Using calls to estimate the abundance of inshore dolphins: a case study with Pacific humpback dolphins Sousa chinensis

Authors


Sofie Van Parijs, Norwegian College of Fisheries Science, University of Tromsø, 9037 Tromsø, Norway (e-mail sofievp@nfh.uit.no).

Summary

  • 1Assessing the number of animals in a population is a fundamental requirement for effective wildlife management. Determining this information for cetaceans can be logistically difficult, and the abundance of inshore cetaceans along most of the world's coastline is unknown.
  • 2In this study we illustrate the potential of using sound as a tool for estimating the abundance of inshore cetaceans, using Pacific humpback dolphins Sousa chinensis .
  • 3Acoustic recordings of humpback dolphins were made in conjunction with visual observations of school size from a land-based platform on Stradbroke Island, Australia.
  • 4The mean number of calls recorded per 3-min sample period was regressed against the number of dolphins observed using schools for which more than three sample periods were recorded. The relationship estimated number of dolphins = 2·39  × the mean number of calls per 3-min sample was used to estimate the number of dolphins in schools for which fewer than four samples were recorded. Comparing these results against known group sizes indicated that this estimation technique is acceptable.
  • 5Recordings could be made using remotely deployed hydrophone units or submerged autonomous units to provide information on the occurrence and group size of inshore delphinids within an area.
  • 6Inexperienced personnel could deploy and retrieve recording units. Analysis would require training to recognize sound types from different species, or computer-based sound recognition programs.
  • 7We conclude that simple techniques using phonations can provide estimates of relative abundance for one species of inshore cetacean. The technique is acceptable for groups of fewer than nine individuals and should be developed to assess its usefulness for studying other species.

Introduction

Human impacts on near-shore coastal environments have increased in recent years and are predicted to rise further (Cincin-Sain & Knecht 1998). Coastal fauna affected by increased human activities in inshore waters include small cetaceans such as coastal delphinids. Some human activities (particularly gillnet fishing) pose or may pose significant threats to the viability of some populations (Barlow, Gerrodette & Silber 1997; Taylor & Rojas-Bracho 1999; Dagrosa, LennertCody & Vidal 2000; Slooten, Fletcher & Taylor 2000). Unfortunately, outside North American and European waters, there are few reliable estimates of the abundance of coastal cetaceans. Reliable estimation of animal abundance is one of the very basic requirements for successful management of mammal populations, thus a lack of such data represents a significant gap in the scientific knowledge base and restricts the capacity for successful cetacean management.

Counts can be used to estimate the absolute abundance of animals when the detection function associated with the counts is known (Yoccoz, Nichols & Boulinier 2001). If this is not possible, counts can still provide estimates of relative abundance (or indices). Comparisons of estimates of relative abundance imply constant (but unknown) probabilities of detection (Yoccoz, Nichols & Boulinier 2001). In some cases assessing cetacean abundance can be relatively difficult and expensive. Most estimates of cetacean abundance are derived visually using distance sampling methods (aerial, boat or land based; Buckland et al. 1993a, 1993b; Forney & Barlow 1998) or vessel-based mark–recapture studies using photography to identify individual animals (Hammond, Sears & Berubem 1990).

Acoustic techniques offer a form of detection very different from visual detection. They can be used when estimating the absolute abundance of cetaceans with distance sampling techniques by providing a second, independent, source of detection. This can be used to improve estimates of the probability for visual detection of animals, particularly when the probability of observing animals on the track line (Buckland et al. 1993a) for visual sampling is known, or suspected, to be fewer than one (Chappell, Leaper & Gordon 1996; Barlow 1999). Acoustic studies have been used to record the presence of cetaceans in the absence of visual detection or to track animal movements (Clark & Ellison 1989; Clark & Bower 1992; Clark 1995; Clark et al. 1996; Stafford, Fox & Clark 1998; McDonald & Fox 1999; Norris, McDonald & Barlow 1999; Clark & Ellison 2000; Clark et al. 2000; Thode, Spain & Kuperman 2000; Gedamke, Costa & Dunstan 2001). Most previous acoustic-based studies have produced estimates of relative abundance. Also, most researchers using acoustic techniques to produce estimates of relative abundance have focused on baleen whales (McDonald & Fox 1999; Clark & Ellison 2000; Thode, Spain & Kuperman 2000), harbour porpoise Phocoenea phocoena (Chappell, Leaper & Gordon 1996) and Delphinus sp. (Goold 1998).

