Camera-trapping estimates of the relative population density of Sympetrum dragonflies: application to multihabitat users in agricultural landscapes

Study area

The study was conducted in a farmland area in eastern Fukushima Prefecture, northeastern Japan (Fig. 1). The area is characterized by typical agricultural landscapes, including rice paddy fields, secondary forests, and woodlands. The area was enclosed within the following four sets of coordinates: (37°47′N, 140°34′E), (37°47′N, 140°59′E), (37°30′N, 140°59′E), and (37°30′N, 140°34′E). The annual mean temperature and annual precipitation during 1991–2020 obtained from the Fukushima Local Meteorological Observatory were 13.4 °C and 1207.0 mm, respectively (Japan Meteorological Agency, 2022). The study area was a part of the area targeted by a three-year survey of biodiversity in paddy fields where rice farming was restarted after the lifting of the evacuation order following the Fukushima Daiichi Nuclear Power Plant accident in 2011 (National Institute of Informatics, 2022). The study area included six study plots (Harimichi, Imada, Kanaya, Mimigai, Sakata, and Sugaya), each of approximately 0.7–1.2 ha comprising one or a few rice paddy fields (Fig. S1). We were granted verbal permission for conducting the research on each of the farmer’s private fields. The farmers were not represented by an organization, but were in a cooperative relationship with the Fukushima Agricultural Technology Centre. Rice farming in Kanaya, Mimigai, Sakata, and Sugaya was suspended during the evacuation and restarted between 2014 and 2018; however, rice farming continued without interruption in Harimichi and Imada. The camera-trapping and dragonfly adult surveys in the autumn of 2018 were not conducted in the Imada plot. Additionally, the target rice paddy fields in Kanaya were changed to neighboring rice paddy fields in 2019 (Fig. S1) because the farmers intended to change farming methods.

Camera-trapping system for darter dragonfly detection

Yoshioka et al. (2020) developed the original camera trap system based on the following principles. A perching dragonfly changes the resistance value of an upper light sensor at the tip of a rod-shaped detector relative to a bottom light sensor, changes the difference in signals between the sensors over a certain period, and activates a connected camera to a certain extent, thereby triggering a picture to be taken of the tip of the detector. In addition, if an image is captured, an additional picture is not taken unless the change in the relative signal values between sensors is restored because of dragonfly takeoff. In a previous study, a change of 5% and waiting period of 10 s (8 s for camera rebooting and an additional 2 s for image capture) could adequately detect perching darter dragonflies.

The camera trap used in this study comprised a detector section, processor section, and a connected commercial digital camera (TLC 200; Brinno, Inc., Taipei, Taiwan). Two CDS cells with a dark resistance of 1 M Ω were used as light sensors in the detector section. The upper cell was then covered with an LED silicone cap. The bottom of each cell was colored black using a permanent marker. These cells were connected to a microprocessor via USB cables. An acrylic pipe (outer diameter: 10 mm; inner diameter: seven mm; length: 500 mm) was used to enclose the cells. The top of the pipe was surrounded by a 150-mm-wide masking tape as a transmission diffuser and 20-mm-wide black vinyl tape and was capped with a 0.5-mm-thick polyethylene terephthalate plate using an adhesive.

In the processor section, an 8-bit PIC microcontroller (Microchip Technology Inc., Chandler, AZ, USA) was used as the main microprocessor. Signal values of 0–1,023 were obtained, which were processed to determine the presence or absence of a perching dragonfly based on the resistance values for each CDS cell in the detector section, which decreased with increasing illumination intensity. More details about the algorithm and materials used in the camera trap can be found in Yoshioka et al. (2020).

