Influence of landscape features on the distribution of the vulnerable frog species Mixophyes iteratus in the Tweed Valley, northern New South Wales, Australia

Introduction

At a broad scale, landscape ecology can be defined as the scientific study of ecological processes, usually interactions and patterns. The scale of study may vary depending on the size of the landscape. The interaction between the scale of habitat and species response can depend on the species itself (Mayor et al. 2009; Jackson and Fahrig 2012) and on what is being measured (Moraga et al. 2019). Frogs generally move over small distances (Hodgkison and Hero 2001). Studies on frogs suggest that habitat selection occurs at small scales, with species selecting pool morphology (Bosch and Martínez-Solano 2003; Van Buskirk and Smith 2021), stream morphology (Brown et al. 2021), ground structure (Blomquist and Hunter 2010; Hinderer et al. 2021), vegetation structure (Shuker and Hero 2012; Green et al. 2021), vegetation floristics (Lehtinen and Carfagno 2011), light and temperature (McEwan et al. 2021), water quality (Banks and Beebee 1987; Simpkins et al. 2014) and conspecific presence (Pizzatto et al. 2016). There have been relatively few studies of frog occurrence on the landscape scale, especially in Australia (Hazell 2003). Studies conducted at a broader scale have found that species do appear to be influenced by landscape-scale effects, including vegetation cover (Vos and Stumpel 1996; Mazerolle et al. 2005; Green et al. 2021), wetland cover (Kolozsvary and Swihart 1999), connectivity (Knapp et al. 2003; Ficetola and De Bernardi 2004; Atobe et al. 2014), pond cover (Mazerolle et al. 2005; Green et al. 2021), and surrounding land use (D’Amore et al. 2010). A few studies have illustrated a relationship between species and habitat at multiple scales (Vos and Chardon 1998; Pope et al. 2000; Hazell et al. 2001; Lemckert and Mahony 2010; Brown et al. 2021).

Landscape scale studies involving frogs have focused largely on sampling ponds, lakes or dams. The advantage of these studies is that the sampling area is clearly defined and conforms with landscape models where the habitat is a defined island (see Pulsford et al. 2017 for information on landscape conceptual models). In contrast, there is a lack of landscape ecology on stream-dwelling frogs (Erős and Campbell Grant 2015). This may be due to adult frogs or tadpoles using the stream to disperse through an unfavourable terrestrial ‘matrix’, thereby not conforming to the typical island-matrix relationship.

Mixophyes iteratus is listed as vulnerable under the Biodiversity Conservation Act 2016 (NSW) and the Environment Protection and Biodiversity Conservation Act 1999 (Cth). The only research study investigating landscape scale effects on M. iteratus (Lemckert 1999) showed the species was found at sites with a greater proportion of undisturbed forest within a 1-km diameter circle. This study aims to understand how the species is influenced by landscape scale features, which may help inform conservation strategies. This broad scale study also provides a better understanding of species distribution across the Tweed Valley landscape in north-eastern New South Wales, Australia.

Materials and methods

Mixophyes iteratus (Straughan 1968; Fig. 1) belongs to the Myobatrachidae family and is a large terrestrial species that burrows under leaf litter, logs, or loose soil during periods of inactivity in winter. Adult males are known to reach 98 mm in snout-vent length and females are known to reach 116 mm in snout-vent length. Adults leave shelter around dusk, with males commencing calling as soon as they have left shelter. The breeding season in the Tweed Valley appears to occur from October to April.

Fig. 1.

Mixophyes iteratus. The top images show adults in amplexus. Note the size difference between the sexes. The bottom left image shows an adult female and the bottom right image shows a stream section that is occupied by M. iteratus. Note the pool, wooded stream edge and undercut bank, which are favourable habitat features for breeding. Photos on the left hand side were taken by Michele Lockwood. Photos on the right hand side were taken by Greg Lollback.

This stream-dwelling frog (Fig. 1) is known to select small scale habitat features (Lewis and Rohweder 2005) including stream areas with flat banks that are undercut at the water edge with slow flowing stream pools (Lollback et al. 2021). Riparian vegetation is often either wet sclerophyll forest or rainforest (Meyer et al. 2001). Since tadpole development takes at least 9 months to complete, the species requires permanent streams (Anstis 2013), once metamorphosed, the species will disperse, forage, find shelter and undergo mate selection on land. Due to the dependency of land, landscape scale factors may influence the presence of the species.

