Ouessant is a small island (1541 ha) located 20 km west off the western coast of Brittany, France (48°28′N, 5°5′W). Due to the presence of rare species, high biological diversity and an exceptionally preserved coastal ecosystem, it is highly protected (Supporting Information, Fig. S1). During the last 50 years, the number of visitors on Ouessant has increased dramatically, due to a combination of (i) a general increased desire to explore natural environments, and (ii) the liberalization of passenger transport services in 1990, which resulted in increased ferry passenger carrying capacity (Levrel et al., in press). The annual number of ferry passengers increased from 5000 in 1950 to 150 000 in 2005, with a constant annual increase of c. 2500 passengers during the last 20 years and no signs of levelling-off in the near future (Levrel et al. in press). High season runs from the second week of July to the end of August, with a peak in August (48% of annual visits). Tourism is currently the main source of income on the island. Most visitors take a 1-day excursion to the island; they are mostly interested in the spectacular coastline scenery, which they discover by following paths around the island, and are generally not aware of the presence of endangered species and habitats (C. Kerbiriou unpublished data).
The chough has a scattered distribution, resulting from specific ecological requirements, (i.e. suitable nesting sites: shallow caves in cliffs) and foraging areas (short grassland with low cover, Blanco, Tella & Torre 1998). During the 19th and 20th centuries, the distribution and population sizes of the chough in Europe have declined drastically (Kerbiriou 2001; Burfield & Bommel 2004) and the species is now listed in Annex 1 of the European Union Directive on the Conservation of Wild Birds (79/409/EEC). This strong decrease is thought to result from changes in agricultural practices, notably abandonment of grasslands that used to provide suitable foraging habitats for choughs (Kerbiriou 2001). The western French population of chough is now confined to very few localities in Brittany and seems to have stabilized at a small size (39–55 pairs in 2002, Kerbiriou et al. 2005). The population is limited to coastal sites where short grassland habitat above cliffs is maintained by marine physical factors, such as wind and salt spray, i.e. precisely where visitors like to walk. In particular, choughs are never seen in inland agricultural grasslands, which tend to be undergrazed and too tall for choughs to forage (Kerbiriou et al. 2006a). Birds are typically distributed around the island coastline in pairs and in a few small cohesive flocks with immature birds.
data collection and analysis
We monitored the chough population of Ouessant between 1993 and 2005, focusing on the potential impact of tourism on chough behaviour and demography.
Flush distance was defined as the distance at which a foraging bird or flock will fly off when approached by a person or group of persons. Flush distance was estimated to the nearest 10 m using take-offs caused unintentionally by visitors walking towards the choughs (n = 103) or triggered by a member of the research team to increase sample size (n = 63). We explored the effects of flock size, presence of dependent fledglings, visitor group size, type of disturbance (unintentional vs. intentional) and season on the flush distance using a linear model and analysis of variance.
Seasonal and daily variation in the spatial distribution of choughs
To study feeding habitat choice, we first examined the spatial distribution of choughs in relation to feeding habitat availability. We have shown previously that choughs avoid inland pastures and feed almost exclusively in very short swards (< 5 cm, Kerbiriou et al. 2006a) found exclusively on the coastline. Hence, we surveyed the coastline only, which was divided into 123 squares measuring 250 × 250 m (see Supporting Information, Appendix S1). During the summer in 1993 and 1994, and all year round between 1995 and 2001, each square was routinely surveyed for 10 to 30 min by the same observer at least once a month, yielding a total of ca. 80 000 data points. For each observation, we recorded date, time and number of choughs observed; when choughs were present (n = 8273), we also recorded the behaviour of each individual on first contact (foraging, resting or flying). The reproductive season of the chough (mid-March to early July) was excluded because (i) the bird distribution is controlled mainly by territorial defence (Kerbiriou et al. 2006a), and (ii) the number of visitors is intermediate and concentrated on a few specific dates (public holidays).
Short grasslands (< 5 cm) and paths were mapped from field observations and aerial photographs (IGN 2002), and the map was implemented in a GIS (ARCGIS9·1/ESRI). We also measured the area of feeding habitat in each 250 × 250 m square. We studied the spatial distribution of birds in relation to their feeding habitat (i) in winter, when visitors are virtually absent, and (ii) in summer, during the peak tourist season, by using a Poisson linear mixed model (R, lme4 package), where the number of choughs observed in a square was a function of the area of feeding habitat in this square (m2), time of the day, a random square effect, and the average number of choughs in adjacent squares, to account for possible spatial autocorrelation.
Impact of tourism on foraging behaviour
Simultaneously with bird counts, the number of visitors was recorded on areas about 10 times larger than those defined for chough observation, because visitors tend to move around more than foraging birds. These larger areas (hereafter ‘visitor zones’) are a combination of squares used for chough observation and correspond to the main points of interest on Ouessant (see Supporting Information, Fig. S1 and Kerbiriou et al. 2008).
As for each observation we have information of all bird behaviour, we used the proportion of foraging individuals as a proxy for foraging time, which, we assume, carries information on food intake. To study the impact of tourism on foraging, we first examined annual variation in foraging time and compared the peak tourist season (August) to neighbouring months (see Supporting Information, Appendix S2 for a description of how confounding effects of day length and prey availability were removed).
