By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
*Corresponding author: Australian Centre for Biodiversity, School of Biological Sciences, Monash University, Victoria 3800, Australia. E-mail: firstname.lastname@example.org
Aim We explored whether the anuran amphibian faunas differed among landscapes that are relatively intact (largely covered in forests and woodlands) and others that are completely converted to agriculture. We also used historical data sets to assess the current condition of the anuran fauna in a region predicted to experience, and experiencing, severe drying and warming.
Location Five pairs of landscapes (each of c. 20 km2) – one in each pair being almost completely wooded and the other cleared for agriculture – across a 30,000 km2 region of northern Victoria, Australia.
Methods Sites were repeatedly surveyed in the austral winter–spring breeding seasons of 2006 and 2007, with records collected of numbers of calling males and the presence of egg masses and tadpoles. We characterized the sites using static (e.g. dimensions, surrounding physiognomic characteristics such as tree cover) and labile (e.g. pH, dissolved oxygen) variables. Data were analysed using hierarchical Bayesian models.
Results For calling males, landscape type did not affect densities or species richness measures. The availability of a grassy verge around water bodies was an important predictor for most species, but other labile and static variables seemed not to be important. Fewer than half of the species historically known from the region were recorded. There were no important predictors of occurrence of egg masses or tadpoles. Reproduction effectively may have failed over the period, with fewer than one in four sites showing evidence of egg masses or tadpoles.
Main conclusions The proportion of sites at which some well-studied species (e.g. Crinia signifera, Limnodynastes dumerilii) were recorded has dropped substantially since the 1970s, as have average densities of calling males of Crinia spp. The remnant anuran fauna appears to be dominated by resilient and hardy species with low current diversity. The on-going (12+ years) drought in these landscapes suggests a bleak long-term prognosis for the few remaining species of anuran amphibians.
Land-use change has been identified as a major driver of biodiversity decline throughout the world (Heywood et al., 1995). Systematic understanding of effects is elusive for many reasons, notably differing methods of assessment, survey scope (extent of area covered and grain size of measurement) and levels of survey repetition (Fleishman & Mac Nally, 2007). Moreover, there is a strong bias towards certain groups while others remain under-represented (Stuart et al., 2004). While one expects that different taxonomic groups would respond differently to land-use perturbations, many reptiles and amphibians potentially have great vulnerability to habitat loss and alteration due to their relatively low mobility (Stevens et al., 2006) and life-history characteristics (e.g. dependence on standing or flowing water for reproduction). Lack of research activity may lead to the taxon being given a low profile in conservation planning (Lunney & Ayers, 1993).
Our focus is on anuran amphibians (frogs). There has been extensive concern about the reasons for global declines in amphibians, much of which is disputed but at least some of which is linkable to disease and vulnerability to altered spectra of transmitted solar radiation (Kiesecker et al., 2001; Johnson, 2006). Decreases in landscape connectivity through fragmentation and habitat loss have been thought to cause declines in amphibian assemblages (Lehtinen et al., 1999). Increased vulnerability to the effects of climate change in massively altered landscapes has been linked to declines in even common species of amphibians (Hazell, 2003; Piha et al., 2007). Severe droughts of 2–3 years’ duration have already been identified as causes of large-scale declines in species such as Pseudophryne bibroni and the Litoria aurea species complex in southern New South Wales, Australia (Hazell et al., 2003). With projected increases in the severity and frequency of drought in south-eastern Australia (Nicholls, 2004), the prospects for many anuran species appear dire.
We focused on how the anuran fauna has responded to land-use change associated with long periods of transformation from woodland to agricultural landscapes, and the interaction of land-use change with the apparent rapid aridification of landscapes in southern Australia (Cai & Cowan, 2008). Climate change has already been invoked as a contributor to anuran (and reptile) decline (Whitfield et al., 2007). The region we consider has been extensively examined for other taxa, especially non-aquatic vertebrates (Mac Nally et al., 2002). Part of our original motivation for the study was to explore whether water-dependent vertebrates respond similarly to non-aquatic vertebrates.