During the study, the exterior of the processor section was modified to make it easier to check the status of the traps and (re)boot them. Thus, three versions of the camera trap with different exterior materials for the processor section were used (Fig. S2). Version 1 was characterized by a processor section enclosed in a gray polyvinyl chloride (PVC) pipe to prevent rainwater intrusion (Fig. S2A). Two silicone caps, penetrated by the detector section, were inserted at the top of the PVC pipe. The enclosed microprocessor and its switch were attached. Investigators could not see or touch the microprocessor without removing the PVC pipe, which made it difficult to handle the camera trap. Version 2 (Fig. S2B) included a transparent PVC pipe surrounding the microprocessor and microprocessor switch, which was separated from the microprocessor and enclosed in a PVC pipe with a larger diameter below the PVC pipe enclosing the microprocessor and connected to the microprocessor via cables. Investigators could visually check the LED light on the microprocessor (which blinked if it worked correctly). The switch in the separate larger-diameter pipe could be handled relatively easily. However, the part connecting the microprocessor to the switch was vulnerable to corrosion owing to moisture during long-term use. In version 3 (Fig. S2C), the microprocessor and its switch were enclosed in a translucent 400 mL polypropylene bottle for salad dressing (MT-TORIMATSU Inc.). The bottle was attached upside down to the PVC pipe with a detector. Cables connected the processor and detector parts through the bottle spout. Investigators could quickly check and handle the processor. Preliminary experiments revealed that the microprocessor was adequately waterproof. In addition, versions 1 and 2 used three AA nickel hydride batteries for the processor, whereas version 3 used three AAA nickel hydride batteries. In the survey, the PVC pipe with the detector part was attached to a tripod or woodpile to adjust the height of the camera trap (approximately 2 m), regardless of the camera trap version.

Darter dragonfly survey using camera traps

A camera-trapping survey was conducted from 2018 to 2020 in the study plots. For each year in each plot, three camera traps were set up from the end of August to early September, and data were collected from the end of October to early November (Table S1). To cover a large area within a plot without disturbing farming, three camera traps were located on the levees of the paddy fields so that they were as far from each other as possible within the plot (Fig. S1). Although the traps were removed at the end of the study season every year, the levees where trap were set were consistent across years, and each trap location in each plot was assigned a location ID. When the trap location was changed at the field level, as in the case of Kanaya, a new location ID was assigned. Different versions of the camera trap were equally allocated to each plot, to the maximum extent possible. Every year, the detector parts of each version were randomly allocated to plots and locations to minimize the effects of individual camera traps. Each camera trap was inspected during early October. We manually intercepted the top of the detector and checked it for successful autodetection. In addition, after the transect surveys or typhoon events, the camera traps were briefly checked to ensure that they did not fall or run out of battery power. When necessary, the traps were reset by rebooting camera traps, exchanging batteries, and exchanging camera traps.

All pictures automatically captured by the camera traps and recorded in SD memory cards in the cameras were visually checked, and those correctly capturing darter dragonflies were recorded as darter dragonfly detection events (hereinafter, the frequency of these events is called “camera-detection frequency”). If two pictures were captured within 5 s, the former was removed from the analyses to avoid overestimating dragonfly detection following Yoshioka et al. (2020). Because the signal to wake up the camera was sent a few seconds before the signal to capture a picture, an extra picture was obtained if the camera was already active.

S. infuscatum (noshime-tombo, in Japanese) was recorded based on its characteristic dark spot on the tip of each wing; this species can be easily distinguished from other darter species (Yoshioka et al., 2020). Because the identification of other darter dragonflies at the species level was difficult owing to low image resolution, they were pooled in subsequent analyses. Most other darter dragonflies were likely the autumn darter S. frequens because this species was dominant based on a transect survey, and we did not directly observe darter dragonflies other than S. frequens and S. infuscatum perching on the camera traps during the survey season. These two species are typical darter dragonflies reproducing at rice paddy fields in Japan (Jinguji, Ozaki & Uéda, 2019) and are considered suitable as biodiversity indicator species of paddy fields (AFFRC, NIAES and NIAS, 2012b). Both males and females of these species are known to exhibit perching behavior (Miyakawa, 1994; Watanabe, Matsuoka & Taguchi, 2004), while males do not clearly exhibit territorial behavior (Taguchi & Watanabe, 1986; Sugimura et al., 1999; Watanabe, Matsuoka & Taguchi, 2004). In addition, mark-recapture studies have shown that the recapture rates of these species are generally quite low (Taguchi & Watanabe, 1986; Watanabe, Matsuoka & Taguchi, 2004; Fukui, 2012) and it is considered unlikely for a particular individual to stay perched at the same point for more than a day (Fukui, 2012).