The abundance of small cetaceans (particularly delphinids and phocoenids) in inshore waters has been estimated using distance sampling techniques with vessel-based (Jaramillo-Legorreta, Rojas-Bracho & Gerrodette 1999; Hammond et al. 2002), land-based (Young & Peace 1999) and aerial observer platforms (Carretta, Forney & Laake 1998). For aerial surveys, detection probabilities tend to be reduced, as animals’ submergence time tends to be long relative to the time available for viewing them. Thus reliable estimates of g(0) (trackline detection probability) can be a problem (Cockroft et al. 1992; Laake et al. 1997). Photo-identification-based mark–recapture methods have also been used to provide abundance estimates and other information useful for managing some populations (Slooten, Dawson & Lad 1992; Wilson, Hammond & Thompson 1999; Durban et al. 2000; Thompson et al. 2000).

Environmental conditions (e.g. shallow or muddy waters) and behavioural characteristics associated with some near-shore species (e.g. vessel avoidance, unpredictable surfacing patterns, low inconspicuous profiles at the surface) render the use of visual sampling techniques difficult. In most cases, photo-identification studies are relatively expensive and time consuming. As funds for cetacean management are scarce in most developing nations, new techniques for estimating the abundance of inshore cetaceans are needed. Although acoustic survey techniques may be valuable, most delphinids have a broad acoustic repertoire that varies dramatically based on behavioural context (Herman & Tavolga 1980; Herzing 1996; Janik & Slater 1998; Janik 2000; Van Parijs & Corkeron 2001a). None the less, the relationship between the number of calls received and the number of dolphins within recording range requires further investigation.

Pacific humpback dolphins are distributed throughout near-shore coastal waters of south-east Asia and Australia, from eastern China in the north to New South Wales in the south (Rice 1998). Three species of Sousa are recognized currently (Rice 1998): Atlantic Sousa teuszi, Indian S. plumbea and Pacific S. chinensis. The status of Pacific humpback dolphins is poorly known throughout most of their range (Corkeron et al. 1997; Jefferson & Leatherwood 1997), although recent studies in Hong Kong waters have described their status there in detail (Jefferson 2000). They usually occur in schools of fewer than 10 animals, although occasionally schools of more than 20 individuals are observed, and are generally in shallow, brackish, estuarine and coastal marine waters (Jefferson & Karczmarski 2001). Their abundance has been estimated only in relatively small areas (fewer than 2000 km2) using photo-identification mark–recapture techniques (Corkeron et al. 1997; Jefferson 2000) or vessel-based line transect surveys (Jefferson 2000). Due to their tendency to remain submerged, their turbid habitat, and difficulties in reliably identifying delphinids to species, aerial surveys have proved unsuccessful as a means of estimating humpback dolphins’ abundance over larger areas (10 000 km2 or more; Corkeron et al. 1997; Preen et al. 1997).

Humpback dolphins have a diverse repertoire of sound, consisting of whistles, burst pulse, low frequency narrow band sounds and clicks (Zbiden et al. 1977; Schultz & Corkeron 1994; Van Parijs & Corkeron 2001a,b,c). Their sounds can be distinguished either visually (on spectrograms) by experienced observers or using statistical techniques from sympatric species in Australian waters, i.e. inshore bottlenose dolphins Tursiops aduncus and Irrawaddy dolphins Orcaella brevirostris (Schultz & Corkeron 1994; Van Parijs, Parra & Corkeron 2000). In this study, the relationship between school size and sound production was assessed as an initial step to applying acoustic techniques to estimate the abundance of inshore delphinids, using Pacific humpback dolphins in Moreton Bay, south-east Queensland, Australia, as a case study.

Methods

The study site was located off Amity Point, in the Rainbow Channel (153°26′N, 27°24′E) within Moreton Bay, south-east Queensland, Australia underwater detection of sounds (Fig. 1). Data collection took place from the 5 to 20 August 1999. Pacific humpback dolphins occur frequently in waters in and around the study site. Behavioural and acoustic recordings were made daily between 12:00 and 17:00 h from a land-based platform approximately 2 m above the low tide line, overlooking the study site. Schools in this area were most commonly observed feeding and socializing (Van Parijs & Corkeron 2001a).

