Big Brother is Watching: Studying Insect Predation in the Age of

172 American Entomologist  • Fall 2012

ResearcH

Predation by generalist predators has long been recognized as an important aspect of insect biological control (Symondson et al. 2002), but it is a factor that is difficult to quantify. The

majority of studies assessing predation of insect pests have relied on a sentinel prey approach, where static prey items are placed into crop fields for a set period of time and then assessed for damage or removal. Although these studies provide a measurement of relative predation rates in different experimental treatments or settings, they often do not provide researchers with the identity of predators. This is often indirectly assessed through the use of associated traps, with the assumption that predatory taxa caught in traps are responsible for predation of sentinel prey (e.g., Frank and Shrewsbury 2004a, O’Neal et al. 2005). To provide a more direct link, researchers have also used human observation to establish the identity of predators attacking sentinel (e.g., Pfannenstiel and Yeargan 2002) or naturally occurring prey (e.g., Costamagna and Landis 2007). This can provide information on the behavior and diel periodicity of predators, but the total number of interactions that can be observed is limited because humans cannot be in multiple places at once and observations at night are challenging. Other researchers have also taken predators

collected in traps into the laboratory and used feeding assays to determine if they damage or consume a target pest (e.g., Frank and Shrewsbury 2004b). However, laboratory feeding assays cannot replicate the outdoor environment. More recently, it has become possible to use molecular technology to detect the presence of pest antibodies or DNA in predator guts. This technique allows predation to be quantified in situ, but requires a great deal of technical expertise, as well as the availability of appropriate ELISA reagents (Sunderland et al. 1987, Hagler and Naranjo 1994) or PCR primers (Greenstone et al. 2010, Szendrei et al. 2010). Moreover, gut content analysis cannot be used to examine the behavior, diel rhythm, or frequency of attack of predators.

Video technology could help fill this void by capturing the identity of predators attacking prey while providing detailed information about their behavior. Early behavioral applications of this technol-ogy include developing a better understanding of the kinesiology of animal movement (Wratten 1994). In the realm of insect behavior, video techniques have been used to study movement and foraging behavior of herbivorous insects (Hardie and Young 1997, Storer et al. 1999, Kindvall et al. 2000, Hardie and Powell et al. 2002), parasit-

Big Brother is Watching: Studying Insect Predation in the Age of Digital Surveillance

Matthew J. Grieshop, Ben Werling, Krista Buehrer,

Julia Perrone, Rufus Isaacs, and Doug Landis

ABSTRACT: We describe a novel video system constructed from readily available security equipment for recording insect predator behavior in the field. Our system consists of a multi-channel digital video recorder (DVR), active night vision cameras, a deep cycle marine battery, and a weatherproof housing. The major advantages of these systems over previous generations of video equipment include reduced expense, improved deployment times, faster frames per second, and higher video resolu-tion. We tested our systems in a pair of experiments: the first assessed the effects of ground cover on predation of key pests in blueberries, and the second investigated predator activity in corn and perennial prairie systems. We identified ants as the most frequent predators in all three ecosystems, with notable activity from arachnids, crickets, and mollusks. Interactions between coleopterans and prey were surprisingly infrequent in all three agroecosystems. Ants were observed under both day and night conditions, while other predators were primarily nocturnal. Ground covers did not significantly affect predator activity in blue-berries, but there was a numerical reduction in ant activity on woodchips and weed barrier compared to grass and bare ground. The predator complex observed in blueberry video footage differed considerably from pitfall captures at the same site. In corn fields, prey mortality significantly increased with ant activity, and high prey mortality was associated with long interactions that involved large numbers of ant foragers. Digital videography of insect predators is a very useful tool for identifying key taxa responsible for the removal of sentinel prey and documenting their behavior.

KEYWORDS videography, insect behavior, biological control, blueberry, corn, prairie, video, digital video recorder, DVR, Formicidae

173American Entomologist • Volume 58, Number 3

oids (Allemand et al. 1994), and predators (Fourcassié and Traniello 1995, Meyhöfer 2001, Schenk and Bacher 2002, Merfield et al. 2004, Frank et al. 2007); insect flight behavior (Riley et al. 1990, El-Sayed et al. 2000, Imafuku and Ohtani 2006); and the response of insects to pheromones under laboratory (Willis and Baker 1994, Cardé et al. 1998, Justus et al. 2002, Justus and Cardé 2002) as well as field conditions (Teixiera et al. 2010).

The challenge has been that previous generations of film and vid-eo cameras that were durable enough to withstand field conditions were also bulky and expensive and provided limited recording time on tape-based media. Thus, they were ill-suited to the collection of large amounts of data in remote locations. However, recent advances in video technology have made small, weatherproof cameras and Digital Video Recorders (“DVRs”) widely available and affordable. Commercial surveillance systems offer several advantages for insect behavioral research: they are relatively inexpensive to purchase and customize; they can be placed in remote locations lacking wired power; and they allow footage to be captured from multiple cameras simultaneously. The major limitation of such systems is that while they allow the collection of temporally rich data, observations are spatially limited to the camera’s field of view. These advantages and limitations make such systems well suited to exploring questions surrounding the broad identity and short-range behavior of insects responding to an attractive “subject” (e.g., a flower, prey item, or pheromone lure).

Here, we describe two case studies to show how this technology can be applied to document the identity and behavior of predators in agroecosystems. The first study assessed predation in highbush blueberry. The second study focused on identifying invertebrate predators in two candidate bioenergy crops: corn and native, warm-season prairie grassland.

Materials and MethodsVideo Observation System Design. The video observation sys-

tem consisted of four components: 1) a multichannel digital video recorder (DVR); 2) video cameras with mounting hardware; 3) a power source; and 4) weatherproof housing for components 1 and 3 (Fig. 1). It is important to note that the DVRs and cameras described in this paper represent only a small sample of the many potential models currently available in the security and surveillance market-place, and improved hardware is available each year. At the time this manuscript was produced, many DVRs could be purchased in bundles that included weatherproof cameras and associated hardware and software, at a price range of $300–$800 for a 4–16 camera system.

