1Few studies of pollination success in fragmented systems measure stigmatic pollen load, and those that do often find it unrelated to plant or population density, size or isolation. Reduced reproductive output, however, is commonly reported, probably because incompatible pollen is contributing substantially to pollen loads of isolated flowers.
2We used manipulated floral arrays of a bee-pollinated species (Dianella revoluta) to investigate isolation effects on deposition of outcross pollen, while precluding self-pollen transfer.
3Outcross pollen receipt declined significantly over short distances up to approximately 50 m but even the most isolated flowers received some pollen grains. In contrast, heterospecific pollen did not decline, indicating that the outcross-pollen decline was not due to reduced pollinator visitation. Increased distance of experimental arrays from a nature reserve did not reduce the probability of pollen receipt.
4Many flowers were damaged by flower feeding beetles in the genus Arsipoda, which would be likely to substantially reduce the efficiency with which flowers are converted to fruits. The probability of flower damage from these predators was significantly lower in arrays that were more distant from the nature reserve.
5This study indicates that reduced plant density and increased isolation from a source of outcrossed pollen can lead to a substantial decline in the probability of outcrossing, even when pollinator visitation is maintained at a high level. Depending on the mating system, this process will lead to reduced seed set or increased inbreeding for plants in fragmented habitats, even when pollinator abundance and behaviour are unaffected.
Pollination success of plants can be detrimentally affected by habitat fragmentation, both quantitatively, due to reduced diversity and abundance of pollinators (Aizen & Feinsinger 2002), and qualitatively through decreased deposition of compatible pollen on the stigma because of reduced mate availability (Kunin 1993). Both are strongly affected by the reduced plant density and increased isolation that often results from habitat fragmentation.
Although isolated plants and those in sparse populations are expected to receive less pollinator visits, stigmatic conspecific pollen loads have rarely been found to be reduced. This is likely to be due to plants in small or isolated patches getting a higher proportion of pollen from geitonogamous selfing or closely related neighbours (Waser & Price 1991). Self-pollen transfer may increase because pollinators stay longer and visit more flowers per plant when they do visit isolated plants (Klinkhamer & de Jong 1989). A substantial increase in self-pollen may act to reduce fruit or seed set. In addition, most plant species share pollinators with other species and, particularly at low densities, may receive considerable quantities of heterospecific pollen. This can further erode reproductive performance through stigma clogging (Waser 1978) or pollen allelopathy (Sukhada & Jayachandra 1980).
Although pollen quality and quantity are easy to separate in theory, few empirical data are available (Feinsinger et al. 1991; Kunin 1993). We investigated these issues with Dianella revoluta R. Br. (Phormiaceae), a partially self-compatible perennial herb common in open-mallee woodland of central New South Wales, Australia. In this study we used experimental arrangements of flowers in natural roadside vegetation to directly assess the effect of isolation from mates on outcross pollen receipt. We excluded self-pollen transfer by sealing the anthers of experimental flowers, removed any conspecific donors that were not part of the experimental array, and used heterospecific pollen data to infer pollinator visitation.
Judiciously applied, pollen receipt studies offer important information about pollination that is not captured by pollinator movement data (Thomson & Stratton 1985) or genetic studies. For example, genetic techniques can establish the level of selfing vs. outcrossing or average distances of successful pollen movement (e.g. Stacy et al. 1996) expressed in seeds from fragmented populations, but they cannot tell us about the patterns of pollen movement that underlie those outcomes.
Changes in the quantity and quality of pollen might arise independently of any flower density effects when fragmentation affects native bee populations, but there is no consensus regarding the most likely outcomes. Lower diversity and abundance of native invertebrate pollinators in fragmented sites have been reported (Aizen & Feinsinger 2002) but knowledge of habitat and nesting requirements is lacking for many species and the impact of fragmentation upon those attributes is often unknown (Cane 2001).
