Population stability in an unmanaged population of the green and golden bell frog in northern New South Wales, Australia

Keywords: Litoria aurea, threatened frog, multipopulation monitoring, mark–recapture, program MARK, robust design, temporary emigration, population decline, Yuraygir National Park, coastal lagoon.

Central to threatened species conservation is population monitoring (Legge et al. 2018). Without knowledge of population size and trajectory it is difficult to have an effective conservation program and to be able to assess the effectiveness of management actions. A particular challenge for threatened species with wide geographic ranges is that more than a single population needs to be monitored. Species’ distributions may contract but without multipopulation monitoring, awareness of a deterioration in conservation status may not arise until much later when the species requires strong management intervention (e.g. Reside et al. 2019). Recognition of the need for multipopulation monitoring leads to the important question of which populations, as well as over what period and with what frequency, should be monitored.

In New South Wales (NSW), Australia, the recent approach to threatened species conservation has included the monitoring of multiple populations of a species. The number and location of those populations was decided by panels of species experts under the guiding principle of determining which populations were needed to secure the species’ persistence in the wild in NSW for 100 years (OEH 2013). Projects that included management actions to mitigate threats were subsequently developed by species experts or with their input.

The green and golden bell frog provides a case study of the above approach to species conservation. This species suffered a substantial decline in NSW where the majority of its wide distribution occurred (Goldingay 2008; White and Pyke 2008; Mahony et al. 2013). Its decline appears to be a combination of habitat loss producing small isolated local populations, predation by introduced gambusia fish and mortality due to the amphibian chytrid fungus (Goldingay 2008; Stockwell et al. 2010; Pollard et al. 2017; Klop-Toker et al. 2018). Eight priority populations were selected for this species (www.environment.nsw.gov.au/savingourspeciesapp/project.aspx?ProfileID = 10483). These populations covered the species’ range in NSW and included the remaining tableland population at an elevation >700 m (see Osborne et al. 2008), and several coastal or lowland populations consisting of one on a near-shore island, one at the northern extreme of the range, and several in the central part of the range. Six were located within conservation reserves or within a mix of land tenure that included conservation reserve. The locations varied in landscape context and included urban Sydney (Pickett et al. 2014), private farmland on the southern tablelands (Osborne et al. 2008), part of a Newcastle industrial site (Hamer and Mahony 2007) and bordering farmland and urban development on the south coast (Hamer 2018). One advantage that this species had over many other threatened species in NSW was that it had been subject to >20 years of field research, with previous studies reporting on the monitoring of populations (see White and Pyke 1996, 2008; Pyke and White 2001; Goldingay 2008; Osborne et al. 2008; Mahony et al. 2013). This provided a head-start in terms of where monitoring should occur at the selected sites and what form it should take (i.e. mark–recapture or occupancy-based).

Here we report on monitoring that was conducted within the green and golden bell frog population at the northern end of the range in Yuraygir National Park. The aims of our study were to produce annual estimates of population size and describe population trajectory. From this we may be able to infer population stability. This population had been subject to previous mark–recapture studies that suggested the population was stable based on population estimates from single breeding seasons spaced 17 years apart (Goldingay et al. 2017). Further mark–recapture studies are needed to corroborate this conclusion.

Our earlier study had identified several constraints in describing population size. We estimated population size based on just the male segment of the population. This is a common approach in frog studies because sampling is conducted at the breeding sites where males aggregate to call and females visit periodically to spawn, which commonly leads to many more males than females being captured (Richards and Alford 2005; Phillott et al. 2013; Quick et al. 2015). Female bell frogs mature later than males (Pyke and White 2001; Hamer and Mahony 2007) and may not spend as much time around the wetland sites where males are typically captured. Although not critical to estimating population size and trajectory, analysing recapture data of females may provide some important insights.

