Conservation implications of turtle declines in Australia’s Murray River system

Conservation requires rapid action to be effective, which is often difficult because of funding limitations, political constraints, and limited data. Turtles are among the world’s most endangered vertebrate taxa, with almost half of 356 species threatened with extinction. In Australia’s Murray River, nest predation by invasive foxes (Vulpes vulpes) was predicted to drive turtle declines in the 1980s. We assessed populations of the broad-shelled turtle (Chelodina expansa), eastern long-necked turtle (C. longicollis), and Murray River turtle (Emydura macquarii) in the Murray River and some of its associated waterways. Our results suggest that the predicted decline is occurring. All three species are rare in the lower Murray River region, and were undetected in many locations in South Australia. Moreover, E. macquarii had considerable population aging almost everywhere, possibly due to comprehensive nest destruction by foxes. Chelodina longicollis also had population aging at some sites. Sustained low recruitment has potential to lead to collapses as turtles age, which is particularly worrying because it was predicted over 30 years ago and may have already occurred in South Australia. Our results show that turtle declines were not mitigated since that prediction. If the crash continues, a vertebrate guild responsible for considerable nutrient cycling in the aquatic ecosystem will disappear. Our results highlight a worst-case outcome when species declines are predicted, but insufficiently mitigated.


Results
Throughout the study, we caught 36 C. expansa, 251 C. longicollis, and 815 E. macquarii. Recapture rates were low, and we recaptured only two C. expansa, eight C. longicollis, and 18 E. macquarii. However, we re-captured one E. macquarii four times at Lake Mulwala. In South Australia, we captured 8 E. macquarii that appeared to be marked from previous studies, but never re-captured these subsequently.
CPUE and numbers of turtles caught varied across the southern Murray River catchment within each species (Table 1, Fig. 1). Because we accounted for trapping effort in each of the following analyses, we focus only on CPUE in our results, from here forward. Despite species-specific statistical patterns in our results, CPUEs of all three species were generally highest in the mid-Murray region (River Distance ~1500-2000 km), which corresponds to north-central Victoria and south-central New South Wales (Fig. 1). In C. expansa, CPUE was low across all sites (Fig. 1), but significantly increased with increasing distance from the river mouth ( Fig. 2A; Table 1). CPUE of C. expansa did not differ with any other factor tested ( Table 1). The relationship between CPUE and river kilometre was not retained when we removed sites with low or high trap effort (Table S1).
Chelodina longicollis CPUE significantly increased with increasing distance from the Murray River, and these relationship varied across wetland types ( Fig. 2B; Table 1). Specifically, C. longicollis CPUE was higher in pond habitats than in all other types of habitats (Tukey P < 0.05; Fig. 2B). After removal of low-trap-effort sites the relationships between CPUE and distance from the Murray River were no longer significant (Table S1). However, there was still a significant difference in CPUE among wetland types (Table S1), and ponds had higher mean CPUE than did lakes, streams, and lagoons (Tukey P < 0.05), but not tributaries or backwaters (Tukey P > 0.05).
Emydura macquarii CPUE was generally higher than for the other two species (Fig. 1), and varied across wetland types and with river distance (Table 1). Specifically, CPUE increased with increasing distance from the river mouth (Fig. 2C). After accounting for river distance effects, mean E. macquarii CPUE was lower in ponds and streams than in all other habitats ( Fig. 2C; Tukey P < 0.05). Unlike the CPUE results for the other turtle species, the statistical results for E. macquarii were not affected by the removal of sites with low or high trap effort (Table S1).
Proportions of juvenile and large adult turtles varied among wetland types and species. We caught C. expansa in sufficient numbers (at least five) to be included in our analysis at only five locations (Fig. 1), so we did not analyze C. expansa size structures. Two of these sites (tributaries at Echuca and Wentworth) had no obvious bias to juveniles or large adults. Two sites (Reedy and Cockatoo lagoons) had juvenile proportions of 22.2% and 50.0%, respectively. Reedy Lagoon also had 11.11% large adult females, and a backwater at Shepparton had 16 Numbers of juvenile and large adult C. longicollis differed uniquely across wetland types, and both increased as the total number of turtles caught increased (Table 2; Fig. 3A). We caught significantly more large adults than juveniles in pond wetlands (Tukey P < 0.05), but there were no significant differences within any other wetland type (Fig. 3A). Juveniles were also rare in backwaters and streams, but the numbers of large adults in these wetland types varied widely across individual sites (Fig. S2). No other factors were significant (Table 2).