There has been much work on the hydrological effects of such broad land-use change, including the effects of reduced native vegetation cover (Burch et al., 1987), numbers of farm ponds (Schreider et al., 2002) and land management (Potter, 1991). The probable drivers are: (1) generation of large numbers of similar habitats (e.g. farm ponds, drainage channels, reservoirs) (Pringle et al., 2000); (2) greater water permanence (i.e. change from transient bodies to purpose-built farm dams), favouring some taxa and disfavouring others (Hazell, 2003); and (3) altered ‘viscosity’ of the landscape to dispersal (Hazell et al., 2003). As vegetation is cleared and agricultural activities increase, the amount of permanent water in the landscape and thus opportunities for dispersal among aquatic habitats probably increase, at least for some taxa. Thus, while fragmentation of forest habitats generally decreases habitat connectivity for many arboreal biota (e.g. birds) (Fahrig, 2003), the associated increase in the number and permanence of aquatic habitats may increase connectivity for many aquatic biota. Of course, this may not be absolutely general because some aquatic taxa may favour cover provided by woodlands (Gagné & Fahrig, 2007) and structures impeding movement, such as roads (Vos et al., 2001a), are more likely to be in higher densities in cleared landscapes. The increase in the number and permanence of aquatic habitats, along with changes in the viscosity of the intervening matrix, may have different effects on the dispersal patterns of various taxa.
We asked three main questions. First, is there evidence that the anuran faunas of wooded and cleared landscapes differ, assuming that the faunas would have been similar prior to the broad-scale modification of the region? Second, given the increasing evidence that climate has changed in the region, especially since 1996, are the anurans successfully breeding under the drying regime? And third, have anuran numbers changed since the onset of this extended arid period?
Our survey design was modelled on a ‘whole-of-the-landscape’ approach (Radford & Bennett, 2007). Such comparisons have been deemed to be urgently needed for anurans (Hazell, 2003). We considered five pairs of landscapes (Fig. 1), using pairs to deal with possible spatial clines, and repeatedly surveyed the anuran faunas at multiple sites within the landscapes. Survey locations were characterized using ‘static’ (e.g. amounts of tree cover in the vicinity of the water body) and ‘labile’ variables (pH, turbidity, etc.). Anurans are often comparatively easy to survey using male vocalizations (i.e. standing crop measures of abundance), but we also monitored egg mass and tadpole occurrence given the importance of reproduction to population viability (Mac Nally, 2007a). We have measures of densities of calling males from studies undertaken in the late 1970s against which we can assess multidecadal changes in the abundance of some species (e.g. Crinia signifera and Crinia parinsignifera) (Mac Nally, 1981). Finally, we interrogated databases for reports of anuran amphibians in the study region from the late 1950s and compare reporting rates from four periods prior to our study to assess whether species have changed in these rates over that time.
The region (30,000 km2, central Victoria, Australia; Fig. 1) is characterized by box and ironbark forests and woodlands (mainly Eucalyptus tricarpa and Eucalyptus microcarpa) on gentle slopes and hills (elevation 150–400 m) and currently is an agricultural–forest mosaic, with only 17% of the original vegetation cover remaining (ECC, 2001). The historical mean annual rainfall ranged from 400–700 mm, mostly falling in winter and spring; summers are hot and dry (Mac Nally et al., 2000). For more than 160 years the region has been much modified by gold mining, timber harvesting and agriculture. The box and ironbark forests and woodlands are open-canopied forests and woodlands of moderate height (10–25 m). The local composition of the forest depends on aspect, elevation, soils and drainage.
Landscapes and sites
We selected five pairs of agricultural and forested landscapes across the region (Fig. 1). Landscapes in each pair were selected to have similar topography, climate, historical vegetation and geology. Landscapes were selected on the basis of the expected resolution of patterns of landscape genetics from work informed by the Molecular Ecology Laboratory at Monash University. The project was constructed to simultaneously explore ecological and genetic information about distributions of amphibians, and also aquatic invertebrates. All water bodies (referred to as sites) were delineated on GIS by using a map layer from the Victorian Government's spatial data division (http://www.land.vic.gov.au/vicmap). The sites were farm dams, forest fire dams, culverts and some natural water bodies. The mean number of dams in the agricultural landscapes was 70 ± 13 (SD); the analogous figure for the forested landscapes was 13 ± 8 (SD). Average cover of remnant vegetation in the agricultural landscapes was 3 ± 1% (SD); for forested landscapes, the figure was 86 ± 14% (SD).
Ten potential sites within a circle of 2.5 km radius (landscape area = 19.6 km2) were selected for each landscape; from these, eight were selected based on on-site inspections (total water bodies = 80). Selection criteria for inclusion were: (1) no revegetation and/habitat enhancement had been undertaken; (2) the site was not subject to draining through irrigation pumping; (3) there was no heavy use by stock; and (4) the site was not close to farm infrastructure.
