thereby showing a difference between the two methods.For Terceira, samples from the two methods were clustered only by habitats (Figure 4).

SPECIES RICHNESS SURROGACY (PRESENCE/ABSENCE
DATA) The pooled species richness includes species that were collected only by suction sampling, only by pitfall trapping, by both methods and by direct search.Even after Bonferroni correction, correlations between pooled species richness and either suction or pitfall species richness were statistically significant, for several different groups of species (see Table 3).For the herbivores, both suction and pitfall estimates of species richness were well correlated with pooled species richness.However, suction estimates of species richness of sap-feeders and pitfall estimates of chewing insects had higher levels of r 2 than alternative sampling method (Table 3).For the predatory groups, only webbuilding spiders were better sampled by suction (r= 0.80; p = 0.0016).Wandering spiders, groundbeetles and rove-beetles were better sampled by pitfall traps, with high proportions of the variance (90, 94 and 86% respectively) being explained (see Table 3).

Discussion
The interpretation of data obtained by a single technique may introduce bias or even be speculative (Sunderland et al., 1995).Pitfall trapping and suction sampling have a long history of use in ecological research (see Southwood, 1978;papers in Toft & Riedel, 1995;Southwood & Henderson, 2000).Indeed, there is already a substantial literature comparing arthropod sampling methods in terms of arthropod abundance in specific sites or habitats and often of species groups of arthropods (e.g.Disney et al., 1982;Schotzko & O´Keeffe, 1989;several papers in Toft & Riedel, 1995;Broatch & Vernon, 1997;Costello & Daane, 1997;Hammond, 1997;Buffington & Redak, 1998  Table 2.-Número de especies de los distintos grupos taxonómicos y ecológicos colectados mediante el uso exclusivo de trampas de caída o técnica de succión, y el uso combinado de ambos métodos.Se presentan también los valores de un índice de disimilitud (1-Índice de similitud de Sorenson).Boiteau & Osborn, 1999;Churchill & Arthur, 1999;Standen, 2000).The study described here was conducted in the context of estimating species richness and composition in different habitats and islands and to give insight into the importance of using, simultaneously, two methods of sampling arthropods in agroecosystems, to assess herbivorous and predatory groups.
METHODOLOGICAL ASPECTS: FIELD WORK AND SAM- PLE SORTING The study was carried out under logistically difficult circumstances on three different oceanic islands and at high altitude pasture sites where access was commonly difficult.It therefore required less labour-intensive sampling techniques.In this contest, the "Vortis" suction sampler had considerable advantages, as it is fully portable, fast to operate and reliable, with few mechanical problems during the two years of operation.Its main advanta-ges are undoubtedly its light weight and ease of retaining and removing the samples, which enables an operator to take 20 two minutes samples (four 30 seconds subsamples) in about one hour of continuous work.Thus, at least three sites located relatively near each other could be sampled daily and at a reasonable level of sampling intensity, assuming that the weather conditions remain constant.The amount of debris collected was highly dependent on the nature of the substrate, but since the pastures had a dense cover of non-flowering perennial grasses (mainly Anthoxanthum odoratum, Holcus lanatus, Agrostis castellana and Poa trivialis), the samples were very clean.Another important advantage of the "Vortis" suction machine is its ability to sample even very small juvenile spiders and immature insects (aphids, thrips, Lepidoptera larvae, etc.) and small adult insects (e.g.male coccids) with great accuracy and without damage.Of course, sorting these samples is more time consuming than sorting  pitfall samples that are composed mainly of adults of more active species, but gives a more detailed assessment of the arthropod community.If time is a constraint, the subsampling of collected samples is always an option.Although suction sampling is less effective for large, nocturnal, active arthropod species and possibly some spiders that build webs on the ground, the ability to sample juvenile stages is an attribute of considerable importance especially when samples cannot be taken throughout the season to coincide with adult stages.
In terms of labour, suction samples, even without debris, are more difficult to sort than pitfall samples, since there were more small specimens.Moreover, suction sampling is very dependent on weather conditions, since it requires dry vegetation and low wind conditions to be effective.This was critical in the hyper-oceanic and wet climate of the Azorean islands, allowing sampling only on some days and in the afternoon when the vegetation was dried by the sun.Pitfall trapping is less sensitive to the weather, since the set-up and recovery of traps can occur even in bad weather conditions, though at least two site visits are necessary for a single sample.The method is far less effective in sampling juvenile and small arthropods.

