Comparative use of active searches and artificial refuges to detect amphibians in terrestrial environments

Study area

We conducted our study in the temperate eucalypt woodlands of south-eastern Australia, and predominantly within the critically endangered white box Eucalyptus albens, yellow box E. melliodora and Blakely's red gum E. blakelyi grassy woodland and derived native grassland ecological vegetation community (Fig. 1). We included five monitoring programmes in the study, encompassing two water catchment management areas in Victoria and four bioregions in NSW and southern Queensland (Thackway & Cresswell 1997). Thus, our entire study area encompassed five geographically and climatically distinct regions: (i) North East and Goulburn Broken catchment areas in Victoria (hereafter called NE Victoria), (ii) NSW Riverina bioregion, (iii) NSW South-west Slopes bioregion (hereafter called SWS), (iv) a small-scale monitoring programme within the NSW South-west Slopes bioregion (hereafter called Nanangroe) and (v) Nandewar, New England Tablelands and Brigalow Belt South bioregions in northern NSW and southern Queensland (hereafter called the North-west Slopes – NWS; Table 1). The entire region extends from Warwick in southern Queensland (28°01S 152°11E) to Merton in southern Victoria (36°58′ 145°42′) and spans a latitudinal gradient of 1180 km (Fig. 1). The average annual rainfall across the study area ranges from 696 mm in the north, peaking in the summer months (Warwick weather station No. 41525), to 710 mm in the south, peaking in the winter months (Alexandra weather station No. 88001). The average annual minimum and maximum summer temperatures range from 17.9–30.0°C in the north to 11.9–29.3°C in the south. The average annual minimum and maximum winter temperatures range from 2.9–17.9°C in the north to 2.5–11.2°C in the south (BOM 2017).

Our study area encompassed a large proportion of the temperate eucalypt woodland ecological vegetation community in south-eastern Australia. This broad vegetation type once formed a relatively continuous band of vegetation on fertile soils west of the Great Dividing Range from approximately 27° S in southern Queensland to the lower south-east of South Australia (Lindenmayer et al. 2005). Currently, more than 95% of the temperate eucalypt woodland has been cleared and converted to agriculture (Yates & Hobbs 2000; Lindenmayer et al. 2010). For this reason, the majority of remnant vegetation on private property in our study area is used for livestock production purposes and remains in a modified condition.

Experimental design and survey protocol

We established 821 survey sites, primarily on private property, across the study area as part of five biophysical monitoring programmes (Table 1). Twenty-eight sites were located in Travelling Stock Reserves in NSW, six sites were located in conservation reserves in Victoria, and 16 sites were located in State Forests in the Nanangroe region near Gundagai in southern NSW. Each site consisted of a 200 × 50 m search area. Grazing management varied at each site and included areas under set stocking, rotational grazing (e.g. spring – summer grazing exclusion) or total grazing exclusion. Between 1999 and 2017, we conducted 5808 site visits across the entire study area.

At each site, we surveyed amphibians using time- and area-constrained (20 min × 1 ha) active searches of natural habitat and inspections of artificial refuges (Fig. 2). Artificial refuges were placed in arrays and consisted of four timber railway sleepers (1.2 m in length), four terracotta or concrete roofing tiles (423 × 265 mm), and one double stack of 1 m² corrugated steel sheeting (Michael et al. 2012). Two arrays were established at each site within the same 1 ha search area, placed 100 m apart and checked 3 months after deployment. The total amount of time to inspect both refuge arrays at each site was 5 min. Active searches included raking through leaf litter, lifting logs and surface rocks and inspecting exfoliating bark of mature trees.

For all monitoring programmes, four to six people visited between eight and ten sites per day. In total, we inspected 1642 arrays (consisting of 6568 roof tiles, 6568 timber refuges and 1642 corrugated steel stacks) between five and nine times over a 19-year period (1999–2017), representing a single survey every 2 years. All of the artificial refuges were placed flat on the ground in terrestrial environments without disturbing surrounding vegetation. However, during above average rainfall years (2011 and 2012), many monitoring sites located in the NSW Riverina were inundated due to local flooding. In all regions, many roofing tiles were damaged by livestock and periodically replaced, and in 2010 all of the original timber refuges (fence palings) in the SWS and Nanangroe were replaced with recycled timber railway sleepers for comparison with other monitoring programmes. We completed surveys between August and December and between 09.00 and 16.00 hours on clear, sunny days. To standardize detections, the order in which sites were surveyed was rotated to ensure each site was surveyed at different times of the day, and by different observers.