Site characteristics (covariates)
There were two sets of site variables: (1) ‘static’ values between years (or values only measured once); and (2) ‘labile’ values between years. Static values were: (1) amount of grassy bank (m); (2) area of macrophytes (m2); (3) whether or not the site dried out over the sampling time; (4) basal area of canopy-level trees in a 5-m annulus around the perimeter of the pond (m2); (5) percentage canopy cover over the water from surrounding vegetation; (6) total perimeter (mean and maximum) of the wet area; and (7) type of landscape – agricultural or forest. Labile environmental variables were measured during six visits (three visits in 2006 and three in 2007). We measured water turbidity (Secchi disc) and conductivity, dissolved oxygen and water pH using a Horiba ES51 water-quality device (http://www.austscientific.com.au/es-51.html). Measurements were not always conducted for every visit at each site so we averaged values for each year.
The site perimeters of wetted areas ranged from 30–280 m. Each site was visited in a ‘systematic random’ order six times throughout the austral winter and spring, when most frog species are active: August, September and October 2006 and July, August and October 2007. By systematic random, we mean that for each round of surveys the five pairs of landscapes were ordered randomly for visitation. Either the agricultural or the forested landscape was surveyed first and the other in the pair second. In subsequent survey rounds, the order was reversed. Within each landscape, sites were visited in random order. This design ensured there should be no systematic biases among pairs, between agricultural and forested landscapes, and among individual sites within landscapes. Survey visits were undertaken from c. 14:00 h during the day and from just after dusk for night surveys. Equal numbers of day and night surveys were undertaken for each site. In each site, all frog calls were monitored over a 20-min period (Hazell et al., 2001) and we attempted to distinguish individual males to provide abundance measures (i.e. numbers of individuals of each species calling within 20 min).
After this period, a thorough search of the banks and nearby surrounding area was carried out for the presence of egg masses; we used flashlights for night surveys. We surveyed intensively for eggs because we were seeking material for undertaking landscape genetics analyses. The presence of tadpoles was established both visually and by using a small sieve net. We also undertook another three surveys for macroinvertebrates (not described here) in which we made careful observations of tadpoles either in sieve samples or visually, and extensively searched for egg masses in 2007. We did not distinguish among species for tadpoles or egg masses because these could not be identified reliably in the field (most species had syntopic congeners with similar egg-masses and tadpoles). For all 480 visits, we recorded air temperature, wind conditions and cloudiness using the Atlas of Victorian Wildlife (DNRE, 2001) field data recording environmental guidelines [wind: 1 calm; 2 light, leaves rustle; 3 moderate, moves branches; 4 strong, impedes progress; cloud: graduated scale 0–7 (clear–completely overcast)].
We describe analyses in line with the questions posed in the Introduction. First, we deal with the important issue of spatial autocorrelation, which can affect inferences if present. Second, we explore which factors influence numbers and distributions of calling males. We present analyses of breeding outputs (egg masses and tadpoles) and conclude with analyses of the current occurrence of species compared with historical data.
We linearly correlated two sets of distances using the ‘mantel’ facility in the ‘vegan’ package (Oksanen et al., 2006) in R (R Development Core Team, 2006), which provided a 95% interval (by permutation) for the correlation coefficient. The first distance was assemblage similarities. These were obtained as follows. We summed the records for each species of anuran over the six rounds. There were 58 ponds (out of a total of 80) with one or more records. For these 58 ponds, we calculated the Canberra distances using the Vegan package. The second set of distances were pair-wise geographical separations of the 58 sites. If spatial autocorrelation is present, then there should be a discernible correlation between these two sets of distances.
Distributions of adult males
For calling adult males, we considered these response variables: numbers of (1) species, (2) individuals of all species, (3) Crinia signifera, (4) Crinia parinsignifera, and (5) Limnodynastes tasmaniensis; records for other species were too sparse to analyse. These were analysed using hierarchical Bayesian, Poisson models. These response variables were related to site-specific variables that varied from survey to survey (labile covariates) and others that were static over the duration of the project.
We used a Poisson regression model because the response variables were non-negative counts data, which usually are best modelled using Poisson variates (McCullagh & Nelder, 1989). Hierarchical models are very flexible and allow incorporation of potential effects in natural ways. Flexibility includes use of specialized interfaces for inferring the importance of covariates on response variables. We used ‘reversible jump’ Markov chain Monte Carlo sampling (RJMCMC; Lunn et al., 2006) with the ‘Jump’ interface (Lunn et al., 2005) for winbugs software (see below). The RJMCMC approach effectively provides posterior probabilities for inclusion of covariates in models that best predict variation in the response variables.
