Wildlife ResearchCSIRO PublishingPO Box 1139 (150 Oxford St)Collingwood, Vic. 3066, Australia

Telephone: +61 3 9662 7622Fax: +61 3 9662 7611Email: publishing.wr@csiro.au

Published by CSIRO Publishing for CSIRO and the Australian Academy of Science

w w w . p u b l i s h . c s i r o . a u / j o u r n a l s / w r

All enquiries and manuscripts should be directed to:

Wildlife Research

Volume 29, 2002© CSIRO 2002

P u b l i s h i n g

© CSIRO 2002 10.1071/WR01063 1035-3712/02/020221

Wildlife Research, 2002, 29, 221–224

Escape behaviour of feral horses during a helicopter count

Wayne L. LinklaterA and Elissa Z. CameronB

Ecology, Institute of Natural Resources, Massey University, New Zealand.ACurrent address and correspondence: Center for Reproduction of Endangered Species,

Zoological Society of San Diego, PO Box 120551, San Diego, CA 92112-0551, USA. Email: wll@mweb.co.zaBCurrent address: Mammal Research Institute, Department of Zoology and Entomology,

University of Pretoria, Pretoria 0002, South Africa.

Abstract. Animal escape behaviour in response to aircraft could influence the accuracy and precision of aerialestimates of population size but is rarely investigated. Using independent observers on the ground and in the air, werecorded the behaviour of 17 groups, including 136 individually marked horses (Equus caballus), during ahelicopter count in New Zealand’s Kaimanawa Mountains and compared the helicopter count with a ground-basedmark–resight estimate in the same area (20.5 km2). The helicopter induced running and changes in group size andcomposition in all horse groups that travelled from 0.1 up to 2.75 km before leaving the ground-observer’s view.One-tenth of marked horses were not counted and a quarter counted twice. The possible double-counting of a further23 (17%) could not be confirmed because only two of the three observers’ records concurred. Thus, the helicoptercount over-estimated the marked sub-population by at least 15% and possibly by up to 32%. The helicopter count(228 horses) was 16.9% larger than the mark–resight estimate (195, 95% CI = 157–234). We identify thecharacteristics of the helicopter count that stimulated horse escape behaviour and discuss how it should beconsidered in the design of aerial population-estimate methods.Horse behavi our duri ng an aer ial countW. L. Li nkl ater and E. Z. Cam eronWR01063W. L. Linklat er and E. Z. Cameron

Introduction

The accuracy and precision of aerial population estimatesvary with observer experience, aircraft type and altitude,weather conditions, season, vegetation, and animal activity,mobility, grouping and orientation (e.g. Caughley 1974;Frei et al. 1979; Kufeld et al. 1980; Gasaway et al. 1985;Wolfe 1986; Bleich et al. 1990; Bodie et al. 1995). Thebehavioural response of animals to being counted from theair is less often investigated than other factors that mightaffect an estimate such as vegetation cover and technique.Nevertheless, whether animals characteristically seek orbreak from cover, freeze or move, and disperse or grouptogether in response to an aircraft may have a profound effecton the accuracy and precision of population estimates(e.g. Bleich et al. 1990; Seber 1992). The anti-predatorbehaviour of the horse (Equus caballus) is characterised bygrouping and running to escape. Thus, if disturbed byaircraft, feral horses tend to break from cover if they are in it,run, and form into larger aggregations. We investigated theinfluence of the helicopter on feral horse behaviour during anaerial count conducted in New Zealand’s KaimanawaMountains and discuss its implications for the design ofaerial monitoring methods.

