Improved Campephiline detection: An experiment conducted with the Magellanic Woodpecker

Abstract Woodpeckers can be difficult to detect, as they are often cryptic, secretive, occurring in low densities, and wary of humans. Several methods exist to detect woodpeckers (e.g., playback surveys, passive point counts), yet no research has established which technique best detects these elusive picids. Thus, we designed an experiment to determine which of three methods best results in a detection of Magellanic Woodpeckers (Campephilus magellanicus), and if weather variables influence detection probability. Mostly during austral summers 2015–2017, we (a) used a drumming device to simulate a double‐knock (i.e., territorial acoustical signal), (b) broadcasted a territorial call, and (c) passively listened (control) for Magellanic Woodpeckers. We conducted our experiment on Navarino Island, Chile, where the Magellanic Woodpecker is the sole picid. The drumming device most effectively influenced the likelihood of a woodpecker detection. The odds of a woodpecker responding to a double‐knock were 2.14 times more likely than responding to either a call or control. Moreover, the odds of a woodpecker detection decreased by 42% as wind increased by one category and decreased by 40% for every additional month (i.e., October–March), which was expected because woodpeckers become less territorial as the breeding season progresses. As Campephilus woodpeckers communicate via drums or double‐knocks, using a drumming device likely will be an effective technique to detect not only Magellanic Woodpeckers, but other woodpeckers within the Campephilus genus in Central and South America.

may be more easily detectable by their drums or vocalizations (e.g., Drever, Aitken, Norris, & Martin, 2008;Vergara et al., 2017), not all woodpeckers express the same easily detectable behaviors or only may be more easily detectable seasonally. Additional factors such as small population sizes (Haig, Belthoff, & Allen, 1993), large home ranges (Tanner, 1964), low densities (Vergara et al., 2017), and steep and varying topography increase the difficulty of detecting woodpeckers. Therefore, passive methods (i.e., no detection device used) are often less reliable than active methods (i.e., use of detection devices). Accordingly, the best detection method may depend on several factors, including species-specific behavior, habitat type, and season.
Various survey techniques have been used to estimate woodpecker abundances or densities. Such techniques include a variable-belt-width transect method (multiple species ;Lammertink, 2004), playbacks of calls and drums with territory mapping (Black Woodpecker [Dryocopus martius]; Kosinski & Kempa, 2007, Pileated Woodpecker [D. pileatus]; Renken & Wiggers, 1993), only playbacks (Pileated Woodpecker; Drever et al., 2008), and passive point counts followed by an active survey method (multiple species; Kumar & Singh, 2010, Magellanic Woodpecker [Campephilus magellanicus]; Vergara et al., 2017). Despite various methods to detect and estimate woodpecker abundances or densities, to our knowledge, the most effective technique encompassing both calls and drums has never been reported.
Related to the likely extinct Imperial (C. imperialis) and Ivorybilled (C. principalis) woodpeckers, the Magellanic Woodpecker The Campephilus genus contains 12 large-sized species (Winkler & Christie, 2002) that are native to the Americas. The Magellanic, however, is endemic to old-growth forests of southern South America (Short, 1970) and is listed as endangered or vulnerable throughout its Chilean distribution (Servicio Agrícola y Ganadero [SAG], 2015).
MAWOs live in family groups of 2-5 individuals (Chazarreta, Ojeda, & Lammertink, 2012) with an average home range size of 1 km 2 (Ojeda & Chazarreta, 2014). Particularly during the breeding season (i.e., mid-late austral spring to early-mid summer; Ojeda, 2004), MAWOs are highly territorial; disputes among family groups occur at home range boundaries or within territories (Soto et al., 2016). Disputes include aggressive behaviors such as chasing, double-knocks, recognition calls (Soto et al., 2016), and supplanting (i.e., hopping/dancing-like moves between woodpeckers on the same tree, A. L. Wynia, personal observation). Adult males are more aggressive, dominant (Chazarreta et al., 2012), and more frequently conduct a double-knock (A. L. Wynia, personal observation), yet MAWOs generally travel with their family group (Ojeda, 2004); therefore, woodpecker families are often detected instead of individuals. Although this woodpecker is an important keystone species and of local conservation concern (Ojeda & Chazarreta, 2014), no standard technique has been established to detect and monitor populations.
Here, we address the following five questions: (a) Which of three detection methods (i.e., call, double-knock, passive listening) is most effective in detecting MAWOs? (b) Which month is best to detect woodpeckers? (c) Does weather influence the likelihood of a woodpecker detection? (d) Does a specific detection method elicit a specific response type? (e) Does woodpecker detection time differ among detection methods? Importantly, this study only accounts for detection probability (i.e., the likelihood of detecting a woodpecker using three different methods) without accounting for imperfect detection (e.g., Royle, Nichols, & Kéry, 2005). To account for this, researchers could deploy transmitters on a subset of woodpeckers and conduct the detection experiment with known woodpecker locations to determine their detectability; that is, given a woodpecker is present, does it respond to different detection techniques and at what distances?
We designed an experiment to determine which of three detection methods would best elicit a MAWO detection. We predicted that the likelihood of a woodpecker detection would be higher with a drumming device (i.e., wooden, acoustical lure device used to simulate a double-knock [i.e., territorial acoustical signal, Short, 1970]; Figure 2) than either a playback or passive listening, because drumming resonates louder and farther than playbacks, especially in windy environments (Vergara et al., 2017, A. L. Wynia, personal observation). Thus, we also predicted that wind would decrease the likelihood of a detection, because sound attenuates more rapidly in windy conditions. Importantly, we used this drumming device as opposed to broadcasting a recorded double-knock with a speaker as the device could produce a louder sound that resonates more than anything broadcasted with our speaker (e.g., Vergara et al., 2017); this mimics the reality that a MAWO's double-knock can be detected farther through a forest than its call. Our main objective was to devise a detection technique that could then be used to estimate MAWO abundances or densities to better monitor population changes. Using active detection techniques to increase detection probability and estimate species abundances or densities is not uncommon (e.g., Jakob, Ponce-Boutin, Besnard, & Eraud, 2010;Michalczuk & Michalczuk, 2006;Vergara et al., 2017). This research can provide valuable information for conservation and land managers that should assist in further protecting the MAWO, its habitat, and by association, co-inhabitants as well.
Also, our results likely can provide a detection technique applicable for other Campephilus species.