In E. macquarii, numbers of juvenile and large adult turtles also varied uniquely among wetland types, and both increased with the total number of turtles caught (Table 2). We caught significantly more large females than juveniles within both backwaters and lagoons (Tukey P < 0.05; Fig. 3B). We also caught more large females than large males within lagoons (Tukey P < 0.05; Fig. 3B). There were no within-habitat differences in the numbers of large male, large female, or juvenile turtles in lakes or tributaries (Fig. 3B), and no other factors were significant. Large females were common at almost all sites we trapped (Fig. S3), but backwaters and lagoons had very high proportions of large adult females (Fig. 3B). Juveniles were rare across all sites, but were relatively common in lakes (Fig. 3B). Large males were consistently distributed throughout all wetland types (Fig. 3B). Notably, locations and wetland types with high proportions of large females, as a consequence, had low proportions of males of all sizes, which may indicate a sex-ratio bias towards females, especially in backwaters and lagoons. We did not catch enough E. macquarii in stream or pond wetland types to include them in this analysis.

Discussion
Australia is now at the stage where the effects of invasive species and habitat destruction are being observed through species declines and extinction events. Across Australia, nearly half of all turtle species are listed as vulnerable or worse by state or federal agencies 49 . Turtle declines of up to 91% have been observed in sections of the Murray River 33 . A recent novel disease has also caused the near-extinction of an Australian freshwater turtle (Myuchelys georgesi), and may be a symptom of deteriorating water quality and climate change 28 . Here, our results provide further evidence that freshwater turtle populations in southern Australia are declining.
Catch-per-unit-effort data indicate that overall turtle abundance decreases as the Murray River approaches the ocean. Notably, we caught zero or very few turtles of any species at several locations in South Australia, including Calperum Station, Ramco Lagoon, and Currency Creek. Turtle numbers (relative to trapping effort) were higher in the mid-and upper-Murray regions. However, 10 of the sites where we caught zero turtles of any species were far upstream on the Murray River or its tributaries, and were much colder and faster flowing than other sites. These locations may not be suitable for turtles due to their ectothermic physiology. Even at the sites nearby where we did catch turtles, CPUE of E. macquarii was always low, possibly because our trapping exceeded their current range and extended beyond their preferred habitats 50 . By contrast, C. longicollis was occasionally very abundant in upstream, rocky stream sites, so it is unclear whether we extended beyond their preferred range or whether they are present but only at low densities at the sites where they were not captured.
Emydura macquarii are relatively abundant in some habitats in the mid-Murray region, but their numbers (relative to trapping effort) drop precipitously further downstream. In 8 of 10 sites in South Australia, we caught fewer than 10 E. macquarii, despite trapping efforts that were consistent with, or even exceeded, those further upstream. All of these downstream sites were near to other locations where we caught turtles, are well within the reported range of E. macquarii, and many probably supported large turtle populations in the past 48 . Across our study, E. macquarii was most abundant in backwater, lagoon, lake, and tributary habitats connected or close to the main channels. The habitat associations we detected largely follow the published preferences of the species 50,51 .
Chelodina longicollis exhibited higher mean CPUE in ponds disconnected from the main channel than in other sites. In addition, CPUE increased with increasing distance from the Murray River, but this relationship was not robust and disappeared when we removed sites with very low and high trapping effort from our statistical model. Low CPUE close to the river may indicate that the species has declined close to the river 33,44 . However, our results are also consistent with the species' reported ecology. Chelodina longicollis has a broad distribution across the region and is not as tied to permanent water as are other species of turtles 50 . It prefers temporary wetlands in much of the Murray catchment, and is also found in both temporary and permanent wetlands far from the larger rivers of the region 50 . It is capable of migrating long distances overland to access/emigrate from ponds as they fill or dry 52,53 . Thus, C. longicollis may still be present in relatively high numbers farther from the river. Whether the differences in CPUE that we detected are due to its habitat preferences or localized declines is unclear, but a long-term comparison does support the hypothesis that a decline may be occurring in the mid-Murray 33 . Chelodina expansa CPUE was low across the entire catchment, and increased with increasing distance from the Murray mouth. The latter trend was not statistically robust, but we caught only one individual in all trapping downstream of Mildura, VIC. Chelodina expansa was recently present in this region 34,36 , so our data may reflect Figure 2. Mean catch-per-unit-effort (CPUE; turtles per trap-hour) of (A) Chelodina expansa increased slightly with increasing distance from the Murray River mouth, regardless of wetland type, but this relationship was not retained in reduced models (Table S1). (B) CPUE of C. longicollis increased with increasing distance to the Murray River, and was higher in ponds than in other types of wetlands. The relationship with distance from the Murray River was not retained in a reduced model, but the difference between ponds and other wetlands was (Table S1). (C) CPUE of Emydura macquarii increased with distance from the Murray River mouth and also significantly higher in backwaters, lagoons, lakes and tributaries (BLLT) than in ponds and streams (PS). Lines on each graph represent best-fit regressions identified in log-linear ANCOVAs (Table 1). our comparatively low sampling effort. We also focused primarily on wetlands adjacent to the river, and did not trap in the river itself, where C. expansa may be more abundant 34,36 . We caution that C. expansa's historically low density also means that even proportionally large declines could be difficult to detect. Thus, although C. expansa may not have declined in the mid-Murray, it may have always been present in relatively low numbers. Sites where we caught the highest numbers of C. expansa again reflect previously-published habitat preferences 50,51 , including backwaters and lagoons with direct connections to the main channel of the Murray River.