Methods

Study animal and site

The study area (20.5 km2) was the Argo Basin that surrounds a sectionof the Southern Moawhango River in the south-western KaimanawaRanges, New Zealand. It is a landscape of river flats and rolling to steepcountry vegetated by grass- and shrub-lands below plateaux and brokenescarpments on all sides. The origins, size, behaviour and ecology ofthe Kaimanawa feral horse, and the vegetation, topography and climateof its range, are described in detail elsewhere (Linklater et al. 2000;Cameron et al. 2001). The behavioural ecology of Kaimanawa feralhorses has been compared with that of other feral horse populations(Linklater 2000) and found to be similar. Horses were reliably andindividually identified by two adjacent 5×8 cm freeze brands on theirdorsal right rump and/or by documented variations in their colorationand white markings that could be clearly differentiated from a distance(e.g. Cameron and Linklater 2000; Linklater and Cameron 2000).

Observations of helicopter counting and horse behaviour

The helicopter counts were designed to census the horse population,rather than generate an estimate of population size based on sampling(Rogers 1991; Department of Conservation 1995). The helicopter countwas conducted from a Hughes 500 helicopter. The range of theKaimanawa feral horses was divided into count strata based loosely onwater catchments and delineated by geographical features such asescarpments, rivers and mountain ridges. The horses in each stratumwere counted by flying approximately parallel paths backward andforward across each stratum, moving from one path to the next adjacent

222 W. L. Linklater and E. Z. Cameron

path in sequence. In this way the helicopter moved systematically in aserpentine pattern beginning at one side of each stratum and ending atthe opposite side in an attempt to count all horses present. Thehelicopter was guided by a global positioning system and paths were300 m apart across the study area. The helicopter was flown atapproximately 60 knots ground speed and at approximately 60 m abovethe ground. Two counters in the helicopter were linked by two-wayintercom and when a group of horses was seen they counted the numberin the group. If necessary, they requested the pilot to circle a group ofhorses so that group size could be confirmed. Counters gave each groupa unique number and marked its location on a 1 : 50 000 scaletopographical map (Department of Conservation 1997).

In the morning of the helicopter count a ground observer recordedthe location and size of marked bands and individual horses in the ArgoBasin river valley (the central study area: Linklater et al. 2000) on a1 : 25 000 scale topographical map. Immediately prior to the helicoptercount the observer obtained a vantage with an approximately 300° view250 m above the Basin’s floor, which allowed the observer to follow themovement of those horses during the helicopter count. The groundobserver recorded their behaviour and movements, and their locations(on 1 : 25 000 scale topographical map) particularly whenever thehelicopter passed over or near them. An aerial observer was in thehelicopter during the count. This observer also recorded the locationand size of any marked bands they were able to identify from thehelicopter, as it passed near or over them, and whether or not the groupwas counted. The ground and aerial observers, but not the counters,were familiar with the unique marks of individual horses in thepopulation from previous work (e.g. Cameron et al. 1999; Linklateret al. 1999). The aerial observers and counters could communicate byintercom but the aerial observer did not contribute to the counting ofhorses and used the intercom only to confirm which groups werecounted and their recorded size.

When the records of the composition and location of a horse groupby ground and aerial observers concurred, but the group was notcounted, a ‘non-count’ was confirmed. When the aerial observeridentified a group of horses more than once and it was counted on eachoccasion then a double-count was recorded. The double-count wasconfirmed only if the records from the counters and observers concurredas to the identity and location of the marked group when it was countedon both occasions. The records of the movement and location of eachhorse group made by the ground and aerial observers were compared toconfirm each group’s identity on both occasions that it was counted. Adouble-count was confirmed if the location and size of the group asrecorded by the counters and aerial observer also concurred. When therecords of the location and composition of a group of horses by aerialand ground observers concurred on both occasions that it was counted,but this location did not concur with that recorded by the counters, it wasregarded as a possible but unconfirmed double-count.

Mark–resight population estimate

A sample of marked and unmarked animals in the Argo Basin on25 July 1996 was conducted two days prior to the helicopter count. Wewalked a circular route recording the size of all groups of horses and theidentities of marked individuals in them. The location of markedbreeding groups (bands) and bachelor males inside or outside the mark–resight area was known from regular resight sampling on average every9 days during the entire study (Cameron et al. 2001; Linklater et al.2001). Therefore, whether or not a marked band or bachelor male wasavailable to be resighted was known and closed-population mark–resight techniques could be used.