System Component 1: Digital Video Recorder. A multichannel DVR is the heart of the video observation system (Fig. 2a). The two DVRs we used were a 4-channel system (QH25DVR) and a 16-channel system (model QSDR16RTCB), both manufactured by Q-SEE (Digital Peripheral Solutions Inc., Anaheim, CA). These systems provide a video frame rate of 30 frames per second on each channel, recorded on a desktop computer hard drive embedded in the DVR. At this rate of data collection, 24 h of video takes up approximately 1 gigabyte (GB) of hard drive space. Each channel is recorded simultaneously as a separate video file, and a channel, date, and time stamp can be placed on each frame (Fig. 3).

Fig. 1. Digital video system schematic. Component 1: digital video recorder; Component 2: camera frame and camera; Component 3: 12V battery/power supply; Component 4: weatherproof housing.

Fig. 2. Photographs of one of the Digital Video Recorders (DVR) and security cameras used by the authors. (a) A four-channel DVR with a removable hard drive bay attached with brackets to the top; this setup allows hard drives to be swapped in the fi eld. (b) A fully assembled night vision camera with a close-up lens attached using a PVC adapter; this reduces the focal length of the camera, allowing it to be used to capture high-resolution images of insects. Note the ring of LEDs surrounding the factory lens assembly; the added close-up lens must be fl ush with the rubber ring housing the camera’s lens to reduce glare from LEDs. (c) A security camera with the factory lens protector removed and, from left to right, a silica gel pack to reduce moisture; a disassembled lens adapter showing the PVC ring; a threaded mount for the lens that fi ts inside the PVC ring; and a close-up lens.

Suggested modifications for systems like these include the instal-lation of a removable hard drive bay (Kingwin K450-1074, Kingwin, Walnut, CA) to facilitate changing drives in the field and a modifica-tion of the power supply so that the system can be powered by a 12V source. For example, we glued aluminum brackets to the top of the DVR case to allow the removable hard drive bay to be attached to the top of the unit (Fig. 2a). We drilled a hole in the DVR case to allow the hard drive data and power supply cables to be attached from the DVR main board to the removable hard drive bay. (Some newer DVRs store data on removable memory cards, eliminating the need to add extra hard drive capacity). We then modified the power supply by cutting the factory power cable beyond the transformer and soldering in a power adapter, which can be attached to any 12V DC power supply (system component 3). For DVRs that do not run on 12V DC, an inverter could be added to convert the DC current provided by the battery into 110V AC.

System Component 2: Cameras and Camera Stands. Our cameras were provided as part of a security bundle designed for indoor or outdoor surveillance (Fig. 2b, c) and included active night

174 American Entomologist • Fall 2012

vision provided by 12 infrared Light Emitting Diodes (LEDs). Such cameras can typically be purchased individually for $50–$120. The cameras themselves are attached via 75 ohm coaxial cable to the DVR as well as to the system power supply (component 3). The fo-cal length of these cameras is typically between 5 and 20 m, as their original purpose is to provide security for parking lots, homes, and retail stores by recording the activity of human-sized subjects. Thus, the stock setup of these cameras is less than optimal for recording insect behavior, a scenario in which the subjects of interest are much smaller than humans and cameras are likely to be placed <25 cm from the subject. The addition of a secondary lens is an affordable way to address this issue. For example, we replaced the protective lens assembly of the camera with an adapter consisting of a piece of PVC pipe that was milled to fit the camera housing on one end and fitted with a 52 mm photo adapter ring on the other (Fig. 2c). With the ring in place, a variety of commonly available close-up lenses can be used to shorten the focal length of the camera. For example, we used macro close-up lenses, which are inexpensive and provide clear pictures of a variety of arthropods when cameras are placed 10-20 cm from the subject (Fig. 3).

There are two important considerations when performing this modification. The first is that the final assembled product must be made weatherproof. To protect our cameras, we wrapped Parafilm around the outside of the camera barrel before slipping on the lens assembly. After placing the lens assembly on the camera, we then sealed it to the camera by wrapping its bottom edge with either Parafilm or electrical tape to provide an outer seal. We also placed a silica gel pack next to the camera’s primary lens before attaching the secondary lens, which helped reduce fogging; the gel pack can be wedged between the outer wall of the camera casing and the LEDs, away from the camera’s field of view. The second issue is that if nighttime observations are being made, the lens must be absolutely flush with the camera lens assembly to prevent reflection from the infrared LEDs (Fig. 2b).

The digital surveillance cameras used in our systems come com-plete with a pre-drilled mounting plate and a universally adjustable bracket, allowing them to be easily positioned. The camera mount itself can consist of any number of designs and will vary greatly depending on the needs of the user. The camera stands used in the studies were fabricated out of either 1.9 cm (3/4”) metal conduit or schedule 40 PVC (wall thickness = 2.9 mm).

System Component 3: Power Supply. Cameras and DVRs require a power supply. At field sites that have access to an electrical grid, the DVR can simply be plugged into a power outlet using an exten-sion cord. However, if deployment to a remote location is required, an alternative power supply is necessary. Many DVRs and cameras

on the market are directly powered by 12V DC, allowing automotive or deep cycle batteries to be used as power supplies. Our 4- and 16-channel DVR and camera systems draw approximately 2 and 4 amps at 12V DC (respectively) when operating at full load. Thus, a deep cycle battery (e.g., a marine battery) with 90 amp hours capac-ity provided us with 1-2 days of continuous recording. Wiring the battery into the system can be accomplished by cutting the power supply wires leading from the stock AC/DC transformer and attaching a positive and negative battery terminal connector wired in its place. The system can be made AC/DC adaptable through the addition of power connectors that allow the user to connect the system to the stock AC power supply or a DC supply from a battery or generator.

Battery life limits the total amount of footage that can be collected without making repeat visits to a site. While most DVRs will allow the user to program recording times, this will not alter the load placed on the battery because the system’s power remains on even when cameras stop recording. This can greatly cut into deployment time, forcing the user to recharge batteries more frequently than might be desired. Two potential solutions for this problem are the use of a DC timer that cuts the power supply to the DVR and cameras when they are not in use, and the integration of solar panels into the system. To date, the authors have successfully used timers (e.g. Novus TM-619 Miami, FL), but have yet to fully explore the solar option. Should a solar array be added to the system, it is critical that it be run through a charge controller to ensure that the battery is not overcharged. The major advantages to a solar charger would be the elimination of the need to service the battery and the potential to use a lower capacity (and therefore smaller and less expensive) battery.