SAMPLING EFFICIENCY, RICHNESS ESTIMATORS AND ACCUMULATION CURVES
In spite of the fact that none of the observed species accumulation curves reached an asymptote other measured indices were useful in evaluating the performance of the investigated sampling methods.Looking for the four richness estimators they performed differently, Jackknife 1 and Chao 1 generating large estimates of species richness, Bootstrap giving a lower estimate of species richness and Michaelis-Menten mean (MMMean) clearly underestimating species richness.Since the available estimators are probably underestimating  ) and significance level of the correlation between log pooled species richness and the suction and pitfall log species richness for a range of taxa.The highest paired correlations are shown with the r2 value in bold.
true species richness in most studies (Chiarucci et al., 2003), all but Michaelis-Menten mean estimator gave probably realistic estimates of species richness.This contrasts with the study of Toti et al. (2000), in which Chao 1 gave unrealistic large estimates and MMMean performed better.Interesting is to compare the estimates of each sampling technique used separately ( were the most reliable estimators with the current datasets, whereas, suction sampling was more effective for herbivores (all species), sucking insects and spiders, and pitfall trapping was more effective for chewing insects, predators and spiders.Therefore, sampling spiders with pitfall or suction was equally effective in terms of estimating true species richness.However, as showed by the Dissimilarity Index (Table 2) pitfall and suction are sampling a different set of the spider community (see also below).
The number of singleton species is usually the key factor in most the above-mentioned indices.If a sampling protocol was efficient than the curve of the singletons should level or decrease with increasing collecting effort (Toti et al., 2000).This was the case of pitfall trapping with predatory guilds and chewing insects, suction sampling with predatory guilds and joint sampling methods for all but chewing insects.Therefore, me may conclude that the use of both sampling methods was useful with a general high sampling efficiency, in average more than 85%.Together with a similar performance of the species accumulation curves for most of the investigated ecological groups, demonstrates that both sampling methods (pitfall and suction) yielded a high percentage of the present species in the pastures of the three studied islands.
Looking for the ecological and taxonomic status of singletons in our data sets, for instance in spiders for which we have a good knowledge of species composition in other Azorean habitats (Borges et al., unpublished data), most of the singletons are species particularly common in other habitats (e.g.native forest both in soil and canopy).Therefore, most our spider singletons are probably tourists in managed pastures and consequently we may have reached a near perfect sampling completeness for spiders.
For the several studied arthropod groups pitfall was as good as suction sampling in terms of sampling efficiency as measured by the inventory completeness and the percentage completeness indices.In terms of species accumulation, pitfall performed even better than suction for chewing insects (see Figs. 1c and 2c) and as well as suction for spiders (see Figs 1d and 2d).Interestingly, when both methods were jointed in the analysis, percentage completeness only slightly improved for phytophagous guilds.In terms of sampling intensity (number of specimens per species) pitfall performed better for predacious groups, whereas suction was better for all herbivores and sucking insects (Table 1).Jointed methods yielded a higher sampling intensity particularly for chewing insects.

SURROGACY
In terms of community composition, suction sampling was more effective for sucking insects, whereas pitfall trapping was preferable for chewing herbivores.Indeed, a higher proportion of sucking insects was sampled exclusively by suction sampling, whereas more chewing species were sampled only by pitfall trapping.For spiders, more webbuilding species were only sampled by suction sampling, whereas most of other species were sampled only by pitfall trapping.Moreover, our results indicate that species richness of some groups, e.g.herbivorous arthropods (all species), insect sap-feeders and web-building spiders is better explained by the suction data set, whereas pitfall data were more informative for predatory arthropods (all species), chewing insects, wandering-spiders, carabids and staphylinids.Thus, neither Vortis suction sampling, nor pitfall trapping described perfectly the species composition and richness of pasture arthropods.Indeed, a few species were only collected by direct searching.However, the cluster analysis (Fig. 4) shows greater similarity in species composition between the two sampling methods than between sites (particularly in Terceira) and, thus, if time was a major constraint it would be possible to use a single method to give a fair reflection of the community.Topping & Sunderland (1992) reported a failure of pitfall traps to reflect the community composition of spiders adequately.However, our results are not in total agreement since: i) pitfall trapping was complementary to suction sampling in describing species composition of arthropods (see also Standen, 2000); ii) species richness estimated from pitfall traps was correlated with pooled species richness for several ecological and taxonomic arthropod groups.
Another interesting subject is related with the species richness ratio of herbivores to predators (124:115).We should expect a much higher number of phytophagous species related to predators than obtained and the explanation for this result is explained not by an inefficacy of sampling but by insular disharmony and geological age of the islands (see Borges & Brown, 1999).
Since each method has its advantages and disadvantages, we are in agreement with Standen (2000) in that pitfall trapping and suction sampling (Vortis) should optimally be considered as two complementary sampling methods in describing the biodiversity of pastures, and not as exclusive methods.In fact, a suction device in grassland will not, by itself, give absolute estimates of density, because careful ground search after suction sampling usually results in more species being added (Buffington & Redak, 1998).However, suction samples were such good estimates of the density of arthropod species that it has been possible to make good predictions between the attributes of the pasture vegetation structure and herbivorous insects and spider densities (see Borges & Brown, 2001).
Our results have shown that pitfall trapping can be used in community ecology studies with some fidelity, and when used together with an absolute density sampling technique (e.g.suction sampling), the final result can be highly accurate.It seems that, unless we are only interested in absolute density estimates, for which suction will be a more reliable method, pitfall trapping may be a more practical and inexpensive method of describing species composition.
The present and growing demands for habitat characterization highlights the need for the critical awareness of the strengths and weaknesses of the various sampling techniques most likely to be used by land managers, farmers, advisory services and conservation bodies.Grasslands are currently a particular focus at attention because of the large areas under intensive management, the need to assess the impact of such management on biodiversity and the success of measures being undertaken to ameliorate the effects or actively restore biodiversity.The results given in this paper may therefore be of value in the assessment of agri-environment schemes being implemented in various parts of the globe.