Associations between individual bachelor males are highly variableand could be considered independent sightings (Linklater et al. 2000).However, the membership of bands is stable, and therefore individualswithin a band are not independently sighted. Thus, bands, rather than

their individual membership, were treated as the sampling unit.Consequently, population estimates for the Argo Basin were calculatedseparately for bands and individual bachelor males. The size of thebachelor male population was estimated using NOREMARK mark–resight software and the Lincoln–Petersen estimate procedure (White1996). The population of horses in bands was generated by multiplyingthe estimate of the number of bands by the average size of marked andunmarked bands sighted or not-sighted in the following way:

(ms × Sms) + (mn × Smn) + (us × Sus) + (un × Sun)

where m = number of marked bands sighted (s) or not sighted (n),u = number of unmarked bands sighted (s) or not sighted (n), andS = average size of marked bands sighted (ms), marked bands notsighted (mn), unmarked bands sighted (us) or unmarked bands notsighted (un).

Where the probability of sighting marked and unmarked bands isthe same (see assumption (a) of the mark–resight method below) thenun = 1 – (mn/(ms+mn)) × us and the Sun is assumed to be the same as theaverage size of all marked and unmarked bands sighted during themark–resight event (i.e. the average size of all bands sighted, Sall = 6.71members per band). A total population estimate was arrived at byadding the population estimate of horses in bands to the estimate of thepopulation of bachelor males. The 95% confidence interval for the totalpopulation estimate was calculated by the square-root of the sum ofvariances from the band and bachelor male population estimates. Thevariance of the band population estimate was calculated from theaverage and variance in the number and size of marked and unmarkedbands sighted and not sighted in the following way:

Var(bands) = (uns2 × Var(Sall)) + (Sun

2 × Var(band)) – (Var(band) × Var(Sall))

where Var = variance.NOREMARK was used to generate Var(band) using the number of

marked bands available and sighted, and the number of unmarked bandssighted, as for the calculation of the bachelor male population andestimate variance.

Mark–resight estimates assume that (a) all bands or bachelor maleshad the same probability of being marked, (b) the marking of bands orbachelor males does not affect their re-sightability, and (c) all markedbands and bachelor males were correctly identified. The freeze-brandedindividuals were gathered for marking by helicopter muster in June1994. Other helicopter musters of the Argo Basin in 1995 gathered morethan 80% of resident horses and individuals returned to bands andhistorical home ranges after release (Linklater, unpublished data). Theremainder of marked bands were ‘marked’ by description of the uniquefeatures of their individuals. Thus, approximately 90% of bands in theresight area had marked members and provided high marked : unmarkedratios in resight events (unmarked bands and bachelor males neverconstituted more than half or three-quarters, respectively, of the totalnumber observed and occasionally all bands and bachelor malesobserved were marked). Freeze brands were small relative to the size ofthe horse and did not change horse visibility. Visibility was possibleacross the entire resight area and all bands and bachelor males in groupswere identified with binoculars or telescopes and by approaching themif necessary since they were habituated to the close proximity ofobservers (e.g. Cameron and Linklater 2000; Linklater and Cameron2000). Therefore, the assumptions of the mark–resight method weresatisfied.

The outermost coordinates of 889 groups of horses recorded during38 other mark–resight events (Cameron et al. 2001; Linklater et al.2001) were used to construct a minimum convex polygon (20.5 km2) ofthe area sampled by a resight event. The area sampled by the mark–

Horse behaviour during an aerial count 223

resight search is defined in this way because the minimum convexpolygon around the perimeter of the bands sighted during the search isrepresentative of the extreme values of the detection function and, thus,accurately describes the area that we visually sampled during oursearch. The number of horses counted from the helicopter within theboundary of the mark–resight area was determined by overlaying theminimum convex polygon on a copy of the map on which countedgroups were marked during the helicopter count.