Figure 1.

A map of Moreton bay, south-east Queensland, Australia, showing Stradbroke Island and an enlarged version of the study site within the Amity channel. The map of Amity channel shows the land (dark grey area), the surrounding sandbanks (light grey area) and the acoustic limit of the study site (striped area). An X marks the site from where visual observations and acoustic recordings were made.

The Rainbow Channel is part of the natural system of channels that is characteristic of tidal deltas. Permanent channel markers used for ship navigation were used as reference points in this study. The channel is approximately 800 m wide and bordered by shallow sandbanks, which are exposed for 1–2 h either side of low tide. Strong tidal currents run through the channel on incoming and outgoing tides. A rock wall was built to prevent the rapid rate at which the sediment is eroding the coastline at Amity Point. An artificial reef lay 150 m to the south-west of our study site.

The topography and vegetation around the study site limited the horizontal scanning range of our visual observations underwater detection of sounds (Fig. 2a). The acoustic limit of the study site was determined qualitatively by detecting the sound of the engines from passing vessels. Vessel noise in this study was comparable in amplitude and frequency composition to recorded dolphin signals. During the passage of single vessels (n = 63) a cut-off line (i.e. when the engines could no longer be heard) was determined and a compass bearing was taken of the observed vessel. These bearings were used to determine the acoustic limits, i.e. range of acoustic propagation (Fig. 2a). The acoustic limits were always within the visual limits, therefore observations were restricted to the area within the acoustic limits. The area encompassing the acoustic limits was divided into a 200 × 200-m grid and quadrants were numbered (Fig. 2b).

Figure 2.

A map of the study site within the Amity channel indicating: (a) the acoustic (AL) and visual limits (VL) of the study site determined using vessel engine noise; and (b) a 200 × 200-m grid overlaid onto the area encompassing the acoustic limit enabling rough distance estimates to be provided of each observed school.

observational methods

Visual observations of school size, composition, location and surface behaviour using scan sampling techniques were recorded for 5 min, starting when a school of dolphins entered the study area until it left. The quadrant location of each school within the grid was determined during each 5-min sample using fixed channel marker buoys as visual marks within the channel or on the sandbank (Fig. 2b). Schools observed outside the acoustic range were not used in these analyses. School composition was defined as either a single animal or any aggregation of dolphins where a member of the aggregation was within 10 m of any other member of the aggregation in the first 5 min of observation (modified from Corkeron 1997). An identification (ID) number was assigned to each school. When the number of individuals within a school changed, a new school ID number was assigned.

acoustic methods

Acoustic recordings of underwater sounds were made for the first 3 min of each 5-min behavioural observation period. A 3-min sample period was chosen in order to define a clear cut-off between one sample and the next. Recordings were made using a High Tech Inc. hydrophone (model HTI-96-MIN; Gulfport, MI; sensitivity −170 dB, flat frequency response 5 Hz–30 kHz, add ±1·0 dB) and a digital audio tape recorder, TCD–D8 (frequency response 5 Hz–22 kHz ± 1·0 dB). The hydrophone was suspended from the pier on a 1-m pole between 0·5 and 1·5 m under water during all recordings. All observations and recordings were made in Beaufort sea states < 3 without rain.

acoustic analyses

Acoustic recordings were digitized and displayed as spectrograms using the BatSound analysis PC software program (Pettersson Elektronik A.B. 1996). Sounds were separated into four different categories, very poor, poor, medium and good, based on the quality of the spectrograms (Fast Fourier Transforms, dt 10 ms, d.f. 102 Hz, FFT size 512). Sounds were classified as good quality when all details of the spectral contours were clearly visible. Medium quality was defined as a sound where most of the spectral contours were distinctly visible and the sounds could still clearly be classified according to type. Poor quality was when most of the spectral contours were distinctly visible and the sounds could still be classified according to type. Sounds were classed as very poor when it was impossible to distinguish any spectral characteristics. Poor quality sounds were excluded from the analyses.