The primary question addressed by this study was does the quantity of outcross pollen received by stigmas decline with distance from a conspecific pollen source? Secondly, if there is a decline, does it reflect a reduction in the quantity or quality of pollinator visits? Thirdly, because fragmentation is expected to affect native pollinators, we ask if patterns of pollen receipt are affected by the distance between small patches of vegetation and the nearest large patch, such as a nature reserve.
We used D. revoluta as a model study species for two major reasons. Firstly, previous experiments with this species established effective techniques for adjusting the density of flowers in the field, and for virtual emasculation that prevents flowers from shedding pollen without reducing their attractiveness to pollinating bees (Duncan et al. 2004). Secondly, previous work on the same species in the same region suggests that it might be sensitive to flower density effects, because plants in isolated roadside vegetation had lower fruit set efficiency than those in nature reserves (Cunningham 2000a).
Materials and methods
Dianella revoluta is a rhizomatous, lilioid, perennial herb, widely distributed in south-eastern Australia. Flowers open for one day only, typically from 08.00–09.00 to 15.00–17.00 (Duncan et al. 2004). In our sites Dianella revoluta is pollinated by a small number of female native anthophorine apid and halictid bee species (Duncan et al. 2004), which are solitary and ground nesting (Dollin et al. 2000). The nectarless flower rewards bees with pollen that can only be collected by vibrating the anthers (‘buzz pollination’). Some pollen is transferred to the stigma of the same flower during the collecting visit, some is transferred to the stigmas of other conspecific flowers, and the remainder is consumed by the bees or otherwise lost from the D. revoluta population (Duncan et al. 2004). The species is only partially self-compatible due to late-acting self-incompatibility. Selfed flowers set fruit at about 24% the rate of outcross pollinated flowers, but a single pollinator visit may deposit at least twice as much self as outcross pollen (Duncan et al. 2004). Outcross pollen does not outperform self-pollen prior to ovule fertilization (‘prepotency’, Lloyd & Schoen 1992), thus high levels of self-pollen deposition carry a potential reproductive cost due to usurped ovules (Duncan et al. 2004). Pollinators are required to set fruit. Bagged inflorescences showed no pollen transfer from anthers to the stigma, and no fruit was set (D. H. Duncan, unpublished data).
The stem of the inflorescence is naturally rigid and remains rigid when inflorescences are cut and placed in a vase. Inflorescences in a vase will continue to open normally presented flowers over at least a 2-day period, suggesting that inflorescence hydraulics were not unduly affected. Because D. revoluta flowers are nectarless, effects of inflorescence cutting on nectar production are not relevant.
Emasculation of flowers was necessary to preclude self-pollen transfer in our experiments. True emasculation (removal of anthers) was not appropriate because the anthers of Dianella species are important for pollinators both as a visual cue (Bernhardt 1995) and for gripping onto while vibrating the anthers (D. H. Duncan, personal observation). We developed a technique for virtual emasculation by blocking the terminal pore of anthers using craft glue (PVA) (Duncan et al. 2004). Observations suggest that this technique did not affect visit frequency or behaviour of pollinators (Duncan et al. 2004).
The work was undertaken in roadside vegetation around Gubbata Nature Reserve, NSW, Australia (33°38.01′ S, 146°33.36′ E). The reserve is approximately 162 ha (Mueck et al. 1987) and consists of shrubby, open mallee woodland dominated by trees of the genus Eucalyptus. Dianella revoluta is very common in the understorey at this site. The roadsides have similar floristic composition to the reserve but are more disturbed, largely due to soil disturbance associated with maintenance of road verges. The roadside strips comprised linear remnants that were 10–20 m wide, on either side of a road 6–20 m in width.
Mean annual rainfall in this area is approximately 452 mm (Australian Bureau of Meteorology data). Rainfall in inland Australia is, however, notorious for high interannual variability (Nicholls 1991). In the two field seasons of this experiment (Spring 2001, 2002) rainfall was lower than average, which appeared to cause a decline in flowering intensity of many species in the second season.