All frog surveys were restricted to the Tweed Valley, located in north-east New South Wales, Australia. The valley is part of an ‘erosion caldera’ that is derived from the Wollumbin Volcanic complex, which was active approximately 23 million years ago. Erosion of this volcanic complex has left escarpment and hills that enclose much of the valley from three sides, while the Pacific Ocean hems in the eastern extent. The Tweed Valley is approximately 1319.7 km² and is subtropical, with a mean minimum and maximum temperature of 14.5°C and 25.8°C, respectively. Mean yearly rainfall is 1574.4 mm (www.bom.gov.au).

Woody vegetation cover in the Tweed Valley is approximately 56% and is comprised mostly of wet sclerophyll forest and rainforest (based upon State Government of NSW and Department of Planning and Environment 2022). Camphor laurel (Cinnamomum camphora) is an invasive weed that was introduced after the expansion of dairy farming in the Tweed during the late 1800s and early 1900s (Stubbs 2012). It is now a naturalised canopy tree within the Tweed Valley, often coexisting with a rainforest understorey.

Results

There was a total of 40 transects sampled, with an average length of 726.6 m (s.d. = 334.3 m) and a minimum and maximum length of 200 and 1580 m, respectively. Thirteen transects were surveyed three times, 25 transects were surveyed twice, and three transects were surveyed once. A total of 15 transects were surveyed within the 2019/2020 breeding season, which provided data for (Lollback et al. 2021). A further 11 transects were surveyed within the 2020/2021 breeding season, and 14 other transects were surveyed during the 2021/2022 breeding season.

Surveys were conducted over a range of abiotic conditions during the breeding season. Average ambient temperature during surveys was 22.6°C (min = 15.8°C, max. = 27.3°C), average relative humidity was 92.1% (min. = 71.7%, max. = 100%), and average rainfall in the previous 24 h was 4.5 mm (min. = 0 mm, max. = 45.8 mm). It rained during 30 out of 90 surveys and it rained 44 times in the 24-h period before surveys were conducted. Average cloud cover was 46.6% (min. = 0%, max. = 100%), average moon availability was 17.5% (min. = 0%, max. = 100%), average moon phase was 48.5% (min. = 0%, max. = 100%), and average air pressure was 1055.3 hPa (min. = 979.3 hPa, max. = 1055.3 hPa).

Site (transect) location was spread across the Tweed Shire; however, there were more sites in the western half of the Shire due to the prevalence of cane farms and brackish streams closer to the coast. Site elevation varied from 1 m to 272 m (mean = 87 m), with 12 sites in national park and the average distance from national parks was 3.7 km. There were four sites located on second order reaches, 14 sites on third order reaches, 16 sites on fourth order reaches and six sites on fifth order stream reaches.

Mixophyes iteratus was detected at 21 of the 40 transects, during 45 out of the 90 surveys. During these surveys, there were 158 adults and 80 juveniles/sub-adults detected. The average number of M. iteratus found at occupied sites during a survey was 1.12 frogs per 100 m (min. = 0.12 frogs per 100 m, max. = 8.5 frogs per 100 m). The average distance a frog was found from the stream edge was 3.2 m (s.d. = 2.7 m, n = 145), with a minimum and maximum distance of 0.0 m and 15 m, respectively. In the 2019/2020 breeding season, M. iteratus was found on 73.3% of transects, while M. iteratus was detected on 45.4% and 35.7% of transects in the 2020/2021 and 2021/2022 seasons, respectively. Spatially, M. iteratus was found more often in the western half of the Tweed Shire. Frogs were found from 26 m to 203 m in elevation and across all orders of stream surveyed. A total of 70% of national park sites were occupied, while 46% of sites outside national park were occupied. Average frog count was 0.35 (0.64 s.d.) frogs per 100 m in transects outside national park, and 1.56 (2.55 s.d.) frogs per 100 m in transects inside national parks.

Occupancy analysis – scale of effect

The scale of effect where the landscape variable had its strongest effect on occurrence, was investigated before a final suite of occupancy models were analysed. The scale of effect differed between buffer types across vegetation types is in Fig. 2. For vegetation categories, the scale that consistently had the lowest QAIC values was 3200 m for either linear stream buffer or circular buffer. The relationship between scale and QAIC differed with vegetation type but was similar for the proportion of native vegetation and the proportion of total vegetation because these two variables were correlated. This was also the case with the two linear buffers.