System Component 4: Weatherproof Housing. A weatherproof housing to protect the DVR and power supply is necessary if the system will be deployed in areas that experience rainfall, irrigation, or other precipitation events. The design of this housing can take many forms. The design described in this manuscript consisted of a 208 L polypropylene drum, previously used for food grade prod-ucts. We have also successfully used plastic storage boxes, which are commonly available at many retail outlets. Such containers provide a nearly airtight container with a water-resistant floor and lid; they are inexpensive (<$20); and the soft and easily workable material of which they are made allows for the cutting of access and venting ports needed to complete the system. Venting is necessary to main-tain temperatures within the DVRs’ operating range (<60°C for our systems). For venting, we cut holes in the housing using a hole saw drill bit and inserted 6 cm round soffit vents that were sealed into place with a bead of plumber’s putty (Fig 1).

Analyzing Video. After video footage was collected, the hard drives were returned to the laboratory for analysis. The authors’ data processing technique consisted of maintaining several additional DVRs in the laboratory that were networked to personal computers and monitors for viewing. Analyzing video collected in the field is labor intensive, but can be sped up considerably by watching multiple video clips simultaneously at increased playback speeds (Fig. 4). For example, if the viewer watches four frames simultaneously at 4x real speed, he or she can view 16 hr of field data in 1 hr of real time. Typically, during the first viewing of video data, observers created a log of behavioral “events,” including date, time, video frames, and the type of behavior observed during the event. This information was then used to extract the frames of interest and export them to a file type recognizable by common video players (e.g., .avi, .wmv, or .qt formats). The use of digital video stored on hard drives greatly

Fig. 3. Images captured with the authors’ video system showing (a) a slug and harvestman and (b) ants and a harvestman feeding on sentinel waxworm larvae during daylight and nighttime hours, respectively. Prey were placed in a Petri dish surrounded by a vertebrate exclosure made of hardware cloth.


streamlined this process and made creating backup video files a simple procedure. Once a complete catalog of events was created and corresponding video clips were extracted, the insect behaviors of interest were further quantified and described using freely avail-able video software (e.g., http://www.videolan.org/).

Using Field Video Data for Predation AnalysisBlueberry Pest Predation. We used our surveillance systems to

record predatory arthropods visiting sentinel cranberry fruitworm (Acrobasis vaccinii Riley) pupae and Japanese beetle eggs (Popillia japonica Newman). Cranberry fruitworm pupae were collected from the field, while Japanese beetle eggs were obtained in the laboratory from field-collected adults. We ran our experiment at a highbush blueberry (Vaccinium corymbosum) planting located at the Michigan State University Trevor Nichols Research Center near Fennville, MI.

We established a completely randomized block design to test

the effect of four types of blueberry row cover on ground-dwelling arthropod diversity and removal of sentinel prey. We replicated treat-ments four times with a single replicate of ground cover covering the soil along a 14 m blueberry row. A buffer strip of three blueberry rows treated with herbicide were maintained around each treatment row. Four ground cover treatments were sampled: untreated ground (grass and weed cover); herbicide-treated ground (bare ground); wood chip mulch; and a plastic weed fabric (barrier) (Dewitt 3P 5 oz. weed barrier fabric, Sikeston MO USA). We measured arthro-pod diversity using a pitfall trap placed within each blueberry row (replicate). We placed sentinel prey (either 5 cranberry fruitworm hibernaculae or 10 Japanese beetle eggs) within the planting at times when each pest is present in this life stage at this location. To accomplish this, fruitworm hibernaculae and Japanese beetle eggs were placed in the field from 5 August 2008 to 12 August 2008 and 19 August 2008 to 26 August 2008, respectively. Sentinel prey were protected from vertebrate predation by placing them in 15 cm x 15 cm x 6 cm wire cages constructed of 6.4 mm hardware cloth. Prey were evaluated for removal/damage and pitfall traps were emptied after seven days.

We collected video observations using four 4-channel DVRs that collected data from 16 cameras (1 camera per replicate), which were each focused on a sentinel prey arena (Fig. 5a,b). One of the four camera replicates suffered data loss due to interference from high-tension power lines running over the planting. We therefore restricted analysis to the three remaining replicates. Predation of sentinel prey was recorded for the initial 72 hr of deployment. Camera stations were placed with 6 m between each camera within a set of cameras (four per DVR) and 6 m between each set of four cameras. Prey items were placed on Petri dishes filled with sand and centered within each cage. Cameras were fitted with +4 macro lenses to provide adequate resolution of prey items and visiting arthropods.

We reviewed the video data in the laboratory over a three-month period and collected information on arthropod visits, including the number of visitors and the time and duration of each visit. Visiting arthropods were assigned to one of seven morphotaxa: Coleoptera, Formicidae, Gryllidae, Diptera, Chilopoda, and Arachnida. A visit was

Fig. 4. One of the authors watching multiple video clips of predation at high speed for the bioenergy study

Fig. 5. Photographs showing the authors’ camera systems set up in the fi eld. (a) A plastic barrel provided weatherproofi ng for a DVR and power source, which were connected to (b) cameras focused on sentinel prey in blueberries. Note the camera wires elevated above the ground in (a) to reduce the risk of rodent damage. (c) Predation in bioenergy crops was captured by attaching cameras to adjustable booms affi xed to a PVC frame. This frame allowed us to capture video in three positions; only video from the bottom camera (circled) was used for this experiment. (d) A camera (circled) set up in prairie habitat.

http://www.videolan.org/


defined as an incidence in which a potential predator came within one centimeter of sentinel prey items. (Ants were always considered to be a single organism, given their eusocial behavior.) We measured the composition of arthropods visiting prey by calculating the proportion of the total visits to each camera made by each morphotaxon, and compared this to the composition of predator captures in pitfalls. In addition, we recorded the number of events by study hour to quantify diel rhythms of activity. Finally, we analyzed the effect of ground cover type on total visits by each morphotaxon using an ANOVA with data normalized by a log10-transformation where appropriate. Data analyses were performed using the R statistical computing language (R Development Core Team 2012).