Another constraint of our earlier study was that the sample area in which frogs were captured was approximately half of a large coastal lagoon. All of the lagoon provided high quality habitat to frogs so population estimates might represent only 50% of the total population. Bell frogs are highly mobile (Hamer et al. 2008) and, due to subsampling of the lagoon, we suspect that temporary emigration occurs and needs to be estimated. Temporary emigration is an aspect of animal behaviour that may arise because some individuals move out of a sample area or otherwise change their behaviour so they are temporarily unavailable for capture. Not accounting for this might bias population estimates. The Robust Design (Pollock 1982) is a population model that allows temporary emigration to be estimated. Our earlier research suggested that male individuals had a probability of 0.39 to undertake temporary emigration between sample months in one breeding season (Goldingay et al. 2017). Further research encompassing several breeding seasons should clarify this.

This study was conducted at Station Creek in Yuraygir National Park, in north-east NSW, where two previous investigations have been conducted into green and golden bell frog population ecology (see Goldingay and Newell 2005; Goldingay et al. 2017). Another smaller population occurs 15 km north at Diggers Camp. The primary wetland occupied by the bell frogs at Station Creek was a large coastal lagoon that measured 40–70 m wide and 700 m long (5.4 ha). This lagoon always retained water during this study but had been observed to reduce to a low level in 2004 (see Goldingay and Newell 2005). Water depth was commonly ~1 m but varied seasonally. A small swamp (30 m by 60 m) that was connected to the lagoon underwent annual drying out during the summer. The lagoon and swamp were fringed by tall sawsedge (Gahnia clarkei), jointed twig rush (Baumea articulata) and cumbungi (Typha orientalis) (Fig. 1). Four ephemeral ponds (~25 m by 25–50 m) were located within nearby sand dunes but did not retain water during this study. Rainfall at the nearest weather station at Wooli had an annual mean of 1361 mm (Bureau of Meteorology; www.bom.gov.au). It varied over the three years of this study and the year before (2014: 1362 mm; 2015: 1878 mm; 2016: 1095 mm; 2017: 1714 mm).

Fig. 1.  The coastal lagoon at Station Creek is fringed by expanses of (a) cumbungi and (b) jointed twig rush. Images: S. Jacomy.

Surveys were conducted over three nights in each of 3–4 months in three frog breeding seasons (September–March) from November 2015 through to December 2017. One additional survey period in 2015, two in 2016 and one in early 2018 were conducted but were excluded from analyses due to very low captures or because they did not extend for three nights. Frogs captured on these nights appear in the overall summaries but were not used in the modelling described below. This meant that the modelled data in the third year were from October–December 2017 but for convenience we refer to this period as 2017/18 (Year 3).

Surveys were conducted at night by two or three people traversing the swamp and lagoon several times until ~0100 hours. Only the southern 50% of the lagoon was sampled each night. The swamp was surveyed on foot whereas, after the initial survey nights in 2015, the lagoon was traversed by kayak. That is, searching for bell frogs occurred from within the water with occasional traverses along the bank. Survey personnel systematically traversed and searched the fringing vegetation, and open water, using head-torches to detect frog eye-shine, which also led to the detection of frogs unaided by eye-shine. Male frogs were often, but not always, calling at the time of detection (Fig. 2).

Fig. 2.  A male green and golden bell frog detected on an algal mat away from the lagoon edge. Image: D. Newell.

Some frogs were captured directly from the kayak but in a majority of cases personnel disembarked from a kayak to effect a capture. Frogs were captured by hand with a single-use plastic bag inverted over the hand. The bag was pulled over the frog upon capture. The location was recorded by GPS so the frog could be returned later in the night to the point of capture and released.