Statistical analysis

We modelled the probability of detecting any frog species separately for each bioregion using a Bayesian generalized linear mixed model (GLMM) with a Bernoulli distribution and logistic link function. We modelled detection probability rather than abundance due to differences in sampling area between active searches and artificial refuges methods. We completed our analysis in R (Team 2017) using the brms (Bürkner 2016) package. The models we considered included the following terms: Capture method (active search, tile, timber and tin); linear and quadratic effects of rainfall in the 3 months prior to the survey being conducted (termed recent rainfall); linear and quadratic effects of rainfall in the 4–12 months prior to survey (termed early season rainfall); and linear and quadratic effects of time since the artificial substrates were deployed (placement time). All continuous variables were standardized to have zero mean and standard deviation one. Sites were split into northern and southern clusters for the North-west Slopes (NWS) region as there was a large latitudinal gradient.

We used default priors in the brms package and ran the Markov chain Monte Carlo (MCMC) with four chains, for 10 000 iterations with the first 2000 used as burn-in and a thinning factor of 8, giving 4000 MCMC samples for inference. We used standard MCMC convergence diagnostics and all chains showed adequate mixing (Gelman & Rubin 1992). One hundred and sixty-two models were considered for the NWS region and 81 models were considered for the other four regions. We used the leave-one-out cross-validation information criteria (LOOIC) (Vehtari et al. 2016, 2017) to choose the most parsimonious model within two LOOIC units of the best fitting model. We report posterior means and 95% credible intervals.

We also modelled the probability of detecting individual frog species in each bioregion where there was sufficient detections to warrant further statistical modelling using the same terms as described above. We modelled the following species in the corresponding bioregions: spotted marsh frog Limnodynastes tasmaniensis (all bioregions), eastern sign-bearing froglet Crinia signifera (Nanangroe and NWS), inland banjo frog L. interioris (SWS), barking marsh frog L. fletcheri (NSW Riverina), eastern banjo frog L. dumerilii (NE Victoria), Peron's tree frog Litoria peronii (SWS) and smooth toadlet Uperoleia laevigata (NWS).

Summary statistics

We recorded a total of 3970 individuals representing 18 frog species from two families, Myobatrachidae and Hylidae (Appendix S1). The spotted marsh frog L. tasmaniensis was the most abundant species, accounting for 67.75% of all observations and was detected using all four survey methods across all five regions. Four additional species (Crinia parinsignifera, C. signifera, L. dumerilii and Lit. peronii) were detected in all regions using at least one survey method. Frog species richness increased along a latitudinal gradient with the northern bioregion supporting, on average, twice as many frog species as sites in southern regions (Appendix S1).

Effect of survey method on amphibian detections

Table 2 gives the overall detection rates for the presence of any frog species in each of the five regions by capture method. Overall, active searches and timber substrates produced the highest detection rates for the presence of any frog species. Active searches were the most effective method for detecting frogs in Nanangroe, a combination of active searches and timber substrates was most effective for detecting frogs in the SWS and NWS, whereas timber substrates were more effective at detecting frogs in NE Victoria and NSW Riverina (Table 2). Of all survey methods, roofing tiles were the least effective method for detecting frogs in all regions.

Effect of placement time and rainfall on amphibian detection rates

The best fitting model for the presence of any frog species in Nanangroe, SWS and NWS was characterized by an interaction between capture method and a quadratic effect of substrate placement time. The model for Nanangroe included an interaction between the linear component and capture method, whereby detection rates from active searches increased steadily over time, whereas detections beneath refuges were consistently low (Fig. 3). The models for the SWS and NWS regions revealed both interactions between the linear and quadratic components and capture method. We provide the results of the LOOIC model selection for the each of the five bioregions for the presence of any frog in Appendix S2 and the various temporal trajectories for each region are illustrated in Figure 3.

Early season rainfall (encompassing 4–12 months prior to survey) was positively associated with frog detections in the SWS, coinciding with a lag time in peak frog detections approximately 8 years after refuge deployment. By contrast, we found no rainfall effects for Nanangroe. In NWS, recent rainfall (within 3 months of a survey) and early season rainfall (4–12 months prior to survey) had positive effects on frog detection with peak detections occurring within 2 years after refuge deployment. These results indicate that an optimal amount of rainfall for detecting any frog species in this region is at least 215 mm 3 months prior to conducting a survey, and 704 mm earlier in the season (at least 4–12 months prior to a survey).

