Spatio-temporal variation in cat population density in a sub-Antarctic environment

Polar Biol (2002) 25: 90±95 DOI 10.1007/s003000100316

O R I GI N A L P A P E R

Ludovic Say á Jean-Michel Gaillard á Dominique Pontier

Spatio-temporal variation in cat population density in a sub-Antarctic environment

Accepted: 14 July 2001 / Published online: 1 September 2001 Ó Springer-Verlag 2001

Abstract We used the walked-line transect method for estimating the density of cats and coecients of variation of density estimates in 4 contrasted sites on the main island of IÃles Kerguelen between 1998 and 2000. Density estimates varied from 0.44‹0.15 cats per km2 to 2.42‹0.23 cats per km2 according to site and period. Coecient of variation of density estimates ranged from 11.92% to 34.76%. The line transect method was, therefore, an ecient method for monitoring the density of the cat population in a sub-Antarctic environment characterised by short vegetation. Our results suggest that cat population size at the main island of IÃles Kerguelen (the total number of cats expected is around 7,000) is much lower than previously thought.

Introduction The domestic status of the cat, Felis catus, as well as its capability of controlling rodents have led to either accidental or deliberate introductions of this species in most parts of the world (Fitzgerald and Turner 2000). Cats have very ¯exible behaviour (Liberg et al. 2000) that allows them to survive and multiply on islands when totally or partially deserted by humans. They are present on all continents and in at least 118 of 131 main groups of islands (King 1985). Insular ecosystems more than continental ecosystems have su€ered from the introduction of non-native species: 93% of the 176 bird species and subspecies extinct since the beginning of the seventeenth century were living on oceanic islands (King L. Say (&) á J.-M. Gaillard á D. Pontier U.M.R. C.N.R.S. no. 5558 ``BiomeÂtrie et Biologie eÂvolutive'', Universite Claude Bernard Lyon I, 43, boulevard du 11 novembre 1918, 69622 Villeurbanne, France E-mail: [email protected] Tel.: +33-4-72431337 Fax: +33-4-78892719

1985). Cats are among the most notorious and harmful introduced predators for the many bird species that reproduce on islands (van Aarde 1980; Fitzgerald and Veitch 1985; Johnstone 1985; van Rensburg and Bester 1988). For example, on Marion Island, about 4 cats introduced in 1949 to control house mice, Mus musculus, resulted in a population of 2,139 individuals 25 years later which were killing approximately 450,000 burrowing petrels per year (van Aarde 1980). At IÃles Kerguelen, cats are involved in the decline of several bird populations (burrowing petrels in particular, Jouventin et al. 1984). It was estimated that cats killed more than 1.2 million individual birds per year in the 1970s (Pascal 1980). The high reproductive output of cats (up to 2 litters of 3±4 kittens per year, Derenne 1976), their longevity (8 years, Pascal 1980), the absence of any predator or competitor (Johnstone 1985) and the low level of parasitism that occurs on Kerguelen (Fromont et al., in press), all accounted for the rapid colonisation of the main island by cats, with a population growth rate higher than 40% (Derenne 1976; Pascal 1980). Using sampling methods based on direct counts or hunting, the population was estimated at 1,750 cats in 1971 (Derenne 1976) and 3,500 cats in 1977 (Pascal 1980), although eradication programs were implemented in the 1970s (Pascal 1980). Using an exponential population growth model, the population size was expected to reach several tens of thousands of cats in the early 1990s (Pascal 1980; Chapuis 1995). The long-term monitoring of the cat population of the main island of IÃles Kerguelen began in 1995. The main objectives were: (1) to estimate whether the population is still in expansion or has reached equilibrium, and (2) to analyse the relationship between cat density, habitat structure, climatic conditions and prey availability. Answering these questions requires close monitoring of the cat population density. Although several methods have been used to estimate abundance of cats (e.g spotlight counts and live or dead captures), they often led to large coecients of variation and therefore to a low precision (e.g. Derenne 1976; Jones and Coman

91

1982; Brothers et al. 1985; Short et al. 1997; Edwards et al. 2000). Here, we used the line transect method to estimate cat density on IÃles Kerguelen. This method has been successfully applied to estimate population densities for a large number of vertebrates (Buckland et al. 2000). We report the results of 2 years of monitoring the cat population in di€erent parts of the main island.

