The study was conducted at Passo do Lontra (19°34′19.01” S, 57°01′53.05” W) in the world’s biggest tropical wetland, the Pantanal, in the state of Mato Grosso do Sul, Brazil in August 2018. Data were collected during three consecutive days from 7:30 to 11:00 am and from 1:00 to 4:30 pm during the dry season of a flooding site. We randomly assigned 38 independent plots of 1 m2 located in our study site and counted the number of L. canescens individuals, inflorescences and the number of yellow flowers per plot. Plots were either located in the sunlight or in the shade.
Lantana canescens flowers are united in inflorescences with fertile yellow flowers and infertile white flowers that are two to three days old. Flowers are zygometric and petals are lobed. Flower production starts during the dry season in August and lasts until the beginning of the wet season in January (Pott and Pott 1994). The plant produces scents (Pino et al. 2011), that seem to be attractive to a large set of insects, especially to bees (Pott and Pott 1994; Boff et al. 2013).
Each of the 38 plots was observed by two collectors for 10 min and all floral visitors were collected. Rarefied species accumulation curve and variation in species composition between plots was evaluated by non-metric multidimensional scaling (NMDS) with the R package vegan (Oksanen et al. 2019). We calculated the Bray-Curtis distances between all pairwise combinations of plots by considering the relative frequency of visits.
Individual and populational aspects
We considered the total number of inflorescences (floral display) per plot (n = 38) as a parameter for long-distance attraction (lda). Thus, the higher the number of inflorescences (whorls) inside a plot, the bigger the floral display.
The effect of floral color pattern on the frequency of flower visitors was assessed first by calculating the ratio between yellow and white flowers in 39 randomly chosen inflorescences within our 38 study plots. The number of yellow flowers varied between inflorescences. It ranged from zero “0” (absence of yellow flowers) to a maximum of eight yellow flowers per whorl. In 25 of the 39 inflorescences we found fewer yellow flowers than white flowers. In the remaining 14 inflorescences the number of yellow flowers was equal or higher than the number of white flowers (Fig. 1). On average the number of white flowers in the population was three times higher than the number of yellow flowers (t-test: t = 4.26, df = 38, P < 0.001), in other words 75% of the flowers were white and 25% yellow.
We used a linear model (lm: number of yellow flowers per inflorescence ~ number of white flowers per inflorescence) to assess the ratio between yellow and white flowers per inflorescence in the whole population, using the car package in R (Fox and Weisberg 2019). The residuals of this model were used to estimate short-distance attraction (sda) which indicates the number of yellow flowers in relation to the number of white flowers. Higher residuals indicate a higher number of yellow flowers (yellowing effect of flowers in the plot), lower residuals indicate a higher number of white flowers, and a residual value of 0 indicates an equal number of white and yellow flowers (Electronic Supplementary Material 1).
Disentangling long and short distance pollinator attraction
We performed two analyses of covariance (ancova) to test the effect of each independent variable, number of inflorescenses (lda), ratio between yellow/white flowers (sda) and their interaction with sunlight, on the visitation rate. Interactions between lda and sunlight as well as the interaction between sda and sunlight were both not significant (F1,34 = 0.046 with P = 0.83 and F1,34 = 0.791 with P = 0.38, respectively) and were removed from the models. Furthemore, we tested the effect of lda on the number of flower visits in sunny and in shaded plots with linear regression models (lm: number of visits per plot ~ lda + number of floral visitor species in sunny plots; and lm: number of visits per plot ~ lda + number of floral visitor species in shaded plots, respectively).
Structural equation model (SEM)
SEM provided a visual representation of our system with statistical key parameters that describe the relationships between the response variable (visitation rate) and each predictor variable (sda, lda, and sunlight) using the lavaan and semPlot R packages (Rosseel 2012; Epskamp and Stuber 2017).