Results

Before the helicopter count the ground observer located andconfirmed the identity of 17 groups with marked individuals(bands and associations of bachelor males) in the Argo Basin.The aerial observer identified and recorded the locations ofan additional 4 groups with marked individuals during thehelicopter count of the entire study area. Observations by theground and aerial observers show that the helicopter inducedan escape response, which included running, in all 17 of thegroups monitored by both observers. Horse groups travelledan average linear distance of 1 km and up to 2.75 km andcrossed an average of 3.5 helicopter paths and up to 10helicopter paths in response to the helicopter. These areminimum estimates of distances travelled since most (n = 14)horse groups disappeared from the ground observer’s viewstill running from the helicopter. Fifteen (88%) of the groupstravelled far enough to move across into adjacent helicopterpaths. Six (35%) of the groups crossed into adjacent countingstrata. The composition of 13 (76%) groups changed duringthe helicopter count by mixing with, and separating from,other groups with the consequent temporary gain or loss ofindividuals.

Comparisons between the records of the counters and twoobservers show that, of the 136 marked horses locatedimmediately prior to the helicopter count, 34 (25%) werecounted more than once, a further 23 (17%) may have beencounted more than once, and 13 horses (9.6%) were notcounted. Therefore, the helicopter count overestimated themarked sub-population by at least 21 (34 – 13: 15.4%) andpossibly by as many as 44 horses (34 + 23 – 13: 32.4%). Thehelicopter count yielded 228 horses and was 16.9% largerthan the mark–resight estimate of 195 (95% CI = 157–234)within the 20.5-km2 mark–resight area.

Discussion

The escape behaviour of Kaimanawa horses in response to thelow-flying helicopter was typical of that observed in otherungulate populations where it is also known to confoundreliable population estimates (e.g. Bleich et al. 1990). Threefeatures of the helicopter count in the Kaimanawa Mountainsappeared to motivate horse escape behaviour. Firstly, adjacenthelicopter paths were too close together, being 300 m apart.Therefore, running horses could travel into flight pathssubsequently taken by the helicopter before the helicoptercould complete the previous one. Secondly, the helicoptermoved systematically from one flight path to the next in

sequence across each count stratum. Therefore, some horseswere inadvertently herded into flight paths and strata not yetcounted. Thirdly, the helicopter flew about 60 m from theground. Low flying is likely to increase the escape responseof the horses and prevent the easy recognition of groups thathad already been counted because it resulted in changes ingroup size and composition as groups mixed and separatedduring escape. Thus, counters unable to follow the movementsof individuals during the count could not differentiate betweencounted and uncounted groups during the melee that resulted.The helicopter counts have these characteristics because theywere designed to census the population rather than samplefrom it to estimate population size. Absolute counts requirelow flying and intensive and systematic coverage of thelandscape that are more likely to motivate, and less likely todetect, horse escape behaviour than sampling regimes such asmark–resight (e.g. Caughley & Grice 1982; Pollock 1991;Bowden and Kufeld 1995) or line-transect (e.g. Hone 1988;Buckland et al. 1993) techniques.

A single retrospective count–remove–recount check of ahelicopter count in 1997 of the type described here suggestedthat it provided a reliable estimate over the entire population(Linklater et al. 2001) despite the double-counting of fleeinghorses that we describe here. Linklater et al. (2001) describeunder-counting from the helicopter in other parts of thepopulation’s range where either horse densities weresignificantly lower (i.e. 0.9 horses km–2) than the studypopulation described here (i.e. 5.2 horses km–2: Linklateret al. 2000) or where the horses were at similar densities(i.e. 5.0 horses km–2) but had not previously experiencedhelicopter mustering and capture. In contrast, the populationstudied here had recent experience of muster by helicopterand capture on at least 2 previous occasions (1994 for freeze-branding and 1995 for an immunocontraceptive trial:Cameron et al. 2001). Population management usinghelicopter mustering may encourage escape behaviour in thesame horses during subsequent aerial monitoring and affectpopulation estimates from a helicopter. The effect of densityis consistent with horse flight behaviour because groupsaggregate during flight, flight behaviour is contagious andgroups break from cover. Thus, at higher densities the flightresponse of one group is more likely to motivate the flightresponse of other groups, and thus also increase their visibilityand encounter or detection rate with observers. Thus, negativeexperience with helicopters (i.e. aerial mustering) may needto be minimised if aerial population monitoring is to be used,particularly where horse densities are high.