Model specification for the Poisson regressions for adult counts was:
The response variable for pond i in survey j (Rij) has a true Poisson-distributed value µij. The latter may depend on year [j(YEAR) associates survey j with YEAR = 2006 or 2007] and individual survey temperature (T), windiness (W) and cloudiness (C). There is a random effect for each site associated with its landscape type (agricultural or forested, ei(LS)); we allowed the variances () for the landscape types potentially to differ. There also are random effects for each site (eij), which depend hierarchically on the static and labile variables. We included landscape type as a binary variable with the static variables rather than an explicit term in model 1 because of correlations with some static variables (e.g. basal area of canopy-level trees). We assigned uninformative normal priors for the bs and c. The σ terms were given uninformative uniform priors on the interval (0.01, 5). We used Bayesian estimation to derive posterior means and 95% credible intervals for parameters.
We assessed whether model parameters differed from 0 by calculating the proportion of the posterior probability distribution that exceeded 0 posterior probability mass (PPM). Given that we had no prior expectation that parameters differed from zero, a PPM > 0.92 corresponds to posterior odds of 10, which is regarded as strong evidence that the parameter exceeded 0 (Kass & Raftery, 1995). Similarly, if PPM < 0.08, this is strong evidence that the parameter is less than 0 (posterior odds < 1/10; note: we inverted posterior odds for negative coefficients for ease of interpretation). For the RJMCMC parameters, the Jump interface yields a probability of inclusion [Pr(inc)] in the best models for each parameter, which translates directly into a posterior odds = Pr(inc)/[1 – Pr(inc)], assuming no prior expectation that a variable is or is not a predictor.
We determined the adequacy of model fit by using posterior predictive assessment (we refer to this as PP-fit) (Gelman et al., 1996). This involves setting up monitoring nodes that simulate samples from the µij-values, ζij. Standardized deviates appropriate for a Poisson-variate are and for the observed and simulated values respectively (Agresti, 2002). These values are summed over all ij combinations. A good model fit has the observed value in the middle of the simulated distribution, so a posterior probability of PP-fit = 0.5 is the objective.
Linear densities of calling males of Crinia spp.
We had access to data on earlier density estimates of calling males of C. signifera and C. parinsignifera from the late 1970s (Mac Nally, 1981), from which we could make comparisons with current estimates of densities of these two species. To do so, we first computed the maximum count per site over the three visits for 2006 and 2007 for the two species. We then divided this value by: (1) total pond perimeter; and (2) length of grassy verge, given its predictive importance (Results).
Analyses of tadpoles and egg-masses
Given that we surveyed sites asynchronously, we chose to represent evidence of reproductive outputs as presence of egg masses or of tadpoles. We used hierarchical Bayesian, logistic models, which are often used for binary data (Fleishman et al., 2001). Therefore, the data were presence/absence (binary) and model 2 was used to relate evidence of reproduction to the covariates:
The model now estimates the probability of presence of egg-masses or tadpoles in site i in survey j (pij). Previously described terms and approaches are as for model (1); note absence of observational conditions (windiness, temperature and cloudiness).
Retrospective analyses of reporting rates of anuran species
To gauge whether reporting rates of species have changed over recent decades, we obtained data from the Atlas of Victorian Wildlife (DNRE, 2001). This is a heterogeneous compilation of records from government agencies, universities and museums. We restricted the interrogation to the area in the bounds shown in Fig. 1 and to records from the comparable months of the year to our study (July–October). The earliest data were from 1958. To establish a measure of search effort in the database, we first identified unique combinations of locations and dates, which we regarded as unique sampling events. By inspection of the data, we noted that there often were multiple entries on such events for some species; we counted such multiples as a species’ presence for the sampling event rather than an abundance measure. A large fraction of unique events recorded just one species, suggesting targeted surveys for that species possibly in relation to other purposes (e.g. population analyses or targeted species collections). Therefore, we pruned the data such that only events in which multiple species were recorded were included, reasoning that these data would more likely be general surveys. We blocked the data into several time slices to facilitate comparisons. These were: 1958–77 (n = 22 unique, multispecies surveys), 1978–89 (n = 55), 1990–96 (n = 43) and 1997–2003 (n = 28). For our surveys, we took an ‘optimistic’ measure of presence of a species at any site in any of the six surveys, which is likely to be inflated relative to the historical occurrence rates.