All sounds visible as spectrograms (good, medium and poor quality sounds) were counted. A variety of sound types was observed during analyses including broadband clicks, burst pulses and frequency-modulated sounds (whistles) (Van Parijs & Corkeron 2001a). All sound types visible as spectrograms, except for broadband clicks, formed discrete units and were therefore easy to quantify. In order to quantify broadband clicks, we defined a broadband click as consisting of a sequence of several individual clicks. The broadband click ended when there was a period of silence between clicks of more than 0·3 s. When overlapping sounds were observed, if these were different sound types and clearly distinguishable from each other, they were counted individually. However, if any confusion existed, either when similar sound types overlapped or the two different types were not easily distinguishable from each other, only one sound was counted. Our aim was to provide a conservative estimate of group size.

statistical analyses

The total number of sounds was calculated for all 3-min samples recorded for each school of dolphins. Schools for which there were fewer than four samples per school were not used in initial analyses. Means and standard deviations of the number of sounds produced by each school were calculated. The relationship between the mean number of sounds produced by each group and the number of dolphins known to be in the group was tested using linear regression. The regression was weighted by the inverse of the standard deviation of each sample and the regression line was forced through the origin. Regression analyses were carried out using the R statistical analysis program (Ihaka & Gentleman 1996). The regression formula derived from this analysis was used to generate predictions of the number of dolphins in groups for which fewer than four samples were recorded, and the results were compared with observed group sizes.

underwater detection of sounds

The propagation properties of humpback dolphin sounds within the study area were examined. Recording sessions when only one school was present within the study area were used for these analyses. The recording level of the DAT recorder was set manually to 10 at the start of the study and was never adjusted throughout the study period. This enabled us to associate all recorded sounds within a sample with a quadrant location on the grid. For all burst pulses and whistles within a sample, the peak in the power spectrum (dB) was measured at the frequency with greatest energy (kHz). Power spectra were plotted against location (numbers 1–13 on the grid), distance from the hydrophone (divided into four categories: 0–200 m, > 200–400 m, > 400–600 m, > 600–800 m) and tidal state (divided into four intervals: 1 = low tide ± 1·5 h, 2 = incoming tide, 3 = high tide ± 1·5 h, 4 = outgoing tide). Broadband clicks and some burst pulses were excluded from this analysis as their frequency range exceeded the frequency range of the recording equipment (5000–22 000 Hz) (Van Parijs & Corkeron 2001a).

Results

A total of 316 3-min samples was recorded from more than 69 h of observation during 16 days. Eighty-one schools were sighted, varying in school size from a single individual to 15 individuals (Fig. 3a). Schools included one (n = 42), two (n = 22), three (n = 1) or no calves (n = 15). For all observations, schools spent 83% of their time in quadrants 4 (n = 52), 5 (n = 42), 8 (n = 55), 9 (n = 68), 12 (n = 24) and 13 (n = 21), between 0 and 400 m from the recording site (Fig. 3b). Schools spent only 1% of their time in quadrants 6 (n = 1) and 10 (n = 3), between 600 and 800 m from the recording site.

Figure 3.

Histograms showing (a) the number and range of school sizes observed throughout this study and (b) the number of schools observed in each quadrat.

Twenty-seven schools were recorded for four or more 3-min sampling periods. The initial regression of number of vocalizations against group size was significant (F1,26 = 157, P < 0·001, adjusted R2 = 0·8524; Fig. 4a). Examination of studentized residuals (Venables & Ripley 1999) indicated that three data points, all from groups of 10 or more animals, provided the worst fit to the regression (Fig. 4b). A better fit was obtained after removal of these points from the regression (F1,23 = 273·2, P < 0·001, adjusted R2 = 0·919; Fig. 4c) with no clear trend in the distribution of studentized residuals. The regression coefficient was 2·39 (± 0·14 SE).

Figure 4.

Scatter plots. (a) The regression of mean number of vocalizations against group size ( F1,26  = 157, P < 0·001, adjusted R2  = 0·85) using the full data set of schools for which four or more 3-min acoustic samples were obtained. The regression line ( y  = 2·67 x ) is shown. (b) Residual plot of (a). (c) The regression of mean number of vocalizations against group size ( F1,23  = 273·2, P < 0·001, adjusted R2  = 0·92) using the reduced data (see text for details). The regression line ( y  = 2·39 x ) is shown.

The regression formula was:

image(eqn 1)

where yi is the estimated number of dolphins in school i and xi is the mean number of calls received in 3-min samples recorded for school i.