Untagged individuals were subcutaneously implanted with a passive integrated transponder (PIT) tag and the entry point sealed with Vetbond adhesive. Each frog was weighed with a Pesola balance while inside the plastic bag and the bag weight subsequently subtracted. Snout–vent length (SVL) was measured by gently pressing the frog flat on top of a ruler that had been placed on a horizontal surface. Arching of a frog’s back may lead to inaccuracies in measurement. Photographs were taken of one thumb (nuptial pad if present), the flank and dorsum. Frogs were identified as adult male, adult female or subadult. Identifying males as adults is important for the purposes of population modelling because the number included will influence the value of the parameters estimated (Goldingay et al. 2017). Adult males were identified by the presence of a black nuptial pad on the top of the thumb on the front foot. The pad darkens as a male frog matures from a subadult and is black when breeding. The SVL at which males reach maturity is likely to vary among individuals and also geographic locations. Late in the breeding season the pad may lack dark pigmentation so SVL was also used to distinguish adult males at that time. The smallest individual with a dark nuptial pad in 2015/16 (Year 1) was 53 mm long (Goldingay et al. 2017) but in 2016/17 (Year 2) one measured 48 mm. Late season frogs that were <53 mm long and with no obvious pigmentation where the pad develops were classed as subadults. Five individuals that measured 54–56 mm in length with light pad pigmentation in November–December (mid-breeding season) were classed as subadult males. The mean size of individuals with blackened pads was 57 mm (range 48–64 mm) and weighed 13 g (range 6–19 g). Only 7% of males with black pads measured <53 mm in length. Previous studies of this species have used the SVL of the smallest male with nuptial pads to set a threshold for adulthood, as well as to distinguish males from females (i.e. those adults without nuptial pads) (Hamer and Mahony 2007: 42.8 mm; Pickett et al. 2014: 45 mm).

Individuals without any obvious nuptial pad and >53 mm long were classed as females. Only four were <58 mm long at first capture, with three of these seen across two breeding seasons, which enabled confirmation of sex. The average size at first capture of females was 67 mm (range 54–76 mm) and 22 g (range 8–39 g). The size at which females become reproductively mature is difficult to determine. Females may take two years to mature (Hamer and Mahony 2007; Pyke et al. 2008) so an unknown number classed as adult females may have been non-reproductive. No males reached 20 g in weight but 76% of females were >20 g, with the smallest of these 62 mm in length.

We constructed capture histories to investigate two separate aspects of the species’ population ecology at Yuraygir. The first was to estimate annual population size for the study area in each of three years to determine whether there was any evidence of population decline. The second was to produce monthly estimates of capture probability and population size, with an explicit intention to examine temporary emigration. Temporary emigration occurs when an individual captured in one session is not available for capture in the next session, either because it has moved out of the sample area or changes its behaviour and avoids capture. Temporary emigration may occur between years for long-lived amphibians (e.g. Muths et al. 2006). In our study we had the minimum number of sample years to investigate at an annual temporal scale and relatively few individuals were captured across years. Pickett et al. (2014) found that annual models that included temporary emigration had no support with 5–6 years of survey data of the green and golden bell frog. However, monthly surveys are an appropriate temporal scale in which to investigate temporary emigration in the green and golden bell frog, which is well documented to be quite mobile within a breeding season (see Goldingay and Newell 2005; Hamer et al. 2008). Furthermore, in complex habitat such as occurs at Station Creek it is likely that frogs may become unavailable for capture even though they remain close to the breeding habitat. If temporary emigration is not estimated then capture probabilities will be underestimated and population sizes overestimated (Muths et al. 2006). This is relevant to visual encounter surveys that aim to provide an index of abundance over relatively few survey nights (e.g. Bower et al. 2014), but also to single season estimates of survival probability and abundance in the green and golden bell frog (Hamer and Mahony 2007).

The Robust Design population model of Pollock (1982) has a specific structure where capture histories are divided among primary sample sessions which are subdivided into secondary sample sessions. This enables modelling to assume that a population is closed within primary sessions but open between primary sessions. This then allows temporary emigration to be estimated between primary sessions. We used the parameterisation of Huggins (1989) for the Robust Design model as implemented in program MARK (White and Burnham 1999). This model estimates apparent survival (s), initial capture (p), recapture (c), temporary emigration (γ or g) and population size is a derived estimate.