Fig. 2.

Scale of effect graphs for landscape variables and M. iteratus occupancy, including the proportion of camphor laurel, proportion of rainforest, proportion of native vegetation, proportion of all vegetation, proportion of road, slope of stream and stream sinuosity. The x-axis is distance of stream or radius of circle (m). The y-axis is QAIC value.

The relationship between occurrence and non-vegetation variables was different to that between occurrence and vegetation variables. The scale of effect for the proportion of road, stream slope and stream sinuosity occurred at short distances, often less than 800 m of circle radius or stream length. It is important to note that the QAIC difference between scales and buffer types for each variable was less than two, suggesting differences were minimal.

GAM analysis – scale of effect

The scale of effect when using standardised maximum frog count occurred at various scales, depending on the landscape variable (Fig. 3). The scale of effect for the proportion of camphor and the proportion of all vegetation, the proportion of road and sinuosity was 3200 m radius/distance of stream. Out of five variables, circular scales performed better than stream buffers. There were no linear response variables that continually decreased or increased along scale.

Fig. 3.

Scale of effect graphs for landscape variables and M. iteratus maximum count, including the proportion of camphor laurel cover, proportion of rainforest cover, proportion of native vegetation cover, proportion of all vegetation cover, proportion of road cover, and slope of stream is stream sinuosity. The x-axis is distance of stream or radius of circle (m). The y-axis is percentage deviance explained.

Final suite of frog count models

As well as the previously mentioned occupancy analysis variables that correlated, the proportion of road within a 3200-m radius circle correlated with the proportion of native vegetation in a 1600-m radius circle (ρ = −0.85, P = 0.00), the proportion of total vegetation within a 3200-m radius circle (ρ = −0.89, P = 0.00) and the number of upstream stream heads (ρ = 0.72, P = 0.00). Correlated variables were removed from the suite of models.

The top performing model explained frog counts with an interaction between longitude and latitude, suggesting some spatial clumping in counts across the Shire (Table 2). The model fit is in Fig. 4. The top six models were reasonably close in performance and four of these models used longitude and/or latitude to explain frog counts. Other top performing models had either the shortest distance to a stream head or the proportion of road within a circle with a radius of 3200 m (Fig. 5). As with the occupancy analysis, other landscape scale variables were poor predictors of the number of frogs counted per 100 m of transect. The top model was used to model frog count across the Tweed Valley (Fig. 6). Modelling mirrored counts in the south-western and southern part of the Tweed Valley but was likely a poor predictor of density in the northeast.

Table 2.GAM model performance of models where the standardised maximum count of frogs (per 100 m of transect) is the response variable.

Model	Deviance explained by model (%)
X*Y (te)	54.7
Shortest distance to stream head	47.5
X + Y	45.3
Road c_3200	44.4
Y	40.1
X	40.1
AHD	28.1
Camphor c_3200	25.5
Slope 200	20.5
Sinuosity 3200	18.2
stream order	17.6
RF 100_400	17.3
Distance to NP	15.8
Longest distance to stream head	13.1
Join NP	3.04

*,  interaction term; te, full tensor product smoothing was used when modelling the GAM. Models are ranked by the percentage of deviance explained. Variables used to model maximum frog count and shown in the table below include longitude (‘X’), latitude (‘Y’), shortest distance to stream head, road cover within a circle that has a radius of 3200 m, elevation (‘AHD’), camphor laurel cover within a circle with a radius of 3200 m, (‘Camphor c_3200’), stream gradient over a distance of 200 m (‘Slope 200’), stream sinuosity over a distance of 3200 m (‘Sinuosity 3200’), stream order, rainforest cover within a 100 m stream buffer for a length of 400 m (‘RF 100_400’), distance to a conservation area (‘Distance to NP’), longest distance to stream head and whether a site was within a conservation area (‘Join NP’).

Fig. 4.

The best performing GAM model, maximum count = X*Y, showing residuals and diagnostic information.

Fig. 5.

The second and fourth best performing GAM models showing residuals and diagnostic information. (a) GAM with the proportion of road within a circle that has a radius of 3200 m, (b) GAM with the shortest distance to an upstream stream head, (c) diagnostic plots for (a), and (d) diagnostic plots for (b). The triangles in (a) and (b) are residuals and the dashed lines are 95% confidence intervals.