Equation 1 was applied to the remaining 54 schools for which data were recorded. One school (a singleton) produced no calls during the sample period and so was excluded from analyses. When only one 3-min sample was recorded, the point sample was used as the ‘mean’ in the regression equation. Twenty-two of 53 (41·5%) school sizes were estimated correctly, and 42 of 53 (79·3%) school estimates were within ± one animal ( Table 1 ).

Table 1.  Results of applying the regression formula inline image ( yi is the estimated number of dolphins in school i and xi is the mean number of calls received in 3-min samples recorded for school i ) to 53 schools for which data were recorded but fewer than four 3-min samples were collected. SS, known school size from visual observations; ESS, estimated school size derived from the regression; n/a, cells for which no data could exist
SSESS < SS − 1ESS = SS − 1ESS = SSESS = SS + 1ESS > SS + 1
1n/an/a711
2n/a1443
324321
403312
511110
600220
700200
810000

Whistles and burst pulsed sounds showed transmission loss with distance, with significant differences between distances (anova: SS = 81 578, F3,1983 = 4983, P < 0·0001; Figs 5a and 6a) and between tidal states. The least transmission loss occurred during high tide (anova: SS = 32 122, F3,1983 = 351, P < 0·0001; Figs 5b and 6b). Quadrats 1, 2, 3, 6, 10 and 11, which were furthest away, showed the greatest transmission loss compared with the quadrats closer to the hydrophone, with significant differences between quadrats (anova: SS = 82 675, F3,1983 = 1398, P < 0·0001; Figs 5c and 6c).

Figure 5.

Plots of power spectra (–dB) of humpback dolphin whistles against (a) distance (0–200 m, > 200–400 m, > 400–600 m, > 600–800 m), (b) tidal state (1 = low, 2 = incoming, 3 = high and outgoing tide) and (c) quadrant location.

Figure 6.

Plots of power spectra (–dB) of humpback dolphin burst pulses against (a) distance (0–200 m, > 200–400 m, > 400–600 m, > 600–800 m), (b) tidal state (1 = low, 2 = incoming, 3 = high and outgoing tide) and (c) quadrant location.

Discussion

This study shows that it is possible to use the sounds of Pacific humpback dolphins to determine both their occurrence within an area, and to estimate their school size. As the regression analysis explained 92% of the variation in the data, the mean of all sounds produced by schools using our sampling design provides an acceptably precise indicator of school sizes for groups between one and nine individuals. This covers the sizes of most humpback dolphin schools (Jefferson & Karczmarski 2001).

We aimed for simplicity in developing all phases of this study. Counting the total number of sounds, rather than having to differentiate between sound types, means that detailed discrimination between call types was not necessary. Among the delphinids, the type of sounds produced by schools varies with behavioural state (Herman & Tavolga 1980; Herzing 1996; Janik 2000; Van Parijs & Corkeron 2001a). However, as behavioural state would be difficult to infer from acoustic analyses, we ignored behavioural states in our analyses. Further analytical simplicity was achieved by using a simple regression approach, rather than more sophisticated methods (Venables & Ripley 1999 chapters 6, 8 and 9). This resulted in a very simple predictive equation of the relationship between call numbers and dolphin school size.

It could be suggested that the test of the predictive equation, using data from schools where fewer than four 3-min samples were obtained, indicates that the technique is not particularly successful. We argue against this for two main reasons. First, estimating school sizes of cetaceans using visual techniques (the method of choice in most studies) can be equally unreliable or worse (for examples from humpback whales Megaptera novaeangliae and bottlenose dolphins, respectively see Corkeron et al. 1994; Corkeron 1997). Our visual estimates of humpback dolphin school size were unusually reliable. We had animals in close view from a land station by a team of at least three people, generally for extended time periods. Secondly, using schools for which few data points were available should reduce the precision of the technique, as means estimated from two or three points tend to be less reliable estimators than means from larger sample sets.

Dolphin whistles and burst pulses within the study area were attenuated more with increasing distance and decreasing tidal state. Within the study area the main behavioural activities of schools were feeding and socializing (although other activities were also observed; Van Parijs & Corkeron 2001a), and 83% of the schools occurred within 400 m of the recording site. This area encompasses the artificial reef and a rock wall that support a high diversity of fish species (Davie & Hooper 1998; Davie 1998). As dolphin schools occurred predominantly in areas where sound attenuation was limited, it is likely to have had a minimal effect on the results of this study.