We used manipulated arrays of flowers to examine the effect on pollen receipt of reduced flower density and increased isolation from a conspecific pollen source. We treat isolation from a conspecific pollen source and plant density to be naturally confounded variables because on the most local scale, both are determined by interplant distance. Experimental arrays were deployed in sections of roadside vegetation, both sides of which had been cleared of all Dianella inflorescences for an 800 m length. We placed vases at the 0, 20, 50, 100, 200 and 400 m points in each section (we refer to these points as ‘stations’) with the zero point being at the end closer to the reserve. Thus, with increasing distance from the 0 position, isolation from the pollen source increased and D. revoluta flower density declined.
Arrays were deployed at three roadsides, with each roadside hosting an array immediately adjacent to the nature reserve (‘near’) and another 1–5 km away (‘far’) (Fig. 1). The location of the ‘far’ arrays was selected to ensure that the structure and composition of the vegetation was similar to that in ‘near’ arrays and that the location was more than 1 km from the reserve or any other substantial patch of native vegetation. The land surrounding the reserve and the roadside vegetation was sown to wheat or pasture and therefore both devoid of native plants and poor habitat for native bees. For the ‘near’ arrays the reserve provided a large potential source of D. revoluta pollen and pollinators. The next nearest D. revoluta were in the roadside vegetation 400 m past the end of the array. For the ‘far’ arrays the nearest source of D. revoluta pollen and pollinators was the roadside vegetation adjacent to the 0 station, and the next nearest D. revoluta were in the roadside vegetation 400 m past the end of the array. Replicating arrays adjacent to, and remote from, the nature reserve allowed us to test if pollen delivery was influenced by proximity to the nature reserve. In late October 2001 arrays were set up at the three ‘near’ sites and two were repeated on a second date. All three far sites and two of the near sites were used again in 2002, with three or four replicates run at each over a period of 6 days between mid October and early November of 2002 to give a total of 16 arrays (Table 1).
Table 1. Number of arrays per site and year, and number of flowers contributed by each array to the GLM analysis. Loss of stigmas to flower feeding beetles (Arsipoda sp.) reduced sample size unevenly across the arrays
No. of arrays
No. of flowers
Lake Road (near)
Lake Road (far)
NE Track (near)
NE Track (far)
Slant Road (near)
Slant Road (far)
On experiment days we collected D. revoluta inflorescences from sites more than 5 km away from the study area. Inflorescences were collected early in the morning while flowers were still opening and before pollinator activity had commenced. Cut inflorescences were immediately placed in a vase and visitors were excluded while the flowers opened. Anthers were blocked on these newly opened flowers. Each vase was given sufficient inflorescences to provide six open flowers, and one vase was placed at each station, with all arrays being established before 10.30. Because the 2002 season had less flowering activity, we provided additional D. revoluta flowers so that the amount of conspecific pollen available near the zero point was approximately equal (by visual estimate) to that observed in the previous season. This was achieved by placing a vase of 15–20 open, non-emasculated flowers approximately 5 m from each ‘0’ station, outside each array (near and far).
Experimental inflorescences were left open to visitors until 15.00 (corresponding with the time that D. revoluta flowers naturally begin to close), after which time flowers were collected and stored in 70% ethanol. In the laboratory, stigmas were excised, rinsed for 10 minutes in distilled water, and then stained for 3 minutes in modified Calberla's solution (Ogden et al. 1974). Dianella revoluta pollen was easily distinguished from other species by morphology and level of staining. We counted the number of D. revoluta pollen grains and heterospecific pollen grains on each stigma, using a light microscope at 20× magnification. Because all the pollen deposited on the stigma arrives via pollinators (Duncan et al. 2004), we assume that the magnitude of pollen receipt (conspecific + heterospecific) broadly scales with the amount of visitation.