Accurate and repeatable estimates of population size,distribution and growth are necessary for the appropriatemanagement of populations. For the management ofcharismatic species, like wild horses, rigorous estimates ofpopulation size and change that can be defended inscientific, public and legal fora are imperative ifmanagement is to proceed with minimal hindrance

224 W. L. Linklater and E. Z. Cameron

http://www.publish.csiro.au/journals/wr

(e.g. Symanski 1996). The use of aircraft for estimatingpopulation size is commonplace (Seber 1992) but how theyare used varies regionally (New Zealand: Rogers 1991;Australia: Caughley and Grice 1982; Hone 1988; Pople et al.1998; North America: Gasaway et al. 1985; Bodie et al.1995; Bowden and Kufeld 1995; Pojar et al. 1995). Theliterature has so far ignored the problem posed by feral horsebehavioural responses to low-flying aircraft on thepopulation estimates using them. Although more work hasbeen done in the USA to validate an aerial counting methodfor feral horses (e.g. Garrott et al. 1991), differences fromthe true number were still occasionally large (e.g. 41% to112% detection: Garrott et al. 1991, p. 645, Table 2). Theinaccuracy, and low precision, of some aerial estimates mayresult, at least in part, from increases in the rate at whichhorses are counted twice or not counted due to their escapebehaviour. In New Zealand, large fluctuations in the numberof horses in adjacent areas between counts was attributed tounstable home ranges and disturbance by army trainingactivities (Rogers 1991) but could instead simply reflecthorse behaviour during helicopter counting (see alsoLinklater et al. 2000). Precautionary investigations of theinfluence of aircraft on horse behaviour are an important stepduring the design of aerial monitoring methods.

Acknowledgments

This work was funded by Department of Conservation(DOC) contract No. 1850 to Massey University and theEcology Group, Institute of Natural Resources (INR). We aregrateful for the assistance of the DOC, particularly Mr BillFleury (Wanganui Conservancy), who provided theopportunity. Our thanks also to Kevin Stafford, Institute ofVeterinary, Animal and Biomedical Sciences, Ed Minot(INR) and Clare Veltman (DOC) who supervised our Ph.D.s,during which this study was conducted. We thank staff ofOperations Branch, HQ; Property Management Section; andWaiouru Support Company, 4th Logistics Battalion, of theArmy Training Group, Waiouru, for logistical support andpermission to work in the Army Training Area. We alsothank Mr John Tulloch (Poronui Station) and fellowmusterers for assisting with branding of the study horses. Wethank Richard Barker (Mathematics and Statistics,University of Otago) for advice on the use and analysis ofmark–resight methods and data. Finally, we thank the threeanonymous reviewers.

References

Bodie, W. L., Garton, E. O., Taylor, E. R., and McCoy, M. (1995). Asightability model for bighorn sheep in canyon habitats. Journal ofWildlife Management 59, 832–840.

Bleich, V. C., Bowyer, R. T., Pauli, A. M., Vernoy, R. L., and Anthes, R.W. (1990). Responses of mountain sheep to helicopter surveys.California Fish and Game 76, 197–204.

Bowden, D. C., and Kufeld, R. C. (1995). Generalized mark–sightpopulation size estimation applied to Colorado moose. Journal ofWildlife Management 59, 840–851.

Cameron, E. Z., and Linklater, W. L. (2000). Individual mares biasinvestment in sons and daughters in relation to their condition.Animal Behaviour 60, 359–367.