We used an analogue of a factorial model to analyse these data. Data were modelled as binomial variables (r presences, n surveys), which necessitates a logit transform (McCullagh & Nelder, 1989). The model is:
The ps are the estimates for probabilities of occurrence, which are modelled as a species-specific initial value (αi, value for the first time slice) and a sequence of deviations (for each species) from the initial probability, βij. The model coefficients are constrained such that the αs and βs are random effects across all species in each time slice, hence the common σs in equations 3. This means that patterns for species that are rare, or for which there are few data, can be analysed in concert with the rest of the anuran fauna and assemblage-level inferences made (Mac Nally et al., 2008). For each species, we computed all pair-wise comparisons between ps (i.e. pim–pin, m≠n), for which we could derive posterior probability distributions. We used the posterior odds inferential approach described above.
Software and general modelling information
We fitted all models in winbugs, version 1.4 (Spiegelhalter et al., 2003). Parameters were estimated from three MCMC (Markov chain Monte Carlo) chains of 100,000 iterations after 20,000 iteration burn-in periods. We checked MCMC mixing and convergence using the ‘boa’ package (Smith, 2006) in R.
General characteristics of the anuran fauna
Calling males of only seven species were recorded over the six survey rounds. These were: C. signifera (288 total records, present in 35 water bodies out of 80), C. parinsignifera (580, 46), Litoria ewingi (21, 10), Limnodynastes tasmaniensis (120, 24), Litoria peroni (3, 2), Limnodynastes dumerili (28, 9) and Neobatrachus sudelli (1, 1).
There was little evidence of important spatial relationships in the anuran fauna. The Mantel correlation was rPearson = 0.04 (0.02–0.06; 95% CI). Therefore, there was no need to explicitly incorporate spatial separations into model 1.
Distributions of calling male anurans
Poisson models for each of the five response variables were good fits to the data (all 0.46 ≤ PP-fit ≤ 0.61; Table 1). We discuss only those effects that were important for posterior odds ≥ 10. Observation conditions had important effects on counts, with higher temperatures and more windy conditions depressing counts for the five response variables. Counts for C. signifera were higher in overcast conditions, while those for C. parinsignifera were lower under higher cloud cover (Table 1). Counts were substantially lower for all response variables, apart from Limnodynastes tasmaniensis, in 2007 than in 2006.
Table 1. Details of model parameters (means ± SD) for analyses of response variables (all data are for records of calling males). Entries in italics imply means < 0; entries in bold imply means > 0 based on posterior odds (see text).
Individuals of all species
PO, posterior odds; PP-fit, adequacy of model fit by using posterior predictive assessment.
−0.07 ± 0.01
−0.12 ± 0.01
−0.14 ± 0.02
−0.12 ± 0.01
−0.10 ± 0.03
−0.70 ± 0.19
−0.54 ± 0.11
−0.35 ± 0.18
−0.63 ± 0.14
−1.53 ± 0.50
0.02 ± 0.02
−0.02 ± 0.01
0.09 ± 0.02
−0.06 ± 0.02
0.00 ± 0.04
−0.33 ± 0.16
−0.39 ± 0.25
−0.32 ± 0.18
−0.58 ± 0.32
−0.54 ± 0.60
0.06 ± 0.12
0.05 ± 0.13
0.02 ± 0.09
0.04 ± 0.15
−0.22 ± 0.38
−0.23 ± 0.18
−0.43 ± 0.27
−0.04 ± 0.12
−0.55 ± 0.32
0.02 ± 0.15
−0.23 ± 0.25
−0.22 ± 0.31
−0.04 ± 0.18
−0.27 ± 0.51
−0.11 ± 0.36
0.01 ± 0.05
0.007 ± 0.09
0.01 ± 0.07
0.01 ± 0.12
0.07 ± 0.22
0.13 ± 0.17
0.33 ± 0.27
0.02 ± 0.08
0.60 ± 0.37
0.12 ± 0.30
Grass verge cover
0.48 ± 0.17
0.94 ± 0.26
0.25 ± 0.35
0.92 ± 0.36
1.32 ± 0.40
0.01 ± 0.07
0.01 ± 0.11
0.07 ± 0.21
0.01 ± 0.13
0.13 ± 0.23
0.03 ± 0.08
0.04 ± 0.13
0.03 ± 0.17
0.03 ± 0.16
0.34 ± 0.42
Tree basal area
0.10 ± 0.14
0.22 ± 0.26
0.79 ± 0.41
0.10 ± 0.20
0.08 ± 0.21
0.01 ± 0.09
0.002 ± 0.13
0.04 ± 0.20
−0.03 ± 0.16
−0.03 ± 0.19
0.04 ± 0.18
0.07 ± 0.27
0.14 ± 0.42
−0.03 ± 0.26
−0.13 ± 0.36
There was little evidence that landscape type influenced counts of any of the anuran species (or aggregated variables such as richness), with all BF ≤ 2 (Table 1). The availability of grassy verge was the main predictor of elevated counts for all response variables apart from counts of C. signifera, but values even for that species were higher when grassy verge availability was greater (Table 1). Counts for C. signifera were positively associated with tree basal area within 5 m of the water (Table 1).
