|Variation among populations—A total of 60 pollinator spe- cies (26 lepidopterans, 23 hymenopterans, and 11 dipterans) were recorded in the 1460 3-min censuses in 1996 at the 15
L. latifolia populations. The number of species recorded per locality (SOBS, Table 2) was considerably lower and varied one order of magnitude among sites (range = 3–30 species re- corded per site; Table 2). Mean number of flowers visited per
3-min census (±SD; all species pooled) varied also consid- erably among populations, from 0.7 ± 2.9 to 14.2 ± 17.4 flowers/census, thus a 20-fold variation (x2 = 350.7, df = 14, P K 0.001; Kruskal-Wallis ANOVA). Across populations, SOBS was directly correlated with mean flower visitation per census (r = 0.675, N = 15, P = 0.006), which suggests that population differences in SOBS largely reflect differences in pol- linator visitation. This is supported by flower-based rarefaction curves for the 15 populations studied (Fig. 2), which reveal that SOBS was a poor index of population-level differences in pollinator species richness. For example, the pollinator assem- blages of populations 3 and 7, with SOBS = 30 and 25 species, respectively, were actually considerably less diverse than that of population 1, with SOBS = 18 species.
Flower-based rarefaction curves show that, after accounting for population differences in pollinator visitation frequency, populations of L. latifolia still differ widely in pollinator spe- cies richness (Fig. 2). Confidence intervals have been omitted from Fig. 2, but those of extreme populations (e.g., popula- tions 1, 3, 7, and 12 vs. 5, 8, 11, and 14) are largely non- overlapping. SRAR100 could be estimated for 14 populations, and ranged between 2.9–13.8 species (Table 2). The expected number of pollinator species implicated in the visitation of 100 flowers thus varied nearly five-fold in the set of L. latifolia populations studied. The correlation between SRAR100 and SOBS across populations was positive and statistically significant (r
Note: SOBS = observed species richness, the total number of pollinator species recorded during all the censuses on a particular population. SRAR100 = rarefaction-estimated species richness, obtained from the flow- er-based rarefaction curve for each population (Fig. 2) as the y-value predicted for x = 100 flowers visited. * Site names, location (as X-Y UTM coordinates to the nearest km on European 1979 map datum sys- tem, UTM zone 30S), and elevation: 1, Arroyo Aguaderillos, 510–4201,
1180 m; 2, Arroyo Amarillo, 505–4193, 1380 m; 3, Arroyo de los Ubios, 508–4199, 1235 m; 4, Caballo de Acero, 514–4195, 1450 m; 5, Collado del Calvario, 510–4200, 1425 m; 6, Cruz de Quique, 504–
4194, 1290 m; 7, Cuevas Bermejas, 513–4203, 1210 m; 8, Las Cana- lejas, 522–4215, 1440 m; 9, Las Navillas, 508–4198, 1170 m; 10, Pista de Los Escalones, 536–4222, 1520 m; 11, Prados de Navahondona,
504–4190, 1540 m; 12, Presilla de T´ıscar, 500–4182, 1190 m; 13, Puerto de T´ıscar, 497–4183, 1180 m; 14, Raso del Tejar, 511–4203,
1040 m; 15, 250 m SE of Arroyo Aguaderillos, 511–4201, 1210 m.
¶ Non-estimable, because total flowers visited <100.
Fig. 2. Flower-based pollinator rarefaction curves showing the expected accumulation of pollinator species with increasing number of flowers visited, obtained separately for 15 populations of Lavandula latifolia studied in 1996 in the Sierras de Cazorla and Segura, southeastern Spain. Each curve is the average of 50 randomizations without replacement of the censuses conducted on each population (see Table 2 for census- and flower-based sampling ef- forts). Numerals identify populations mentioned in the text.
Fig. 3. Flower-based pollinator rarefaction curves for five L. latifolia populations studied on two consecutive years. Each curve is an average obtained after
50 randomizations without replacement of the censuses conducted on each population and year. Dashed lines denote ± 1 SD around the mean. Sampling effort for 1996 as shown in Table 2; for 1997, number of censuses and total flowers visited in parentheses: population 1 (65, 306), population 3 (65, 549), population
5 (60, 746), population 9 (60, 163), population 14 (70, 1124). Note different scale of horizontal axis for populations 1 and 9.
= 0.623, N = 14 populations, P = 0.02), although its mag- nitude and significance level were enhanced when the influ- ence of pollinator visitation was accounted for by partialing the correlation on mean visitation rate per census (partial r =
0.783, N = 14, P = 0.001).