In the NSW Riverina and NE Victoria bioregions, we found additive effects of capture method and time since substrate deployment (Fig. 3), indicating that differences among capture methods were constant over time. The ranking of capture methods for NSW Riverina (from best to worst) was timber railway sleepers, corrugated steel, active search and roofing tiles, whereas the ranking for NE Victoria was timber railway sleepers, active search, corrugated steel and roofing tiles. We found a quadratic relationship between detection and recent rainfall (peak = 210 mm) for NSW Riverina and a positive relationship between frog detections and early season rainfall in NE Victoria (Fig. 3). Overall, across most bioregions, frog detections using active searches and several different types of artificial refuge peaked simultaneously, irrespective of when refuges were first deployed.

Species-specific detection rates

We provide the overall detection rates by capture method for seven common frog species in each region in Table 3. Active searches, timber railway sleepers and corrugated steel were most effective at detecting L. tasmaniensis in all regions, except NSW Riverina, where timber and steel substrates yielded the highest detection rates. The probability of occurrence for L. tasmaniensis differed between the northern and southern parts of the NWS region, although capture patterns by method over time were identical. Active searches were most effective for detecting C. signifera and U. laevigata, whereas timber was effective at detecting Lit. peronii and a combination of timber and steel was effective for detecting Lit. interioris, Lit. fletcheri and L. dumerilii (Table 3). Limnodynastes tasmaniensis was the only common species detected in all regions, so for this species we provide the various temporal trajectories by region in Figure 4. We provide the results of the LOOIC model selection for L. tasmaniensis and the six other abundant frog species in Appendices S3 and S4.

Effect of survey method on amphibian detections

Overall, we recorded more than 3000 individuals from 18 species of the 25 frog species predicted to occur within the study area (Cogger 2014). Frogs were detected using all survey methods, although detection rates varied according to method and region. Generally, active searches and timber refuges were equally effective in detecting a variety of frog species, whereas roofing tiles generally performed poorly in all regions. Artificial refuges may provide a more standardized method for detecting amphibians across sites than active searches because levels of natural microhabitats can be highly variable, especially in modified landscapes, but using refuge requires financial resources (e.g. approximately AU$4.00 per railway sleeper) and labour to deploy them before surveys commence. The time between refuge deployment and first survey also needs to be taken into account because detection rates are influenced by rainfall patterns, and this may potentially preclude their use in short-term studies, especially those conducted during droughts or below average rainfall years. Refuges also may be disturbed by livestock, wildlife poachers or damaged by strong winds meaning they may need to be regularly replaced or repositioned.

Our findings suggest that artificial refuges can produce similar if not higher frog detection rates than active searches in some bioregions. Furthermore, the time required to inspect refuge arrays (approximately 5 min/site) is four times less than the time required to obtain similar frog numbers by actively searching natural habitat over a larger search area (1 ha). For example, we recorded a maximum detection rate of 20% beneath timber in the NSW Riverina as opposed to a 16.7% detection rate using active searches in the high rainfall NWS bioregion. Thus, while costs are associated with establishing artificial refuges, once established, they require less survey effort to return similar results to active searches.

Timber substrates, particularly the recycled railway sleepers used in this study, provide amphibians with suitable terrestrial refugia in agricultural landscapes because they are solid, and hence, not easily disturbed or damaged by livestock. Damage caused by livestock trampling and breaking roofing tiles used in this study was evident and a potential reason why detection rates beneath this type of refuge were comparatively low. Weathered timber refuges also provide a variety of microhabitats, including vertical holes and subsurface cracks and cavities, where both terrestrial and arboreal species were often found sequestered. Timber refuges also retain soil moisture and create deep soil cracks as ambient air temperature increases (Michael et al. 2004), providing humid microclimates during dry periods and ideal subterranean conditions for burrowing species. We suggest that future studies and amphibian monitoring programmes consider using timber railway sleepers and sheets of corrugated steel as complimentary methods for obtaining estimates of frog abundance in terrestrial environments because they are durable and provide a standardized method for comparing across sites, although we acknowledge that occupancy patterns are likely to be influenced by a range of biotic and abiotic factors (Hoare et al. 2009; Thierry et al. 2009). Timber refuges also may prove to be more cost-effective in the long-term for detecting frogs in any given area as the time required to inspect refuges is considerably less than the time required to search for frogs in their terrestrial habitat, an important consideration in environmental assessments. Thus, artificial refuges provide a useful tool for understanding amphibian use of terrestrial environments and provide a robust standardized method for evaluating frog occupancy patterns, although frog detection rates using both active searches and artificial refuges are likely to be influenced by natural variation in levels of suitable terrestrial shelter sites in the surrounding landscape.