To test for the possibility that pollen might reach the most distant stations within an array indirectly, via intermediate stations, we established reduced arrays, with stations only at 0 and 400 m. This was repeated four times over 3 days and three sites. The difference in mean number of D. revoluta pollen on stigmas of flowers from reduced arrays was similar to the full arrays (data not shown). We concluded therefore that long-distance pollen transfer via experimental flowers was negligible.
The stigmatic pollen load data set was comprised of D. revoluta and heterospecific pollen grain counts for all site × date × distance replicates. Flea beetle (Arsipoda spp.) attack consumed almost 40% of stigmas from the experimental arrays; therefore we recorded the number of stigmas lost per station (range 0–6) for separate analysis. Pollen load data were adjusted for stigma loss by treating chewed flowers as missing values. Generalized linear models (GLMs, 1McCullagh & Nelder 1989) were fitted to the data using GenStat for Windows, Release 6.2 (VSN International Ltd, Hemel Hempstead, UK). For pollen count data we used a log linear model with variance proportional to the mean. For the stigma loss data we used a logistic link function and variance proportional to that which would have been appropriate if a binomial distribution had been assumed. Both these models allow for the variability between stations to be greater than it would be if individual plants at a station were completely independent. For graphs we present fitted means and standard errors from the models unless otherwise specified.
For model fitting we added main effects sequentially in the order: distance from nearest conspecific pollen source (‘Distance’, six levels), proximity to reserve (‘Proximity’, two levels), Year (two levels), and Site (three levels). For pollen counts sampling date (‘Date’, eight levels) was also included. Date was nested within Year because individual dates in the first year were not repeated in the second year and differences between years were of greater magnitude. We then used stepwise selection to add interaction terms that were significant at the 0.1% level and had a mean deviance ratio greater than 4. These selection criteria were adopted to avoid the inclusion of spurious terms as a result of selection bias.
We obtained stigmatic pollen load data from 455 flowers, approximately 60% of the total number deployed in the experiment. The remainder were lost to flower chewing Arsipoda beetles. Outcross conspecific pollen loads ranged from 0 to 170 grains and fewer than 10% of flowers had no conspecific pollen at all. There was a greater range in heterospecific pollen loads (0–410 grains) but a larger proportion of flowers (almost 20%) had no heterospecific grains. Pollen of at least six species contributed frequently to the heterospecific pollen load, including Senna artemisioides (Caesalpiniaceae), a sympatric buzz-pollinated species, and Helichrysum sp. (Asteraceae). We did not identify all pollen grains to donor species as our focus was on D. revoluta pollen.
conspecific outcross pollen load
We found a significant decline in D. revoluta outcross pollen load with increasing distance from the zero point of the arrays (P < 0.001, Table 2 and Fig. 2a). In other words, conspecific pollen load declined as isolation increased and D. revoluta density decreased. The largest difference in mean values occurred at the shortest distance classes, with relatively little further decline with distance beyond 50 m. A few flowers at 400-m stations had very high pollen loads, accounting for an increase in overall mean between 200-m and 400-m sampling points. The decline with distance was consistent over the three roadsides; however, there were differences amongst locations in the rate of decline (distance × site, P < 0.001, Table 2). Such natural variation between sites is to be expected.
Table 2. Summary of generalized linear models (GLMs) of outcross conspecific pollen per stigma, heterospecific pollen per stigma, and stigma loss per station, from manipulated roadside arrays of D. revoluta. Main effects and significant interactions with the highest deviance ratio values are displayed. *0.05 > P > 0.01; **0.01 > P > 0.001; ***P < 0.001
Surprisingly, flowers in remote arrays (> 1 km from the nature reserve) had higher conspecific pollen loads on average than those in adjacent arrays (P < 0.001, Table 2, Fig. 2b). This pattern does not contradict the observed decline in pollen load with distance within arrays, as it reflects a broader geographical scale. Higher mean pollen loads in remote arrays were consistent across three roadside comparisons; however, the difference was smaller in Lake Road data (Fig. 2b). There was a significant distance–proximity interaction term (P = 0.040) that reflected a stronger decline of D. revoluta pollen with distance in remote arrays compared with adjacent ones.
