PICT: A low cost, modular, open-source camera trap system to study plant-insect interactions

1. Commercial camera traps (CTs) commonly used in wildlife studies have several technical limitations that restrict their scope of application. They are not easily customizable, unit prices sharply increase with image quality and importantly, they are not designed to record the activity of ectotherms such as insects. Those developed for the study of plant– insect interactions are yet to be widely adopted as they rely on expensive and heavy equipment. 2. We developed PICT (plant– insect interactions camera trap), an inexpensive ( < 100 USD) do- it- yourself CT system based on a Raspberry Pi Zero computer

Despite tremendous progress in the fields of pollination biology, quantitative genetics, comparative biology, phylogenetics and genomics, the paucity of empirical data from natural history studies limits progress in understanding pollinator-driven evolution (van der Niet, 2021).
Conventional studies of plant-insect interactions typically involve the collection of data using direct (e.g. Suetsugu, 2019;Tang et al., 2020;Varma & Sinu, 2019) or indirect observations (e.g. Boyer et al., 2020;Johnson et al., 2011). However, because observations are time-intensive, limited by environmental conditions and logistics, they are not conducted over large spatiotemporal scales and often underestimate the importance of furtive organisms compared to larger or slower ones (Micheneau et al., 2006). Furthermore, the presence of a human observer and the need to illuminate the study organism at night may influence its behaviour (Opp & Prokopy, 1986).
Camera trap (CT) technology can greatly advance the study of plant-insect interactions by providing a convenient replacement to classic human observations. This technique have gained popularity because it allows for non-intrusive observations at large spatiotemporal scales and constant sampling effort (Rovero & Zimmermann, 2016;. Recently, camera trapping of insects has become an active field of research and development but important technical limitations still persist (Pegoraro et al., 2020;Preti et al., 2021). First, although it has been reported that the passive thermal infrared motion sensor of commercial CT systems can be activated by large flying insects (Houlihan et al., 2019;Johnson & Raguso, 2016), most ectotherms, such as reptiles, amphibians and invertebrates, do not trigger motion sensors (Hobbs & Brehme, 2017). Moreover, the initial trigger delay has been deemed excessive in many cases, especially in hot environments Meek & Pittet, 2012). To circumvent these problems, researchers have developed CT systems relying on active motion detection based on pattern recognition or changes in the successive frames captured by a camera (Barlow & O'Neill, 2020). This technique has proven to be efficient for obtaining data on insect visit frequency, visit duration and for modelling insect activity (Barlow et al., 2017;Steen, 2017). However, applying an on-the-fly motion detection algorithm to filter the video stream during recording increases power consumption and does not allow one to estimate the rate at which motion events fail to be detected.
Second, camera characteristics of commercial CT systems often limit the number of taxa that can be accurately identified, especially when taxonomically relevant traits are subtle. Image resolution can often be modulated in the camera settings, but shutter speed decreases as resolution increases, hence decreasing the sharpness of moving animals. Image quality is ultimately limited by the quality of the sensor and the lens, neither of which are interchangeable in most cases (Meek & Pittet, 2012;Rovero et al., 2013). Most CTs use wide-angle fixed-focus lens that are set so that the depth of field ranges from infinity down to a few metres. As a result, these models are not suited for macro-photography. Finally, the cost of CT units is often the limiting factor in terms of the number of sensors that can be deployed simultaneously, and therefore, the statistical power of the analysis. Currently, a mid-range CT costs 200-500 USD (Rovero et al., 2013;Wearn & Glover-Kapfer, 2017). The unit price of motiontriggered CT systems designed for insect monitoring range from 400 EUR (Pegoraro et al., 2020) to several thousands of euros (Danaher et al., 2020;Houlihan et al., 2019).
Here, we propose a new system, called plant-insect interactions camera trap (PICT), that overcomes the above shortcomings. We report results from the deployment of this system under two conditions where manual observation is impossible: (a) in places where an observer cannot remain for long periods of time (pollinator visitation in the canopy of the African ebony tree) and (b) when the time scale involved is too large (low visitation rates of pollinators of an African epiphytic orchid).
PICT contrasts with other solutions by its increased portability, reduced cost and low energy use hardware that does not require heavy and bulky lead batteries to operate. Low-energy consumption is mainly achieved by separating the recording and analysis steps.
By providing enough memory to the camera and using an efficient H264 compression algorithm, we can record high definition videos continuously in the field and use a computer to search for the frames of interest later in the lab.

| DE S I G N AND A SS EMB LY
PICT consists of four main components, a single-board computer, a micro SD card, a camera and a USB power bank battery (Figure 1).
A practical guide with detailed instructions for constructing PICT as well as the control programs and codes are available online as Supporting Information . allow for large-scale deployment and the acquisition of novel insights into the reproductive biology of plants and their interactions with difficult to observe animals.