Effect of time since deployment and rainfall on frog detection rates

A key assumption associated with using artificial refuges is that they provide a reliable and standardized tool for detecting species in their natural habitat. However, factors such as sampling intensity (Marsh & Goicochea 2003), time of sampling (Lettink & Cree 2007), weather variables (Hoare et al. 2009), refuge thermal properties (Thierry et al. 2009) and placement period can influence detection rates (Batson et al. 2015). We are not aware of any studies that have investigated the use of artificial refuges or cover boards to monitor herpetofaunal abundance over timeframes longer than 10 years. In our two longest monitoring programmes (SWS and Nanangroe), we found interactions between survey method and refuge placement period. In Nanangroe, detection rates by any method were low prior to 2010 (during a prolonged drought period), thereafter increasing using actives searches while simultaneously decreasing beneath refuges. Similarly, in the SWS, detection rates by any method were low prior to 2010, increased substantially beneath refuges during 2011 (coinciding with above average rainfall events), whereas detections using active searches steadily increased. In these two bioregions, fence palings were used for the first 9 years, thereafter replaced with railway sleepers due to the rapid rate at which the fence palings degraded. The replacement by a different type of timber refuge may have resulted in improved detection rates in the SWS, although increased detections post 2010 were not evident in Nanangroe. Prior to 2010, the probability of detecting any frog species was low and was likely influenced by the Millennium drought, a decade long period associated with declines in amphibian abundance in south-eastern Australia (Scheele et al. 2012; Mac Nally et al. 2014). During drought years, many frog species also spend long periods sequestered below-ground, thereby reducing detectability.

In NE Victoria, NSW Riverina and NWS regions, detection rates beneath refuges peaked approximately 2 years after deployment and then sharply declined in 2015/2016. These peaks occurred simultaneously across all survey methods suggesting that lag times in detection are influenced by weather-related variables rather than survey method. Thus, by examining frog detection rates between methods, we are able to separate the temporal effects of placement time from the stochastic effects of rainfall and conclude that time since refuge deployment had little influence on frog detections in this study. Several studies have reported increased frog abundance along a rainfall gradient (Woinarski et al. 1999), with increased detection rates at breeding sites attributable to recent rainfall (Paltridge & Southgate 2001; Penman et al. 2006) as well as long-term rainfall patterns (Trenham et al. 2003). Variables such as water levels, water temperature and long-term weather patterns, including drought-flood cycles, can affect breeding activity, community structure and population dynamics (Dostine et al. 2013; Wassens et al. 2013; Mac Nally et al. 2014). However, little is known about the influence of climate patterns on amphibian movements and their use of terrestrial habitats (Hazell 2003; Ocock et al. 2014). As frog movement and dispersal behaviour occur more frequently during the breeding season, and under favourable environmental conditions, (Pittman et al. 2014; Westgate et al. 2018), encounter rates with terrestrial refuges are also predicted to increase during these periods.

Species-specific responses to artificial refuges

The most abundant and widespread species responsible for driving overall frog detection patterns was L. tasmaniensis, a pond-breeding species that utilizes a wide variety of terrestrial microhabitats for over-wintering and foraging (Barker et al. 1995), and for migrating between ephemeral, rain-fed wetlands (Wassens et al. 2013) and farm dams (Hazell et al. 2004). The broad terrestrial habitat requirements of this species (Barker et al. 1995), were reflected in the capture rates beneath both natural (logs and rocks) and artificial refuges across all five regions. Species-specific dispersal ability and dependence on waterbodies for breeding are thus likely to explain differences in capture rates among species with different life-history traits and ecological requirements. For example, stream-dwelling species that were predicted to occur in the study area were rarely or never detected, whereas several burrowing and pond-breeding species (e.g. Notaden bennettii and Pseudophryne bibroni) were only occasionally detected. Furthermore, small cryptozoic species such as U. laevigata and C. signifera were more likely to be detected beneath natural substrates such as surface rocks and logs than artificial substrates. Therefore, species capable of using a wide variety of natural, semi-natural of artificial wetlands are more likely to encounter artificial refuges than species restricted to permanent waterbodies, have poor dispersal ability or specific habitat requirements. This bias towards capturing wide-ranging habitat generalists suggests that artificial refuges may have limited application for procuring records of sedentary, range-restricted or stream-breeding frog species, traits that are shared by many threatened Australian frog species (Hero et al. 2006), unless artificial refuges are placed specifically along wetland or stream margins. The use of artificial refuges to survey and monitor stream-dwelling species in Australia requires further research.