Between-year comparisons of rarefaction curves for the five
populations studied in 1996 and 1997 are shown in Fig. 3. The
shape and slope of the curves for the two years were quite
similar in three populations (1, 3 and 9). In the other two
populations, differences between curves were moderate (pop-
ulation 14) to large (population 5), and the two sites differed
in the sign of the change between years. In general, differences
among populations in pollinator diversity tended to be larger
than differences between years for the same population and,
at least in some populations, pollinator diversity seemed a rel-
atively constant, population-level trait.
Biases and pitfalls in assessing pollinator diversity—Irre- spective of whether they pertain to individual shrubs (Fig. 1) or populations (Fig. 2), the shape of flower-based rarefaction curves obtained in this study was similar to those customarily obtained in investigations of species diversity of ecological communities (Gotelli and Colwell, 2001). Pollinator species are first quickly added as the number of flower visits increases, and the rate of species addition declines progressively as the curve approaches an (expected) asymptote. Similar patterns were obtained when rarefaction curves were scaled to number of censuses rather than number of flowers (graphs not shown). These findings confirm Ollerton and Cranmer’s (2002) predic- tion that the relationship between pollinator diversity and sam- pling effort should most likely be saturating rather than linear. Consequently, using species/sampling effort ratios to correct for differences in sampling effort among the groups being compared (e.g., species, populations, regions, studies) is prob- ably inappropriate in most instances. Gotelli and Colwell’s (2001, pp. 384–385) detailed arguments on the various pitfalls associated with using this kind of ‘‘category-subcategory ra- tios’’ to compare species diversities thus seem to apply also to estimates of pollinator diversity, and will not be repeated here.
With just a few exceptions (e.g., plant 2 in Fig. 1; popula- tions 8 and 14 in Fig. 2), all pollinator rarefaction curves com- puted in this study were far from reaching any apparent as- ymptote. Despite the rather large number of censuses con-
ducted and flower visits recorded, sampling effort was still clearly insufficient to reveal all pollinator species that interact with L. latifolia individuals or populations and, therefore, sim- ple counts of the number of pollinators species observed (SOBS) are not valid descriptors of the actual pollinator diversity of either plants or populations. SOBS values for individual plants or populations depart to variable degrees from true species richness set by expected asymptotes because of differences among plants and populations in both species-accumulation slopes and flower-visitation frequencies. This has the impor- tant implication that observed variation in SOBS represents a distorted version of actual differences among plants or popu- lations in pollinator diversity. This distortion is clearly illus- trated in this study by weak or nonsignificant correlations be- tween SOBS and S100RAR values, and further highlighted by the strong positive correlations found between SOBS and flower vis- itation rate in both the among-plant and among-population comparisons.
In the present study, sampling effort (number of pollinator censuses) was held relatively constant across groups under comparison. Despite this, however, variation in flower visita- tion rate alone explained as much as 48% and 45% of among- plant and among-population variance in SOBS, respectively. In the only other study known to me that evaluated the influence of sampling effort on SOBS, Ollerton and Cranmer (2002) found that 36% of variance among plant communities in mean SOBS was accounted for by differences in sampling effort (number of days of observation). Taken together, these figures mean, on one hand, that holding sampling effort constant across groups under comparison is not sufficient to guarantee reliable pollinator diversity estimates. And on the other hand, that raw SOBS figures may ultimately become almost meaningless as de- scriptors of pollinator diversity when subject to the combined effects of broad variation in both sampling effort and polli- nator visitation and neither of these two factors is adequately taken into consideration. This confirms, in the context of pol- linator diversity studies, the long-known general principles ap- plying to the measurement of species diversity in ecological communities, that comparing species richness without refer- ence to a taxon sampling curve is problematic at best and that comparing raw taxon counts for two or more assemblages will generally produce misleading results (Gotelli and Colwell,
2001; and references therein). As with conventional species diversity measurements, pollinator taxon sampling curves emerge as the most reliable method to compare pollinator spe-
cies richness among individual plants, populations of the same species, or species.