Predation in Corn and Prairie grassland. Predation of insect larvae was examined in corn and warm-season grassland to docu-ment potential predators in these candidate bioenergy crops (Fig. 5c, d). Waxworm larvae, Galleria mellonella L. (Lepidoptera: Pyralidae) were used as surrogates for pest larvae and pupae that are present on the ground. Predation was observed at two stations spaced 50 m apart at each of seven corn and grassland sites located in southern lower Michigan. Five grassland sites were planted with a mix of forbs and prairie grasses such as Andropogon gerardii Vitman and Sorghastrum nutans (L.) Nash; two were planted as monocultural stands of switchgrass, Panicum virgatum L. At each station, a Petri dish was sunk so that the top was level with the soil or surface litter and protected from vertebrates with a circular (15 cm diameter) piece of hardware cloth (1.25 cm mesh) (Fig. 3). The inner surface of Petri dishes was painted with flat grey paint to reduce glare. Four live waxworm larvae were affixed to bottom of each petri dish by pressing the tip of each worm’s abdomen onto a drop of cyanoacrylate glue (Loctite, Henkel Consumer Adhesives, Avon, OH). At each station, a camera fitted with a +4 macro lens was trained on sentinel prey for 24 hrs. Observations at each site began 16 July 2010 and continued as weather permitted, with the last site visited on 5 August 2010. The total period of observation varied from 20–25 hr per site, given time constraints and the need to move equipment between sites on successive days.

Data were first used to examine broad patterns of visitation by different invertebrates. To accomplish this, we used total interaction hours to quantify activity of invertebrate morphotaxa at each sta-tion. Morphotaxa were identifiable to family (for insects) or order (for non-insect invertebrates). Interaction hours, quantified as the product of a video clip’s duration and the number of individuals active during that time, were calculated for each clip containing visitation by a morhpotaxon. For clips with a single individual, interaction hours were equal to the duration of the individual’s visit. For clips with multiple individuals, we recorded the number present at 10 equally spaced time intervals and calculated the total area under the resulting curve to obtain an estimate of interaction hours. Ants were again considered to be a single organism given their eusocial behavior. Interaction hours were then summed across all video clips at a station and averaged across the two stations at a site. For graphing, mean interaction hours for daylight (between sunrise and sunset) and nighttime hours were calculated to visualize broad trends in diel activity.

We also used camera data to quantify the frequency of interac-tions and distinguish taxa that acted as predators from those that were secondary feeders or never interacted with prey. For each morphotaxon, we measured the number of clips per station in which an individual 1) was the first predator feeding on sentinel larvae; 2)

fed on sentinel prey after another predator had already attacked it; or 3) was present in the field of view but did not interact with prey. An interaction was recorded only if an organism was in the field of view for one minute, and was considered over when the individual(s) were absent from the field of view for at least five minutes. These parameters allowed us to count instances where a single individual repeatedly left and reentered the field of view as a single interaction, and to eliminate clips where an organism passed through but did not interact with prey. For each morphotaxon, the frequency of clips with each type of interaction was averaged across the two stations at a site prior to analysis.

Next, we used camera data to further investigate predation of sentinel prey by ants. Prey mortality was measured by recording the number of sentinel larvae dead at each station at the end of video recording. We investigated ant predation rates in corn by relating prey mortality to total ant interaction hours accumulated at each station, using quadratic regression to quantify this relationship (Ne-ter et al. 1996). Data from grassland were not used, given that prey mortality was near 100% at all stations. The curve was fit with and without outlying data points from two video stations in corn, where ant activity was probably underestimated because there were a large number of small, unidentifiable organisms (likely small ants) in the field of view. Finally, we used data from both grassland and corn to determine how the number of ants recruited to prey varied with the duration of individual ant-prey interactions. (Parameters delimiting unique interactions are described above.) For this analysis, we only included clips in which ants were clearly feeding on sentinel larvae. To complete this analysis, we recorded the number of ants present in the field of view at 10 equally spaced time points spanning each interaction. We then used the maximum number of ants present as an index of ant recruitment to sentinel prey. This index was loge-transformed prior to analysis and related to the duration of each interaction using linear regression. All analyses were conducted in JMP vs. 8 (SAS Institute 2008).

ResultsBlueberry Pest Predation. There were no significant differences

detected for the number of visits to hibernaculae or number of individuals collected in pitfalls between ground covers for Formici-dae, Coleoptera, Gryllidae, Arachnida, or Chilopoda (P > 0.05). Ants accounted for the majority of visits to sentinel hibernaculae on all ground cover types, but were observed less frequently on woodchips than on grass, bare ground, or weed barrier (Fig. 6). Arachnida and Gryllidae were primarily observed in the woodchip treatment, while prey visits by Coleoptera were rare in all treatments (Fig. 6). Pitfall trap captures revealed a very different picture of potential predator community structure, with ants collected only in the grass cover treatment. Coleoptera made up a much higher percentage of pitfall trap captures than video observations (Fig. 6).

Japanese beetle eggs were visited much more frequently than cranberry fruitworm hibernaculae, with ants once again making the majority of visits to both sentinel prey types for all but the wood chip ground cover (Fig. 7). Gryllidae were observed most frequently in the woodchip cover and infrequently or not at all in the other ground covers (Fig. 7). Coleoptera were infrequently observed across all ground covers, while Arachnida were observed as frequently as ants on the weed barrier (Fig. 7). Pitfall data once again looked quite dif-ferent, with Coleoptera making up a comparatively large proportion of specimens and ants making up a small proportion of visits among


all four cover types (Fig. 7). Gryllidae were collected on all four cover types in similar numbers. No significant differences were detected for the number of events or number of individuals collected among ground covers for Formicidae, Coleoptera, Gryllidae, Arachnida, or Chilopoda. (P > 0.05).

Formicidae was the only morphotaxon observed visiting hiber-naculae with sufficient frequency to allow estimation of the diel rhythm: visits occurred almost exclusively during the photophase (Fig. 8). In contrast, Japanese beetle eggs were visited frequently by Formicidae, Coleoptera, Gryllidae, and Arachnida. Ants visited Japanese beetle eggs during both the photophase and scotophase, with slightly more visits during the scotophase (Fig. 9). Coleoptera, Gryllidae, and Arachnida visited Japanese beetle eggs almost exclu-sively during the scotophase (Fig. 9).

