Murray cod is a long-lived (maximum 48 years) apex predator that matures at around five years of age (Anderson et al., 1992; Koehn, 2009). Mature fish display limited movements and high site fidelity for much of the year, but do migrate during spawning in late winter/early spring (cued by either temperature or day length) (Koehn et al., 2009; Tonkin et al., 2022). Movements can also be associated with shifts in home range for recovery and dispersal (e.g. Thiem et al., 2017; Lyon et al., 2019; Tonkin et al., 2022).

Murray cod forms breeding pairs and lays adhesive eggs on hard substrata in spring, at temperatures greater than 15°C (Humphries, 2005; Koehn and Harrington, 2005; Koehn and Harrington, 2006), with the eggs then being guarded by the male fish (Rowland, 1998). The survival of fertilized eggs declines by 50% if water temperatures fall below 14.8°C, with 100% mortality occurring at 13°Cm (Todd et al., 2005). Larvae hatch after approximately 5−13 days, then disperse either actively or via downstream drift (Humphries, 2005; Koehn and Harrington, 2005; Koehn and Harrington, 2006) before settlement in the main river channel (Koehn and Harrington, 2006).

This study focussed on a reach of the lower Mitta Mitta River, a highly regulated floodplain river of south-eastern Australia ( Figure 1 ) and typifies the many regulated reaches across the MDB. The study reach is approximately 48 km in length and makes up almost half the length of the regulated lower reach with a flow regime largely governed by releases from a large reservoir, Lake Dartmouth (36°30’S, 147°30’E). The Lake has a depth of 161 m and capacity of 3 856 GL and is the largest water storage in south-eastern Australia’s MDB. Construction was completed in 1979, with the purpose of augmenting irrigation water supplies to the Murray River system and providing electricity generation. Water is released from two outlets, 62 m and 121 m below Lake Dartmouth’s full supply level. The water within the lake stratifies, with a thermocline generally occurring at 5 to 10 m below the surface from October to January, and up to 40 m in August to September (Ebsary, 1990). As a result, water releases from Lake Dartmouth reduce water temperatures in the reach of the Mitta Mitta River between the dam and Lake Hume (approximately 100 km) during spring and summer by up to 12°C below natural levels (Ryan et al., 2003; Todd et al., 2005).

Environmental studies demonstrated substantial changes in the aquatic fauna in the lower Mitta Mitta River post construction/operation of the dam, including a shift from a ‘warm water’ native fish population to one dominated by introduced ‘cold water’ species, particularly brown trout Salmo trutta (Koehn et al., 1995). Of particular note was the apparent disappearance by 1992, of self-sustaining populations of Murray cod, Trout cod (M. macquariensis) and Macquarie perch (Macquaria australasica) (Koehn et al., 1995). Nevertheless, recreational fisher reports and fish surveys conducted since 2010 recorded an increasing number of Murray cod in the lower Mitta Mitta River (Lieschke et al., 2013). This has prompted waterway managers to reconsider their requirements for managing this river reach for improved ecological and recreational benefits while still achieving irrigation transfers and hydroelectric generation.

Existing infrastructure (only two water outlets) allows only a narrow window of opportunity to release water with no or minimal reductions from natural temperatures, occurring just three times in the past decade (Murray-Darling Basin Authority, unpublished data). The study reach receives inflows from several smaller unregulated tributaries but these have very little influence on downstream water temperatures during sustained periods of above-minimum water releases from Lake Dartmouth. From 2006 – 2018, the reach has also been stocked with 10 000 – 40 000 hatchery-produced Murray cod fingerlings each year (January and February liberation).

The first three years of the study (May 2013–May 2016) coincided with prolonged periods of high spring flows in the Mitta Mitta River. During this period, water was released from below the thermocline meaning that temperatures in the Mitta Mitta River (recorded at Tallandoon) during the spring spawning period for Murray cod (October and November) only occasionally exceeded the 50%-mortality threshold temperature for larvae of 14.8°C (Todd et al., 2005) and frequently dropped below the 100%-mortality threshold of 13°C ( Figure 2 ). Discharge in early 2016 was also high throughout January, with a gradual reduction to < 1 000 ML day^-1 in March ( Figure 2 ). Water temperatures in early 2016 were much higher than the previous two summers, even during high-flow periods, increasing in early January from 15°C to a high of 23.4°C in early March.