Fig. 6.

Estimated count of M. iteratus per 100 m of third order stream within the Tweed Shire. Estimated abundance is based upon the best-performing GAM model. M. iteratus survey site presence/absence (stars) is also shown. The basemap is sourced from Google LLC.

Discussion

This study looked at the distribution of M. iteratus throughout the Tweed Valley landscape at 40 transect sites with differing elevations, proportions of vegetation cover, stream morphology types and distance to conservation reserves. The vulnerable M. iteratus was detected at 21 of these sites. Previous surveys of the species within the Tweed Valley have been limited (Goldingay et al. 1999), resulting in much uncertainty regarding its distribution within the study area. While the vulnerable status of the species is justified, the species is much more broadly distributed across the Tweed Valley than previously thought. Goldingay et al. (1999) found M. iteratus occupied sites less than 300 m in elevation. All survey sites used during this study were located less than 300 m in elevation and stream reaches above this elevation are likely to be first order gullies that are too steep and rocky and lack permanent water to provide quality habitat for the species. While according to the analyses there was no distinguishing association between M. iteratus occupancy or counts and national parks, there does seem to be a mildly favourable association in abundance/occupancy and national parks. National parks in the Tweed Valley are generally exempt from ongoing vegetation clearing and can contain more mature vegetation, have greater canopy cover and less disturbance than private land. These features likely create more plant litter for burrowing and foraging habitat, while providing a more stable microclimate, better water quality and habitat that is more flood resilient for M. iteratus. National parks within the Tweed Valley are often situated in steep, higher elevated land that is unfavourable for grazing or cropping. While this type of topography can also be restricting in regard to habitat availability for M. iteratus, national parks often encompass the heads of streams where M. iteratus populations have been found to occur and they may act as source populations where tadpoles and juvenile frogs become washed downstream during high flow events.

Analyses suggested that there was no strong relationship between M. iteratus occupancy and landscape features and this was reflected in only small differences in QAIC value when investigating scale of effect. Moraga et al. (2019) hypothesised that occurrence, which generally takes generations to change, should be influenced at broad scales. They also hypothesised that trends in occurrence should occur at larger scales than trends in abundance. They found that there was a scale of effect regarding wood frog (Lithobates sylvaticus) occurrence and road density; however, the relationship between occurrence and the scale of forest cover was not clear. Additionally, scale of effect for wood frog abundance did not occur at a much smaller scale than occurrence. Jackson and Fahrig (2012) reasoned that the scale of effect is less likely to exist in gap-avoidant species and although M. iteratus requires a forested riparian zone, the species likely can travel through forest gaps by using the stream to disperse. Many streams in the Tweed Valley have a continual thin section of riparian vegetation, including some occupied sections of stream (<70 m wide). Mixophyes iteratus frog count also displayed a larger scale of effect than occupancy for cover of some vegetation types, proportion of road and stream sinuosity, which dispels the hypothesis postulated by Moraga et al. (2019).

Not surprisingly, the response to large scale effects appears to be species specific (Kolozsvary and Swihart 1999; Ficetola and De Bernardi 2004), as has been the case with Australian frog species and elevation, distance to large reserves (Lemckert 1999), solar radiation, moisture level (Lemckert and Mahony 2010) and area of native canopy within 1 km radius (Hazell et al. 2001). Lemckert (1999) did find an association between occupancy of M. iteratus and percentage of undisturbed forest within 500 m. Most of the sites surveyed within the Tweed Valley had a history of disturbance, so this variable could not be tested on a large scale. However, disturbance level was investigated at a small scale in Lollback et al. (2021) and found not to be an important factor in influencing M. iteratus occupancy. Nevertheless, it was deemed an influencing factor when assessed by Lewis and Rohweder (2005). Agents and measures of disturbance, including weed prevalence, fire history, agricultural history, and erosion may influence frog occurrence and the importance of disturbance should not be dismissed. However, our study would need to have different site selection methodology to test this hypothesis suitably and would be better reflected by measuring variables at a smaller scale. Nevertheless, undisturbed sites would likely contain better tree cover, leaf litter coverage, understorey cover along banks and water quality and some disturbance may also be measured at a larger scale (e.g. land tenure). In contrast, drought is a stochastic process that often occurs over a wider scale than a single landscape and can indeed influence occupancy by causing localised extinction to species that are dependent upon waterways. This study occurred just after a drought, but included sites were surveyed more than two decades ago (Goldingay et al. 1999) and were found to be still occupied by M. iteratus (see Lollback et al. 2021 for further details).