Owing to external factors, such as adverse weather and unanticipated internal factors, the period for which the camera traps worked effectively varied. To standardize the dragonfly detection frequency among camera traps, the occasion index (i.e., the period during which each camera trap effectively worked each year) was also required. To calculate this value, the dates on which pictures were taken were obtained; history of camera trap setting, resetting, and collection were checked; and periods starting with the camera trap (re)setting and ending with the last picture before resetting (or successfully capturing a picture through an operation check at the end of the survey) were extracted. “Last pictures” included pictures with an appropriate view of the tip of the detector, regardless of the presence of a dragonfly (pictures captured from a fallen camera were invalid because such traps were considered ineffective). Although the exact date when camera traps became ineffective could not be determined, the date of the last picture before resetting is typically used in camera trap monitoring (Fukasawa et al., 2016). In this study, autodetection (irrespective of whether the dragonflies were correctly detected) typically occurred daily when the camera trap was active (Yoshioka et al., 2020). Second, the length of each period was calculated as the difference between the date of (re)setting the camera trap and the last date when a picture was captured before reset or the date of camera trap collection. Finally, the length of each segment was summed as the number of occasions per camera trap per year. Thus, the occasion index at the location in the year can be formulated as follows: ${\displaystyle \text{occasion}=\sum _{i=1}\left(d_i^{last}-d_i^{set}\right)}$ where $d_i^{last}$ is the date of the last picture in the i-th segment and $d_i^{set}$ is the date of the (re)setup of the camera trap.

For example, consider the case in which a camera trap was set up on September 5, reset in response to a technical issue identified on September 19, and collected on October 31 with depleted batteries. The last picture before September 19 was captured on September 11, and the last picture before October 31 was captured on October 30. In this case, the occasion index for the trap was (September 11 − September 5) + (October 30 − September 19) = 6 + 41 = 47. If a camera trap remained active without any issues, the number of days between the date of setup and the date of collection corresponded to the occasion index.

Transect survey of adult darter dragonflies in autumn

A line-transect survey was conducted as a classical method for measuring the relative population density of adult dragonflies (Bouwman et al., 2009; Pearce-Higgins & Chandler, 2020). This survey was conducted in the autumn of 2018, 2019, and 2020. Each study plot was surveyed twice (once each in September and October) from 9:30 to 17:00 h on days with mild wind and no measurable precipitation. According to the number of paddy fields in the plots, 8–16 line transects of 10 × 2 m were set along the levees of the rice paddy fields in each plot. Four line transects per rice field were set, except for the Mimigai plot, which had one large rice paddy field. During the line-transect survey, the investigator slowly (about 0.15 m/s) walked in one direction while searching for dragonflies within the line transect, and the number and species of darter dragonflies were recorded for each transect. When relatively small and inconspicuous individuals, such as females of S. kunckeli (maiko-akane in Japanese) or S. eroticum (foot-tipped darter), were observed, a sweeping net of 50 cm in diameter was used to capture and evaluate them. For each year and plot, the adult-density index by line transects for S. infuscatum was obtained for analysis as follows: ${\displaystyle \text{adult-density index by line transects}=\left(\frac{n_{inf1}}{n_t}+\frac{n_{inf2}}{n_t}\right)/2,}$ where n_inf1 and n_inf2 are the total numbers of observed S. infuscatum in the first and second surveys, respectively, and n_t corresponds to the total number of line transects in the plot per survey. In the same way, the adult-density index by line transects of other darter dragonflies (i.e., darter dragonflies other than S. infuscatum) was calculated.