Our sampling consisted of three consecutive nights of sampling in different months over the breeding season. We had three sample months in 2015/16 (Year 1), four in 2016/17 (Year 2) and three in 2017/18 (Year 3). The male data collected in Year 1 have previously been published (Goldingay et al. 2017) but are included here to enable more detailed multiyear analyses. We conducted what we refer to as annual modelling to investigate annual population size. To do this we collapsed the nightly capture histories into one record per survey month so months became secondary sessions with years as primary sessions. Therefore, we assumed that the population was closed within a year but open across years. We modelled capture and recapture as year-specific for males and females but we also allowed it to vary by month for the subadults due to some convergence issues. The temporary emigration parameter (γ) was set to 0. Preliminary models that included γ showed a poor fit to the data. Instead, temporary emigration was investigated in monthly modelling where we used sample months as primary sessions and nights as secondary sessions. In this we constrained apparent survival (hereafter survival) to be able to vary among years but not across months within a given year, which was a plausible simplifying assumption and facilitated a focus on monthly variation in other parameters. We hypothesised that differences in behaviour between males and females (e.g. males aggregating and calling at the edge of waterbodies versus females visiting the edge of waterbodies to select a mate) would lead to higher estimates of temporary emigration in females compared with males.

In the annual modelling we specifically investigated whether there was a difference across years in survival and capture parameters. Rainfall during the main 5-month breeding period (October–February) averages 632 mm at the nearest location, Wooli (Bureau of Meteorology; www.bom.gov.au). Rainfall in this 5-month period during our study varied but was below average in each of the three years (Year 1: 456 mm; Year 2: 503 mm; Year 3: 511 mm). Inclusion of a rain covariate was analogous to a year effect. We tested models where parameters varied by year, were constant across years, or one year (e.g. 2015/16) differed from the other two years. In the monthly modelling we investigated the influence of rainfall for the 5-day period before a capture session on capture probability.

Temporary emigration is modelled with two parameters: γ′ represents the probability that a previously marked individual that was not recorded in the previous session remains unavailable for capture in the next, and γ″ represents the probability that a previously marked individual that was not available in the previous session becomes available for capture in the next. These parameters can be modelled as no emigration (i.e. equal to 0), constant over time (.), time-varying (t), or random (γ′ = γ″). The latter means the probability of being available for capture in the current session is the same for those available or unavailable for capture in the previous session.

We modelled males and females separately. Preliminary modelling suggested that the smaller sample size of females relative to males led to some important insights for the females being obscured if modelled together. We began with all parameters set to be constant across all occasions. We then modelled variations of the survival parameter. We then used the best model of that parameter and modelled the capture and recapture parameters. The best model with the three parameters was then used for modelling the two gamma parameters. If models did not converge, as indicated by the parameter estimates, they were deleted. Where estimates of individual parameters were similar across time periods we ran another model with those parameters set to be the same. We ran models again with variation in the survival parameter to ensure the correct survival model was selected. Models were ranked using AIC (Akaike Information Criterion) corrected for small sample size (AICc) with the top model having the lowest AICc. We relied on ΔAICc from the top model to compare models. If models differed by <2AICc we treated them as equally plausible. Models in which ΔAICc was >4 were considered much less plausible. We also used the AICc model weight to compare the top two models, with that for the top model divided by that of the second model to indicate the relative strength of the support for the top model. The top model was used to generate parameter estimates.

The number of frogs captured varied over the three years (Table 1), though there were more capture sessions in Year 2. The number of adult males captured was higher in Years 2 and 3 than in Year 1. A much higher number of females was captured in Year 2 than in Year 1 or 3. The number of subadults declined substantially over the three years, from 103 in Year 1 to 29 in Year 3. Among males, 27% from Year 1 were recaptured in Year 2, 5% from Year 1 were recaptured in Year 3, and 15% from Year 2 were recaptured in Year 3. Among females, 21% from Year 1 were recaptured in Year 2, and 8% from Year 1 were recaptured in Year 3, but only 2% from Year 2 were recaptured in Year 3. Overall, a higher percentage of male frogs was captured in a subsequent breeding season compared with that of females (24% versus 11%). Three males recorded as adults when first captured survived for at least two years. Of those in the unsexed subadult class, 13% were recaptured in a subsequent breeding season. Three subadults were detected two years after they were tagged. We recorded few (<10) juvenile (i.e. <40 mm) or metamorphling frogs in any of the three years.