Mixophyes iteratus appears to be robust to habitat fragmentation within the Tweed Valley may be because the species can disperse via steams, thereby avoiding ‘matrix’ type land unlike pond-breeding species where fragmentation has been shown to influence occupancy (Knapp et al. 2003; Ficetola and De Bernardi 2004; Mazerolle et al. 2005; D’Amore et al. 2010; Green et al. 2021). Many of the streams within the Tweed Valley do not display fragmentation of ‘longitudinal connectivity’ (Erős and Campbell Grant 2015) in the form of dams, flooding or extreme changes in water quality, with connectivity along the streams being maintained (Fig. 7); and many of the streams within the Tweed Valley contain a narrow corridor of vegetation, which is enough to disperse across land and provide breeding habitat. Although heavily disturbed from past clearing, weed invasion, cattle grazing, land shaping, and vegetation modification, the Tweed Valley still contains approximately 56% woody vegetation cover (taken from Office of Environment and Heritage 2012 dataset). This likely allows the distribution of M. iteratus across the Tweed Valley landscape to be broad, especially in the western and south-western sections of the Valley.

Fig. 7.

(a) Transect locations in the landscape showing view from Border Ranges National Park overlooking Mebbin National Park, the Limpinwood area and Wollumbin Red stars, locations where Mixophyes iteratus was detected; white star, location where M. iteratus was not detected. (b) View of a survey location within Doon Doon area. (c) View of a survey location within the Crystal Creek area. (d) View of a survey location within the Chillingham area. Images (b), (c) and (d) show approximate transect locations (red line) where M. iteratus were detected.

This is not the only study to indicate some landscape scale relationship with frog count. Like this study, da Silva et al. (2011) found a relationship with frog abundance and roads and Atobe et al. (2014) found a relationship with connectivity and frog count. Roads may cause isolation and increase mortality (Fahrig et al. 1995; Vos and Chardon 1998; Goldingay and Taylor 2006), but road influence would likely occur on a small scale, unlike this study where the scale of effect was at 3200 m diameter. It is more likely that sites with large counts occurred in larger forest patches with fewer roads at a large scale (for example, national parks). These forest patches tended to have better quality habitat that could support higher abundance. This may be the case even if forest cover was not a good predictor of abundance because the important factor may not be patch size, but only if the patch is small or large. Landscape features can indirectly influence species by shaping or coinciding with smaller habitat scale attributes (Ficetola et al. 2011). For example, the edges of large forest patches may offer a buffer from disturbance to core vegetation (Saunders et al. 1991) or large patches may indicate minimal historical disturbance. Either way, the result is better quality small-scale features in large patches.

The best predictor of maximum M. iteratus count was the interaction between easting (X) and northing (Y), with higher counts occurring around Mebbin National Park in the south-west of the Tweed Valley (Fig. 6). While this area was not without disturbance, stream structure, and surrounding vegetation was of generally good quality. Areas near Mount Jerusalem National Park also revealed good counts. Nevertheless, the best landscape predictor could only account for approximately 57% of count variation. This is to be expected due to the potential dynamic behaviour of abundance and the number of factors that can influence it (Krebs 1994).

Stream features within Mooball National Park appeared typical of what M. iteratus would occupy. However, like the other eastern sites of the Tweed Shire, the species was absent. It appears that while small scale factors may influence occupancy and abundance, there is also a large-scale spatial influence on the species. This influence may have initially occurred well before the arrival of Europeans and the clearing of forest and may be more aligned with distribution expansion and topographical features not measured here.

The absence of strong large-scale patterns with relative abundance and occupancy likely reflects species ecology and pre-clearing (late 1800s) distribution within the Tweed Valley. It is known that M. iteratus displays a strong selection for small-scale stream features (Lollback et al. 2021) and these features were likely more common across the Valley, including the lower elevation areas where stream velocity is slower. It is likely that M. iteratus was once a common frog species in the Tweed Valley, especially the southern and western sections, before broadscale land clearing occurred.

Supplementary material

Supplementary material is available online.