Transect survey of exuviae of darter dragonflies in early summer

Population growth of darter dragonflies may be related to the quality of the local aquatic environment rather than to the abundance of adults with high mobility that oviposit at poor-quality aquatic sites (Raebel et al., 2010). Therefore, sampling exuviae of emerged adults is a highly effective method for surveying viable populations of dragonflies and the quality of local aquatic environments, although it has some disadvantages, such as low detectability (Pearce-Higgins & Chandler, 2020). This method has been recommended for surveys of darter dragonfly densities in rice paddy fields in Japan (Baba, Kusumoto & Tanaka, 2019; Mitamura et al., 2013; Nakanishi et al., 2022). In this study, Mitamura et al. (T. Mitamura, N. Matsuki, A. Yoshioka, K. Tabuchi, 2020, unpublished data) collected a part of the dataset to survey the biodiversity in rice paddies after the Fukushima nuclear disaster. The data set for darter exuviae was obtained following a sampling protocol based on a survey and evaluation manual for indicator animals of functional agrobiodiversity compiled by the national government of Japan (AFFRC, NIAES and NIAS, 2012a; Mitamura et al., 2013). In this survey, four 10 m line transects per rice field were set along the levees of the rice paddy fields in each plot. The investigators visited each plot twice a year during the daytime, from late June to early July in 2018–2020. During the line-transect survey, an investigator walked slowly in one direction while searching for and collecting exuviae of dragonflies on rice plants up to the third row of rice hills from the edge of the levees (corresponding to a width of approximately 1 m). The collected exuviae were identified to the species level in the laboratory. The number of exuviae in the two surveys was summed for each transect in each year (hereinafter, “exuviae-density index by line transects”). This index was expected to show the relative density of the exuviae rather than the absolute density.

Statistical analyses

To examine whether the camera-detection frequency of adult darter dragonflies reflected the adult-density index by line transects, a generalized linear mixed model (GLMM) with a negative binomial error distribution was constructed. In the GLMM, the camera-detection frequency of S. infuscatum (n = 51) was the response variable, adult-density index by line transects of S. infuscatum was the explanatory variable, and the log of the occasion index was the offset term. The year, study plot, location ID of a trap nested within a study plot, and trap version were included as random effects. The adult-density index by line transects of S. infuscatum, was log-transformed before the GLMM to obtain a more likely relationship between the density of dragonflies and the camera-detection frequency as suggested by Yoshioka et al. (2020). That is, a larger abundance of dragonflies would not always correspond to a higher camera-detection frequency when the number of camera traps is smaller than the abundance of dragonflies because only a small portion of individuals can occupy the detector in a short term. Therefore, log transform to reduce the weight of larger values of the adult-density index by line transects was desirable to obtain a more proportionate relationship with camera-detection frequency. Because the data of S. infuscatum included zero values, 0.5 was added before log transformation (i.e., log(adult-density index by line transects of S. infuscatum + 0.5)) following the recommendation of Yamamura (1999). Similarly, the effect of the adult-density index of other darter dragonflies on their camera-detection frequency was examined.

In addition, relationships between exuviae surveyed in early summer, as an indicator of the quality of the local aquatic environment, and adult darter dragonflies detected by camera traps in the autumn were examined. The exuviae-density index by line transects of S. infuscatum for each line transect in each year was divided by two to obtain the mean of the two surveys and averaged to obtain the mean exuviae density of S. infuscatum (the explanatory variable). The effect of the mean density of S. infuscatum exuviae on the camera-detection frequency of S. infuscatum in the same year was examined using a GLMM with a negative binomial error distribution (n = 51). The log-transformed occasion index was included as the offset term, and the year and location ID of a trap nested within the study plot were used as random effects. The effect of the camera-detection frequency of S. infuscatum on the exuviae-density index of the transects of S. infuscatum in the following year was also examined. The camera-detection frequency divided by occasion index was averaged in each plot for each year as the mean detection frequency per occasion. A GLMM with a negative binomial error distribution was then constructed using the exuviae-density index by line transects of S. infuscatum for each transect in each year as a response variable (the sum of values from two surveys was used rather than the mean because integer values were required for the GLMM with negative binomial) and the mean detection frequency per occasion of the previous year as an explanatory variable (n = 116). In the model, year and transect nested within the rice paddy fields and study plots were included as random effects. Similarly, the relationships between the exuviae-density index of other darter dragonflies and detection frequencies of these species were examined. The GLMM was implemented using the glmmTMB function in the glmmTMB package version 1.1.2 for R version 4.0.3 (Brooks et al., 2017; R Core Team, 2020). See Code S1 for details of the GLMM.