Table 1.  Number of individuals in different age–sex classes across all surveys
This includes some captured on nights excluded from the analyses. Subadults includes males and females smaller than adult males in size (mostly <53 mm). Females includes subadults and adults mostly >58 mm long. The number of individuals recaptured in another year are shown in parentheses. Three males and two females from Year 1 were recaptured in Year 3

The top model for males estimated survival as constant across years (0.43 ± 0.19), but capture probability varied with year (Table 2). There was a high initial capture probability (p) (0.59 ± 0.08) in the first year, but a lower probability in Years 2 and 3 (0.10 ± 0.05). The recapture probability (c) was lower in Years 1 and 2 (0.22 ± 0.03) compared with Year 3 (0.33 ± 0.06). This model had 2.8 times more support than the next model (Table 2). The mean number of males in the population was estimated to be ~60 in year 1 but then >200 in Years 2 and 3. These estimates had wide confidence intervals but suggest the number in the population was >100 males in Years 2 and 3 (Fig. 3a).

Table 2.  The top four annual models for male, female and subadult bell frogs
Temporary emigration and immigration were set to 0 in all models so these parameters do not appear. S, survival; p, capture; c, recapture; (.), constant; yr, different in each year; t, time (i.e. session)-varying; =, equivalent; w, model weight; L, model likelihood; k, number of parameters

Fig. 3.  Estimates of the number of (a) males, (b) females and (c) subadults in the population each year. Values are means ± 95% confidence intervals. Note the different scale to the y-axis in (c).

The top model for the females estimated survival as constant across years (0.45 ± 0.23), but capture probability varied with year (Table 2). The capture probability was low and equal across years (0.06 ± 0.02), and equal to the recapture probability in Years 1 and 2, but different to recapture in Year 3 (0.35 ± 0.12). This model had 9 times more support than the next model (Table 2). The mean number of females in the population was estimated to be between 100 and 200. These estimates had wide confidence intervals but suggest the number in the population was >60 females in each year (Fig. 3b).

The top model for subadults estimated survival as constant across years (0.14 ± 0.04), with capture probability equal to recapture and varying across all nightly sessions, from a low in the first session in Year 1 (0.01 ± 0.01) to a high in the second session in Year 2 (0.38 ± 0.06) (all sessions in sequence by year: Year 1, 0.01 ± 0.01, 0.06 ± 0.03, 0.09 ± 0.05; Year 2, 0.26 ± 0.05, 0.38 ± 0.06, 0.09 ± 0.03, 0.12 ± 0.03; Year 3, 0.03 ± 0.02, 0.21 ± 0.06, 0.31 ± 0.08). This model had 24 times more support than the next model (Table 2). The model had low precision to estimate the number of subadults in Year 1 but high precision in Years 2 and 3, when 90–130 subadults were estimated (Fig. 3c).

The top model for the male data had a model weight of 1.00, indicating no support for the second model (Table 3). Survival was estimated as constant across years (0.72 ± 0.03). The probability of initial capture and recapture were time-varying but some sessions had equivalent estimates. Initial capture probability varied across nights from a low of 0.05 ± 0.03 in Session 8 to a high of 0.78 ± 0.05 in Sessions 6 and 9 (other numbered sessions: 1 and 7 = 0.38 ± 0.05; 2–5 = 0.20 ± 0.04; 10 = 0.61 ± 0.10). The probability of recapture varied across nights from a low of 0.01 ± 0.01 in Session 4 to a high of 0.38 ± 0.05 in Session 1 (other numbered sessions: 2 = 0.20 ± 0.04; 3 = 0.29 ± 0.12; 4–6 and 8–10 = 0.01 ± 0.01; 7 = 0.11 ± 0.06). The probability of temporary emigration was estimated by random movement (γ′ = γ″) as 0.54 ± 0.06. Modelling these parameters separately provided no improvement to model fit.