Predation in two bioenergy cropping systems. Omnivorous invertebrates were dominant predators in both corn and prairie grassland. Overall, nine morphotaxa of invertebrates were captured

on film (Fig. 10 a,b). Of these, ants (Hymenoptera: Formicidae) and crickets (Orthoptera: Gryllidae) spent more time interacting with prey than other taxa in corn, while ants and slugs were the most common visitors in grassland (Fig. 10 a,b). Harvestmen (Opiliones) and ground beetles (Coleoptera: Carabidae) also visited prey at lower levels in both habitats, while other beetles (besides carabids), flies (Diptera), grasshoppers (Orthoptera: Acrididae) and isopods were detected but spent minimal time interacting with prey (Fig. 10 a,b).

Our data highlight differences in the types of interactions in which different taxa engaged. Ants, slugs, crickets, harvestmen, carabids, and grasshoppers were observed feeding on live prey prior to at-tack by another predator, while other taxa only approached prey after attack by another predator (non-carabid beetles and diptera) or visited but never fed on prey (isopods, Fig. 10 c,d). Our findings also highlight differences in the temporal nature of interactions. For example, harvestmen engaged in frequent interactions with prey, particularly in corn (Fig. 10 c,d), yet spent less total time interacting

Fig. 6. The average proportion of pitfall captures (left column) and video observations (right column) comprised of formicids, arachnids, gryllids, coleopterans, chilopods, or other taxa visiting cranberry fruitworm hibernaculae in different ground cover treatments applied in a highbush blueberry planting. Numbers on wedges show the average number of individuals captured on video or in pitfall traps.

Fig. 7. The average proportion of pitfall captures (left column) and video observations (right column) comprised of formicids, arachnids, gryllids, coleopterans, chilopods, or other taxa visiting Japanese beetle eggs in different ground cover treatments applied in a highbush blueberry planting. Numbers on wedges show the average number of individuals captured on video or in pitfall traps.


than ants (Fig. 10 a,b). This suggests that harvestmen engaged in many short interactions with prey, while ant interactions were less frequent but were sustained over a longer time.

Camera data also suggested that ants were important predators in corn, and that longer interactions involved the recruitment of large numbers of foragers. Prey mortality in corn increased with the total duration of ant interactions at a station (Fig. 11a), and curves

fit with and without outlying data points both suggest a significant relationship between prey mortality and ant activity (Fig. 11a; with outlier: F2,11 = 5.06, P = 0.03, R2 = 0.48, without outlier: F2,9 = 90.4, P< 0.0001, R2 = 0.95). The best-fit curve (without outliers) suggests that it took ants three hours to kill a single larva and 15 hours to kill all four larvae at a station (Fig. 11a). These longer interactions involved a large number of ant foragers. The maximum number of ants observed in a clip increased exponentially with the clip’s dura-tion (Fig. 11b), with up to 150 individuals recruited to feed on prey in longer interactions.

DiscussionAs the old adage goes, “seeing is believing,” and the use of

field-deployed digital surveillance systems allows researchers to observe wild insect behavior in much greater detail than was previously possible. As in human surveillance, video data are par-ticularly useful because they can simultaneously capture definitive evidence pointing to what attacked prey, when and where predation occurred, and how it happened. Cameras can move investigators beyond establishing circumstantial linkages between the presence of a predator and predation to unambiguously determine the taxa involved. For example, our data showed that although beetles were more commonly collected in pitfall traps than ants (i.e., they were in the vicinity of sentinel prey during the experiment), ants were more active as predators of pests in blueberries (Figs. 6 and 7). Ants were not only important for predation of blueberry pests; they also dominated predator activity in both corn and grassland (Fig. 10). Ants

Fig. 8. The average number of ants visiting cranberry fruitworm hiber-naculae across all blueberry ground covers by study hour. Grey boxes denote the nighttime/scotophase.

Fig. 9. The average number of Formicidae (a), Coleoptera (b), Gryllidae (c) and Arachnida (d) visiting Japanese beetle eggs across all blueberry ground covers by study hour. Grey boxes denote the nighttime/scotophase.


are known to be important predators in tropical systems (Philpott and Armbrecht 2006), orchards (Huang and Yang 1987, Morris et al. 2002), and golf courses (Lopez and Potter 2000); our data also suggest they are important in small fruit crops, field crops, and grass-land. A recent sentinel prey study performed at the blueberry site also pointed towards ants as the primary predator of Japanese beetle eggs and cranberry fruitworm hibernaculae (Mason and Isaacs, per-sonal communication). Video data revealed other surprises as well, including observations of grasshoppers and slugs feeding on insect larvae (Fig. 10). While it is clear these organisms feed on both plant and animal matter (Lockwood 1988, Barker and Efford 2004), they are not often discussed in the context of predation of pest insects. This highlights that digital surveillance can broaden the cast of usual suspects (e.g., coccinellids and carabids) by providing evidence of unexpected predator-prey interactions—a result echoed by other research (Pfannenstiel et al. 2008).

Video data allow investigators of predator-prey interactions to determine the specific times when predation occurs. For example, in addition to documenting that ants could be important predators, our video data revealed that they were active during both daylight and nighttime hours, in contrast to crickets, carabids, and harvestmen,

which were more strictly night-active (Fig. 9 and 10 a,b). Interestingly, there appeared to be a change in the diel activity of ants on Japanese beetle eggs and fruitworm hibernaculae: ants visited hibernaculae primarily during daylight hours, but visited P. japonica eggs during both the photo- and scotophase (Fig. 8 and 9a). This could be due to differences in the identity of ant species visiting these prey or seasonal changes in their behavior.

Because cameras are tied to a spatial location, they can also be used to examine where predation by different predators occurs. For example, data from blueberry suggest that small-scale differences in microhabitat can affect predation in blueberry: ants were less active in blueberries surrounded by woodchips than in other ground covers (Figs. 6 and 7). Similarly, Frank et al. (2007) used video technology to document differences in the predator communities within differ-ent strata of vineyard vegetation, showing that different predators were active in the ground and canopy. Such data could potentially be combined with information on pest life history to help understand the types of predators that will attack pests as they move between different microhabitats to complete their life cycles.

