Optimal floral display (the number of simultaneously open flowers) is frequently viewed as a balance between pollinator attraction and avoidance of pollinator-mediated interfloral selfing (geitonogamy). However, the most beneficial display size may be affected by pollinator abundance, pollinator identity, and other environmental variables. We determined the effects of individual floral display and patch size on pollination outcomes in natural populations of a mixed-mating annual plant, Clarkia concinna. We predicted that individual plants with larger floral displays would have both greater pollinator visitation and more geitonogamy. Additionally, we tested whether the effects of floral display would be greater in smaller patches of plants. Larger individual floral displays increased total geitonogamy, but the effect of display on geitonogamy differed among pollinator taxa. Visitation rates were lower in small patches for some pollinators. However, individual display size did not affect per flower visitation rate in small or large patches. Therefore, if we only consider the expected tradeoff between pollinator attraction and geitonogamy, the number of simultaneously open flowers appears excessive. In this species, large floral display size requires explanations beyond increased pollinator attraction, such as temporal constraints on flowering imposed by a limited growing season and the benefits of additional low-quality offspring.

In the springs of 2014 and 2015, we conducted five studies on the floral display size, patch size, and pollinator assemblage of Clarkia concinna in northern Napa, southeastern Lake, and northwestern Yolo Counties, California. Here, Clarkia concinna grows in mixed serpentine chaparral and riparian woodland communities. Mixed serpentine chapparal is dominated here by evergreen shrubs such as Arctostaphylos viscida, Ceanothus jepsonii, and Quercus durata, and includes a Pinus sabiniana overstory. Riparian woodlands are comprised of a fairly dense but generally low canopy of Populus fremontii, Aesculus californiaca, Cercis occidentalis, Salix laevigata and additional medium-sized trees, with a shrub layer of Rhus trilobata, Rosa californica, and others. These habitats grade into each other in locations where the rocky serpentine hillsides meet ephemeral or permanent streams. In both habitat types, Clarkia concinna populations occur patchily on steep slopes and road cuts with shallow soil and limited additional surrounding vegetation. In both communities, Clarkia concinna plants grow to similar sizes and are visited by similar species—all floral visitors with more than five total observations in the study were seen in both chaparral and riparian woodlands. Studies 1, 4, and 5 were conducted at 6 populations in both plant communities (Table 1). Studies 2 and 3 occurred at Davis Creek, a very large population (> 100,000 plants) growing in riparian woodland in the Donald and Sylvia McLaughlin Natural Reserve. All studies in both years were restricted to patches in peak flowering, to minimize temporal changes in pollinator abundance (Thompson 1981; 1982). Studies 1, 2, and 3 looked for an overall effect of floral display size on pollinator attraction and geitonogamy by comparing of pollinator movement, pollen analogue (pigment) deposition, and seed traits. Studies 4 and 5 tested whether the effect of floral display size depended on patch size by comparing visitation and pollen deposition between small and large patches.

Clarkia concinna is a self-compatible annual plant native to a variety of habitats throughout California’s Northern Coast Ranges with disjunct occurrences in the Northern Sierra Foothills. Individuals produce 1 to more than 100 four-merous, fuchsia-colored, lobed and clawed flowers (average of 6 in natural populations) on leafy compound spikes. Flowers are protandrous and self-compatible, and fruits are elongate capsules containing an average of about 50 ovules (range 1-124; Groom 1998). Seeds germinate at the beginning of the winter rains and plants bloom for a few weeks in May or June. Across the species range, mean precipitation during its growing season (October-June) varies from approximately 500 mm in the southeast to 1800 mm in the northwest.

Study 1) How does floral display size affect pollinator movement?