Table 3.  The top four models used to estimate parameters based on monthly primary sessions
The top model is also included with temporary emigration set to zero. Survival was modelled as year specific. S, survival; g′ (or γ′), temporary emigration and remains unavailable; g″ (or γ″), temporary emigration and becomes available; p, capture; c, recapture; (.), constant; yr, different in each year; t, time-varying; =, equivalent; rain, total rain in 5-day period before

The estimated abundance of males varied across months (Fig. 4a). There was a decline in abundance across the breeding season in all years. The monthly estimates were 30–45 individuals in Year 1 and 30–60 individuals in Year 2. In Year 3 it was ~50 individuals with one estimate >100 but this had a very large confidence interval. This estimate is for the session when the probability of capture and recapture were at their lowest. When temporary emigration was set to zero in the top model, there was a much poorer fit to the data (Table 3). The probability of survival was estimated to be 0.67 ± 0.03, and the mean of the monthly population estimates was 1.5 times higher.

Fig. 4.  Monthly estimates of the number of (a) males and (b) females in the population. Values are means ± 95% confidence intervals. Note that the y-axis scales are different and the CI for males in October 2017 extends to 359.

The top model for the female data had 2.9 times more support than the second model (Table 2). It estimated survival as constant across years (0.46 ± 0.29). The probability of initial capture (0.37 ± 0.08) was estimated as constant across all nights. The probability of recapture (0.06 ± 0.02) was much lower and was also estimated as constant across all nights. Temporary emigration was constant across sessions with a probability of becoming available for capture after being unavailable of 0.79 ± 0.07 and a probability of remaining unavailable for capture of 0.92 ± 0.05. The estimated abundance of females varied across months, from <5 in the first month up to 23 in the last month (Fig. 4b). Mean abundance was commonly ~15–20 individuals. When temporary emigration was set to zero in the top model, there was a much poorer fit to the data (Table 3). The probability of survival was estimated to be 0.21 ± 0.10, and the mean of the monthly population estimate was 4.5 times higher.

Amphibians are particularly vulnerable to extinction under current environmental conditions. Since 1980 Australia has lost at least four frog species (Woinarski et al. 2019) and as many as another 46 frog species are at risk of extinction (Hero et al. 2006; Gillespie et al. 2020). This requires management action and intervention on a range of fronts but also requires long-term monitoring to inform and guide management responses. Few published studies of Australian frog species span more than four years and those that do are often those documenting the local extinction of species (e.g. White and Pyke 1996; Richards and Alford 2005; Gillespie et al. 2015).

Our study of a green and golden bell frog population in Yuraygir National Park now spans 20 years, though only the beginning and end of this period have been covered in detail. Our surveys over three years at the end of this period suggest a stable population. We do not know what the population levels were in the intervening period when monitoring did not occur but it seems implausible that the population could have declined severely in that period and recovered without any intervention. We infer that this unmanaged population has remained stable over a 20-year period. This compares with several of the other key green and golden bell frog populations, which may not be stable and where management intervention has occurred. One precinct at Sydney Olympic Park that showed a severe decline in 2012 received a boost with the release of 11 500 captive-bred tadpoles (Bower et al. 2014). On Kooragang Island, Newcastle, 8200 captive-bred tadpoles were released into unfenced ponds and wetlands in 2011/12 to provide population support (Klop-Toker et al. 2016), and the south coast periurban population had eight ponds constructed in 2013 to offset potential effects of a nearby highway expansion (Hamer 2018).

An important motivation for estimating population size is that it can enable conservation managers to weigh up management priorities. Small populations may warrant management intervention or perhaps to be triaged out of consideration (McDonald-Madden et al. 2008). Large populations may allow a lower allocation of resources or allow other management actions to be devised (e.g. provide a source for translocations). Determining whether a population is large or small requires detailed investigation and comparison among other studies of the species.