Cameron, E. Z., Linklater, W. L., Stafford, K. J., and Veltman, C. J.(1999). Birth sex ratios relate to mare condition at conception inKaimanawa horses. Behavioural Ecology 10, 473–475.

Cameron, E. Z., Linklater, W. L., Minot, E. O., and Stafford, K. J.(2001). Population dynamics 1994–98, and management, ofKaimanawa wild horses. Science for Conservation No. 171. viii +165 pp.

Caughley, G. (1974). Bias in aerial survey. Journal of WildlifeManagement 38, 921–933.

Caughley, G., and Grice, D. (1982). A correction factor for countingemus from the air, and its application to counts in Western Australia.Australian Wildlife Research 9, 253–259.

Department of Conservation (1995). Kaimanawa Wild Horses Plan.Department of Conservation, Wanganui Conservancy, Wanganui,New Zealand.

Department of Conservation (1997). Kaimanawa wild horse populationestimate: draft summary report. Wanganui Conservancy, Wanganui,New Zealand.

Frei, M. N., Peterson, J. S., and Hall, J. R. (1979). Aerial census of wildhorses in western Utah. Journal of Range Management 32, 8–11.

Garrott, R. A., Siniff, D. B., and Eberhardt, L. L. (1991). Growth ratesof feral horse populations. Journal of Wildlife Management 55,641–648.

Gasaway, W. C., Dubois, S. D., and Harbo, S. J. (1985). Biases in aerialtransect surveys for moose during May and June. Journal of WildlifeManagement 49, 777–784.

Hone, J. (1988). A test of the accuracy of line and strip transect estimatorsin aerial survey. Australian Wildlife Research 15, 493–497.

Kufeld, R. C., Olterman, J. H., and Bowden, D. C. (1980). A helicopterquadrat census for mule deer on Uncompahgre Plateau, Colorado.Journal of Wildlife Management 44, 632–639.

Linklater, W. L. (2000). Adaptive explanation in socio-ecology: lessonsfrom the Equidae. Biological Reviews 75, 1–20.

Linklater, W. L., and Cameron, E. Z. (2000). Tests for cooperativebehaviour between stallions. Animal Behaviour 60, 731–743.

Linklater, W. L., Cameron, E. Z., Minot, E. O., and Stafford, K. J.(1999). Stallion harassment and the mating system of horses.Animal Behaviour 58, 295–306.

Linklater, W. L., Cameron, E. Z., Stafford, K. J., and Veltman, C. J.(2000). Spatial and social structure and range use by Kaimanawawild horses. New Zealand Journal of Ecology 24, 139–152.

Linklater, W. L., Cameron, E. Z., Minot, E. O., and Stafford, K. J.(2001). Estimating Kaimanawa feral horse population size andgrowth. Science & Research Internal Report No. 185. 40 pp.

Pojar, T. M., Bowden, D. C., and Gill, R. B. (1995). Aerial countingexperiments to estimate pronghorn density and herd structure.Journal of Wildlife Management 59, 117–128.

Pople, A. R., Cairns, S. C., Clancy, T. F., Grigg, G. C., Beard, L. A., andSouthwell, C. J. (1998). An assessment of the accuracy of kangaroosurveys using fixed-wing aircraft. Wildlife Research 25, 315–326.

Rogers, G. M. (1991). Kaimanawa feral horses and their environmentalimpacts. New Zealand Journal of Ecology 15, 49–64.

Seber, G. A. F. (1992). A review of estimating animal abundance. II.International Statistical Review 60, 129–166.

Symanski, R. (1996). Dances with horses: lessons from theenvironmental fringe. Conservation Biology 19, 708–712.

White, G. C. (1996). NOREMARK: population estimation from mark–resighting surveys. Wildlife Society Bulletin 24, 50–52.

Wolfe, M. L. (1986). Population dynamics of feral horses in westernNorth America. Equine Veterinary Science 6, 231–235.

Manuscript received 13 June 2001; accepted 30 April 2002