Recent studies have frequently considered the prevalence of generalized plant–pollinator relationships in nature (e.g., Her- rera, 1996; Waser et al., 1996; Olesen, 2000), and analyzed plant–pollinator interaction networks at the local or regional plant community level in relation to hypotheses on pollinator generalization or, more generally, plant adaptation to pollina- tors (e.g., Memmott, 1999; Dicks et al., 2002; Olesen and Jor- dano, 2002; Bascompte et al., 2003). These investigations have often relied on plant community-level compilations of polli- nator species and/or raw pollinator species counts gathered from preexisting studies (but see, e.g., Memmott, 1999; Dicks et al., 2002; Nakano and Washitani, 2003), have used uncor- rected pollinator species counts (SOBS) to measure degree of plant generalization, and have generally paid little or no atten- tion to the potential influence of sampling biases inherent to, and artifacts derived from, using that kind of data (but see Ollerton and Cranmer, 2002). Findings of this study on L. la- tifolia suggest that comparative analyses using raw SOBS values as measures of pollinator diversity are prone to suffer from artifacts caused by heterogeneities in sampling effort, polli- nator visitation, or some complex combination of these. This is exemplified by Ollerton and Cranmer’s (2002) investigation on latitudinal trends in pollinator specialization. These authors found a significant latitudinal trend in pollinator generalization when raw SOBS values were used, but the pattern vanished when they accounted for differences among sites in sampling effort.
Artifacts derived from neglecting or inadequately account- ing for heterogeneities in sampling effort and pollinator visi- tation frequency may likewise affect in unpredictable ways some community-wide analyses of plant–pollinator networks. In these studies, unaccounted differences among species and communities in sampling effort and/or pollinator visitation may lead to analyzed pollinator spectra being uncorrelated or weakly correlated with actual pollinator spectra across the spe- cies or communities involved in the comparison. Furthermore, unappreciated correlations across species or plant communities between apparent pollinator species richness (SOBS) and polli- nator visitation frequency are apt to lead to spurious conclu- sions whereby correlates of pollinator abundance are errone- ously interpreted as correlates of pollinator diversity. Disen- tangling the relative contributions of variations in pollinator abundance and pollinator diversity to observed variation in apparent species richness (SOBS) should become a priority of plant–pollinator community studies. Meanwhile, some conclu- sions of these investigations are to be treated with caution until their robustness to underlying pollinator sampling inadequa- cies and hidden abundance—diversity correlations is tested and verified.
Generalization as a local and individual property—At the regional level, the pollination system of Lavandula latifolia undoubtedly qualifies as highly generalized. About 85 species of dipteran, hymenopteran, and lepidopteran pollinators were recorded in an earlier investigation conducted in the same area studied here (Herrera, 1988), and a total of 60 pollinator spe- cies from the same three major insect groups were recorded in the 15 populations studied in 1996. The present study has shown, however, that extensive generalization is not an in- variant, species-level property of L. latifolia. Populations vary broadly in degree of pollinator generalization and, within the
highly generalized population of Arroyo Aguaderillos, indi- vidual plants are quite variable in their degree of generaliza- tion. There, the range of SRAR100 values for individual plants in
1991 (2.8–11.1 pollinator species) was strikingly similar to the range of population-level values for the 15 populations studied in 1996 (2.9–13.8 species). This suggests that, after accounting for variation in sampling effort and pollinator visitation, the magnitude of the variation in degree of pollinator generaliza- tion occurring at the scale of tens of meters (among shrubs of the same population) may be as large as that occurring at the scale of tens of kilometers (among populations).
An analysis of the correlates of observed variation in degree of pollinator generalization falls beyond the scope of this pa- per, but available information allows for at least some tentative interpretations. Variation in SRAR100 was not significantly relat- ed to variation in mean corolla tube length at either the among- population (r = —0.146, N = 14 populations, P = 0.62; C. M. Herrera, unpublished data) or within-population levels (r
= 0.109, N = 12 plants, P = 0.74; data in Table 1 and the Appendix in Herrera, 1995). In contrast, location effects seem to account for a significant fraction of observed variation. Pol- linator diversity of populations growing adjacent to permanent streams (mean SRAR100 ± SD = 9.7 ± 2.8 species, N = 4 populations) was nearly double that of populations on arid slopes (mean SRAR100 = 5.1 ± 1.8 species, N = 10 populations; x2 = 6.48, df = 1, P = 0.01, Kruskal-Wallis ANOVA; C. M. Herrera, unpublished data). In Aguaderillos in 1991, SRAR100 of individual shrubs was inversely correlated with their daily mean solar irradiance (r = —0.581, N = 12 plants, P = 0.037; data in Table 1 and Appendix in Herrera, 1995). Taken to- gether, these relationships suggest that the variable degree of pollinator generalization in populations and individuals of L. latifolia may be more parsimoniously explained in terms of abiotic factors influencing insect diversity at the landscape and microsite spatial scales than in terms of variation in some floral trait limiting the range of pollinators. It is not surprising that in the dry, hot summer typical of the Mediterranean-type cli- mate of my study region, L. latifolia populations contiguous to the few permanent streams (see also Herrera, 1988) and shrubs that occupy relatively shadier locations in the forest understory have the most diverse insect pollinator assemblages because of the more benign microclimates. In an abiotically driven scenario of this kind, microclimatic factors extrinsic to the plants might be more important than intrinsic plant features in determining the extent of pollinator generalization of L. la- tifolia populations and individuals.