One possible explanation for the decline in D. revoluta pollen with distance within arrays would be a decline in pollinator visitation. To test this idea we used the presence and abundance of heterospecific pollen to broadly infer visitation. We plotted means of D. revoluta conspecific pollen load against distance for four levels of heterospecific pollen load (Fig. 3). These four levels of heterospecific pollen represent the observed range of values on a log scale. Conspecific and heterospecific pollen loads were positively correlated; at any given distance higher D. revoluta pollen was associated with high heterospecific pollen loads (Fig. 3). However, for any given value of heterospecific pollen, there was a decline in D. revoluta pollen with distance from the conspecific pollen source. In other words, the proportion of D. revoluta pollen in the total stigmatic load was greater at stations closer to the pollen source, but the abundance of heterospecific pollen at all distances indicated there was no decline in flower visitation.
mean heterospecific pollen load per flower
There was a positive effect of distance on heterospecific pollen, such that flowers at more distant stations received more heterospecific pollen than those at the nearest stations (Table 2, P < 0.001). The low deviance ratio compared with other main effects reflects that although significant, the pattern of increase was not pronounced. Most importantly, that heterospecific pollen loads did not decrease with distance suggests that flowers at the most isolated stations within an array received at least as much visitation as those at shorter distances. Similarly, mean heterospecific pollen loads were higher at flowers located away from the nature reserve (far), compared with those in adjacent (near) arrays (Table 2, P = 0.041). The most important factor driving variation in heterospecific pollen load was year, with higher values in 2001 (Fig. 4, P < 0.001). This pattern reflects the reduced amount of flowering for many plant species in the 2002 season.
Large differences in the size of stigmatic pollen load among dates within years (Table 2, P < 0.001) were common to both conspecific and heterospecific pollen load data. We were not concerned to understand the factors underlying this variation, but note that it did not appear to correspond to variation in weather data (temperature maxima and minima, wind speed, rain; Data from Bureau of Meteorology, not shown). Most importantly, there were no interactions between date and any other main effects indicating that although the amount of pollen varied substantially according to date, it did not alter the pattern.
In the 2001 season approximately 25% of stigmas were removed by flea beetles; however, loss was considerably greater in 2002, with almost 50% of stigmas chewed off (Table 2, P = 0.004). Stigma loss was significantly higher in roadside arrays adjacent to the nature reserve compared with arrays far from the reserve (fitted means for stigma loss ± SE, 54 ± 4% for near, 27 ± 4% for far, P < 0.001, Table 2). There was significant variation between sites, with arrays in NE Track having higher levels (> 50%) of stigma loss than the other two roadsides, which lost approximately 30% (P < 0.001). There was no decline in stigma loss with distance within arrays (P = 0.242), suggesting that density of D. revoluta flowers at that scale (there were more stations at shorter distances) had no effect on the flower-chewing flea beetles.
We found that the probability of outcross pollen transfer declined over distances between 0 and 400 m in the arrays. Frequent pollinator movements between D. revoluta and other co-flowering species, as indicated by high heterospecific pollen loads, suggest that much pollen may have been lost to the stigmas of other coflowering species, making conspecific pollen transfer ineffective over longer interplant distances. This pattern suggests an important role for conspecific flower density in maintaining outcrossing. In contrast, pollinator visitation to flowers was relatively unaffected by reduced density of the target species, and native pollinating bee populations were sufficient to visit most D. revoluta flowers in this fragmented landscape. Evidence for this comes from the high heterospecific pollen loads, even on flowers at isolated stations within arrays, and high levels of heterospecific pollen in arrays more than 1 km from the reserve. In contrast, flower predators damaged more flowers adjacent to the reserve compared with those in far arrays, suggesting that the fragmented landscape was less favourable for flea beetles than pollinating bees.
