Evaluation of field and analytical methods for estimating the population size of burrow-nesting seabirds from playback surveys

Abstract

Capsule Acoustic playback methods have been widely used to survey burrow-nesting petrels but playback stimuli typically yield low response rates. Digitally manipulated recordings of the calls of the European Storm Petrel were created with the aim of producing an acoustic “super-stimulus” which could be used to elicit a higher response rate to improve population census methods. However, the manipulated recordings elicited a lower response than those that were unmanipulated. We also compared two methods for estimating population size of European Storm Petrels based on playback responses and found that the du Feu method is more widely applicable and precise than the standard method widely used in the UK and Ireland.

Nocturnal burrowing-nesting Procellariiformes are notoriously difficult to census. James & Robertson (1985) first advocated the use of diurnal playback of conspecific song, to locate ‘apparently occupied sites’ (AOS). A standardized playback method (henceforth referred to as the ‘Standard Method’) was developed by Ratcliffe et al. (1998), Fowler (2001) and Mitchell et al. (2004) and was applied during the Seabird 2000 census, to provide the first systematic population estimates of the European Storm Petrel Hydrobates pelagicus, Leach's Storm Petrel Oceanodroma leucorhoa and Manx Shearwater Puffinus puffinus for Britain and Ireland (Mitchell et al. 2004). This Standard Method has subsequently been widely applied to survey burrow-nesting Procelariiformes in a variety of species and circumstances (Ambagis 2004, Rayner et al. 2008, Barbraud et al. 2009, Ryan & Ronconi 2011).

The Standard Method requires the calculation of a response rate measured over a minimum of 7 days at sample calibration plots within a colony, to derive a correction factor, which is then applied to the number of responses recorded on one-off (single) visits over the whole study site (e.g. a specific island) to give a population estimate. However, despite the substantial effort and expense invested in developing the playback methodology, the response rates from petrels are usually low. For example, it has been found when surveying the European Storm Petrel that only around 31% of playbacks elicited a response from nests that were known to be occupied (Ratcliffe et al. 1998). Similarly, Berrow et al. (2000) reported that 10% of White-chinned Petrels Procellaria aequinoctialis did not respond at all to playback surveys carried out weekly over an entire breeding season on South Georgia.

Response rate has also been found to differ significantly among years within a colony, and markedly among colonies. Ratcliffe et al. (1998) found the overall diurnal response rate of European Storm Petrel on Mousa (Shetland) in 1997 was 0.25, which is dramatically lower than the 0.75 rate recorded at the same colony in 1992. Likewise, the response rate recorded in 1999 on Priest Island (NW Scotland) was 0.41, whereas in 2004 this figure had dropped to 0.27. Response rates also vary with the quality and origin of the recording of the tape; Brown (2006) found that tape quality had a significant effect on the number of responses elicited from European Storm Petrels, and Ratcliffe et al. (1998) found that European Storm Petrels on Mousa were more likely to respond to a recording made on Great Skellig (SW Ireland) than to a recording made on Mousa, but response rate did not differ significantly between the recording made on Mousa and recordings from the UK Channel Islands, Skokholm Island (Wales) or St Kilda (NW Scotland). Studies on other species have also reported that some individuals are more likely to respond to playback surveys than others (Amy et al. 2010) and response rates are also known to vary in relation to volume of the playback recording and time of day/night and season (Ratcliffe et al. 1998). These authors found the highest response rate for the European Storm Petrel to occur at night (midnight response rate of 0.63 ± 0.002), but recognized that storm petrel census work would generally take place during daylight given the hazardous locations of most breeding colonies. During daylight hours Ratcliffe et al. reported a lower average response rate of 0.25 (± 0.01).

Another potential limitation of the playback methodology is that in some areas, nests of several individuals may be aggregated under a single boulder or in adjacent crevices and difficult to differentiate (personal observations). Consequently, the numbers of breeding birds in colonies where nests are aggregated are likely to be underestimated.