Our annual population modelling suggests that our sampled area contained 175–425 adult frogs so the overall population was probably ~350–850 adults, due to our sampling of only half of the available breeding habitat. This indicates a larger population than previously suspected (see Goldingay and Newell 2005). This estimate suggests that the population is as least as big as that at Sydney Olympic Park across three precincts, where modelling suggested a population of 350–500 adults (Pickett et al. 2014). Hamer and Mahony (2007) estimated a male population on Kooragang Island at Newcastle of 1950 males using the Cormack–Jolly–Seber (CJS) method, which provides higher estimates compared with the Robust Design and a proportion of the number estimated included subadults (Goldingay et al. 2017). Furthermore, a within-season (i.e. year) CJS estimate that does not account for temporary emigration is also likely to overestimate the true number (see below). Therefore, it is difficult to compare across these studies but our modelling suggests that subadults would account for another 180–800 frogs across the whole lagoon per year at Station Creek. The range in the estimates reflects interannual variation in adult abundance, most likely due to variation in annual rainfall which will be a characteristic of this and other green and golden bell frog populations (e.g. Goldingay and Lewis 1999; Pickett et al. 2014). Rainfall was above average in the first year (2015) of this study but dropped to 80% of average in the second year. This may explain the large number of subadults in Year 1 that was followed by a much lower abundance in Year 2.

Population studies of the green and golden bell frog have documented sex ratios of captured animals to be heavily skewed towards males (Goldingay and Newell 2005; Hamer and Mahony 2007; Pickett et al. 2014). Modelling of detection has supported the hypothesis that the sex ratio is close to parity but females have much lower detection probability (Pickett et al. 2012), potentially due to higher detection of males due to their calling behaviour. We had sufficient captures of females to model them separately. Annual modelling confirmed that the capture probability of females was indeed low in the first two years (0.06) but high in Year 3 (0.35) compared with that of males (initial capture 0.10–0.59; recapture 0.22–0.33). This modelling suggested that the abundance of males and females was equivalent in Year 2 but unequal and reversed in the other two years. The confidence intervals for the female estimates were wide so any difference in abundance compared with male abundance could simply reflect different behaviour in relation to their availability for capture. The monthly modelling showed differences in the abundance of the sexes around the lagoon; there were frequently >40 males but <20 females. There was a large difference in rates of monthly temporary emigration between males (0.53) and females (0.79), although the probabilities of both were high. The high values likely reflect behaviour that makes individuals temporarily unavailable for capture, such as movement among different microhabitats, but in the case of females may reflect behaviour where they only move into zones where they could be captured when they were motivated to spawn. Therefore, we conclude that the evidence suggests that the sex ratio is likely to be close to parity.

One novel aspect of our study is that we were able to model individuals first captured as subadults. This suggested that annual apparent survival of this class was low (0.14) compared with annual apparent survival of males (0.43) and females (0.45). Capture probability of subadults varied between low (0.01) and high (0.38). The abundance estimates indicate that a much higher number of subadults was present in Year 1 compared with Years 2 and 3. The high number in Year 1 aligned with the increase in males and females in subsequent years. This high number suggests that conditions before Year 1 were particularly optimal for breeding and subsequent recruitment of juvenile frogs. Rainfall during the breeding period (October–February) averages 632 mm at Wooli, the nearest weather station. In the year before our study (2014) rainfall in this period was 73% above average. During our study it was 23–19% below average. Our study area contains several ephemeral ponds but none filled for breeding to occur during the three breeding seasons of our study. The coastal lagoon where our sampling was conducted contains seven native fish species, including three that may feed on tadpoles of the bell frog (see Pyke and White 2000). Consistent with breeding being low when confined to the lagoon, we observed very few metamorphlings during our surveys. However, the number of subadults captured in Years 2 and 3 indicate that successful breeding does still occur in the lagoon. The variation in juvenile recruitment appears unlikely to destabilise the wider population due to a high probability of survival (males: 0.43; females: 0.45) of adults to a second breeding period and juvenile recruitment still occurring in low rainfall years.