K E Y W O R D S
behavioural ecology, digital video recording, DIY camera trap, e-ecology, low-cost technology,

plant-insect interaction, pollination biology, Raspberry Pi
To protect the components from natural elements, PICT is sealed in a food storage case of about 1 L in volume. Each component inside the case is fixed in place by adhesive Velcro ® strips. A mount with a standard ¼ in screw is glued onto the case to allow PICT to be fixed to a standard camera mount. At the time of writing, the full cost of building one operational unit is less than 170 USD. The components needed for a PICT with functionality comparable to a retail CT, that is, without a mount, battery or memory card, would cost less than 100 USD (Table 1).
The camera is operated through the picamera Python package (https://picam era.readt hedocs.io/) installed on a Raspberry Pi Zero, which is a credit card-sized, low-cost, high-performance single-board computer. All the Raspberry Pi models with an integrated Wi-Fi controller can provide the functionality required, but we recommend on which the operating system, programs and data are stored. It is powered through a 5V mini-USB port that can be supplied by a standard lithium-ion power bank ( Figure 1).
We used the 5-megapixel Raspberry Pi Camera Module v1 (OmniVision © OV5647 sensor), based on a 2,592 × 1,944 photosites, ¼ in format sensor. It comes in customized versions with (a) an embedded 3.3V power output that can be connected to a nearinfrared LED without need for soldering, (b) a 3.6-mm lens with a diagonal field of view of 75 degrees and adjustable focus distance, (c) no embedded infrared filter, improving lens speed and allowing illumination of the night scene with IR light. To illuminate the scene, we used one 850-nm infrared LED equipped with an onboard photoresistor to decrease light intensity with increasing ambient light.
An onboard resistor can be tuned to control the photoresistor ambient light threshold toggling the infrared LED. Near-infrared light is preferred because it is invisible to animals thereby not influencing behaviour. Insects' photoreceptors have a large spectral sensitivity range, but the maximal observed peak absorption wavelength is 630 nm (Briscoe & Chittka, 2001). Positive phototaxis of insects to larger wavelengths has been observed but intensity decreases with increasing wavelength (van Grunsven et al., 2014;Wakakuwa et al., 2014), and is relatively small at 850 nm, as shown for a Coleoptera (Meyer, 1976) and a Hemiptera .  (Barnes, 2020).

| P OWER CON SUMP TI ON AND DATA S TO R AG E
Low power consumption is essential to avoid the need for heavy or bulky batteries and to provide autonomous operation times that exceed the duration of the targeted phenomenon (the duration of anthesis for instance). To reduce the power drawn by the PICT by about 0.13W, we deactivated the components that are not needed for our application: the HDMI port, Bluetooth and activity LEDs.
We used an electronic multimeter (RuiDeng UM25C) to measure the power drawn by a PICT under various operating conditions.
The observed power load of each of the components and for different camera settings is given in Table 2. We found that both frame rate and resolution settings have a substantial effect on power use (Table 2; Figure S1). We used a resolution of 1,296 by 972 pixels and 15 frames per second (FPS) to achieve the lowest possible power consumption and storage needs while not affecting the ability to identify insects. With these settings and with Wi-Fi switched off at night, the PICT will draw only 0.76 and 1.87 W, respectively, during the day and at night. This theoretically permits continuous filming for over 72 hr with a 30,000-mAh (111 Wh) power bank, as was confirmed during field deployment. With these settings, PICT would be able to run for almost 9 days if recordings are performed during the day only and the IRD LED is not connected ( Figure S1).
We advocate the application of motion detection algorithms as a post-processing stage rather than in situ because the processing of the video stream to filter out still sequences is computationally expensive. The additional power drawn will directly depend on the algorithm complexity. before the storage media get saturated. Furthermore, we noticed no compression artefact when reducing file size by a factor of 2 using a higher compression level, thus allowing for further increase in storage efficiency if needed.

| Deployment and data processing
To assess the performance of PICT in the field, we studied two plant species with contrasting habits, pollination ecologies and floral characteristics: the African Ebony tree Diospyros crassiflora Hiern and the epiphytic orchid Cyrtorchis letouzeyi Szlach. & Olszewski.
Diospyros crassiflora is a commercially valuable ebony tree native to the rainforests of Central Africa that can reach 25 m in height. Until this study, the identity of its pollinators was unknown (Deblauwe, 2021). Staminate and carpellate flowers are found on different plants, a character known as dioecy. We considered as potential pollinators all insects that entirely enter the narrow opening

| Results
We  insects' social and predator-prey interactions, the effect of (micro) climate change on their activity or herbivory and plant phenology.

ACK N OWLED G EM ENTS
This study is part of the Congo Basin Institute's Ebony Project generously funded by UCLA and Bob Taylor

D I SCL A I M ER
The authors declare to have no connection whatsoever with the brands and commercial entities cited in this manuscript. The brands cited in the text and accompanying practical guide are only for illustration.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.

DATA AVA I L A B I L I T Y S TAT E M E N T
A practical guide with computer code and step by step instructions for building PICT, using in the field and post-processing movies is archived at https://doi.org/10.5281/zenodo.4139839 .