isolation effects on outcross dianella revoluta pollen movement
This study demonstrated isolation effects on outcross pollen receipt in manipulated field arrays of D. revoluta. The observed decline was manifest in two seasons with different weather conditions and consistent across roadside arrays located adjacent to, and distant from, a nature reserve. We are aware of only one other study that examines plant-density effects on outcross pollen deposition; Waites & Agren (2004) similarly show that the proportion of outcross pollen on stigmas was greatest in bigger populations of Lythrum salicaria, a species with heteromorphic incompatibility. In contrast, studies that examined total conspecific pollen load (self + outcross) have often found no relationship between plant density measures and stigmatic pollen load (e.g. Aizen 1997; Bosch & Waser 1999; Molano-Flores & Hendrix 1999; Caruso 2002; but see Heithaus et al. 1982). However, declines in seed set at low density in the same studies led the authors to conclude that an increased proportion of self-pollen in sparse populations was one likely explanation (e.g. Bosch & Waser 1999; Molano-Flores & Hendrix 1999).
It is not possible to discriminate a decline in pollen availability from a decline in pollinator availability unless pollinator service is monitored (e.g. Groom 1998; Gigord et al. 1999). However, consistently high heterospecific pollen loads rule out pollinator limitation as a cause of reduced outcross pollen loads in this study. We concur with Kunin (1993) that effects on plant reproductive traits are more likely to be mediated through reduced delivery of compatible outcross pollen. Although fragmentation might affect some specialized pollinators, most plants, D. revoluta included, are visited by a range of pollinators, and therefore have insurance against declining availability of one species (Ashworth et al. 2004). Pollinator limitation has, however, been demonstrated in other species. There is also evidence linking habitat fragmentation with decreases in local and cumulative diversity of pollinator fauna (reviewed in Aizen & Feinsinger 2002), and pollinator limitation could become more important if such effects prove to be widespread (e.g. Allen-Wardell et al. 1998).
Our study tested for isolation effects over relatively long distances, with only one station per array within 5 m of the pollen source (i.e. the zero point). This is a more sparse arrangement than in many other studies (e.g. Kunin 1993). The largest differences in pollen receipt between stations, however, occurred within 50 m and often in the first distance interval between 0 and 20 m (Figs 2 and 3). On the one hand we show strong effects over short distances that are applicable to many natural populations, but on the other hand our study suggests that flowers at the most isolated distance station (400 m) still received some outcross pollen. Solitary halictid bees are thought to forage over distances of hundreds of metres (Eickwort & Ginsberg 1980), but the high levels of heterospecific pollen deposited might have created an expectation that all D. revoluta pollen would have been lost to co-pollinated species during intervening flower visits.
larger scale effects on pollen movement and pollinators
When comparing data from arrays adjacent to reserve (‘near’) to those more than 1 km away (‘far’) we found no evidence of lower pollinator activity with increasing distance from the nature reserve. Previous work (Duncan et al. 2004) suggested that the pollinator fauna was relatively homogenous across our study area over several seasons. Solitary anthophorine apid and halictid bees, the major pollinators of Dianella revoluta, are central place foragers (Roubik 1989) and the females may return to their nest in between foraging bouts. If the pollinators were only rarely nesting in roadside vegetation but commonly foraging from a home base within the reserve, the amount of pollen on stigmas in far arrays would have been expected to be lower on average than at a corresponding distance in arrays adjacent to the reserve. In fact, we found that mean pollen loads were higher in distant arrays, possibly reflecting an increase in the probability of an individual flower being visited by foraging bees as floral density of the target species decreases. ‘Near’ arrays may thus be less attractive to bees relative to the larger floral resource of the reserve, whereas flowers in ‘far’ arrays are probably the largest floral resource locally available. Our results suggest that native pollinators are commonly active within roadside remnants in this area and that it is not unusual for them to move between flowers that are 400 m apart. This pattern of pollinator behaviour makes it less likely that habitat fragmentation will completely disrupt the connectedness among patches (Murren 2002).