The results of our analyses on the proportions of juveniles and large adults within turtle populations are striking, particularly for E. macquarii. Even in the habitat types where E. macquarii is most abundant, large adult females (>250 mm SCL), on average, account for the largest proportion of the population. Thus, most E. macquarii populations are strongly biased to large adult females. In contrast, juveniles made up very small proportions of most populations, which suggests that recruitment rates have been very low in recent years. Our observation of low juvenile numbers is similar to those reported in other demographic studies of E. macquarii 9,33 , and highlights that many populations of E. macquarii are aging. The trend is possibly a result of ongoing, consistently high rates of nest predation by foxes, which may nearly eliminate juvenile recruitment 15,22,54 . Population aging may also be caused by other factors that reduce recruitment but have not yet been investigated in turtles. The southern Murray catchment is heavily impacted by altered flow regimes 55 and invasive carp 56 . Resulting changes water turbidity and macrophyte abundance have been linked to changes in E. macquarii diet 57 , and we suspect that the loss of macrophytes might also increase predation of hatchling and juvenile turtles by fish and wading birds. Regardless of the cause, our results suggest that ongoing, consistent low recruitment may drive turtle declines in the Murray River, which could lead to a population collapse across the catchment 9,48 .
Several locations are clear exceptions and exhibit relatively high rates of juvenile recruitment. Lake Eppalock on the Campaspe River, Lake Meran, Serpentine Weir on the Loddon River, and Clayton Bay at Lake Alexandrina all exhibited relatively high numbers of juvenile E. macquarii. All but Lake Meran also exhibited low numbers of turtles caught, and relatively low CPUE. Lake Eppalock, Lake Meran, and Serpentine Weir experienced complete or near-complete drying during the Millennium Drought 58-61 . Near Clayton Bay, an infestation with the Australian tubeworm (Ficopomatus enigmaticus) was reported in the Lower Lakes region in 2008. This emergent condition in turtles is due to high water salinity, which occurred during the Millennium Drought 35 . The worms form calcareous tubes on a range of artificial and natural surfaces and restrict the ability of turtles to swim effectively. Either complete drying or tubeworm encrustation could have killed a number of adult turtles at these sites, which may account for our low CPUE. Interestingly, comparisons of turtle body size against published growth-curves 62 indicate that the juvenile turtles caught at these sites hatched since the end of the drought in 2010.
Assuming that fox predation on turtle nests is a major driver of the low numbers of juveniles we observed 15,22,54 , it is possible that the Millennium Drought caused such severe crashes in the adult E. macquarii populations at these locations that their nests are difficult for foxes to find (e.g. functional response 22 ). At very low densities, nests may experience some relief from fox predation as foxes switch to other prey 22 , and at least a limited recovery of the population via juvenile recruitment may be possible, until nest abundance is again high enough for foxes to target. It is also possible that the numbers of turtles we caught are too low to accurately reflect population size structures, particularly at Clayton Bay and Lake Eppalock. Our age structure analysis for C. longicollis did not detect strong geographic or climatic differences. Large adult turtles (>220 mm SCL) were much more common than juveniles within pond habitats, but we detected no other differences. Ponds had relatively large numbers of C. longicollis, so it is possible that high densities of turtle nests might attract higher levels of fox predation on eggs at these sites 22 , which would reduce recruitment. Alternatively, some of the ponds included in our sampling completely dry occasionally, and it is possible that hatchling and juvenile turtles are less able to avoid the impacts of wetland drying than are adults, via estivation or migration, though this has not previously been reported 52 . In contrast to E. macquarii, we caught relatively high numbers of juvenile C. longicollis at some lagoon, lake, and tributary habitats throughout the Murray catchment. In another study, C. longicollis juveniles were rare during the Millennium Drought (1996-2010) in the lower Murrumbidgee River, but have been caught with increasing frequency since 2013 51 . Together, these results suggest that C. longicollis populations experience unique patterns in juvenile recruitment, and are not as consistently at risk of population aging as are E. macquarii. The ability of C. longicollis to migrate long distances overland to exploit temporary wetland habitats 52,53 may allow a greater degree of terrestrial dispersal of turtles of all age groups than occurs in other species, which could buffer against population aging. The same migratory ability may also have allowed adult C. longicollis to persist in environments that experienced drying during the Millennium Drought, especially Serpentine Weir.