An advantage to using sound for estimating seasonal occurrence and abundance is that acoustic data recording can be automated and operated remotely. Numerous researchers are currently engaged in developing reliable and robust techniques, using aerial or boat observations and acoustic recordings singly or in combination, with the aim of improving the detection, quantification and spatial relationships of cetacean species (Croll et al. 2001; DeNardo et al. 2001; Gordon 2001). Various acoustic recording techniques could be used for deployment in remote areas, such as single hydrophones or sonobuoys, multiple hydrophones in an array or autonomous bottom units (Clark, Ellison & Beeman 1986; Clark & Ellison 1988; Freitag & Tyack 1993; Frankel et al. 1995; Clark & Fristrup 1997; Clark & Ellison 2000). These have provided information on the presence or absence of some cetacean species within a given area as well as more detailed information on movements and relative abundance (Charif, Clapham & Clark 2001). Croll et al. (2001) combined the detection of blue Balaenoptera musculus and fin B. physalus whale sounds with visual surveys and related these to resource distribution. Sonobuoys with data storage capacities (Hayes et al. 2000) or autonomous remote bottom recorders (Clark, Borsani & Notarbartolo-di-Sciara 2002) seem likely to provide a viable option for estimating the relative abundance of inshore cetaceans.

The DAT recording equipment used in this study limited the observed sounds to those occurring within the frequency range of 5–22 kHz. Although several inshore species produce sounds that range above 22 kHz, the aim of this study was to use equipment that is commercially available and relatively simple to use. All sounds recorded in this study are detectable using equipment recording to 16 kHz, allowing the use of cheaper and more readily available standard audio-cassette recording equipment.

Another advantage of using sound for detection relates to species identification, which can be problematic in aerial surveys (Preen et al. 1997) or when relying on reports from relatively inexperienced observers (e.g. fishermen, management agency staff, members of non-government organizations). In cases where enough data are available, sympatric delphinid species can be distinguished by their sounds (Steiner 1981; Schultz & Corkeron 1994; Rendell et al. 1999), so acoustic analysis can provide an indication of the species present within an area. However, in order to identify sounds reliably to species, inexperienced personnel would have to be trained by experienced bioacousticians to recognize different sound types, to pass recordings to experienced personnel for analysis, or sound recognition programmes would need to be developed.

In this study we only tested the efficacy of acoustic techniques to estimate the size of humpback dolphin schools, but the technique should be applicable for other species that exhibit similar acoustic repertoires. This method would have to be re-evaluated if it were to be used on inshore dolphin species with very different acoustic repertoires, for example Irrawaddy dolphins Orcaella brevirostris (Van Parijs, Parra & Corkeron 2000) or Cephalorhynchus spp. (Dziedzic & De Buffrenil 1989; Dawson & Thorpe 1990; Dawson 1991).

Little is known about the abundance, distribution and habitat use of most species of inshore cetaceans. As these species are probably under significant and growing pressure from anthropogenic alteration of their environment, there is a need to overcome this paucity of information. In this study we show that a relatively simple acoustic technique can be used to provide reliable estimates of group sizes of humpback dolphins, despite these animals’ complex acoustic repertoire. We suggest that these techniques could be used in several ways. Hydrophone and recorder packages (of whatever sophistication possible) could be deployed by people working in coastal areas (e.g. researchers, ecotourism operators, management agency staff, members of fishermen's co-operatives or non-government organizations). Analysis of tapes will provide estimates of the number of schools using the area and of school size. With an appropriate design (Yoccoz, Nichols & Boulinier 2001) for package placement, estimates of the relative abundance of inshore cetaceans could be obtained from the many areas where at present no data are available. We also suggest that the usefulness of this technique for studying other species be assessed. However, it is important that all new applications of this technique be validated using visual techniques, at least for a subsample of the data.

Acknowledgements

We would like to thank Jan Taylor for providing access to the field site. This work was funded by grants from the Earthwatch Institute, the Natural Heritage Trust ‘Coasts and Cleans Seas’ programme and the Sea World Research and Rescue Foundation Inc. We are grateful to the R core development team for providing such a powerful Open Source tool. Fieldwork was carried out under permit from the Queensland Environmental Protection Agency, and with ethical approval from James Cook University. Thanks go to Chris Clark, John Gould and two anonymous referees for substantially improving and providing valuable input into the manuscript.

Ancillary