Finally, video footage captures predator behavior to describe how predation occurs. Specifically, videography captures the frequency,

Fig. 10. The duration and frequency of interactions between different inverte-brate taxa and sentinel waxworm larvae at prey stations in corn (a,c) and prairie grassland (b,d). Interaction hours in corn (a) and grassland (b) were calculated by integrating the total time predators were observed interact-ing in video clips and the number of predators active during each clip for daylight and dark periods. For each station in corn (c) and grassland (d), we also measured the mean frequency of video clips in which an invertebrate entered the field of view but did not contact prey (“visit only”), was the first organism attacking a prey item (“first feed”), or fed on a prey item after another predator had at-tacked it (“second feed”). Error bars show +1 S.E.M. for the means of total interaction hours (a,b) and frequencies (c,d).


Table 1. Comparison of different methods of quantifying predation by generalist invertebrate predators.

Camera observation

Molecular gut content analysis

Visual observation

Type of data Identity of predators yes yes yes Temporal distribution of predation events yes no yes Spatial location of predation events yes yes* yes Predation rates yes yes yesCharacteristic of technique Effects on predator behavior low low potentially high Potential to alter prey behavior low (sessile prey) to high

(restrained, active prey) low low (sessile prey) to high

(restrained, active prey) In-field labor low low high Labor in lab high high low Temporal richness of observations high low intermediate Spatial extent of observations low high high Time required to develop technique intermediate high low Equipment cost intermediate intermediate low Cost to process data high high low*Only if predators are confined to a specific location

duration, and sequence of predation events and can be used to describe rates of these behaviors. For example, our data show that ants engaged in infrequent but long interactions with prey, while harvestmen made short but frequent visits (Fig. 10). In addition, both taxa attacked live prey prior to visits by another predator, sug-gesting they are more than just secondary feeders (Fig. 10 c,d). Our data also allowed us to examine recruitment of ants to prey over time and were used to parameterize a function that described changes in prey mortality as a function of the duration of ant-prey interactions (Fig. 11). This highlights the benefit of collecting data that inher-ently include information about the timing of events: it can allow researchers to describe rates of different behaviors (in this case ant recruitment and predation), as well as incidences of multi-predator interactions such as intraguild predation (Meyhöfer 2001). This could be especially useful for developing field-based functions that describe predator behavior, which could be used as components of predator-prey models (e.g., Bianchi and van der Werf 2004). These models can be used to explore the impacts of different habitat management techniques on biological control agents and explain why a technique succeeds or fails (Bianchi and van der Werf 2004).

While video surveillance can collect useful data, it has distinct advantages and disadvantages, which means it should be used to complement (rather than replace) other approaches to predator-prey studies (Table 1; also see Weber et al. 2008 for an insightful discussion about visual observation in general). The advantages of digital surveillance of insect behavior over previous generations of video recording include improved video resolution, improved availability and economy of equipment, and the ability to record over long periods of time. Advantages over traditional sentinel prey studies include the ability to record the broad taxonomic identity of predators as well as the time and duration of predation events (Table 1). One disadvantage of digital videography is that it provides a small spatial window of observation. Digital recorders and day/night cameras allow the observer to place a stationary experimental subject under constant surveillance. The data collected are a total census of what transpired around the subject. What is not provided is coverage of any events that may have taken place outside the camera’s field of view (which, in our example studies, was limited to less than a square meter). In contrast, a human observer might only be able to

focus on a single subject for a few hours at a time, but can shift his or her field of view to encompass a much greater spatial extent. Thus, it can be said that digital video techniques provide temporally rich but spatially poor data (Table 1).

Another major challenge when using digital video techniques is that field setup is quick, but processing video footage in the lab requires a significant investment of time. A simple four-treatment study involving four spatial replicates will yield sixteen days of footage (384 hours) for every study day. This translates to someone spending a great deal of time viewing video to identify clips in which the behavior of interest occurs. We have found that, once trained, a skilled video technician can process between two to six video frames at a time at 4x real speed, thus viewing 8–24 hr of field-collected video per hour. Even so, a day of video from the simple 16-camera study we conducted in blueberry (yielding 384 video hours) would require 2–6 full workdays (16–48 person hours) to initially process. In addition, a more detailed analysis of the interactions identified in this initial viewing requires additional time. It is likely that future developments in signal processing will allow computers to pre-identify clips of interest and reduce processing time. At present, the authors have not found an automated solution to this problem due to issues of changing field of view and depth of field, as well as false positives produced by movement of foliage and shadows. However, with continued increases in affordable computer power and the development of new visual algorithms, it is likely that this limitation will be overcome in the not-too-distant future.

A final advantage and disadvantage of videography concerns the opportunity for observation to alter natural insect behavior. An advantage of video cameras is that they eliminate any effect of hu-man presence on animal behavior. Such observer effects have been documented in a variety of systems (e.g., Baker and McGuffin 2007; also see references in Wade et al. 2005). On the other hand, prey must be confined to the camera’s field of view and placed in a visible position. This could alter predator-prey interactions if, for example, restraining active prey reduces its ability to defend itself, or if prey naturally occurs in hidden locations (such as beneath the soil surface) not accessible to current video technologies.

Given these advantages and limitations, we would suggest that videography can best enrich our understanding of predator-prey


interactions when it is used in combination with other available techniques (Weber et al. 2008). For example, molecular gut content analysis can be used to assay large numbers of predators to see which attack prey under natural conditions (Sheppard and Harwood 2005, Greenstone et al. 2010, Szendrei et al. 2010). Video data can then be used to determine where, when, and how these predation events oc-curred. Alternatively, human observation could be used if predation events are rare and widely dispersed, or if labor is not available to process video. The combined use of these techniques will uncover new insights about predator-prey interactions, allowing us to intel-ligently engineer agroecosystems to support key predators that were previously unknown and incorporate unexpected interactions (e.g., predation by pest herbivores) into models of food webs to examine their implications.

The applications presented in this paper only touch upon the potential uses of this technology. Ongoing experiments by the au-

thors include examinations of the effect of microclimatic conditions on insect response to pheromones, documentation of foliar and soil predators of key insect pests, identification of pollinators and disease vectors, and an examination of insect trap design efficiency. Additional applications are likely only limited by the researchers’ creativity and imagination.

AcknowledgementsWe would like to thank Brenda Sattersthwaite and Elizabeth

Kenyon for their assistance in analyzing blueberry data and Aaron Balogh, Erymas Birru, and Liz Racey for help conducting the bioen-ergy study. This work was funded in part by the DOE Great Lakes Bioenergy Research Center (DOE BER Office of Science DE-FC02-07ER64494), DOE OBP Office of Energy Efficiency and Renewable Energy (DE-AC05-76RL01830), the NSF Long-term Ecological Re-search Program, and the Michigan Agricultural Experiment Station.

