We analysed the difference in movements within and out of the land use types using binomial Generalised Linear Mixed Models (GLMMs). We used the “lme4” package (Bates et al. 2015) with transect nested within site as random effects. Where groups of species were analysed, species was also included as a random effect. An animal was coded as either having moved from one trap array to another (or not). Each response (moved or not) was classified by the land use type in which it was captured (fixed effects). The response variables analysed in the models were recaptures of: frogs, reptiles, the two most recaptured frog species (spotted marsh frog Limnodynastes tasmaniensis and smooth toadlet Uperoleia laevigata), and the single common reptile species (Boulenger’s skink Morethia boulengeri). We first analysed the recaptures that occurred within the
recaptures that occurred in a different survey to the previous capture (a month or more). This allowed us to differentiate movements over a few days from movements that occurred over a month or more.
There can be seasonal influences on abundance and movement behaviours (e.g. Brown et al. 2005; Paltridge and Southgate 2001). We used binomial GLMMs to test for differences in movement according to the month of capture to determine if there were seasonal effects, with movement coded as having moved or not as the response variable, month as the fixed effect, and transect nested within site and species as random effects.
We also used GLMMs with a negative binomial distribution to determine if there were differences in the distance to the nearest water body between remnants and paddocks, and between the four paddocks types (mean distance to water = 256m, sd = 174, see Appendix S3). For these models, the distance to water was the response variable, the land use types were the fixed effects, and the random effect was site. We used a negative binomial distribution for these models to account for differences in mean/variance relationships. We created subsets of our data using the “dplyr” package to analyse the response variables separately. We determined significant trends in the models using the Anova function from the package “car” (Fox and Weisberg 2011). All analysis was performed in R (R Core Team 2014) and plots were drawn with ggplot2 (Wickham 2009).
To answer our second question (does the distance travelled by recaptured individuals vary between land cover types?), we first visualised the range of distances moved by individuals by plotting the distance travelled by reptiles and frogs that were captured in the same survey round and in different survey rounds. We then modelled the difference in movement distance using linear mixed models by applying the “lme4” package (Bates et al. 2015) with transect nested within site as random effects. We modelled the same reptile and frog response variables as described above. When analysing groups of species, we included species as a random effect. Prior to analysis, we ln(x+1) transformed distance values as there were a large number of zeros and a few individuals that moved long distances.
To answer our third question (does the direction of capture indicate future movement direction?), we tested whether the side of the drift fence that an individual was captured on indicated future movement direction. We labelled one side of the drift fence R and the other P, so that an animal caught on the “R” side was assumed to be moving towards a remnant and an animal caught on the “P” side was assumed to be moving
towards a paddock. We then coded each movement as either “True” - being in the predicted direction (i.e. continuing straight along the transect in the direction inferred from initial capture side), or “False” – an animal moved in the opposite direction to the predicted direction (i.e. moving along the transect in the opposite direction to what was inferred from initial capture side). Some recaptures were classified as NA as the recapture trap was in a different transect and the direction was ambiguous.
5.4 Results
We recorded 2,378 captures of 11 different frog species, and 1,186 captures of 28 reptile species. Approximately 51% of frog captures and 60% of reptile captures were in remnants compared to the paddocks. Among the sites that contained all four paddock types for frogs captured in the paddocks, 25% occurred in pasture, 18% in coarse woody debris, 34% along fences, and 23% in plantings (among the sites that contained all four paddock types) (Fig. 2). For reptiles, 15% of captures in the paddocks were in pasture, 17% were in the coarse woody debris, 23% were along fences, and 44% were in plantings (Fig. 2).
Figure 2. Capture rates of reptiles and frogs by land use type. a) Captures in remnant vs paddocks, b)
captures in the four paddock types (only sites with plantings are displayed due to uneven sampling).
We recaptured 353 individuals at least once during the study, comprising 243 frogs and 114 reptiles (see Appendix S1 for species nomenclature). There was a total of 450 recapture events, with 70 animals caught multiple times (out of a total of 3,564 captures). Most animals were recaptured at the same trap array as their previous capture (total: 76.3%, reptiles: 78.9% and frogs: 74.7%) and therefore no movement was detected for these animals (Table 2). Of the 19 species which were recaptured, 13 species were
recaptures could not be matched to their original capture due to ambiguity in the VIE markings. The different transect types were at different distances to the nearest water body ( 2 = 45.433, df = 3, p<0.001), with coarse woody debris transects the furthest and
pasture transects the closest to water bodies (Fig. 3). There was no difference between the paddocks and the remnants in their distance to water (p = 0.42).
Figure 3The distance to the nearest water body for each paddock type. Estimates are plotted onto
the original scale and error bars represent 95% confidence intervals