Together, our data provide evidence for regional declines in freshwater turtles in the Murray River catchment. All three species were detected at low numbers, relative to trapping effort, throughout most of South Australia, but E. macquarii populations appear to be at particular risk. Although relatively more abundant at many locations further upstream, E. macquarii may be at risk of a collapse even at many of these sites because the majority We present the data for both species as percentages for clarity, but we accounted for the number of turtles caught in each site using a log-linear ANCOVA on the raw numbers of each sex-size group, rather than on percentages. Error bars represent ± 1 SE, and letters indicate pairwise significant differences between size/sex number within each habitat type. Mean percentages do not sum to 100% because we did not include the percentages of intermediate-aged turtles in these analyses. of E. macquarii populations are heavily biased to older females, and juvenile turtles are uncommon. These size structures may be caused by low rates of recruitment. Low recruitment may be due to high rates of fox predation on nests, or to changes to the environments that hatchlings require once they reach the water. Incubating and hatching eggs in captivity to avoid foxes altogether would test the role of foxes in the decline, following Spencer and Thompson 15 . If recruitment improved after release of captive-incubated hatchlings at multiple sites along the Murray River, then foxes might be assumed to be the major cause of turtle losses in recruitment at a landscape scale. Alternatively, if released hatchlings were not recaptured and recruitment remained low even after hatchling release, then other factors, like macrophyte loss, food web disruption, or water quality issues, may be responsible for low recruitment rates. It is also possible that low CPUE is caused by both lack of recruitment and by high adult mortality. The causes of the decline thus require additional study. Despite turtle declines in South Australia being predicted 30 years ago, that prediction was not followed by extensive study or successful management of freshwater turtles. If further surveys show continued declines in turtles, a major vertebrate guild responsible for considerable nutrient and mineral cycling in the aquatic ecosystem 48,63 , could eventually become extinct. Our results suggest that reassessments of the conservation status and management plans of these species are needed. Emydura macquarii is listed as Vulnerable in both South Australia and Victoria, and Chelodina expansa is listed as Vulnerable in South Australia and Endangered in Victoria, but none of these species are listed in New South Wales. Furthermore, none are listed at the federal level. By contrast, C. longicollis is widespread away from the major waterways of the Murray River system, and further assessments are needed to examine its status away from the catchment, where our study was focused. Finally, it is worth noting that, despite the many threats they face, freshwater turtles are sometimes considered resilient taxa because many species are able to survive in poor quality habitats where other taxa cannot [64][65][66][67][68] .
Long-lived species like turtles are often neglected as at-risk species in need of conservation action, due to the "perception of persistence" 63 . Even in species with relatively high abundance, low rates of recruitment may be catastrophic once the majority of individuals present die of old age 63 . Often it is only when populations crash or a crash is imminent that they become listed for conservation action, yet the life history of long-lived species means that they give us considerable warning. In turtles from the Murray River, this population crash was predicted almost 30 years ago, and our data indicate that it is in progress.