References CitedAllemand, R., F. Pompanon, F. Fleury, P. Fouillet, and M. Bouletreau.

1994. Behavioural circadian rhythms measured in real-time by auto-matic image analysis: applications in parasitoid insects. Physiological Entomology 19: 1-8.

Anathakrishnan, T.N. 1994. Functional Dynamics of Phytophagous Insects. Science Publishers, Lebanon, NH, USA.

Baker, R.L., and M.A. McGuffin. 2007. Technique and observer presence affect reporting of behavior of damselfly larvae. Journal of the North American Benthological Society 26: 145–151.

Barker, G.M., and M.G. Efford. 2004. Predatory gastropods as natural enemies of terrestrial gastropods and other invertebrates, pp. 279-404. In G. M. Barker (ed.), Natural enemies of terrestrial molluscs. CABI Publishing, Cambridge, MA.

Bianchi, F.J.J.A., and W. Werf. 2004. Model evaluation of the function of prey in non-crop habitats for biological control by ladybeetles in agri-cultural landscapes. Ecological Modelling 171: 177-193.

Cardé, R.T., R.T. Staten, and A. Mafra-Neto. 1998. Behaviour of pink boll-worm males near high-dose, point sources of pheromone in field wind tunnels: insights into mechanisms of mating disruption. Entomologia Experimentalis et Applicata 89: 35-46.

Costamagna, A.C., and D.A. Landis. 2007. Quantifying predation on soybean aphid through direct field observations. Biological Control 42: 16–24.

El-Sayed, A.M., J. Godde, and H. Arn. 2000. A computer-controlled video system for real-time recording of insect flight in three dimensions. Journal of Insect Behavior 13: 881-900.

Fourcassie, V.J.L., and J.F.A. Traniello. 1995. Ant search behaviour analysis with a video frame grabber. Insectes Sociaux 42: 249-254.

Frank, S.D., and P.M. Shrewsbury. 2004. Consumption of black cutworms, Agrotis ipsilon (Lepidoptera: Noctuidae), and alternative prey by com-mon golf course predators. Environmental Entomology 33: 1681-1688.

Frank, S.D., and P.M. Shrewsbury. 2004. Effect of conservation strips on the abundance and distribution of natural enemies and predation of Agrotis ipsilon (Lepidoptera: Noctuidae) on golf course fairways. Environmental Entomology 33: 1662-1672.

Frank, S.D., S.D. Wratten, H.S. Sandhu, and P.M. Shrewsbury. 2007. Video analysis to determine how habitat strata affects predator diversity and predation of Epiphyas postvittana (Lepidoptera: Tortricidae) in a vineyard. Biological Control 41: 230-236.

Fry, S.N., M. Bichsel, P. Muller, and D. Robert. 2000. Tracking of flying insects using pan-tilt cameras. Journal of Neuroscience Methods 101: 59-67.

Greenstone, M.H., Z. Szendrei, M.E. Payton, D.L. Rowley, T.C. Coudron, and D.C. Weber. Choosing natural enemies for conservation biological control: use of the prey detectability half-life to rank key predators of Colorado potato beetle. Entomologia Experimentalis et Applicata 136: 97-107.

Hagler, J.R., and S.E. Naranjo. 1994. Qualitative survey of two coleopteran predators of Bemisia tabaci (Homoptera: Aleyrodidae) and Pectinophora

Fig. 11. (a) Mortality of sentinel prey increased non-linearly with ant ac-tivity and (b) the maximum number of ants observed in a video clip was exponentially related to the clip’s duration. Prey mortality is the number of waxworm larvae/station dead after 24 hr in the field. Lines were fitted with least-squares regression; two outliers (circled) were excluded to fit the regression line in (a) but are included here for reference.


gossypiella (Lepidoptera: Gelechiidae) using a multiple prey gut content ELISA. Environmental Entomology 23: 193-197.

Hardie, J., M. Holyoak, N.J. Taylor, and D.C. Griffiths. 1992. The combina-tion of electronic monitoring and video-assisted observations of plant penetration by aphids and behavioural effects of polygodial. Entomologia Experimentalis et Applicata 62: 233-239.

Hardie, J., and G. Powell. 2002. Video analysis of aphid flight behaviour. Computers and Electronics in Agriculture 35: 229-242.

Hardie, J., and S. Young. 1997. Aphid flight-track analysis in three dimen-sions using video techniques. Physiological Entomology 22: 116-122.

Huang, H.T., and P. Yang. 1987. The ancient cultured citrus ant. BioSci-ence 37: 665-671.

Huckschlag, D. 2008. Development of a digital infrared video camera sys-tem for recording and remote capturing. European Journal of Wildlife Research 54: 651-655.

Imafuku, M., and T. Ohtani. 2006. Analysis of coordinated circling and linear flights of a lycaenid butterfly species. Naturwissenschaften 93: 131-135.

JMP 2008. JMP 8.0.2 computer program, SAS Institute, Cary, NC.Justus, K.A., and R.T. Cardé. 2002. Flight behaviour of males of two moths,

Cadra cautella and Pectinophora gossypiella, in homogeneous clouds of pheromone. Physiological Entomology 27: 67-75.

Justus, K.A., S.W. Schofield, J. Murlis, and R.T. Cardé. 2002. Flight be-haviour of Cadra cautella males in rapidly pulsed pheromone plumes. Physiological Entomology 27: 58-66.

Kindvall, O., G. Nordlander, and H. Nordenhem. 2000. Movement be-haviour of the pine weevil Hylobius abietis in relation to soil type: an arena experiment. Entomologia Experimentalis et Applicata 95: 53-61.

Lockwood, J. A. 1988. Cannibalism in rangeland grasshoppers (Orthoptera: Acrididae): Attraction to cadavers. Journal of the Kansas Entomological Society 61: 379-387.

Lopez, R., and D. A. Potter. 2000. Ant predation on eggs and larvae of the black cutworm (Lepidoptera: Noctuidae) and Japanese beetle (Coleoptera: Scarabaeidae) in turfgrass. Environmental Entomology 29: 116-125.

