Measurement of habitat characteristics
Malaysian plovers defended a rectangular 100–300-m long, 5–40-m wide beach section and adjacent mudflats (20–400 m exposed at low tide). We characterized 200-m long breeding territories and random sites that included the mudflat, beach and vegetation backing the beach.
At each 200-m beach section, we measured the beach width at three transects 60 m apart. Percentage cover of ground cover (height < 0·5 m), medium vegetation (height 0·50–5 m) and tall vegetation (height > 5 m) was estimated visually along three 50-cm wide transects running perpendicular to the tide line and extending 20 m into the vegetation, from the beginning of the vegetation line on the upper shore of the beach (Daubenmire 1959). In 2005 we also used the same method to estimate the percentage ground cover on the beach because vegetation may reduce an incubating plover's predator-detection abilities (Wiebe & Martin 1998; Amat & Masero 2004).
In 2005, there were strong correlations between the different height categories of vegetation cover backing the beach (Pearson's r= 0·13–0·65). Thus we used factor analysis and reduced the three variables (short, medium and tall vegetation) into two orthogonal variables that accounted for 70% of the variance. Component 1 described territories with dense medium vegetation cover (correlations with independent variable and principal components r= 0·88) and low percentage ground cover (r = −0·29). The second component described areas with high percentage cover of large trees (r = 0·86) and low percentage ground cover (r = −0·49). These two components were used in the logistic habitat model.
Large (4–6-cm wide carapace) Ocypode (ghost) crabs shared both the mudflat and beach with Malaysian plovers. Previous studies have suggested that Ocypode crabs predate plover eggs and chicks (Watts & Bradshaw 1995; Gregory-Smith 1998). Thus at each sampling section we recorded the number of Ocypode burrows (diameter greater than 5 cm) in three randomly selected 12-m2 quadrats on the beach.
Small Scopimera (bubbler) crabs are a major prey item for Malaysian plovers (M. Yasué, unpublished data). At each sampling section, we measured the width of the mudflat where crab burrows were present. We paced the mudflat and visually estimated the proportion of the area with high, medium or low crab densities. These proportions were used to calculate weighted average crab densities for the section. At each of the three density categories we counted all burrows that had a 2–10 mm diameter in two 0·46-m2 randomly placed quadrats. We then multiplied the width of the mudflat with weighted average burrow densities to calculate a crab abundance estimate for the sampling section. Crab sampling was conducted between 08:00 and 13:00 at 0·7–1·0 m (based on published tide table values) on 15–25 July in 2004 and 2005. Although direct Scopimera counts would have yielded more accurate estimates of prey availability than burrow counts, our method was necessary because of the large number of sites that needed to be sampled over a short period. Moreover, burrows were a good indicator of crab availability because crabs created burrows at every low tide period to feed at the surface (Takahashi, Suzuki & Koga 2001).
We counted people and dogs on the mudflat, beach or up to 20 m in the vegetation behind the beach along 500–800-m long sample sections. These counts were standardized to an average number of people or dogs per kilometre at each site. Repeated counts were conducted at sampling sections on different times, days and month (n = 9–29 section−1). Census sections were larger than 200 m because people frequently walked or drove along the length of the beach and using 200-m would have yielded many zero counts, making comparisons between breeding territories difficult.
Human visitation rates are likely to increase dramatically in tourist areas during Thai holidays and in less tourist-orientated areas during short-term periods of high fish stocks. Maximum rather than average human visitation may be important for plover breeding success. Consequently we also qualitatively ranked the human disturbance level within a 500-m radius of each nest site based on the number of houses, restaurants, hotels, roads, shrimp ponds, trails or other human-made structures. Scrubland with no human use was ranked 0, and areas with interconnected resorts and paved roads were ranked 5.
We used a factor analysis to reduce these three correlated human disturbance variables (Pearson's r= 0·33–0·54) into one variable for each year that accounted for 69% and 61% of the variation (Fig. 1a,b). All three variables were positively correlated to the principle component (component matrix factor loadings = 0·69–0·85).
Figure 1. Scatter plots showing the relationship between instantaneous human (left) and dog (middle) counts and qualitative human disturbance ranking (right) to the human disturbance principal component for 2004 (top) and 2005 (bottom). Although human disturbance rankings were integers, we displaced the x-axis by 0·1 or 0·2 so that so it is easier to differentiate the site symbols on the graph.