These low and unpredictably variable playback response rates require correspondingly large and uncertain correction factors which result in large confidence limits in the calculated population estimate and a lack of precision in the estimate of population trends over time. For example, playback data from Priest Island indicated a substantial decline in nest occupancy of the European Storm Petrel between 1999 and 2001 from 4370 to 2670 AOS, whereas population estimates based on mark–recapture of birds mist-netted early in the season suggest little change in the numbers in flight above the colony at night (Hounsome et al. 2006).

To address these methodological issues, a playback study was carried out on the population of European Storm Petrels on the island of Skomer (Wales, UK) and had two aims:

1.	to produce a ‘super-stimulus’ digital recording that would elicit a higher and more consistent response rate of nesting European Storm Petrels, resulting in improved census precision
2.	to evaluate the analytical techniques for extrapolating playback response rates to whole-island population estimates.

COMPARISON OF RESPONSE RATES TO DIFFERENT ACOUSTIC STIMULI

Between 21 and 31 May 2008, visits were made at night (between 23:00 and 03:00) to the main European Storm Petrel colonies on Skomer Island, to identify occupied burrows and to test the response rate of petrels to four different European Storm Petrel playback stimulus recordings. The recordings were: (1) a recording of a male and female duet from Skomer Island, (2) a recording of male song from Skokholm Island (approx. 3 km south of Skomer Island), (3) a recording of a male song obtained from the British Trust for Ornithology (provenance unknown) and (4) an additional recording of male song (provenance unknown).

The two recordings that elicited the highest response were selected and using the goldwave freeware program (http://goldwave.com), a set of seven playback stimuli of the European Storm Petrel male songs were created from (i) the recording from Skokholm Island, and (ii) the recording from the British Trust for Ornithology (provenance unknown). The burrow songs of male storm petrels are comprised of a series of ‘purr’ elements, interspersed with higher-pitched ‘ter-chick’ sounds (Robb & Mullarney 2008). Previous studies have indicated that the pitch of a song and its composition may influence response to playback in a wide range of bird species (Naguib et al. 2002, Christie et al. 2004). To test this, the original recordings (henceforth known as ‘Skokholm control’ and ‘BTO control’) were edited to remove either the ‘ter-chicks’, and/or ‘purr’ elements, and to increase or decrease the pitch of the song by one octave.

Between 26 June and 6 July 2008, 30 previously identified occupied burrows were visited in a randomized order between 05:00 and 09:00 GMT and one of the seven playback stimuli was played for 10 s, close to the burrow entrance. Each recording was played a total of 47 times over the period of 7 days. A total of 51 responses were recorded over the 7-day period.

A significant difference in response rate was found among the seven playback stimuli (χ² = 23.25, d = 6, P =< 0.001). The playback stimulus which elicited the highest response rate was the unedited song recorded from the nearby island of Skokholm, followed by the unedited song recording of unknown provenance obtained from the BTO (Fig. 1). The digitally manipulated recordings of songs elicited weaker responses from burrows on Skomer Island than did the unmanipulated control songs. Thus, there is currently no evidence that digitally manipulated songs result in increased response rates for petrel playback surveys. Likewise, a previous study of the Wilson's Storm Petrel Oceanites oceanicus found that digitally manipulating the frequency of elements in songs did produce some significant differences in response rate among experimentally manipulated songs, but manipulated songs did not elicit higher response rates when compared to unmanipulated songs (Bretagnolle & Robisson 1991). However, our results do add further support to the recommendations of Ratcliffe et al. (1998) and Brown (2006) that the content and provenance of recordings are important factors to consider before commencing any population census.

Figure 1. European Storm Petrel response rates to playback of manipulated and control recordings.

RESPONSE RATES FROM CALIBRATION PLOTS

The playback stimulus eliciting the largest number of responses from this initial survey (i.e. the Skokholm control recording) was then used to estimate the response rates within two calibration plots on the island, following the recommended playback methodology for this species (Mitchell & Dunn 2004). Two statistical techniques were applied:

a.

The recommended Standard Method based on the methodology of Fowler (2001) and developed by Mitchell et al. (2004) and Brown (2006), whereby calibration plot asymptote estimates are made by plotting cumulative detected AOS from each calibration plot against visit number, followed by the reciprocal transformation of both axes. The reciprocal of the y-intersect is the estimated asymptote for that calibration plot (Fowler 2001).

b.

