Estimation of turtle catch per unit effort
We conducted our field experiment in February–April 2018 at two wetland complexes near Murray Bridge, South Australia, selecting two study sites at each complex (Supplementary Fig. S3). At each site, we estimated turtle population density using catch-per-unit-effort (CPUE; Supplementary Table S1). We conducted three 3-day rounds of turtle trapping using a combination of fyke and cathedral traps, baited with offal. Up to eight traps were deployed at a time. We calculated turtle CPUE by dividing the total number of turtles caught (regardless of species) by the total trap-hours. The number of trap-hours was similar across all four sites (average 1685 ± 7.6 SE total trap-hours).
Carp carcass decomposition
After the first and the second trapping rounds, we deployed whole carp carcasses at each site to measure carp decomposition rates depending on turtle accessibility. We placed each carp in a pre-weighed plastic box (340 × 230 × 120 mm), securing it with cable ties. Carp were made non-accessible to turtles in half of the deployments by covering the plastic boxes with 25 × 25 mm mesh (Supplementary Fig. S4). The mesh prevented turtle access to the carp, but was large enough to allow scavenging by crayfish (Cherax destructor) and other freshwater invertebrates. We tied each box to a brick and submerged the boxes around the four study sites ≥ 30 m away from each other, sunk at an average depth of 436 mm (± 13 SE). We used a total of 38 accessible and 40 non-accessible carp, split between our four study sites over two rounds (Supplementary Table S9). Every day, starting from day 2, the box and carp were weighed together with a digital scale. In all measurements, we calculated the wet mass of the carp by subtracting the box weight from the total weight. Carp carcasses were left in the wetlands for up to 10 days, or until they were fully consumed. All work was performed in accordance with DEWNR Permit M26663-1, PIRSA permits MP0085 and ME9902980, and The University of Sydney Animal Ethics Committee approval (project number 2017/1208), observing all relevant guidelines and regulations.
We analysed our data using RStudio 1.1.45633 (packages: “lme4” 1.1-2134, “MuMIn” 1.42.135). To assess whether turtles were important scavengers of our carcasses, we computed a linear mixed model testing whether turtle CPUE and carp access (yes/no) affected the rate of mass loss of the carp carcasses. The turtle CPUE values used were the average CPUE in the trapping round before and after each carp was deployed. We used the rate of mass loss per day as a dependent variable. We included the carp mass before deployment as an independent variable to account for initial mass variation, and we included study site as a random variable. We log-transformed all data before analysis. We assessed model fit by examining predicted versus residual and Q–Q plots, and testing the normality of residuals.
Turtle trapping and experiment procedure
We caught 20 adult male E. macquarii with fyke nets baited with offal at Hawkesview Lagoon, Albury, NSW, in November 2018. The E. macquarii captured at this site belong to the same genetic population as the E. macquarii trapped in South Australia36, therefore we expect behaviours to be similar between the two populations. We focussed on E. macquarii as this is the most common species in the Murray–Darling Basin, and fish carrion is an important part of its diet19,37. The turtles were transported by car to the Experimental Wetlands facility at Western Sydney University, in Richmond, NSW (Supplementary Fig. S5). This facility is comprised of 10 circular mesocosms (0.42 m depth × 2.1 m diameter) filled with 1,450 L of tap water. Each is an independent flow-through system where the water flow is regulated, and was maintained at 1998.6 ml/min (± 149.5 SE) throughout the experiment. Each mesocosm had two cement blocks for the turtles to bask on, and two plastic tunnels for shelter. The experiment was conducted for 40 days, therefore it is a short-term study (Supplementary Fig. S6). Upon arrival at the facility, we placed four adult male E. macquarii turtles in each of five random mesocosms, which means the experimental replication was 5. The remaining five mesocosms were controls and had no turtles. The four turtles comprised an average 5,376.6 g total biomass per pond, each being 3.46 m2. This would result in a biomass of 11,560 individuals/ha or 15,537 kg/ha on average. Kinosternon integrum has been estimated reaching densities of 20,000 individuals/ha in Sonora, Mexico, while Podocnemis vogli may reach 10,300 individuals/ha or 15,450 kg/ha in Venezuela, likely in temporary aggregations38. Emydura macquarii tend to congregate around food sources, therefore we considered four turtles per carp carrion as a realistic density. After 7 days of acclimation, we introduced one carp carcass to all mesocosms, and a second 6 days later. We used one ~ 1 kg carp at a time to simulate a density close to 3,144 kg carp/ha31. The turtles had continual access to the carp, which was their main food source throughout the experiment. The day all carp carcasses were fully eaten in all turtle mesocosms, we removed turtles from their mesocosms and released them at the point of capture. On the same day (day 10), we ended the data collection in their mesocosms, because any further change in water quality here would not have been related to carp decomposition. We continued the daily water quality measurements in the five control mesocosms until all carp were fully decomposed (day 32). This experimental design allowed us to collect water quality data without the need to add turtle food to the mesocosms, which would have biased our measurements once carp were removed from the turtle mesocosms.
We measured water temperature, dissolved oxygen, conductivity, turbidity, phosphate, and ammonia concentration in all mesocosms every morning from the day before the first carp introduction (see Supplementary Materials for equipment used). We also photographed the carcasses daily to estimate their decomposition rate based on a scale (Supplementary Table S10) designed after the decomposition stages described by Benninger et al.39 Due to the short transit time of fish matter in E. macquarii’s gut37, the effects of the turtles’ metabolic wastes on water quality are included in our experiment for carp 1. All work was performed in accordance with OEH Permit SL100401, DPI permit P09/0070-3.0, and Western Sydney University Animal Ethics Committee Animal Research Authority approval A12390, observing all relevant guidelines and regulations.
To assess whether the presence of turtles affected the decomposition of carp we computed a mixed linear model using the repeated measures PROC MIXED procedure using SAS (3.8 University Edition, SAS Institute Inc., Cary, NC, USA). For this model, days to total decomposition/removal was a dependent variable, turtle presence/absence was a fixed effect, and carp number (first or second) was a repeated fixed effect.
We used DO, conductivity, turbidity, phosphate, and ammonia to carry out a principal component analysis (PCA) using PROC PRINCOMP. We conducted a PCA because the parameters are a multivariate response and have potential to covary with each other, which would not be detected in univariate analyses. We considered a parameter loaded onto a PC when the absolute value of its eigenvector was > 0.300. If the same parameter loaded onto more than one PC, we considered it only on the PC where its eigenvector had a higher absolute value.
To test the effect of turtles scavenging on water quality, we computed general linear mixed models (GLMMs), using PCs as response variables, in PROC MIXED. We used a PC as response variable if its eigenvalue was greater than one (Kaiser criterion40). For each of these GLMMs, we included turtle presence (yes/no) and day number after the first carp introduction as fixed effects, water temperature and flow as covariates, and mesocosm ID as a random effect. We computed a model with full interactions first, and then, in absence of four- or three-way interactions, simplified the model to focus on main effects.
Finally, to test the effect of turtle scavenging on each water quality parameter, a GLMM was computed for each (logged) parameter that loaded onto a PC with eigenvalue > 1, i.e. dissolved oxygen, ammonia, turbidity, conductivity, phosphate. For these GLMMs, turtle presence and day number were fixed effects, water temperature and flow were covariates, and mesocosm ID was a random effect.