| Study site
The MAWO is a resident species of Navarino Island, Chile (55°04′S, 67°40′W; Figure 3), the location of this study. Navarino is 2,528 km 2 (Lombardi, Cocozza, Lasserre, Tognetti, & Marchetti, 2010) and part of the Cape Horn Biosphere Reserve, which consists of an extensive archipelago in the Magellanic sub-Antarctic ecoregion at the southern end of South America (Rozzi et al., 2012). Relatively harsh climatic conditions exist throughout the year, and of the few tree species inhabiting Navarino, several are Nothofagus (i.e., southern beech). Moreover, the MAWO is the only Picidae species inhabiting the island (Rozzi & Jiménez, 2014).

| Methods
We conducted this experiment mostly during austral summers 2015-2017 (i.e., varying 3-or 4-month periods between 12 October-12 March). During a pilot study in summer 2015 (i.e., 25 January-12 March), we established 12 forested survey points along the accessible, northern coast of Navarino; these points were sampled for one For the active techniques, we either played a short territorial call (www.xeno-canto.org: XC52601) via a speaker (Altec Lansing Mini H20 model IMW257) at about 55 dB for approximately 10 s or simulated a double-knock with the drumming device. We did not measure the sound pressure level (dB) of the drumming device because it depended on multiple factors (e.g., substrate on F I G U R E 2 Wooden drumming device created to simulate a Magellanic Woodpecker (Campephilus magellanicus) double-knock (i.e., territorial acoustical signal) on Navarino Island, Chile, 2015-2017 F I G U R E 3 Navarino Island, Chile (55°04′S, 67°40′W), where detection methods for Magellanic Woodpeckers (Campephilus magellanicus) were compared in 2015-2017 which the device was placed, user's strength, location hit on the device). The device, created on Navarino from lenga wood (high deciduous southern beech, N. pumilio), had two 40-cm × 19.5-cm sides and two 9-cm × 19.5-cm sides that were inset by 7 cm on each long side ( Figure 2). Two open sides projected the sound produced by the double-knock that we created with sticks found in the forest. We repeated the active techniques three times (i.e., once about every 3.5 min) during a 10-min period or passively listened for 10 min. Using a Kestrel 3000 Wind Meter, we recorded average wind speed (km/hr) and temperature (°C); we also recorded cloud cover (%), start time of each simulation, woodpecker behavioral response (e.g., call, double-knock, visually approach, no response), detection time, and estimated distance from survey point (m) at first detection. We repeated this experiment 3-4 times (i.e., monthly) per field season between 04:45-15:30 local time, as woodpeckers are active and responsive throughout the day (Kumar & Singh, 2010;Vergara & Schlatter, 2004).

| Analyses
We performed all statistical analyses with R statistical software version 3.5.0 (R Core Team, 2018). We set the significance level at 5% and reported 95% confidence intervals (CIs) or limits (CLs), and means with standard errors. If CIs included 0, predictors were not significant. We checked for outliers (there were none) and multicollinearity among predictors (package usdm; Naimi, Hamm, Groen, Skidmore, & Toxopeus, 2014). Our global model was not overdispersed (ĉ = 0.99), nor was there multicollinearity among predictors (i.e., no variance inflation factor (VIF) value was >10).
There was no effect of year (p = .79, CI = −0.42-0.44) on the probability of a woodpecker detection; therefore, all years were combined, and survey points remained the only random effect in our mixed models.
For question 1 on the best detection method, we used a generalized linear mixed model (GLMM, package lme4; Bates, Maechler, Bolker, & Walker, 2015) with a binomial error distribution. The data set included all woodpecker detections during each 10-min survey.
For questions 2 and 3 on month and weather effect, respectively, we used a GLMM with an offset to account for uneven detections per month and used the first detection during each survey. We considered the following variables: temperature, cloud cover, wind speed, detection method, month, and survey time. We created categories for all variables but month (Table 1); wind speed categories followed the Beaufort wind force scale (WMO, 1970). We created our a priori global model based on all independent environmental and temporal variables, detection method, and relevant interactions; we created all possible model combinations (package MuMIn, function dredge) and used an information-theoretic approach with the Akaike Information Criterion corrected for small sample size (AICc; Burnham & Anderson, 2002) to select the best-supported model.
We applied the principle of parsimony if ΔAICc < 2. Additionally, to determine the magnitude of the effect of influential predictors, we computed odds ratios and reported 95% CLs. If CLs included 1, the predictor had no influence on the likelihood of a woodpecker detection.
For question 4 on response type, we used a multinomial logistic mixed-effects model (MLMM; package lme4; Bates et al., 2015) with response types (i.e., call, double-knock, other, and no) as response variables and used all detections during each survey. Finally, for question 5 on detection time, we used a mixed-effect ANOVA (package stats; R Core Team, 2018) and used the first detection during each survey.

| RE SULTS
The drumming device most effectively influenced the likelihood of a woodpecker detection (p = .02). The odds of a woodpecker responding to a double-knock were 2.14 times more likely than responding to either a call or control (Table 2). In general, the number of detections per survey point varied between 0-5, and the type of woodpecker response and number of responses to each detection method varied as well ( Note: Time 0 min indicates a detection occurred within the first min of a survey. Survey points 1-30 were visited seven times across two austral field seasons (2015-2017), whereas points 31-42 were visited three times during one field season (2015). "None" implies no method resulted in a detection. The order listed per row in detection method corresponds to the order in remaining columns. Abbreviations: Ca, call; Co, control; DK, double-knock; F, flying (heard, not seen); NA, not applicable; P, pecking; V, visual. occurred in monthly woodpecker detections (p < .01); for each additional month (i.e., October-March), the odds of detecting a woodpecker (for all methods) decreased by 40% (Table 2). Thus, we were more likely to detect a woodpecker earlier in the breeding season than later (Figure 4).
The best-supported models all included wind as a predictor of a woodpecker detection (Table 4), but the most parsimonious model contained wind only. Specifically, the odds of a woodpecker detection decreased by 42% as wind increased by one category (Table 2, Figure 5).
Regardless of detection method used, there was no difference in woodpecker response type (e.g., call, double-knock, visual, no response) between the control and call (z = 1.78, p = .08) or double-knock methods (z = 1.49, p = .14), nor between the call and double-knock methods (z = −0.28, p = .79). Finally, mean woodpecker detection time (i.e., at first detection) did not differ among detection methods (F 2,30 = 0.18, p = .84). Mean detection times were 3.3 ± 0.67 min for the control, 3.3 ± 0.70 min for the call, and 3.9 ± 0.54 min for the drumming device.

| D ISCUSS I ON
Woodpecker drumming, that is, rapid, repetitive strikes with a bill on a substrate that is not associated with foraging or excavating, is used for long-distance communication with conspecifics in mate selection and territoriality (Stark, Dodenhoff, & Johnson, 1998 and references therein). Given that double-knocking is the main long-distance territorial signal in Campephilus species (Short, 1970), we suggest researchers simulate double-knocks to increase detection probability.
We further recommend the use of a drumming device over broadcasting double-knocks with a speaker, because (a) speakers may not broadcast at the same sound pressure level (dB) and (b) speakers that could broadcast loudly can be more expensive. The speaker used in and is more cost-effective than a speaker.
Another benefit to the drumming device is its simplicity; it never needs to be charged, it will not die in the field, and batteries do not need to be replaced. Moreover, modifications to our drumming device to increase woodpecker detections may better assist researchers in detecting and monitoring woodpecker populations in the Campephilus genus. Modifications could include adjusting device dimensions, using different wood or drumming sticks, or training researchers to increase accuracy of drum mimicry. At a minimum, the drumming device should help establish baseline presence/absence data or contribute to occupancy modeling.
In this study, we report that MAWOs were 2.14 times more likely to respond when we used a drumming device. Similarly, Kumar and Singh (2010) reported that individuals of 11 woodpecker species in India were 2.2 times more likely to respond during a playback survey than during a visual/aural survey. However, they only broadcasted calls, not drums; therefore, the impact of drumming is unknown.
Furthermore, woodpecker territoriality was not discussed, which may have impacted their results.
MAWOs are highly territorial against conspecifics during the breeding season; therefore, an active detection technique (i.e., a drumming device) can elicit a detection more readily than passive techniques. MAWOs often approach the "intruder" (i.e., playback or Previous studies have reported increased woodpecker detections with use of playbacks (e.g., Kumar & Singh, 2010;Michalczuk & Michalczuk, 2006), but surprisingly, there was no difference be-  (Short, 1970), as is common with other woodpecker species (Lammertink, 2004). Their wings, however, produce a flapping sound (Short, 1970), which can be detected and uniquely identified particularly on Navarino (as it is the only picid) in a relatively quiet environment. If audible, these nuances may increase detectability for seasoned researchers, but likely will be missed by untrained or inexperienced investigators. Moreover, a drumming device additionally could assist novice or inexperienced researchers, as less skill is required to identify the species by call or drum because woodpeckers often respond or move toward the observer (Kumar & Singh, 2010 Detecting and monitoring woodpecker populations is particularly important as several species are declining or endangered (Mikusińki, 2006). Notably, little research has been conducted on the Campephilus genus (Ojeda, 2004). Our study suggests that a drumming device is an effective alternative to playbacks to establish baseline population estimates, a primary conservation objective for all Campephilus species, including MAWOs.

ACK N OWLED G M ENTS
We thank G. E. Soto and A. Smiley for the design and G. E. Soto for assistance with creating the drumming device; D. Cardona, S.
Cuadros, J. Malebrán, N. Jordán, and A. Savage for invaluable field assistance; S. Lewis for statistical assistance; J. Bednarz, G. E. Soto, and reviewers and editors for draft improvements; M. Lake for Figure 3 assistance; and the Institute of Ecology and Biodiversity grant ICM P05-002, CONICYT grant PFB-23, Partners of the Americas grant, Omora Foundation, and University of North Texas for providing financial and other support.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
ALW and JEJ conceived the ideas and designed the methodology, ALW collected the data, ALW and VR analyzed the data, and ALW led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

DATA AVA I L A B I L I T Y S TAT E M E N T
Original data and analyses pertaining to this research are available on the Dryad Digital Repository: https ://doi.org/10.5061/ dryad.78dj7t9.