Temporary emigration has not been widely studied among amphibians. Muths et al. (2006) modelled temporary emigration in male boreal toads (Bufo boreas) across seven years and at three locations. They found that the random model described it best and at one location it varied across years from 0.04 to 0.95. Boreal toads are long-lived and may skip breeding in some years as indicated by these estimates. De Lisle and Grayson (2011) studied the Jefferson salamander (Ambystoma jeffersonianum) over four years and estimated temporary emigration (γ″) to be 0.25. These studies are different from ours in that we have investigated temporary migration within a breeding season rather than among seasons (i.e. years) because our species is not so long-lived. Furthermore, we were more interested in understanding temporary emigration from the viewpoint of animals being unavailable for capture (i.e. detection) rather than moving in and out of the wetland. When temporary emigration was set to zero in our top monthly models, the probability of survival was estimated to be lower and the monthly estimates of abundance were higher. This suggests caution is required in relying on other bell frog studies (Hamer and Mahony 2007; Heard et al. 2012) that have estimated survival probability and population size when temporary emigration has not been accounted for. Our sample area consisted of dense stands of emergent aquatic vegetation which are likely to allow animals to evade detection and capture. Temporary emigration was best modelled as random in males (0.54). It was non-random in females and, as predicted, was substantially higher (γ″ = 0.79, γ′ = 0.92). The higher values in females likely reflect differences in habitat use. For example, Valdez et al. (2016) found that females were observed significantly more than expected, and males significantly less than expected, on terrestrial vegetation away from a pond. An interesting comparison is that of the stream-breeding Fleay’s barred frog (Mixophyes fleayi), in which temporary emigration among months in males was random (i.e. γ′ = γ″) and estimated at the much lower value of 0.15 ± 0.05. Perhaps as a consequence, estimates of the probability of capture and recapture were high (0.4–0.95) (Newell et al. 2013; Quick et al. 2015).

We present evidence that the northern unmanaged population of the green and golden bell frog shows population stability. We do not advocate complacency that this will always be the case. A population of the introduced cane toad (Rhinella marina) that is subject to ongoing control occurs 50 km to the north of the bell frog population (Turnbull and MacKenzie 2019). Bell frogs are known to eat other frogs (Pyke and White 2001), so would be expected to occasionally consume toads. Bell frogs have disappeared from locations further north where toads occur (Lewis and Goldingay 1999). Yuraygir National Park has many recreational visitors each year so it is an ongoing concern that toads could be inadvertently translocated into the park. Management authorities need to maintain a watch and act in response to the toad population (e.g. Greenlees et al. 2018) while conducting periodic monitoring of the bell frog population. We previously recommended a frequency of monitoring every 3–5 years unless there was reason to suspect a decline in the population (Goldingay et al. 2017). The basic recommendation is that annual monitoring is not required. An important question is whether visual encounter surveys could replace mark–recapture surveys. Visual encounter surveys may require only 3–5 nights under suitable conditions as opposed to 9–12 nights for mark–recapture as employed in the present study. Bower et al. (2014) found that relative abundance based on visual encounter surveys was highly correlated (0.85) with population size estimated from mark–recapture surveys and based on half the effort. This should be investigated at Station Creek.

A major concern for other populations of the green and golden bell frog is the amphibian chytrid fungus (Stockwell et al. 2015a). This has been detected at low levels in bell frogs in our study population (D. Newell, unpubl. data). The salinity levels in the study lagoon have been recorded at 1.9 ppt and summer water temperatures average 23.8°C (D. Newell, unpubl. data), characteristics likely to reduce the severity of infections in summer (Rowley and Alford 2013; Stockwell et al. 2015b; Clulow et al. 2018). Stability of the bell frog population at Station Creek in Yuraygir National Park is likely to arise from the very large area of breeding habitat (see also Hamer and Mahony 2010), the saline conditions of the waterbody and the warm water temperatures that prevail in summer. These are attributes also identified as influential on the persistence of the related Litoria raniformis (Heard et al. 2015). These attributes may also enable persistence of the green and golden bell frog in East Gippsland, Victoria (Gillespie 1996).

The data that support this study will be shared upon reasonable request to the corresponding author.

The authors declare no conflicts of interest.

This project received funding in seasons 2016/17 and 2017/18 from the NSW Government as part of the Saving our Species conservation program.