The du Feu ‘mark–recapture’ estimate (based on the method of du Feu 1983), which has been previously used to estimate the number of AOS within a calibration plot (Bolton et al. 2010) by considering which sites elicit a response on each day of playback. A ‘mark’ is defined in this context as an AOS from which a response has been elicited, and a ‘recapture’ is an AOS from which a response has been elicited on a previous day. The total number of AOS within the plot is given by.

where n = number of nest locations found; p = population size; r = total number of ‘retraps’.

Reponses from a total of 21 AOS were recorded from calibration plot 1, with a mean of 6.08 AOS recorded per visit (95% CI 5.73–6.43). Reponses from a total of 19 AOS were recorded from calibration plot 2, with a mean of 4.22 AOS recorded per visit (95% CI 4.03–4.41).

The two statistical techniques for estimating the number of AOS within the calibration plot and hence the whole-island population, yielded quite different results (Table 1).

Table 1. Population estimates (with 95% confidence intervals) derived from both calibration plots and whole island estimates using the ‘Standard’ and ‘du Feu’ methods of calculation. The standard method was unable to predict the population of calibration plot 1 as an asymptote was not met after nine visits to the site.

	Estimate of number of AOS
Location	Standard method	du Feu method
Calibration plot 1	No estimate	24
Calibration plot 2	56	23
Calibration plots 1 and 2 combined	77	47
Whole island	107 (91–130)	64 (55–79)

WHOLE-ISLAND POPULATION ESTIMATES

(a) Standard method

Using this method an asymptote could be estimated for calibration plot 2 (of 56 AOS), but no asymptote could be estimated for calibration plot 1, due to the shape of the cumulative AOS curve, which had not levelled out to an asymptote even after nine repeat visits. Calibration plot 2 gave a mean daily plot response rate of 0.075 (bootstrapped 95% CI 0.063–0.089). This rate is thus the probability of a petrel responding to playback. This response rate gives a correction factor for the island of 13.3. Ideally, a mean daily response rate should be calculated from a minimum of two calibration plots, but as this was not possible as the number of detected AOS in calibration plot 1 did not reach an asymptote. The data from the two calibration plots were instead combined and considered as a single plot, giving a mean daily plot response rate of 0.13 (bootstrapped 95% CI 0.11–0.15) and hence a response rate correction factor for the whole island of 7.6.

(b) du Feu method

Using this statistical method, the number of AOS could be estimated for both calibration plots, with an estimated 24 and 23 AOS for plots 1 and 2, respectively. Combining the two calibration plot responses gave a mean daily response rate of 0.22 (bootstrapped 95% CI 0.18–0.26) and hence a response rate correction factor for the whole island of 4.6.

In total, 14 AOS were recorded on the first visit to all suitable and accessible habitats on the island. This value was converted to a whole island population estimate by multiplying by the correction factors derived from the Standard and du Feu methods calculated above. The whole-island population estimates were very different when the two different methods of analysis were used, leading to substantial differences in the whole island population estimate of almost 50%. The du Feu method estimated a lower number of AOS, a higher calibration plot response rate and hence a lower whole-island population estimate with a smaller range in confidence limits, compared to the Standard Method.

In summary, the du Feu method of analysis was found to be more widely applicable than the Standard Method, as it produced population estimates in situations where asymptote estimates of calibration plot AOS total could not be calculated, even after the recommended number of survey visits (9–12). Furthermore, the du Feu method of analysis generated more precise estimates of population size (with smaller confidence limits). These findings add support to the study of Bolton et al. (2010), who found that the du Feu method most accurately estimated the number of known nest sites when compared to the Standard Method.

ACKNOWLEDGMENTS

We thank the Wildlife Trust of South and West Wales for allowing us to carry out the fieldwork, and to the wardening staff on Skomer Island in 2008, in particular Jo and Dave Milborough and Dave Boyle, for their advice, expertise and logistical support. We are grateful to Dr Jonathan Green for his critical comments on this manuscript.