Linear densities of calling males of Crinia spp.
The mean maximum linear density of calling males of either species, in either year, gauged for whole perimeter or grassy verge, did not exceed 0.13 m−1 (Table 2). The maximum value for each species < 0.77 m−1 was achieved only once for grassy-verge calculations for each species. These maxima are misleading because the amount of grassy verge was < 6.5 m and the number of males was only four (C. signifera) and five (C. parinsignifera), respectively.
Table 2. Details of densities of calling males of Crinia signifera and Crinia parinsignifera.
2006 – total perimeter
Mean (m−1) ± SD
0.01 ± 0.02
0.04 ± 0.10
2006 – grassy verge
Mean (m−1) ± SD
0.05 ± 0.09
0.13 ± 0.22
2007 – total perimeter
Mean (m−1) ± SD
0.02 ± 0.04
0.03 ± 0.05
2007 – grassy verge
Mean (m−1) ± SD
0.07 ± 0.16
0.05 ± 0.08
The logistic model fit was average for presence of egg masses or tadpoles (PP-fit = 0.38) (Table 3). There was no evidence of differences among landscapes or years, and none of the labile or static variables were important in predicting breeding outputs (Table 3). The proportions of water bodies with either egg masses or tadpoles were 0.163 (2006) and 0.225 (2007). While the interannual difference was large, evidence of reproductive activity was low overall (in less than 25% of water bodies).
Table 3. Details of model parameters (means ± SD) for analyses of the reproductive response variable, which is the presence of egg masses or tadpoles at each site in a survey round. No predictors were important.
PO, posterior odds; PP-fit, adequacy of model fit by using posterior predictive assessment.
0.18 ± 0.41
0.01 ± 0.08
−0.03 ± 0.12
−0.12 ± 0.32
0.11 ± 0.20
0.00 ± 0.11
Grass verge cover
0.04 ± 0.16
−0.12 ± 0.31
0.22 ± 0.28
Tree basal area
−0.01 ± 0.09
0.01 ± 0.09
0.03 ± 0.29
Retrospective analyses of reporting rates of anuran species
Fourteen species of anurans were in the atlas database: C. parinsignifera, C. signifera, Crinia sloanei, Geocrinia victoriana, Limnodynastes dumerilii, Limnodynastes tasmaniensis, Litoria ewingi, Litoria paraewingi, Litoria peronii, Litoria raniformis, Neobatrachus pictus, Neobatrachus sudelli, Pseudophryne bibronii, Pseudophryne semimarmorata, although there were no records for Litoria raniformis in multiple-species unique surveys and so this species was dropped from analyses.
Recall that our results are optimistic because the number of surveys per site was high (six), which means there is a high chance of detection if a species is present compared with the database surveys, when there was only one visit per site. Only two species appear to have changed little in probabilities of occurrence over the five time slices, G. victoriana and Litoria ewingi (Table 4). The former usually breeds earlier in the year than the months of our survey period, so its rate of detection is expected to be low. Probabilities of occurrence in our surveys were the lowest over the five periods for six species, some of which had probabilities of occurrence over an order of magnitude lower than previously (C. sloanei, Litoria paraewingi, Litoria peronii, N. sudelli, P. bibronii; Table 4). Two of the most widespread species, C. signifera and Limnodynastes dumerilii, were recorded at substantially lower rates than previously (Table 4). Two other widespread species had lower reporting rates in our study than in previous periods, but the latter were more similar to rates recorded in earlier time slices (Table 4). The most pronounced decline was for P. semimarmorata, although this decline appeared to begin many decades ago (years 1978–89 compared with 1958–77, Table 4).
Table 4. Details of posterior comparisons of probabilities of occurrence in the retrospective analysis for each species. Figures in bold indicate periods in which the probability of occurrence was substantially greater than in our study period (2006–07).