Conclusions
This study allows us to conclude that both suction sampling and pitfall trapping techniques were reliable in terms of sampling efficiency getting a good estimate of arthropod species richness in sown and semi-natural pastures of the three Azorean islands.Using the two methods together we reached values of almost 90% completeness in some guilds, which is very good when compared with other studies (Toti et al., 2000;Sorensen et al., 2002).Pitfall and suction alone were reliable in estimating species richness of predatory arthropods (all species) and spiders.Joint data is needed for getting reliable estimates for herbivores guilds.We also found that both suction sampling and pitfall trapping techniques were inadequate by themselves for a detailed description of community composition, and species richness surrogacy of most of the diurnal active grassland arthropods."Vortis" suction sampling is a reliable method for estimating these parameters for most sap-feeding insects and web-building spiders.Pitfall trapping has some advantages for larger, nocturnal and active arthropods, most chewing insects, wandering spiders, ground-beetles and rove-beetles.Thus, suction and pitfall methods should be considered complementary techniques for sampling grassland epigeic arthropod species and for an accurate estimate of total arthropod community structure.If only an assessment of community composition as an indication of grassland sward quality is needed, both techniques may be useful on their own.It is therefore important to tailor the choice of technique to address the requirements of the study.

( c )Fig. 1 .
Fig.1.-Plotscomparing the estimated species richness with the observed species accumulation curve (Sobs), using data from 48 pitfall pasture samples (12 sites x 4 sampling periods) of herbivores insects (a), predatory arthropods (b); chewing insects (c), spiders (Araneae) (d) and sucking insects (e).The estimators used are Chao 1, Jackknife 1 (Jack 1), Bootstrap and Michaelis-Menten Mean (MMMean) and were generated by EstimateS, version 6.0b1(Colwell, 2000).The two curves at the bottom of the graph plot mean numbers of singletons and doubletons as a function of cumulative number of samples.For all curves, each point is the mean of 100 values based on 100 randomizations of sample accumulation order.

Fig
Fig.2.-Plots comparing the estimated species richness with the observed species accumulation curve (Sobs), using data from 48 suction pasture samples (12 sites x 4 sampling periods) of herbivores insects (a), predatory arthropods (b); chewing insects(c), and spiders (Araneae) (d) sucking insects (e).The estimators used are Chao 1, Jackknife 1 (Jack 1), Bootstrap and Michaelis-Menten Mean (MMMean) and were generated by EstimateS, version 6.0b1(Colwell, 2000).The two curves at the bottom of the graph plot mean numbers of singletons and doubletons as a function of cumulative number of samples.For all curves, each point is the mean of 100 values based on 100 randomizations of sample accumulation order.

Fig. 4 .
Fig.4.-Cluster analysis of the species composition of arthropods from twelve sites.Each site is represented by four symbols in which the first symbol refers to island (S -S.Maria; T -Terceira; P -Pico), the second symbol to site management type (S -sown pastures; N -semi-natural pastures), the third symbol to replicate number and the last symbol to sampling method (suction (S) and pitfall trapping (P).

Table 2 .
-Number of species of different ecological and taxonomic groups collected exclusively by pitfall, or by suction or by the two sampling methods.A dissimilarity index (1-Sorenson´s Index of similarity) is also presented.

Table 3 .
-Coefficient of determination, percentage of variance explained (r 2