Methods
Field sampling. We assessed turtle populations at 52 sites throughout the southern Murray River catchment in January-April of 2015-2017 (Fig. 4). January-April in the Murray region coincides with consistently warm and dry weather and high turtle activity 50 . It is after the spring nesting season for C. longicollis and E. macquarii, but overlaps with the autumn nesting season of C. expansa. All upper Murray sites (river km 1730-2600) were assessed in Jan, 2016; mid-Murray sites (river km 1125-1730) in Feb-Mar, 2015 and 2016; and lower Murray sites (river km 19-570) in Jan-early Mar, 2017. Only 2 sites were assessed in early April, 2016: Wentworth and Euston, but maximum air temperatures at these sites exceeded 30 °C and turtles were active. Most, but not all, of our sites were chosen because they were on opposite sides of a dam, weir, or impoundment, to contribute data to a concurrent study on genetic connectivity among E. macquarii populations (Berman et al. unpubl. data), but were not near the dam or in a weir pool itself. Notably, we did not survey much of the upper Murray ( Fig. 4; near Albury, NSW) because of pre-existing data of the concurrent genetic study, which used a different survey method (Berman et al. unpubl. data). We included a range of wetland types in our assessment, and included sites associated with the Murray River itself, tributaries to the Murray River, and anabranches in the Murray River catchment. We defined wetland types by their physical characteristics: backwaters are unrestricted embayments broadly connected to the main channel of a river or lake; lagoons are embayments that are narrowly connected to the main channel of another, larger water body; lakes are sites trapped near the open shoreline of a large natural lake or large reservoir/impoundment; ponds are small wetlands disconnected from any adjacent waterway; streams are small fast-flowing upland waterways with rocky bottoms; and tributaries are large slow-flowing waterways with sand or mud bottoms.
Our trapping schedule evolved over time. To meet the requirements of our concurrent genetic study (Berman et al. unpubl. data), we initially aimed to trap at each site until at least 30 E. macquarii were caught. In our initial trapping in 2015, we were sometimes able to catch 30 E. macquarii using only 3-4 traps at a time overnight, which resulted in high CPUE relative to low trapping effort at Burramine and Ovens (Fig. 1). As our trapping progressed to other locations, our catch rates decreased. Thus, to balance trapping effort with trapping at as many sites as possible, at the majority of sites we used at least 10 traps at a time, set for at least three overnight periods or until we caught at least 25 E. macquarii. We switched to 25 minimum E. macquarii rather than 30 because in some cases we caught between 25 and 30 E. macquarii within 2 nights, and we decided it was more productive to move traps to another site in our survey to ensure we trapped all sites before the active season ended.
There were several exceptions to this rule due to extraneous circumstances. Landowners of Raukkan-Narrung, Banrock Station, Alexandra, Malmsbury, and King River were only able to allow us access for 1-2 nights, and there were limited sites suitable for setting our traps due to high currents and shallow water at Malmsbury and King River. Our sites in Benalla, Tullaroop, and Cairn Curran had anglers present after 1-2 nights, so we removed traps to prevent them from being stolen, which occurred several times early in our study. Finally, we had a side project at Lake Bonney that allowed us to trap there for 2 weeks (Cameron et al., unpublished data). In overnight trap sessions, we typically set traps in the evening of the first day, and removed traps on the morning of the final day (overall mean trap effort: 576 ± 58 trap-hours, Fig. S1). We acknowledge that our approach is probably too short to determine whether turtles are absent from a site, and do not consider an absence of detection as an absence of turtles. Instead, our aim was to determine a relative index of turtle population characteristics that could be compared across the catchment.
We used three different turtle trap types: cathedral traps, single-wing fyke traps, and crab pots. We used a variety of traps because each trap type is not equally suited to use in every type of habitat. Cathedral traps are best suited for still water >1.4 m deep, while fyke traps are best suited for still or slow-moving water <1 m deep, and crab pots are best suited for very shallow, slow-moving water <0.2 m deep. At most sites, we used at least 10 traps, and usually mixed trap types (3-4 each), while trapping a site (maximum 25). We set cathedral traps in open, still water deeper than 1 m. We set fyke traps in open water 0.2-0.5 m deep, with the wing of the fyke parallel to the nearest shoreline. We set crab pots in water ~0.2 m deep, sometimes nestled within benthic vegetation. Despite their differences, each trap type is capable of catching at least 10 turtles at a time. When trapping in fast-flowing streams, we always placed traps in or on the edge of deeper pools, where currents slowed as much as possible. All traps were baited with a mix of offal (lamb heart, lamb liver, chicken liver). Traps were checked twice per day and were re-baited once per day.