Materials and methods The IÃles Kerguelen are a French possession, situated at 49°20¢S, 70°20¢E. The main island of IÃles Kerguelen (some 6,675 km2) is surrounded by 85 other islands, making an archipelago totalling 7,215 km2. In common with other sub-Antarctic islands, IÃles Kerguelen are cold and wet, with a wind blowing continuously. Summer temperatures average 4±5°C, whilst during the rest of the year temperatures average 1±2°C (MeÂteÂo France, Port-aux-FrancËais). Ice covers nearly one-third of the largest island, called ``Grande Terre'' (Fig. 1). Cats were introduced by early nineteenth century sailors but died out at the beginning of this century (Derenne 1976). The present population was founded by about 4 cats introduced in the 1950s at the time of the establishment of the permanent scienti®c base (Pascal 1980), in order to control other introduced species, mainly rats, Rattus rattus, mice and rabbits, Oryctolagus cuniculus (Derenne 1976). A rapid expansion of the population occurred in following years in the areas surrounding Port-auxFrancËais (Pascal 1980). By 1974, cats occupied 20% of Grande Terre, mainly the Courbet peninsula. In fact, cats seem to be widely distributed over the main islands, except in the southwestern part and above 200 m (Derenne 1976) where rabbits are not present because vegetation is sparse or absent. Line transect sampling was conducted at four sites: Port-auxFrancËais (PAF), Port-Couvreux (PCX) and Ratmano€ (RAT) on the Courbet peninsula, and Port-Jeanne-d'Arc (PJDA) situated on Jeanne-d'Arc peninsula (Fig. 1). No human settlement was present at RAT. A scienti®c station was established in 1950 at PAF. PJDA

Fig. 1 Location of sampled sites on the Courbet peninsula (Kerguelen, Grande Terre): Port-aux-FrancËais (PAF), PortJeanne-d'Arc (PJDA), PortCouvreux (PCX), Ratmano€ (RAT). The centre of the island is occupied by an ice cap. The grey area denotes the distribution range of cats on Grande Terre

was a whaling station on the southeastern peninsula, constructed in 1920 and abandoned in 1929. A family of farmers raised sheep for 2 years (1930±1931) at PCX. The distance between study sites ranged from 20 to 140 km. The 4 sites were assumed to be spatially independent in respect of feral cat movements: home ranges of cats have been estimated to be around 9 km2 (D. Pontier and L. Say, unpublished data). The four sites were situated in coastal regions where cats are generally more abundant (IÃles Kerguelen main island, Derenne 1976; Marion Island, van Aarde 1979). They included the same type of short vegetation (tussock grass, Poa cookii, Azorella selago and Acaena adscendens) but di€ered in terms of number of prey species available (Pontier et al. 1999). Cats' diets consisted mainly of rabbits at PJDA, PAF and PCX, but were more evenly distributed among rabbits, mice and birds at RAT (D. Pontier, L. Say, F. Debias, J. Thioulouse, T. Micol, E. Natoli, unpublished data). We established 1 permanent transect of 3±5.5 km at each site except at PJDA where 2 transects (1 of 3 km called PJDA1 and a 2nd of 5 km called PJDA2) were established (see Table 1). Transects at PJDA and RAT were established along the coast at less than 500 m from the sea. Transects at PAF and PCX were directed towards the interior land and were more protected against marine in¯uences. Transects were linear and stations were identi®ed with numbered and coloured posts at 50-m intervals. Typically, a line transect count was conducted over a 5-day period at di€erent times of the day (between 0430 hours at sunrise and 1730 hours at sunset). The transect was walked at least twice a day except during rain or snow. There was always more than 2 h between two samplings. We considered that this time was sucient to avoid non-independence in the data. We thus did not test for autocorrelation in our data set. Only adult cats were counted. Once a cat was detected, the perpendicular distance of the observed individual from the line was recorded using a telemeter (Leica). When several cats were observed at the same place, the group size as well as the perpendicular distance to the centre of the group was recorded. Transects were surveyed at each site (5-day period) at about 3-month intervals from April 1998 to April 2000. Because of the low number of cats detected during each sampling period for each site, we decided to pool the data over 2 extended periods: from 01.05.1998 to