In order to compare pollinator movement among plants with different floral display sizes, we observed floral visitors to C. concinna during peak flowering in 2014 (May 21^st through June 8^th). Observations took place throughout the day 7:40-19:20. Each observation began with a visitor arriving at a new plant and ended when it left the patch or stopped foraging for approximately 15 seconds. A visit included any contact with the interior surface of the corolla or reproductive parts. Most observed visitor taxa were found to be effective pollinators in a previous study (Miller et al. 2014). We counted the number of flowers visited per plant after the initial flower was visited, giving the total visits for which geitonogamy could occur (consecutive visits). Plants with one open flower could not have consecutive visits and were not included. We used numbered flags to temporarily mark visited plants and counted the number open flowers per plant (floral display size) after the completion of the foraging bout. Rare instances when visitors returned to the same flower or plant after visiting elsewhere were counted as separate visits. We identified all visitors to genus. Visitor preference for either male or female phase flowers could reduce the effectiveness of consecutive visits as a measure of geitonogamy. In initial observations, we generally did not observe such bias in flower choice except for two small solitary bee genera, Cerotina and Lasioglossum. These bees were uncommon visitors that exclusively pollen-collected on male-phase flowers. Emasculation did not reduce visitation in a study of the related Clarkia xantiana, supporting the idea that visitors do not cue in on flower phase (Moeller et al. 2012). Floral display size and consecutive visit data were approximately lognormally distributed, and were log transformed. We tested for a positive linear relationship between the log of floral display size and the log of consecutive visits. To test for variation among pollinator species in this relationship, we selected the eight pollinator genera with more than 50 observations. We compared the log of consecutive visits made by visitors among these genera using an ANCOVA with log of display size as the covariate.

Study 2) How does floral display size affect pollen deposition?

Although high numbers of within-plant transitions strongly indicate the potential for geitonogamy, high pollen carryover or erratic pollinator behavior may reduce the relative amount of within-plant pollen transfer (Geber 1985). We compared the amounts of self and outcross pollen movement across floral display sizes by manipulating display and comparing the deposition of fluorescent pollen pigment. From May 18^th through June 7^th, 2014, we selected 140 focal plants in large patches (greater than 250 open flowers) that had at least 6 male phase and 1 female phase flowers. We marked the most recently female phase flower on the main stem (highest) with thread and removed any other female phase flowers on the plant by pinching off the perianth just above the ovary. We chose plants with no prior pollen deposition on the focal flower stigma. Pollen in this species is large and colorful enough in this species that we could observe deposition with a hand lens. We varied floral display size by leaving 1, 2, 4, or 6 of the male phase flowers on the focal plant and removing additional flowers. Although large plants can have greater than six simultaneously open male-phase flowers, the range of treatment display sizes was comparable with most natural display sizes. We used a pipe cleaner to brush all anthers of the remaining male-phase flowers with blue, orange, green, yellow, or pink fine-grained florescent powdered pigment (self pigment; Series JST-300, Radiant Color, Richmond, CA, USA). We coated the anthers of the 50 nearest male phase flowers with a different color of pigment (outcross pigment). The surrounding treated flowers were within a one-meter radius of the focal plant. Replicates were at least 10 meters apart and pigment colors were changed between adjacent replicates. The presence of pigment did not appear to affect visitation—we observed all common visitors foraging freely on treated flowers. Treatments were initiated between 7:00 and 12:00.

After 6-10 hours, we collected the focal stigma and quantified the amount of self and outcross pigment deposition the same evening using a dissecting microscope. There was no effect of time on pollen deposition (r² <0.01, N = 136, P = 0.81) Three stigmas were lost or eaten, resulting in sample sizes per treatment of 34 or 35. We separately ranked focal and surrounding pigment deposition as: no pigment (0), 1 fluorescent grain to approximately 5% stigma surface covered (1), 5% to 20% coverage (2), 20% to 40% coverage (3), and >40% coverage (4). Pigment as a pollen surrogate may overestimate pollen movement—its smaller size may result in higher amounts of wind transfer (Thompson et al. 1986) However most pigment deposition was associated with pollen deposition, indicating the pigment is an adequate pollen surrogate. We tested whether the proportion of stigmas receiving any pigment depended on display size with c² contingency test. For stigmas that received pigment (N = 105), we compared self and outcross pigment scores among display sizes using Kruskal-Wallis and post-hoc Steel-Dwass All Pairs tests (the nonparametric equivalent of a Tukey test).

Study 3) How does floral display size affect seed size and number?

If large floral displays increase the deposition of self-pollen, then floral display size may be negatively correlated with offspring quantity (fewer seeds matured) or quality (smaller seeds) due to early-acting inbreeding depression. Alternatively, plants may be able to discriminate against self pollen, and display size will not affect seed set (Jones 1994). We tested whether floral display size affects seed weight or number, by selecting 130 total plants on May 19^th and May 30^th, 2014. Focal plants were separated by at least 5 meters and had one unpollinated freshly-female phase flower (confirmed with a hand lens) and 6 or more male phase flowers. On each plant, we removed all other female phase flowers and left 0, 1, 2, 4, or 6 male phase flowers. We also removed buds that were likely to bloom while the focal flower was still open. Because plant size and flower position affect seed traits in this species, we measured the basal stem diameter with calipers (a proxy for size), and the position of each focal flower along the main stem (with position 1 being the lowest and first to bloom). After corollas senesced, we placed a small piece of transparent adhesive tape around the top of the capsule to prevent seed loss. Mature fruits were collected on June 25^th, 2014, and ripe seeds were counted and weighed collectively after air-drying. Twenty-four fruits were missing or eaten prior to collection, resulting in per-treatment sample sizes between 19 and 23 plants. We compared weight per seed and seed number among floral display size classes with an ANCOVA, using stem diameter and stem position as covariates.

