The activity of many animals follows recurrent patterns and foraging is one of the most important processes in their daily activity. Determining movement in the search for resources and understanding temporal and spatial patterns in foraging has therefore long been central in behavioural ecology. However, identifying and monitoring animal movements is often challenging. In this study we assess the use of camera traps to track a very specific and small-scale interactions focused on the foraging behaviour of Heliconiini butterflies. Data on floral visitation was recorded using marked individuals of three pollen-feeding species of Heliconius (H. erato, H. melpomene and H. sara), and two closely related, non-pollen feeding species (Dryas iulia and Dryadula phaetusa) in a large outdoor insectary. We demonstrate that camera traps efficiently capture individual flower visitation over multiple times and locations and use our experiments to describe some features of their spatial and temporal foraging patterns. Heliconiini butterflies showed higher activity in the morning with strong temporal niche overlap. Differences in foraging activity between males and females was observed with females foraging earlier than males, mirroring published field studies. Some flowers were more explored than others, which may be explained by butterflies foraging simultaneously affecting each other’s flower choices. Feeding was grouped in short periods of intense visits to the same flower, which we refer to as feeding bouts. Heliconius also consistently visits the same flower, while non-Heliconius visited a greater number of flowers per day and their feeding bouts were shorter compared with Heliconius. This is consistent with Heliconius having more stable long-term spatial memory and foraging preferences than outgroup genera. More broadly, our study demonstrates that camera traps can provide a powerful tool to gather information about foraging behaviour in small insects such as butterflies.

Experimental cage

The experimental cage consisted of a circular dome of approximately 12 m in diameter and 10 m in height covered with black mesh and located outdoors (Figure 1A). Inside the dome, motion-activated cameras (Mini Wildlife Trail Camera Version 18022, K&F Concept ©, Shenzhen, China) were positioned 30 cm away from a floral resource to be able to detect the small body of a butterfly (Figure 1B). To permit adequate focus at close proximity, extra external lens (half frame of +3.00 glasses, Figure 1B and Supp Figure 1) were added in front of the camera lenses, an adjustment which provided a clear image of a marked butterfly (Figure 1C and Video 1). Cameras were set to have a 0.2 sec trigger time, maximum motion sensitivity, and to record 20 sec videos between 8 am and 4 pm, the period of highest activity for Heliconius.

Each camera was pointed towards one flower in each plant cluster, created by tightly clumping a group of potted plants inside the dome using Palicourea tomentosa (synonyms: Psychotria poeppigiana and Cephaelis tomentosa), which has terminal inflorescences that are capitate with red bracts and yellow flowers (Figure 1C and 2). The pollen of this flower is documented to be highly consumed by Heliconius (Estrada and Jiggins, 2002) and visited by other Heliconinii species that exploit it as a nectar resource (personal observations). This permitted comparable data collection across both pollen and non-pollen feeding species. Palicourea tomentosa blooms are stable and produce new flowers each day at the same spot providing a temporally reliable resource (Coelho and Barbosa, 2004; Valois-Cuesta, López-Perea and Quinto-Valoyes, 2009). During the experiment, a few terminal inflorescences stopped blooming, and the camera position was therefore adjusted to capture another flower in the same cluster. Plant clusters were created to space out the cameras around the dome (Figure 2).

Experimental procedure

At the beginning of each experiment adults were moved to the experimental cage, where butterflies could fly freely, with access to floral and hostplant resources in semi-natural conditions. We performed two experimental trials: i) In the first trial, performed from November to December of 2022, five plant clusters (with ~ 5 available flowers) were created inside the dome in a pentagon shape, 2.8 m between the clusters, each with one camera trap adjacent to a flower (Figure 2). In total, 15 days of data were collected with 58 individuals released inside the dome (16 H. erato, 16 H. melpomene, 14 H. sara and 12 D. iulia). ii) In the second trial, we aimed to reduce competition that was observed during the first trial by arranging fewer flower clusters more distantly from one another, with four plant clusters organized in a rectangle, with 5 to 7 m between clusters (Figure 2). In this trial, butterflies were divided into five groups of two species (total of 19 H. erato, 13 H. melpomene, 11 D. iulia and 11 D. phaetusa), always pairing one Heliconius species with a non-Heliconius species with the same number of individuals. Each group was held in the dome for 10 days, and cameras were active during the last 5 days. Groups were run consecutively between March and May of 2023.

Video and data analysis

During the trials, each feeding event detection by the camera trap produced one video. These videos were reviewed and annotated using the image analysis software Timelapse 2 (Greenberg, Godin and Whittington, 2019). For each video, the individual identity and species of the butterfly present were recorded. The videos also contained information about date, time of the day and temperature. Therefore, datasets of spatial and temporal foraging patterns were obtained by counting video appearances and grouping visits for each individual by date, time and flower cluster (Figure 2). Individuals with less than two recorded visits were removed from the dataset. Activity pattern graphs were made using density in ‘ggplot2’ package (Wickham, 2016) in R (R Core Team, 2023), and consist of a representation of the distribution (using kernel density estimate, a smoothed version of a histogram) of videos recordings for each species and/or sex during time.

Differences in visitation rates/patterns between sex and species were calculated using linear mixed-effect models (‘lmer’) implemented with the ‘lme4’ package in R (Bates et al., 2015) using a poisson distribution for raw data or binomial distribution for proportion data. We included “individual” as random factor when response variable was per individual and “week” as random factor for trial 2 (as stated above, trial 2 was divided in 5 groups, each on a different week), followed by analysis of deviance (ANOVA Type II) and Tukey’s post hoc tests with the packages ‘car’ and ‘emmeans’ in R (Fox and Weisberg, 2019; Lenth, 2023). In addition, Chi-squared tests (R built-in package ‘stats’) were used to compare the absolute number of visits observed to each flower against the expected number assuming random foraging (presumed random probability of 0.2 for trial 1 and 0.25 for trial 2 for each flower) for each individual. Individuals with less then 3 flower visits were removed from the analysis.

Finally, the level of sequence repetition in flower visits was explored to assess whether there was any evidence of potential trap line foraging in this experimental set up. Using individual visit sequence data, we used determinism analysis, a statistical metric to quantify the predictability of sequential behaviours (Ayers, Armsworth and Brosi, 2015). The denominator of the determinism (DET) varies between 0, indicating that the individual never repeats the same sequence, and 1, indicating that the individual always repeats the same sequence. For each individual the DET was calculated using the minimal sequence length of recurrent visits of 3 different flowers (Ayers, Armsworth and Brosi, 2015).