92 Table 1 Density estimates D (in cats/km2) and coecients of variation (%CV) calculated for each transect and period using Half Normal/Cosines (HNC), Hazard Rate/Cosines (HRC) key function and adjustment term [l length of transects, n number of Site

l

Years

Model

PAF

5.5 km

98±99 98±99 99±00 99±00 98±99 98±99 99±00 99±00 98±99 98±99 99±00 99±00 98±99 98±99 99±00 99±00 98±00 98±00

HRC HNC HRC HNC HRC HNC HRC HNC HRC HNC HRC HNC HRC HNC HRC HNC HRC HNC

PJDA1

PJDA2

RAT

PCX

5 km

3 km

5 km

3 km

n 56 67 94 121 103 119 23 38 27

observed objects (single or clusters of cats), P probability of observing an object in a de®ned area, ESW (in metres) e€ective strip width, n/L the number of observed objects divided by the total length of transect line, TG group size of detected adult cats]

P

ESW

n/L

D

TG

%CV

0.23‹0.04 0.28‹0.03 0.49‹0.06 0.47‹0.04 0.82‹0.05 0.78‹0.21 0.43‹0.05 0.40‹0.04 0.62‹0.06 0.57‹0.05 0.89‹0.06 0.80‹0.07 0.52‹0.11 0.52‹0.07 0.44‹0.08 0.46‹0.05 0.87‹0.24 0.90‹0.23

229‹44 279‹27 309‹35 291‹25 393‹23 372‹100 304‹34 278‹30 430‹42 393‹33 611‹39 553‹52 134‹29 135‹18 129‹23 135‹14 362‹99 375‹96

0.68‹0.09

1.47‹0.33 1.18‹0.19 0.88‹0.14 0.93‹0.14 0.90‹0.10 0.95‹0.26 2.26‹0.32 2.42‹0.23 0.65‹0.09 0.71‹0.09 1.14‹0.12 1.27‹0.16 0.87‹0.25 0.87‹0.21 1.74‹0.41 1.84‹0.31 0.46‹0.17 0.44‹0.15

1.02

23.42 16.48 16.67 15.00 11.92 28.91 14.55 14.07 13.81 12.90 11.20 13.09 29.99 24.78 23.97 18.90 36.10 34.76

31.04.1999 for the ®rst, and from 01.05.1999 to 31.05.2000 for the second period. Each extended period comprised a summer (January to April) and a winter (May to December) period. This was done for each transect (PJDA1, PJDA2, RAT, PAF) except for PCX. For this last transect, the low number of records led us to pool data over the 2 years. For each of the ®ve transects, the density of cats as well as con®dence intervals around density estimates were calculated by the distance sampling method (Buckland et al. 1993) using DISTANCE software (Laake et al. 1994). The technique assumes that as distance from the line increases, ability to detect animals decreases. Perpendicular distance data were plotted in histograms (distance class on the x-axis and number of records on the y-axis) and analysed to determine the probability detection function of the perpendicular distances. For that, we compared four key functions and adjustment terms: Uniform/Polynomial (UP), Half Normal/ Cosines (HNC), Hazard Rate/Cosines (HRC) and Negative Exponential/Cosines (NEC). We used AIC and Goodness-of-®t tests for selecting the best-®tting probability function. Lastly, we converted the estimated group density calculated with DISTANCE to animal density by multiplying the density estimate by the average group size recorded for each transect. We compared densities of cats among sites (5 transects), periods (2 periods: 1998±1999 and 1999±2000) and key functions and adjustment terms using a three-way ANOVA performed using GLIM software (Francis et al. 1993).

0.53‹0.06 0.68‹0.07 1.33‹0.12 0.53‹0.05 1.32‹0.12 0.23‹0.05 0.45‹0.07 0.33‹0.08

1.03 1.03 1.01 1.04 1.06 1.04 1.11 1.00

F(4, 3 df)=2.13, P=0.28; key functions e€ect: F(1, 8 df)=0.37, P=0.56]. Precision of density estimates increased with the number of observed cats (Spearman rank correlation: Rho=±2.45, P=0.01 for HRC and Rho=±2.07, P=0.04 for HNC). Thus, except for RAT during the period 1998±1999 and for PCX where the number of observed cats was low, density estimates were characterised by a less than 20% coecient of variation (Table 1). For a given model, the density estimates varied according to the site and the period [interaction sites and periods: F(3, 8 df)=69.42, P