Study 4) How does patch size affect the pollinator environment?

The previous three studies were conducted in relatively large patches of plants (hundreds to thousands of individuals). However, the relationship between display size, visitation, and geitonogamy may depend on patch size. Therefore, we tested for an effect of patch size on visitor frequency and identity with an observational study. During peak flowering in 2015 (May 2^nd to May 19^th), we observed groups of 20 flowers centrally located in varying patch sizes for 121 20-minute observation periods between 10:00 and 16:30. Patches ranged from 20 to approximately 10,000 open flowers, and we did not reuse the same patch on the same day. Patches were groups of Clarkia concinna separated from others by at least ten meters. We identified all visitors to genus and counted the total flower visits and visits per taxa. We used a generalized linear model (GLM) to test for a relationship between the log of patch size and total visits. We also included observation date as a predictive variable. We modelled visits with a Poisson distribution and a log link function. In addition to total visits, we ran separate GLMs for Apis and Bombylius, the two taxa with more than 100 visits in the data set. Finally, for observation periods with at least one visitor (N = 87) we conducted a regression analysis of the Shannon-Weiner species diversity index against the log of patch size, including date as a random variable.

Study 5) Do the effects of floral display size depend on patch size?

If pollinators are less abundant in small patches, plants with small displays may be pollen limited in small, but not large patches. We manipulated both patch size and floral display size to test for a significant interaction between these variables. From May 2^nd to May 17^th, 2015, we selected single unpollinated (confirmed with a handlens) freshly-female flowers on a total 204 focal plants with at least 6 male-phase flowers. For half the plants, we created very small patches (< 50 additional open flowers within 2 meters of the focal plant, mean 14.8, s.d. = 8.9). In order to avoid excessive destruction of plants, we used natural patchiness in plant distribution, supplemented by thinning excess plants within a 10-meter radius of a focal plant. Other focal plants were chosen in large patches (> 250 flowers).  Focal plants in small and large patches received one of 3 display treatments: 1) all additional open flowers and large buds removed, 2) six unmanipulated male-phase flowers left on the plant, or 3) six male-phase flowers left on the plant, but emasculated by removing the anthers with forceps. For the treatments with six unmanipulated and six emasculated flowers, we maintained the display size until the focal flower senesced by removing flowers as they aged and controlling the number of newly opening flowers daily, emasculating them in bud in the emasculated treatment. Unlike studies 2 and 3, we emasculated the focal flower, ensuring any pollen deposition was pollinator mediated. Differences between the emasculated large display treatment (3) and the small display treatment (1) are attributable to differences in display attraction. Differences between the emasculated large display treatment (3) and the unmanipulated large display treatment (2) are due to within plant pollinator movement.

An equal number of replicates per treatment (2 or 3) were initiated on each of 13 days between May 2^nd to May 17^th. Replicates were set up between 7:00 and 10:00, before the bulk of pollinator activity, and ran until the corolla senesced (1-5 days). Concurrently running replicates were spaced more than 10 meters apart. Four focal flowers were eaten, but they were replaced the next day, resulting in even sample sizes of 34 per treatment. We checked focal flowers for pollen deposition with a hand lens and corolla senescence 3 times per day, between 7:00 and 9:00, between 12:00 and 14:00, and between 17:00 and 19:00. We calculated the time to initial pollen deposition and time to corolla senescence to the nearest third of a day. Once the corolla wilted, we collected the stigma. The same evening, we applied basic fuchsin jelly (Kearns and Inouye 1993) to a collected stigma, heated it, squashed it between two microsope slides, and counted the stained pollen grains under a dissecting microscope. We used Cox Proportional Hazards to test for an effect of patch size, display size, and their interaction on time to pollen deposition and time to senescence. We also included date as an independent variable, as among-day differences in weather and location could affect the response variables. We conducted an ANOVA to test for an effect of patch size, display size, their interaction, and the random effect of date on the number pollen grains deposited on focal stigmas.