Secondly, we assessed the correlation between the number of choughs observed foraging and the number of visitors using a Poisson linear mixed model (R, lme4 package), as well as a Generalized Additive Model (GAM, Hastie & Tibshirani 1990, R package mgcv), because we expected a non-linear relationship due, for example, to threshold behavioural responses. Spatial autocorrelation was accounted for as described above.
Finally, we quantified the spatio-temporal decrease in available feeding habitat generated by the presence of visitors. To this end, we used the observed relationship between number of foraging choughs and number of tourists to assess the threshold number of visitors above which birds stop foraging in a given visitor zone. By combining this information and the observed daily number of visitors on the island, we estimated the total area of feeding habitat available for each hour of a day. For each day, this value was summed over all hours of daylight and compared to the total area of feeding habitat to generate a daily spatio-temporal decrease in feeding habitat.
Estimates of juvenile survival rates
Because the peak tourist season on Ouessant occurs simultaneously with the fledging period of the chough, we expected a strong impact of the presence of visitors on chough juvenile survival. Chough breeding success was monitored thoroughly from 1998 to 2005 (on average 12 breeding pairs each year). All accessible juveniles were colour-ringed a few days before fledging (n = 122, representing 72% of fledglings observed between 1998 and 2005). Juvenile survival was estimated through resighting of marked individuals (n = 2972 records), via a square-by-square survey similar to that used to collect behavioural data. Resighting data between Ouessant and the mainland coast (not shown) suggest that dispersal outside Ouessant is possible but occurs rarely (as in Reid et al. 2004) and is unlikely to remain undetected.
Monthly survival was estimated each year between June and December. The date of disappearance of a given individual was estimated accurately, thanks to very high resighting rates, that is, all living individuals were seen at least once every 30 days (between 1998 and 2003) or 60 days (in 2004–2005). We estimated monthly juvenile survival using the Cormack–Jolly–Seber (CJS) model (Pollock et al. 1995) implemented in program mark (White & Burnham 1999). The following covariates were included in the survival analysis: (i) total number of visitors in August (ranging from 27 431 to 42 243 between 1998 and 2005, data from ferry companies and office of tourism), to test the impact of tourism on juvenile survival; (ii) annual productivity (number of fledglings on Ouessant, ranging from 15 to 32) to assess a possible year quality effect (as in Reid et al. 2003a); (iii3) climatic data (monthly rainfall, temperature and number of sunny days; data from the Ouessant meteorological station/Météo France), to investigate whether monthly survival depended on environmental conditions. For details on the goodness of fit, the model selection, and the design matrix see Supporting Information, Table S2.2.
Viability of the Ouessant chough population
We assessed the effects of tourism on chough population viability using two types of population models. First, a deterministic matrix model (computer program ULM; Ferrière et al. 1996) was developed to examine population equilibrium and sensitivity of the population growth rate to demographic parameters (Zambrano et al. 2007). Parameter values were obtained from this or previous experimental studies (see Supporting Information, Fig. S2.3).
Secondly, to examine the joint effects of population regulation (limited number of nesting sites, as suggested by a census of available nesting areas, Kerbiriou et al. 2006b), temporal and environmental variation (tourism), as well as demographic stochasticity, we developed a stochastic two-sex individual-based population model (IBM). The IBM allowed a complete description of sex, age, and reproductive status (nesting versus non nesting) of all individuals (see Supporting Information, Fig. S2.3). Because tourism was shown to strongly affect August juvenile survival (see Results), we modelled the expected August juvenile survival in year t as a function of the number of visitors in August (divided by 1000) the same year, using results from the most parsimonious model of capture–recapture of monthly juvenile survival. The relationship between August juvenile survival in year t, sa,t, and number of visitors in August, xt, takes the form: For the sake of simplicity, we did not incorporate the effect of weather on juvenile survival, which was small compared to the effect of visitor number. Therefore, a and b coefficients used in the above equation were estimates from the survival model including the effect of tourism only (see model selection presented in Supporting Information, Table S3.3). The values of these coefficients were a = 0·29 (SE = 0·073) and b = 10·11 (SE = 2·56). The average juvenile survival rate in year t was thus s0(t) = srsa,t, where sr = 0·509 is the juvenile survival rate for the rest of the year (constant across years). Different scenarios for the variation of number of tourist (xt) through time were investigated to extrapolate the effects of tourism on population dynamics and viability. Scenario A: constant number of visitors; xt was set to the average value estimated over the 8 years study period (32 150); Scenario B: stochastic annual variation in visitor numbers, no deterministic increase; xt was varied stochastically across years, by sampling from a Normal distribution with mean 32 150 and standard deviation 5350 (estimated from data over the study period); Scenario C: deterministic increase in visitor number; xt was a linear function of time, xt = 0·7t + 32 150 (Supporting information, Fig. S1 and Levrel et al. in press), estimated from the observed trend in visitor numbers in Ouessant over the last 20 years; Scenario D: deterministic increase and stochastic variation in visitor numbers; was drawn from a normal distribution with mean xt = 0·7t + 32 150 and standard deviation 5350. In each case, N0 individuals (the current population size, n = 55) were initially present in the population.