Merfield, C.N., S.D. Wratten, and S. Navntoft. 2004. Video analysis of predation by polyphagous invertebrate predators in the laboratory and field. Biological Control 29: 5-13.

Meyhöfer, R. 2001. Intraguild predation by aphidophagous predators on parasitised aphids: the use of multiple video cameras. Entomologia Experimentalis et Applicata 100: 77-87.

Morris, T.I., W.O.C. Symondson, N.A.C. Kidd, and M. Campos. 2002. The effect of different ant species on the olive moth, Prays oleae (Bern.), in Spanish olive orchard. Journal of Applied Entomology 126: 224-230.

Neter, J., M.H. Kutner, C.J. Nachtsheim, and W. Wasserman. 1996. Applied Linear Statistical Models. McGraw-Hill, Boston, MA.

O’Neal, M., E. Zontek, Z. Szendrei, D. Landis, and R. Isaacs. 2005. Ground predator abundance affects prey removal in highbush blueberry (Vac-cinium corymbosum) fields and can be altered by aisle ground covers. BioControl 50: 205-222.

Paulson, G.S., and R.D. Akre. 1992. Evaluating the effectiveness of ants as biological control agents of pear psylla (Homoptera: Psyllidae). Journal of Economic Entomology 85: 70-73.

Pfannenstiel, R.S., W. Booth, E.L. Vargo, and C. Schal. 2008. Blattella asahinai (Dictyoptera: Blattellidae): A new predator of lepidopteran eggs in South Texas soybean. Annals of the Entomological Society of America 101: 763-768.

Pfannenstiel, R.S., and K.V. Yeargan. 2002. Identification and diel activity patterns of predators attacking Helicoverpa zea (Lepidoptera: Noctui-dae) eggs in soybean and sweet corn. Environmental Entomology 31: 232-241.

Philpott, S.M., and I. Armbrecht. 2006. Biodiversity in tropical agroforests and the ecological role of ants and ant diversity in predatory function. Ecological Entomology 31: 369-377.

Prasifka, J.R., N.P. Schmidt, K.A. Kohler, M.E. O’Neal, R.L. Hellmich, and J.W. Singer. 2006. Effects of living mulches on predator abundance and sentinel prey in a corn‚ soybean‚ forage rotation. Environmental Entomology 35: 1423-1431.

R Development Core Team. 2012. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna,

Austria. URL http://www.R-project.org.Riley, J.R., A.D. Smith, and B.W. Bettany. 1990. The use of video equip-

ment to record in three dimensions the flight trajectories of Heliothis armigera and other moths at night. Physiological Entomology 15: 73-80.

Schenk, D., and S. Bacher. 2002. Functional response of a generalist insect predator to one of its prey species in the field. Journal of Animal Ecology 71: 524-531.

Sheppard, S.K., and J.D. Harwood. 2005. Advances in molecular ecology: tracking trophic links through predator–prey foodwebs. Functional Ecology 19: 751–762.

Storer, J.R., S. Young, and J. Hardie. 1999. Three-dimensional analysis of aphid landing behaviour in the laboratory and field. Physiological Entomology 24: 271-277.

Sunderland, K.D., N.E. Crook, D.L. Stacey, and B.J. Fuller. 1987. A study of feeding by polyphagous predators on cereal aphids using ELISA and gut dissection. Journal of Applied Ecology 24: 907-933.

Symondson, W.O.C., K.D. Sunderland, and M.H. Greenstone. 2002. Can generalist predators be effective biological control agents? Annual Review of Entomology 47: 561-594.

Szendrei, Z., M.H. Greenstone, M.E. Payton, and D.C. Weber. Molecular gut-content analysis of a predator assemblage reveals the effect of habitat manipulation on biological control in the field. Basic and Ap-plied Ecology 11: 153-161.

Teixeira, L.A., M.J. Grieshop, and L.J. Gut. Effect of pheromone dispenser density on timing and duration of approaches by peachtree borer. Journal of Chemical Ecology 36: 1148-1154.

Wade, M.R., M.P. Zalucki, and B.A. Franzmann. 2005. Influence of ob-server presence on pacific damsel bug behavior: Who is watching whom? Journal of Insect Behavior 18: 651-667.

Weber, D. C., R.S. Pfannenstiel, and J.G. Lundgren. 2008. Diel predation pattern assessment and exploitation of sentinel prey: New interpreta-tions of community and individual behaviors, pp. 485-494. In P. G. Mason, D. R. Gillespie and C. Vincent (eds.), Procceedings of the Third Interna-tional Symposium on Biological Control of Arthropods, 8-13 February 2009, Christchurch, New Zealand. FHTET-2008-06. USDA Forest Service.

Willis, M.A., and T.C. Baker. 1994. Behaviour of flying oriental fruit moth males during approach to sex pheromone sources. Physiological Ento-mology 19: 61-69.

Wratten, S.D. 1994. Video techniques in animal ecology and behaviour. Chapman & Hall, London, UK.

Matthew J. Grieshop is an assistant professor at the Michigan State Univer-sity (MSU) Department of Entomology, where he serves as the organic pest management specialist. His research program focuses on the development of sustainable pest management tactics for organic farmers. Benjamin P. Werling is a postdoctoral associate in the MSU Department of Entomology at Michigan State University, where he conducts research for the Great Lakes Bioenergy Research Center. His research focuses on the factors that shape insect foodwebs in agroecosystems at scales ranging from individual farm fields to agricultural landscapes. Krista Buehrer is a Ph. D. student in Dr. Grieshop’s laboratory studying the integration of organic hog and apple production. Julia Perrone is a research assistant in Dr. Landis’ lab study-ing insect predation in bioenergy systems. Rufus Isaacs is the Berry Crops Extension Entomologist and Professor at Michigan State University where he directs a research and extension program focused on integrated pest management and crop pollination. Douglas A. Landis is a Professor of insect ecology in the Department of Entomology at Michigan State University and Biodiversity Team Leader in the Great Lakes Bioenergy ResearchCenter. His research focuses on the role of landscape structure in shaping insect-insect and insect-plant interactions.

Stay Connected!If you move, change jobs, or change contact information, be sure to update your member record online at www.entsoc.org by selecting “member resources,” then “member account management,” or email [email protected].

http://www.R-project.org

http://www.R-project.org

http://www.entsoc.org

Documents

Big Brother is Watching: Studying Insect Predation in the Age of