Variation in pollinator composition among populations of the same plant seems to be the rule in nature (e.g., Herrera,
1988; Go´ mez and Zamora, 1999; Thompson, 2001; Eckert,
2002), thus the finding that populations of L. latifolia differed
in pollinator diversity was not unexpected. More interesting
was the finding that in some populations the shape and slope
of pollinator rarefaction curves varied little between years. Al-
though based on rather limited evidence (only five populations
studied over two years), this result suggests that some L. la-
tifolia populations may interact over the years with pollinator
assemblages of a given, locality-specific diversity and that
population differences may remain consistent across years.
Such a pattern would lead to geographically variable oppor-
tunities of adaptation to particular pollinators (Thompson,
1994). The broad variation among populations in extent of
pollinator generalization documented in this study (as mea-
sured by SRAR100) also lends support to Olesen’s (2000) con-
tention that discussions on the evolution of pollinator gener- alization should focus on the population level. Among-popu- lation variability revealed by this study also suggests that sin- gle-population data are probably insufficient to characterize a plant species with regard to its degree of generalization. To this end, both the central tendency and variability of some suitable population-level estimator of generalization (e.g., SRAR100 as used here) obtained at a sufficient number of distinct populations should be used, rather than single figures or pooled averages.
Concluding remarks—This paper has shown that pollinator diversity estimates based on raw species counts may be heavi- ly dependent on aspects related to research design (variation in sampling effort), biological phenomena (differences in pol- linator abundance or visitation rates), or both. If unaccounted for, such effects may combine to mask or distort underlying ecological patterns of interest. As shown in this study, flower- based rarefaction curves applied to data obtained through ran- dom sampling of pollinator activity at flowers are useful for making rigorous comparisons of pollinator species richness among individual plants, populations of the same species, or different species.
The notion of generalization/specialization gradients has been present in the ecological literature for decades (Futuyma and Moreno, 1988), and the study of factors promoting or limiting specialization has played a central role in the devel- opment of entire ecological subdisciplines like, e.g., the study of plant–herbivore interactions (Berenbaum, 1990; Jaenike,
1990). Developments in the study of factors influencing spe- cialization in plant–pollinator systems may likewise catalyze significant improvements in our understanding of the evolution of these mutualistic interactions (Herrera, 1996; Waser et al.,
1996; Go´ mez, 2002). Devising a clearer formalization of con- cepts and developing more rigorous analytical tools are two prerequisites for such advances to take place. More important than this, however, will be to obtain significant amounts of fresh field data for a broad variety of species, plant commu- nities and ecosystems using adequate sampling protocols al- lowing for rigorous comparisons and analyses. New methods of analysis cannot compensate for the current scarcity of re- liable field data on plant–pollinator interactions, as recently stressed by Kay and Schemske (2004), and sophisticated an- alytical tools can hardly redeem biased or otherwise messy pollinator data.
Our understanding of the evolution of plant–pollinator in- teractions will also benefit from a better knowledge of how plant specialization on pollinators varies among individuals of the same population, among distinct populations of the same species and among different species, much in the same way and for the same reasons as progress in the study of plant– herbivore interactions has benefited from recognition of the distinct levels implicated in the evolution of herbivore spe- cialization (Fox and Morrow, 1981; Bernays and Minkenberg,
1997; Bernays and Singer, 2002). As with herbivore speciali- zation on plants, plant specialization on pollinators may be a variable attribute of populations rather than a trait of a species throughout its geographical range and, like in plant–herbivore systems, consideration of pollinator generalization as a local phenomenon will affect the framing of questions about plant– pollinator interactions in both ecological and evolutionary con- texts (Fox and Morrow, 1981).
ARMBRUSTER, W. S. 1988. Multilevel comparative analysis of the morphol- ogy, function, and evolution of Dalechampia blossoms. Ecology 69:
ARMBRUSTER, W. S., C. B. FENSTER, AND M. R. DUDASH. 2000. Pollination
‘‘principles’’ revisited: specialization, pollination syndromes, and the evolution of flowers. Det Norske Videnskaps-Akademi. I. Matematisk Na- turvidenskapelige klasse, Skrifter, Ny Serie 39: 179–200.
BASCOMPTE, J., P. JORDANO, C. J. MELIA´ N, AND J. M. OLESEN. 2003. The nested assembly of plant-animal mutualistic networks. Proceedings of the National Academy of Sciences of the United States of America 100: