Effect of Revisitation Surveys on Detection of Wetland Birds with Different Diel Vocalization Patterns

Wildlife managers aim to obtain the most reliable distribution data from which to base conservation decisions (MacKenzie 2005; Potamitis et al. 2014). When monitoring bird populations, survey timing should ideally be matched to behavior (e.g., movement, diel vocalization patterns) to maximize detection (McCallum 2005; Nielson et al. 2014; Budka and Kokociński 2015). Time and financial constraints have made developing species-specific survey programs a prohibitive task; hence, data from multispecies surveys are more prevalent in species' habitat and management studies (e.g., Hepinstall and Sader 1997; Iglecia et al. 2012; Börger and Nudds 2014; Feldman and Mcgill 2014). Mismatch between survey timing and behavior can result in false absences when imperfect detection is not accounted for, and if used for subsequent analyses may lead to misleading species–habitat relationships and misguided conservation decisions (Dettmers et al. 1999; Drapeau et al. 1999). It is therefore important to identify bias in existing multispecies surveys to detect species with different diel vocalization patterns and to develop protocols to improve species detection for minimal field effort.

Birds exhibit a wide range of diel vocalization patterns, including dawn chorus, irregular vocalizations throughout the day, and nocturnal vocalizations (Staicer et al. 1996; La 2012). Dawn chorus refers to a well-defined pattern of increased song at dawn that steadily decreases around midday (Catchpole 1973; Catchpole and Slater 2008). However, standard multispecies surveys from long-term monitoring programs and studies that cover large geographic regions are conducted so that each location is surveyed once annually, in the morning for short periods to coincide with the songbird dawn chorus (e.g., Robbins et al. 1987; Francis et al. 2009; Jiguet et al. 2012; Matsuoka et al. 2014). Consequently, birds that sporadically vocalize throughout the day such as waterfowl may be incidentally detected or completely overlooked. One way to quantify the ability of standard surveys to detect species is to compare the results to more complete censuses. Comparison studies, however, have largely focused on songbirds (e.g., Buckland 2006; Campbell and Francis 2011; Rempel et al. 2013) or make comparisons relative to other morning-based surveys (e.g., Cimprich 2009; Newell et al. 2013; Baumgardt et al. 2014; but also see Tozer et al. 2006; Budka and Kokociński 2015). Here, we quantified how well point counts and morning acoustic recordings detected the number of occupied sites for birds that exhibit different diel vocalization patterns (songbirds that exhibit a dawn chorus and waterfowl that exhibit irregular vocalizations; Poole 2014) relative to estimates from acoustic recordings conducted throughout 24-h periods.

Waterfowl estimates from point counts and morning acoustic recordings may be considered atypical, because other ground-based waterfowl surveys exist such as flushing lines and line transects (Bibby et al. 2000). However, neither of these latter surveys are ideal for generating the large sample sizes required for species–habitat relationship studies, or in habitats that are difficult to traverse because they are time and labor intensive and disturb breeding activities (Fletcher et al. 2000; Tozer et al. 2006). These difficulties are likely why waterfowl data from large-scale, songbird-based monitoring programs are sometimes used for conservation studies (e.g., Niven et al. 2005; Costanzo and Hindman 2007; Forcey et al. 2011), although we acknowledge that using point counts for surveying waterfowl are not the norm. In this study, we primarily use waterfowl to represent a general group of birds that sporadically vocalize rather than to specifically investigate alternative methods for conducting waterfowl surveys.

One approach to reduce false absences is to repeat sampling at a given site, especially for species with less predictable vocalization patterns. Revisiting stations multiple times, however, can reduce the number of sites surveyed, decreasing spatial variation and statistical power (Verner 1988; Gutzwiller 1993). Automated acoustic recorders are a powerful tool to examine site revisitation because they can be left in the field and programmed to record at set schedules, allowing for repeated sampling without spending extra time traveling back to sites. To our knowledge, few studies have investigated the effect of revisitation timing (i.e., same days vs. different days) on species detection, and none have explicitly examined timing with respect to different diel vocalization patterns (e.g., Buskirk and McDonald 1995; Brooks et al. 2001).

Our objective was to determine passive survey designs that most efficiently detected the number of occupied sites for waterfowl and wetland-associated songbirds that exhibit different diel vocalization patterns. We used a two-fold approach: 1) we quantified bias of point counts (visual and auditory cues) and morning acoustic surveys (auditory cues only) to measure point-scale presence–absence, and 2) we examined which revisitation schedule—same day or different day sampling, as well as level of sampling effort—best maximizes detection for a set of dawn chorusing songbirds and irregularly vocalizing waterfowl species. We used presence–absence data obtained from 10-min acoustic recordings conducted every hour throughout 24-h periods for three consecutive days (720 min) as a standard for comparison. We expected that songbirds with more predictable diel vocalization patterns (i.e., dawn chorus) would return greater number of occupied sites by repeating visits at dawn, across different days, than at different hours within the same day. In contrast, waterfowl that vocalize irregularly would return greater number of occupied sites by repeating visits at different hours within a single day.

One hundred and eight sites were surveyed from late April to June 2011 and 2012 across a variety of bogs, fens, marshes, swamps, and lakes in Algonquin Park and Nippissing District in Ontario, Canada (Figure 1). Acoustic recordings were collected using a Song Meter SM2 (Wildlife Acoustics Inc., Maynard, MA) that has two omnidirectional microphones. At each site, a song meter was strapped to a tree less than 0.5 m in diameter at 1.2–1.4 m in height adjacent to each wetland and set to record in stereo at 22 kHz with 16-bit resolution for 10 min on the hour for three consecutive 24-h periods. Fourteen song meters were rotated among sites on a 4-d schedule as weather permitted. For each of the 41 sites surveyed in 2012, a single unlimited radius 10-min point count was conducted at the edge of each wetland between 0600 and 1100 hours (i.e., dawn and 5 h later for our study location; Robbins et al. 1987). Vocalizations were manually identified to species by using sound spectrograms (Hamming window, fast Fourier transform length: 512) from sound analysis software SYRINX-PC version 2.6 (Burt 2006). A species was considered to have occupied a site if they were detected at least once during any survey. Consequently, there were no issues regarding overlapping vocalizations because species could be detected in later recordings. All recording interpretation and point counts were performed by the same researcher.

Waterfowl study species (Anatidae; irregularly vocalizing birds; Poole 2014) included American black duck Anas rubripes, bufflehead Bucephala albeola, Canada goose Branta canadensis, common goldeneye Bucephala clangula, common merganser Mergus merganser, hooded merganser Lophodytes cucullatus, mallard Anas platyrhynchos, red-breasted merganser Mergus serrator, and ring-necked duck Aythya collaris. Mallard and American black duck are considered indistinguishable by vocalizations alone (Longcore et al. 2000; Sterry and Small 2009), so these data were combined for analyses (referred to as mallard in figures). Songbird study species (dawn chorusing birds; Poole 2014) included alder flycatcher Empidonax alnorum, common yellowthroat Geothlypis trichas, northern waterthrush Parkesia noveboracensis, red-winged blackbird Agelaius phoeniceus, and swamp sparrow Melospiza georgiana. Species were selected beforehand as part of a larger study on empirical tests of previously published coarse-scale distribution models (La 2015).

We examined bias in the number of occupied sites detected by point counts and morning acoustic recordings. This analysis was limited to sites surveyed in 2012 (N = 41) when point counts were conducted. The percentage of occupied sites detected (number of sites detected/total sites) for each species was calculated using three survey methods per site: 1) a 10-min point count, 2) a 10-min morning acoustic recording, and 3) extended acoustic recordings (i.e., 720 min over 3 d). The latter survey was used as an index of what we considered the actual number of occupied sites for which to compare the other two surveys. Morning acoustic recordings were a single 10-min recording selected randomly between 0600 and 1100 hours for each site similar to the time frame of the Breeding Bird Survey (Robbins et al. 1987). The percentage of occupied sites detected among survey methods was compared using Kruskal–Wallis tests. Pairwise comparisons between surveys were made using Wilcoxon rank sum tests with Bonferroni correction. Effect sizes for pairwise comparisons were calculated using the formula r = z/√N, where N was the total number of samples per pair.

We examined the effect of revisitation schedule on the number of occupied sites detected for birds with different diel vocalization patterns. This analysis was limited to species that were detected at more than one site (songbirds, five species; waterfowl, six species) in the extended acoustic recording data (720 min) from all sites. The total number of sites at which each species was present was calculated. We then subsampled the extended acoustic data and determined the number of sites in which each species was detected for the following 10-min, on-the-hour revisitation schedules:

• A

  Different day (30 min): 0800 hours for three consecutive days

• B

  Same day (30 min): 0800, 0900, and 1200 hours within a single day

• C

  Different day (90 min): 0800, 0900, and 1200 hours for three consecutive days

• D

  Same day (70 min): 0700–1200 and 2100 hours within a single day

• E

  Different day (210 min): 0700–1200 and 2100 hours for three consecutive days

Revisitation schedule selection was based on information collected from a previous study that exhaustively analyzed more than 16 million subsampling combinations to best estimate species richness (La and Nudds 2016). For each species, the number of occupied sites detected was converted to a percentage for easier comparisons (number of occupied sites detected/total occupied sites). The percentage of occupied sites detected among survey methods for songbirds and waterfowl, respectively, were compared using analysis of variance (ANOVA). Pairwise differences in the percentage of occupied sites detected among survey methods were examined using Tukey honest significant difference (HSD) tests for multiple comparisons. Effect sizes were calculated using point biserial correlation coefficients. All statistical tests were performed in R version 3.1.0 (R Development Core Team 2014).

In compliance with the data archiving policy (Wenburg 2011), we deposited the raw acoustic recording data (Data A1), the species lists from point counts (Data A2), and the global positioning system (GPS) coordinates for the 108 wetlands surveyed in this study (Data A3) available at http://www.vantla.com/JFWM.

For waterfowl, the percentage of occupied sites detected differed significantly among survey methods (10-min point counts, 10-min morning acoustic recording, and 720-min extended acoustic recordings) for five of eight species: Canada goose, hooded merganser, mallard, red-breasted merganser, and ring-necked duck (Kruskall–Wallis tests: χ² > 12.5; df =2; P < 0.005; Figure 2A). Pairwise comparisons among survey methods for all waterfowl showed that 10-min point counts and 10-min morning acoustic recordings captured 81 and 94%, respectively, fewer occupied sites on average than did 720-min extended acoustic recordings (10-min point counts vs. 720-min extended acoustic recordings, Wilcoxon rank sum tests: P < 0.05; r = 0.28–0.54 and 10-min morning acoustic recordings vs. 720-min extended acoustic recordings, P < 0.005, r = 0.28–0.60). No significant differences were found between 10-min point counts and 10-min morning acoustic recordings (Wilcoxon rank sum tests: P > 0.05).

For songbirds, the percentage of occupied sites detected differed significantly among survey methods for four of five species: common yellowthroat, northern waterthrush, red-winged blackbird, and swamp sparrow (Kruskall–Wallis tests: χ² > 7.9; df = 2; P < 0.05; Figure 2B). Pairwise comparisons indicated that 10-min morning acoustic recordings significantly detected 67% fewer occupied sites on average than did 720-min extended acoustic recordings for northern waterthrush, swamp sparrow, and red-winged blackbird (Wilcoxon rank sum tests: P < 0.05, r = 0.28–0.46). Ten-minute point counts significantly detected 67% fewer occupied sites detected on average than did 720-min extended acoustic recordings for common yellowthroat, swamp sparrow, and red-winged blackbird (Wilcoxon rank sum tests: P < 0.05, r = 0.30–0.53). No difference was found between 10-min morning acoustic recordings and 720-min extended acoustic recordings for common yellowthroat or between 10-min point counts and 720-min extended acoustic recordings for northern waterthrush (Wilcoxon rank sum tests: P > 0.05). No difference was found between point counts and morning acoustic recordings for songbirds (Wilcoxon rank sum tests: P > 0.05).

The total number of sites at which each species was detected varied (alder flycatcher = 16, common yellowthroat = 59, northern waterthrush = 29, red-winged blackbird = 33, swamp sparrow = 73, Canada goose = 33, common merganser = 11, hooded merganser = 15, mallard = 36, red-breasted merganser = 16, ring-necked duck = 28). For all birds, the percentage of occupied sites detected differed significantly among revisitation schedules (ANOVA for waterfowl: F_{4, 25} =12.16; P < 0.0005; for songbirds: F_{4, 20} = 8.87; P < 0.005). Increasing sampling effort increased the percentage of occupied sites detected (Figure 3).

For waterfowl, the most intensive revisitation schedule (different day 210 min) resulted in significantly greater percentage of occupied sites detected than all other revisitation schedules (Tukey HSD: P < 0.005; r = 0.67–0.91) and identified 60% of occupied sites (Figure 3A). No significant differences were detected between the other revisitation schedules (Tukey HSD: P > 0.2). For songbirds, the most intensive revisitation schedule (different day 210 min) resulted in significantly greater percentage of occupied sites detected than all other revisitation schedules (Tukey HSD: P < 0.05, r = 0.73–0.89) except for different day (90 min; Tukey HSD: P > 0.05) and identified 87% of occupied sites (Figure 3B). No significant differences were detected between different day (30-min), same day (30-min), and same day (70-min) revisitation schedules (Tukey HSD: P > 0.05). Significant differences were detected between different day (90-min) and same day (30-min; Tukey HSD: P < 0.05; r = 0.65), but not same day (70-min) or different day (30-min) revisitation schedules (Tukey HSD: P > 0.05).

Point counts and morning acoustic recordings underestimated the number of occupied sites detected for most species, regardless of diel vocalization pattern. Some monitoring programs use shorter point count durations than the 10 min used in this study (e.g., Francis et al. 2009; Bonthoux and Balent 2011); so, underestimates may be more marked than this study illustrated. No significant differences were detected between 10-min point counts and 10-min acoustic recordings.

False absences are an important concern for studies that analyze relationships between occurrence patterns and habitat characteristics at point scales (Dettmers et al. 1999). From a conservation perspective, false absences can result in misidentification of species-at-risk because of apparent “rarity” or selecting larger-than-needed protected areas that are more expensive to manage (He and Condit 2007; Hermoso and Kennard 2012). From an ecological perspective, underestimation will make species–habitat relationships more difficult to ascertain due to lack of samples that indicate presence (Tyre et al. 2003).

Species detection can vary with time of day and species with sporadic vocalizations may be more difficult to detect (Skirvin 1981; Farnsworth et al. 2002). Although several studies have investigated the effect of time of day on species detection (e.g., Ralph and Scott 1981; Verner and Ritter 1986; Gutzwiller 1993; Esquivel and Peris 2008), none have explicitly examined revisitation schedules based on differences on diel vocalization patterns. To our knowledge, only four studies have explicitly examined the effect of revisitation timing (same day vs. different day) on species detection. Buskirk and McDonald (1995) and Brooks et al. (2001) found no differences in number of species detected between same day (2–3 h apart) or different day site revisitation schedules. Drapeau et al. (1999) and Field et al. (2002) found that sampling on different days detected more species than same-day visits. All studies focused on forest birds (where songbirds comprised the majority of species analyzed) and were in the context of calculating species richness rather than species presence.

Similar to Buskirk and McDonald (1995) and Brooks et al. (2001), we found no significant difference between same-day and different-day revisitation schedules for the same amount of sampling effort (Figure 3), regardless of diel vocalization pattern. This result suggests that birds may be similarly detectable in the morning across days compared to within a single day. Same day sampling designs may be more useful for studies that aim to reduce traveling time between sites if using conventional survey methods (e.g., point counts) or have only a handful of acoustic recorders that need to be rotated frequently among sites to increase sample size. However, 30 min of sampling effort detected fewer occupied sites (60 and 16% occupied sites for songbirds and waterfowl, respectively; Figure 3), indicating that more intensive sampling effort is needed to obtain more accurate estimates.

From a practical perspective, studies often need to consider costs, as well as data accuracy when selecting survey methods (Gregory et al. 2004). For songbirds, the majority of occupied sites could be obtained by subsampling acoustic recordings (60–87% for 30–210-min sampling effort, respectively; Figure 3B), but less so for waterfowl (11–60% for 30–210-min sampling effort, respectively; Figure 3A). Poorer performance for waterfowl is likely due to the sporadic nature of their vocalizations (Poole 2014), making targeting specific times for optimal detection more difficult. As a result, accurate estimates of waterfowl from acoustic surveys require more intensive sampling effort, likely throughout 24-h time period, to maximize the likelihood of detection.

Study assumptions include that species were identified correctly and detection radii were similar for all survey methods. Species misidentification can lead to incorrect occupied sites detected. To reduce the likelihood of species misidentification, all recording interpretation and point counts were conducted by the same experienced researcher, and bird vocalizations were compared to known recordings of birds from the study region and were conservatively identified. Differences in detection radii would make direct comparisons among methods more difficult. We acknowledge this difficulty and highlight that no significant differences were found in the number of occupied sites detected between 10-min point counts and 10-min morning acoustic recordings in our study area (Figure 2). Although point counts had a higher potential for detections by having a visual as well as an auditory component, we found that most detections were heard, rather than seen, similar to other studies conducted in forested areas (e.g., Brewster and Simons 2009).

Spatial rarity and detectability should be considered when designing survey protocols. For example, point counts may be sufficient to detect species that are spatially rare but that frequently vocalize during the dawn chorus (e.g., olive-sided flycatcher Contopus cooperi; Altman and Sallabanks 2012; Ministry of Natural Resources and Forestry 2015). For species that are more sporadic in their daytime vocalizations, are visually cryptic, and spatially rare (e.g., yellow rail Coturnicops noveboracensis; Leston and Bookhout 2015; Ministry of Natural Resources and Forestry 2015), 24-h acoustic recordings may be required to ascertain whether a species is present or not. Last, if species are spatially rare, vocalize infrequently, but are more easily detected visually (e.g., bufflehead and common goldeneye; Gauthier 2014; Poole 2014; Holopainen et al. 2015), combinations of point counts and extended acoustic recordings may provide better estimates.

Occupancy modeling is a powerful tool that can be used to estimate true occupancy by using repeated visits, method combinations, and time of detection to calculate detection probabilities (e.g., Farnsworth et al. 2002; MacKenzie et al. 2003; Alldredge et al. 2006). However, overlooking species vocalizing behavior can result in potential bias in occupancy estimates. For example, if a species is not available for detection in the morning, repeated visits or conducting different surveys during that time would not be an accurate representation of detectability. This study demonstrated that survey timing and number of visits are important for accurate detection, particularly for species with sporadic vocalization patterns. Consequently, occupancy modeling methods that target guilds with different vocalization patterns should consider diel vocal behavior, so that surveys can be planned to be sufficiently repeated and include times of day that are most appropriate for data accuracy.

Although acoustic recordings conducted throughout a 24-h period provide the highest number of occupied sites detected of all surveys in this study, they also took the longest time to analyze (approximately 2 h for 720 min of recording). Analysis time to determine species presence may be substantially reduced by using automated classification methods, especially for studies interested in a few target species (e.g., Bardeli et al. 2010; Venier et al. 2012; Potamitis et al. 2014). Automated classifiers have been developed mainly for songbirds (e.g., Anderson et al. 1996; Briggs et al. 2012; Stowell and Plumbley 2014). For non-songbirds such as waterfowl, automated classifiers may be more difficult to develop because their vocalizations are lower frequency (low-frequency noise filters would erase them from recordings) and are structurally short and simple, reducing the number of features available for classification. Nonetheless, automated classifiers for non-songbirds would be a worthwhile endeavor that can substantially increase analysis efficiency of acoustic recordings.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Reference S1. Buskirk W, McDonald J. 1995. Comparison of point count sampling regimes for monitoring forest birds. Pages 25–34 in Ralph CJ, Sauer JR, Droege S, editors. Monitoring bird populations by point counts. General Technical Report PSW-GTR-149. Albany, CA: U.S. Department of Agriculture, Forest Service, Pacific Southwest Research Station.

Found at DOI: http://dx.doi.org/10.3996/082015-JFWM-077.S1 (1029 KB PDF); also available at http://www.r5.fs.fed.us/psw/publications/documents/gtr-149/pg25_34.pdf (1030 KB PDF).

Reference S2. La VT. 2015. Non-dawn vocalizations by birds, survey improvements and scale-dependent habitat selection. Doctoral dissertation. Guelph, Ontario, Canada: University of Guelph.

Found at DOI: http://dx.doi.org/10.3996/082015-JFWM-077.S2 (3487 KB PDF); also available at http://atrium.lib.uoguelph.ca/xmlui/handle/10214/9076 (3.405 MB PDF).

Reference S3. McCallum D. 2005. A conceptual guide to detection probability for point counts and other count-based survey methods. Pages 754–761 in Ralph J, Rich TD, editors. Bird conservation implementation and integration in the Americas: proceedings of the third international Partners in Flight conference. Albany, CA: U.S. Forest Service General Technical Report PSW-GTR-149.

Found at DOI: http://dx.doi.org/10.3996/082015-JFWM-077.S3 (208 KB PDF); also available at http://www.fs.fed.us/psw/publications/documents/psw_gtr191/psw_gtr191_0754-0761_mccallum.pdf (209 KB PDF).

Reference S4. Verner J. 1988. Optimizing the duration of point counts for monitoring trends in bird populations. Berkeley, California: U.S. Forest Service Research Note PSW-395.

Found at DOI: http://dx.doi.org/10.3996/082015-JFWM-077.S4 (894 KB PDF).

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any archived material. Queries should be directed to the corresponding author for the article.

To cite this archived material, please cite both the journal article (formatting found in the Abstract section of this article) and the following recommended format for the archived material.

Data A1. We deposited 72, 10-min raw acoustic recordings conducted hourly for three consecutive days by using song meters for each of 108 wetlands in Algonquin Park and Nippissing District in central Ontario, Canada, from early April to June 2011 and 2012 available at http://vantla.com/JFWM (198 GB FLAC).

Data A2. Species lists obtained by a single unlimited radius 10-min point count conducted at the edge of each wetland at the song meter location between 0600 and 1100 hours (i.e., dawn and 5 h later) for 41 wetlands surveyed in Algonquin Park and Nippissing District in central Ontario, Canada, in April to June 2012 available at http://vantla.com/JFWM (168 KB XLSX).

Data A3. GPS coordinates of 108 wetlands surveyed in Algonquin Park and Nippissing District in central Ontario, Canada, in early April to June 2011 and 2012, grouped by year, available at http://vantla.com/JFWM (4 KB KMZ).

We thank J. Cuthbert, F. Zhang, the Associate Editor of the Journal of Fish and Wildlife Management, and two anonymous reviewers for helpful comments. Thanks also to R. Rempel, J. Jackson, C. Debruyne, B. Steinberg, J. Robinson, and J. Baker for logistical support and to A. Zolderdo and M. Ackert for field assistance. Field research funding was provided by the Ministry of Natural Resources and Forestry, Algonquin Park, Domtar Inc., Mittigoog LP/Weyerhaeuser Company Limited, Resolute FP Canada Inc., and EACOM Timber Corporation. We also thank the University of Guelph, the Government of Ontario, and the Natural Sciences and Engineering Research Council of Canada.

Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.