Contrary to our prior expectations and results from comparable studies (Hazell et al., 2001), we found little evidence of a constitutive difference between the anuran faunas in the agricultural and wooded landscapes. Given that the total counts and counts of each of the most abundant taxa (Table 1) per aquatic water body were little affected by landscape type, the agricultural landscapes probably provide habitat for more anurans than the wooded landscapes by virtue of the high ratio of aquatic habitats in agricultural compared with wooded landscapes [70 ± 13 (SD) (19.6 km2)−1 vs. 13 ± 8 (SD)]. Moreover, for the important predictor variables (Table 1), the average extent of grassy verge in agricultural landscapes was 52 ± 55 m (SD) compared with 11 ± 23 m (SD) in forested landscapes. The analogous figures for basal area of trees (within 5 m of the pond, important for C. signifera) differed little: 0.8 ± 1.7 m2 (SD) (agricultural) and 1.6 ± 1.3 m2 (SD) (forested). The almost five-fold greater amount of grassy verge is not necessarily a reflection of larger aquatic bodies per se because the perimeters were much more similar in magnitude [107 ± 50 m (SD) and 87 ± 47 m (SD) for agricultural and forested landscapes, respectively].
We make several observations regarding the outcomes of our study. These relate to: (1) species composition; (2) changing abundances; and (3) reproduction.
We detected only seven species of anurans (there were records for 15 species in historical databases) and, given the spatial extent and duration of surveys, only three of the species were of even moderate abundance (C. signifera, C. parinsignifera, Limnodynastes tasmaniensis). Even prior to major landscape transformation by Europeans in the 19th century, the region may have been depauperate in anuran species owing to the unpredictability and average low values of precipitation (≤ 700 mm year−1) and comparative lack of topographic variety. We are unaware of the existence of any useful data against which to gauge the pre-European condition of fauna. The region had been extensively transformed decades before any survey data were accumulated, so we have little idea of the anuran fauna from the period before widespread clearances and changes in availability and types of aquatic bodies in the landscape. The lack of a landscape-type effect may reflect a much-winnowed fauna, in which only the most resilient of species (the Crinia species and Limnodynastes tasmaniensis) have persisted through the many changes wrought over the past 200 years. The historical retrospective analyses suggested that the anuran fauna has experienced a severe recent decline.
Using reporting rates, as we have done, provides better resolution than species lists alone. For example, a broadscale, one-off survey in a comparable area of Western Australia, also experiencing severe drying, reported all species known to have occupied the region, with each species apparently as broadly distributed as in the past (Burbidge et al., 2004). However, without prior reporting rates, it is difficult to assess whether the species were declining or not.
Extensive data are available for ecological characteristics and breeding biology for the two species of Crinia from the late 1970s to early 1980s (Mac Nally, 1981). While there now appears to be evidence of climate change in the region (especially reduced precipitation) as early as the 1960s (Cai & Cowan, 2008), precipitation patterns were consistent with historical averages in the late 1970s period, with high rainfall from austral autumn (April) to October in three years out of four (Mac Nally, 1981). In an extensive spatial survey (2600 km2) in the southern part of the region we studied, calling males of C. signifera were recorded in 100% of 207 water bodies of similar character to the ones we describe (Mac Nally, 1985). We found this species in just 35 out of 80 water-bodies a quarter of a century later; the 100% reporting rate is more consistent with the values for those earlier decades shown in Table 4 (> 73%). In the late 1970s, densities of calling males were measured over periods of many months over several years. Average separations were usually c. 1.3 m, corresponding to a density of c. 0.77 m−1, very consistent with a maximum acoustic spatial packing based on play-back experiments (Littlejohn et al., 1985). Average densities now are much less than this. For sites at which C. signifera was present, the average maximum density in 2006 was 0.03 ± 0.02 m−1 (SD) and in 2007, 0.06 ± 0.02 m−1 (SD). Analogous figures for C. parinsignifera were 0.09 ± 0.14 m−1 (SD) (2006) and 0.06 ± 0.06 m−1 (SD) (2007). These data suggest a 10-fold decrease in average densities at sites at which these two species were present and, given the reduction in sites occupied, the decrease is much more severe.