Upon checking traps, we removed any turtles present and immediately re-set the trap, and re-baited it if necessary. After removing turtles from traps, we checked them for existing carapace marks to determine whether they were recaptures, from our own or previous studies 69 . We recorded the identifications of turtles recaptured within the trapping session and immediately released them. Occasionally, we caught turtles that had been marked by a previous study, potentially up to 40 years earlier e.g., Chessman; Thompson 9,33 . We recorded these turtles' existing IDs, but considered them novel to our current analyses. We weighed all turtles to the nearest 1 g on a digital scale. We measured straight carapace length and width with forestry callipers. We marked all unidentified turtles individually by notching their marginal scutes 69 . We recorded the sex of each turtle where possible. The tails of adult and subadult C. expansa and E. macquarii are sexually dimorphic and longer in males. The tails of adult C. longicollis are only slightly sexually dimorphic and other diagnostics are not universally reliable, so we did not record sexes of C. longicollis 47  Data Analysis. We analyzed two metrics of turtle population status at each site: catch-per-unit-effort (CPUE) and the proportions of juvenile and large adult turtles of each species. Both metrics were analyzed separately for each species. We used CPUE as a relative index of local abundance to assess the number of turtles present in a site whilst accounting for trapping effort produced by our design. Because we caught at least 25 E. macquarii at some sites very rapidly, our trapping effort at those sites was relatively small, whereas our trapping effort was larger at sites where we caught turtles at a slower rate. We compared CPUE of each species across sites using log-linear Analyses of Covariance with Poisson distributions (PROC GLIMMIX) in SAS University Edition (SAS Institute, Cary, NC). We accounted for trapping effort by multiplying the number of traps set by the number of hours they were set, disregarding trap type, and including this value as a covariate. We specified wetland type as an independent variable to test for CPUE differences among wetland types, and approximate river distance on the Murray River (distance in km from the river mouth, hereafter referred to as River Distance) as a covariate to test for geographic differences across the Murray Catchment. In this design, multiple sites sampled from the same tributary (e.g., Loddon River) were assigned the same river distance, because the confluence between the tributary and the Murray River was the same for all. We also included distance from the nearest permanent water body (km), distance from the Murray River, mean annual precipitation (mm) and mean air temperature (°C) to test for environmental and climatic effects on turtle CPUE across the catchment. Distance data were extracted from Google Earth TM . Climate data were extracted from Worldclim (Ver. 2, www.worldclim.org) 70  within 30 seconds of latitude/longitude spatial resolution, from 1970-2000. We log-transformed all numerical data in this analysis. Because of the wide variation in trapping effort across sites (Fig. S1), we tested for robustness in our results by running an identical model without sites that had fewer than 100 total trap-hours for any reason, and also without Lake Bonney, because it was an outlier with very high trapping effort. For presentation of the data, we also calculated CPUE by summing the number of turtles caught by all traps at a site and dividing by the number of trap-hours (number of traps multiplied by number of hours traps were set). We calculated CPUE separately for each species, and did not distinguish among sex, body size, or trap type. We used the numbers of juveniles and large adults, relative to total number caught, as metrics of recent recruitment and population aging, because both drought and fox destruction of nests have been suggested to drive losses of recruitment, which may result in population collapses if adults are not replaced as they die of old age 9,33,48 . Furthermore, size-structure bias toward smaller (younger) or larger (older) turtles may suggest different stressors affecting population dynamics, e.g. low recruitment or population bottlenecks. At any site where we caught at least five turtles of a given species, we counted the numbers of juveniles of each species, using straight-carapace length (SCL), following Chessman 33,47 . The maximum cutoff for juvenile SCL was 220 mm for C. expansa, 180 mm for C. longicollis, and 190 mm for E. macquarii. We also counted the number of large adults of each species using arbitrary cutoffs: 350 mm for C. expansa, 230 mm for C. longicollis, and 250 mm for E. macquarii.
Within each site, we tested for differences in the relative numbers of juveniles and large adults using log-linear Analyses of Covariance with Poisson distributions (PROC GLIMMIX) for each species. We included the total number of turtles caught at each site as a covariate to account for differences in the numbers of turtles caught. We included "sex" (juvenile, large female, large male; or juvenile, large adult in C. longicollis) to test for differences in the numbers of each group. As in the CPUE analysis, we included river distance, distance from the Murray River, distance from the nearest permanent water, wetland type, mean annual temperature, and mean annual precipitation to assess geographic and climate relationships in the numbers of juveniles and large adults. We also included the interaction sex*wetland type to test whether numbers of juveniles and large adults differed uniquely within each wetland type. Finally, we included site as a random effect to account for repeated within-site comparisons across sexes. We also calculated percentages of juveniles, large females, and large males (or large adults for C. longicollis) at each site in order to present the data clearly. In all analyses, we determined statistical significance at 0.05 type I error level, and we present data as means ± SE.

Data Availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.