Within roadside arrays we can reasonably extrapolate heterospecific pollen loads to amounts of visitation. Across the whole experimental landscape (near vs. far sites), however, the relative density of co-pollinated species in the immediate floral neighbourhood, which can strongly influence pollination success of target plants (Feinsinger et al. 1991; Kunin 1993), may come into play. Although a floristic survey of vascular plant species revealed no differences in species composition, it did not assess flowering intensity, which might also influence pollinator activity. Heterospecific pollen load must reflect the local floral neighbourhood, at least at the broadest level. In 2002 climatic conditions led to relatively fewer species flowering and, unsurprisingly, there was less heterospecific pollen found on D. revoluta stigmas. There was a less pronounced decline in D. revoluta pollen loads but this may be because we supplied additional pollen_source flowers for the second season.
why so much heterospecific pollen?
Heterospecific pollen deposition may constitute a form of interspecific competition (Rathcke 1983), either from stigma clogging (Waser 1978) or pollen allelopathy (Sukhada & Jayachandra 1980), or might, in some cases, facilitate pollination (Rathcke 1983). Heterospecific pollen was found on over 80% of flowers in this experiment, and in other experiments involving D. revoluta (D. H. Duncan, unpublished data). Although we cannot estimate the actual number of visits flowers received during the course of the experiment, we can estimate the total amount of pollen deposited during the life of a flower (1 day), and see that it does not decrease with isolation.
There is great variation in the amount of heterospecific pollen that species receive, varying from the occasional grain (Snow 1982) to ‘coatings’ (Groom 1998). McLernon et al. (1996) found that the larger a species’ pollen, the more heterospecific pollen it was likely to receive, possibly because stigmatic papillae usually correspond to a species’ pollen dimensions. Stigma of species with large pollen are therefore likely to hold pollen of smaller-grained species but not vice versa. There were indeed more smaller-grained species represented on D. revoluta stigmas, whereas large heterospecific grains were rarer.
High levels of heterospecific pollen suggest that at least some of the flower visitors to D. revoluta have low fidelity. In buzz-pollination, like other animal-pollinated systems, pollinator generalization rather than specialization is the norm (Waser et al. 1996; Harter et al. 2002), although this does not rule out pollinator constancy (preferentially flying between conspecifics) within foraging bouts (Michener 2000), as observed in 43% of flights between flowers of sympatric species by Gross (1992). In our study heterospecific pollen loads were proportionally lower where D. revoluta floral density was highest (Fig. 2), implying increased pollinator constancy.
Although the design of our experiment focused on pollination, the impact of floral predators was clearly substantial enough to have an important impact on reproduction (see also Cunningham 1995, 2000b). Whereas we found significant effects of both distance (0–400 m) and proximity (near vs. far) in pollen load data, flower attack by Arsipoda flea beetles only varied significantly at the larger spatial scale, with significantly less flower attack at arrays distant from the reserve. Although we were unable to relate variation in frequency of flower damage to climatic or floristic factors, there are other examples of florivores or seed predators having reduced impact at more fragmented or isolated sites or at lower plant densities (Silander 1978; Cunningham 2000a). Flower predators, unlike pollinators, appear not to find the roadsides suitable habitat.
We thank Meredith Cosgrove, Kelli Gowland, Kaska Hempel, Martin Henery, Andrea Leigh, Ben Moore and Cassia Read for assistance in the field and laboratory. We thank Tom Weir for identification of flea beetles. This manuscript was improved by the suggestions of Peter Klinkhamer, Tim Heard, Joslin Moore and two anonymous referees. An award from the Australian Flora Foundation to DHD, and ANU FRGS funding to ABN and SAC supported this study. The NSW National Parks & Wildlife Service kindly gave permission to carry out the work in reserves and roadsides of the study area.