Many assessments of change in patterns of biodiversity focus mostly on species presence, reporting rates (proportions of presences in numbers of surveys) or abundances. Some modelling suggests that rates of patch (here water body) occupancy are good measures of metapopulation viability (Vos et al., 2001b). Nevertheless, there has been too little focus on reproductive success notwithstanding a rich conceptual and methodological literature in the population viability arena (McCarthy et al., 2001). One of the main reasons for the lesser focus on reproductive assessments seems to be the intensity of work needed to establish the main demographic parameters for use in modelling frameworks such as vortex or ramas. We have argued elsewhere that simpler approaches than full (meta)population viability analyses may be useful for gauging reproductive success and are more informative about the worth of management actions, at least for birds (Selwood et al., 2009).
Our indicators of reproductive success are limited to the presence of egg masses and tadpoles, yet even these very basic indicators tell a tale – there appears to be widespread failure of breeding across this entire region. The maximum proportion of surveyed water bodies in which either egg masses or tadpoles were observed was less than a quarter. This figure does not differentiate among species, so that the realized reproductive success of the anuran fauna collectively must be commensurately reduced (c. < 0.25/3 water bodies per species, at most, where ‘3’ is for the three most widespread and abundance species and could be replaced by five or seven in the worst case). Unlike abundance measures and species distributional data, there almost certainly are no comparable landscape-scale data for assessing whether the low reporting rate of egg masses and tadpoles that we found is substantially poorer than in the past. Many amphibian species appear to be able to persist with sporadic breeding activity and success (both in few places and in few seasons) (Roberts, 1993; Tinsley & Tocque, 1995; Lauck et al., 2005), but it seems unlikely that successful breeding in as few as or fewer than 10% of water bodies per species, as we report, is likely to be viable over even a relatively short time frame.
Unlike effects that have been documented for birds (Mac Nally, 2007b; Radford & Bennett, 2007) and reptiles (Mac Nally & Brown, 2001) in this region, and amphibians elsewhere (Joly et al., 2001; Ficetola & De Bernardi, 2004), we found no discernible differences in the anuran faunas associated with agricultural landscapes. This situation may imply: (1) anurans are not sensitive to cover of native vegetation; or (2) the fauna has been ‘winnowed’ to such an extent that only generalized and resilient species remain. There are no data from pre-clearance times (beginning in the 1850s), so point (1) cannot be resolved. Point (2) seems more likely given the identity of the species currently found most often in the region in both kinds of landscape, namely the two species of Crinia and Limnodynastes tasmaniensis. It is possible that the kinds of sampling habitats we used (mostly constructed pools and ponds in wooded and cleared landscapes) bias occurrence lists towards widespread generalists. Targeted surveys may be needed to determine whether other anuran taxa that are not so heavily reliant on such lentic water bodies are sensitive to forest cover.
Maintenance of viable populations of anurans in a landscape depends on successful recruitment, which depends on temporal patterns and amounts of water availability (Laurance, 1996). Droughts, especially long ones, can greatly deplete frog populations (Piha et al., 2007). Since 1996, south-eastern Australia has been locked in a severe drought, with a 14% drop in mean annual rainfall combined with an increase in average maximum and mean minimum temperatures (Murphy & Timbal, 2008). Apart from the overall decline in rainfall, the main change has been a sharp decline (−61%) in autumnal falls (Murphy & Timbal, 2008). Autumn rains wet the soil, which primes the landscape for winter and spring rains to produce runoff into surface water bodies upon which anurans rely. Many climate models for south-eastern Australia predict a long-term shift to ‘flashy’ summer rainfall events rather than the autumn–spring peaks (Shi et al., 2008), which are very unlike the precipitation patterns in which the anurans evolved. Therefore, the ecological effects of reduced annual rainfall are further exacerbated by a change in its seasonal availability (Nicholls, 2004).
Most of the species of anurans in the region are autumn–spring breeders, so these shifts in water availability are likely to have significant impacts on population viability. While there are many water bodies in the landscape, the lack of breeding cues (e.g. heavy rains in the cooler parts of the year) seems to have been the overriding factor in generating the apparent steep decline in incidence, abundance and breeding activity.
The work was supported by the Australian Research Council (grant DP0664065). Surveys were conducted under ethics approval no. BSCI/2006/04 from the School of Biological Sciences, Monash University. We thank the following for assistance with the study and/or comments on the manuscript: David Reid, Andrew Bennett, Jarom Stanaway, Briarna Lacey, Geoffrey Brown and Jim Radford. This is publication no. 169 from the Australian Centre for Biodiversity, Monash University.
Ralph Mac Nally is Professor of Conservation Ecology, Director of the Australian Centre for Biodiversity and Theme Leader in Biodiversity and Ecosystem Processes at the Monash Sustainability Institute. His interests include landscape and conservation ecology and ecological computing and modelling.