Download figure to PowerPoint
Habitat selection model
To identify factors influencing territory selection, we compared the habitat characteristics of beach breeding territories to control sections without nests in the four areas (2004 and 2005, breeding:control sections, respectively, n= 55:36 and 79:40). Control sites were selected based on proximity to nest sites and interspersed among nest sites as much as possible to avoid spatial autocorrelation (Legendre 1993). All beach sections without nests that were immediately adjacent to nesting territories were included as control sites. To increase sample size of control sites, other beach sections that were within 2 km of nest sites were also included in our study.
Control sites were visited at least once per month from January to July to ensure that there were no breeding plovers. We did not conduct more frequent checks at these sites because Malaysian plovers generally occupied territories at least 1 month prior to laying a clutch and usually did not change habitats after a failed nesting attempt (M. Yasué, unpublished data). The variables that were included in the habitat selection models were: beach width, the percentage cover of ground, medium and tall vegetation behind the beach, Ocypode burrow densities, Scopimera burrow abundance, the human disturbance factor, and site. As discussed above, in 2005 two principal components were used instead of direct vegetation cover behind the beach and percentage cover of beach vegetation was also included in the analysis. For all independent variables we checked the variable inflation factor (all < 2), tolerance (all > 0·5) and bivariate Pearson's correlation coefficients (all < 0·5) to ensure there was no multicollinearity in the data set.
Breeding success model
To minimize pseudoreplication, we calculated reproductive success per breeding territory for each pair of plovers rather than by each breeding attempt. Thus for pairs renesting during a season, the habitat characteristics for successive breeding attempts were averaged and the number of chicks hatched or fledged was summed and then divided by the number of breeding attempts per pair.
In the breeding success models, we identified habitat characteristics that influenced the number of chicks hatched per breeding attempt and the number of chicks fledged per breeding attempt (npairs for hatch success 2004, fledge success 2004, hatch success 2005, respectively, 49, 49, 71). We used these measures of breeding ‘efficiency’ rather than total eggs hatched or chicks fledged because multiple breeding attempts in a season could entail costs to adult survivorship (Ghalambor & Martin 2000).
We tested the same variables as in the habitat selection model and also included conspecific nest density because if there are density-dependent declines in breeding success, these indirect effects may lead to a greater combined impact on populations than could be attributed to the direct effects of habitat loss alone (Sutherland & Norris 2002). Also, a previous study on Kentish plovers suggested that more nests were predated in areas with higher nesting densities (Page et al. 1983). The nest density estimate was the number of conspecific nests within 200 m of the nest, based on Universal Transverse Mercator (UTM) coordinates obtained at each nest.
For the habitat selection, hatch and fledgling success analysis, we used binary logistic regression and selected the most important variables using log-ratios stepwise backwards elimination (removal if P > 0·10) in SPSS version 11·0, while controlling for site. The dependent variable for the habitat selection model was the presence or absence of nesting plovers. For breeding success, we recorded the number of chicks hatched and fledged per breeding attempt of a pair into high and low categories that had approximately equal sample sizes (high categories for hatch 2004, fledge 2004, hatch 2005 were eggs hatched clutch−1≥ 1·5, chicks fledged clutch−1≥ 0·67, eggs hatched clutch−1≥ 1, chicks fledged clutch−1≥ 0). Although scoring breeding success reduces resolution, this was necessary because data could not be transformed into a normal or a Poisson distribution using standard data transformation techniques. We conducted analyses for the 2 years separately because different independent variables were tested between these years (i.e. percentage beach ground cover in 2005 and factors instead of raw data for vegetation structure in 2005).
To check for the independence of successive nesting territories or control sites, we ordered the 200-m beach sections (south to north) within each site and used scatter plots, runs test and lag-1 autocorrelation functions to quantify the degree of serial autocorrelation in the residuals of our model (Pindyck & Rubinfeld 1998; Keitt et al. 2002; Diniz-Filho, Bini & Hawkins 2003). Data showed no sign of strong spatial autocorrelation (significance values for the runs test and autocorrelation functions were all > 0·05).