River flow and water temperatures in the lower Mitta Mitta River during the final two years of the study (May 2016–May 2017) differed greatly from those in the previous three years ( Figure 2 ). Discharge at Tallandoon in early October 2016 reached the highest levels during the study (14,765 ML day^-1), due largely to high inflows of unregulated tributaries, but then declined by the end of November, with minimal contribution from Lake Dartmouth releases. These flows resulted in water temperatures during the Murray cod spawning period above 14.8°C for more than 50% of the time, peaking at 21.6°C and 23.4°C in 2016 and 2017, respectively ( Figure 2 ). Summer and autumn water temperatures in these years generally ranged between 20–25°C. There were, however, drops in water temperatures with minimums as low as 12.8°C and 14.3°C during November 2016 and 2017 respectively, and 12.9°C and 12.8°C in February 2017 and March 2018, respectively.

Finally, following the results and management recommendations generated during the core data collection period from 2013-2018, discussions with river operators indicated potential for minimal water transfers through the study reach during both spring and summer in 2020/21. This enabled discharge during the core spawning period of Murray cod in spring 2020 to remain low, driven primarily by unregulated tributary inflows. As such, water temperature only briefly dipped below 14.8°C in mid-October, remaining above this survival threshold for the remainder of the spawning period ( Supplementary Material 1 ). Summer discharge also remained low, with water temperatures from summer through to early autumn remaining above 19°C (and usually >21°C).

We used a range of techniques and multiple independent lines of evidence to assess impacts of CWP on key population processes of movement, natal origin, age structure, and stocked fish survival. We then used our leanings to help refine a population model developed for Murray cod (see the sections below) to assess population dynamics and also explore the potential for recovery and persistence of the species using mitigation measures such as flow management and stocking.

Under low flow conditions, the study reach is on average approximately 20m wide, very low turbidity (mostly <10 NTU) and rarely exceeds 3-4m in maximum depth. As such, data generated using standardised boat electrofishing surveys were used for determining population demography. The method was used from 2011–2021, specifically to capture fish for implanting with telemetry tags (see Migration section below), collect juvenile fish for natal origin (see Natal origin section below) and generate data to assess population structure and subsequent growth, survival and year-class strength (YCS). Between five and 40 sites were surveyed in 2011 (five sites; Lieschke et al., 2013), 2013 (seven sites), 2015, 2017, 2018 and 2021 (40 sites each year). In each of these years, surveys were conducted between March and May, a period specifically selected to coincide with periods of low river levels (i.e. periods of minimum irrigation transfers) and minimise the effect of changing river levels on electrofishing efficiency.

Electrofishing at each site consisted of single-pass, boat-mounted electrofishing (Smith Root^® Model 5 KVa generator-powered pulsator), conducted in an upstream direction. This technique has been used extensively across the MDB to assess fish populations, including a basin wide long term monitoring program (Sustainable Rivers Audit; Lieschke et al., 2013). Sites were approximately 100 m long, usually along an outside bend of the river (noting the adjacent bank was usually too shallow to navigate and indeed, hold fish). All wetted accessible habitats were surveyed between the pre-defined downstream start and upstream site end points. Parameters recorded at each site included: electrofishing effort (total seconds fished), distance fished and site depth (average and maximum). All fish captured were identified and counted. For Murray cod, their total length (mm) and weight (nearest gram) were recorded. All Murray cod greater than 200 mm total length were fitted with an external dart or T-Bar tag to provide addition information on estimates of growth, movement, and mortality (from recreational fishing) via catch-mark-recapture data. All fish were released at the site of capture after being processed.

The size structure of fish populations was first explored using size-frequency histograms generated from Murray cod length data collected during each electrofishing survey of the Mitta Mitta River from 2011–2018. This exploration revealed a bimodal distribution of fish in some years. Therefore, to assess temporal changes in the length distributions, we used quantile regressions using the rq() function in the quantreg package (Koenker, 2017) in the statistical program R (R Core Team 2021). First, we assessed whether median lengths changed linearly over time by running a quantile regression that included sample year as the interaction, and then whether the lower 20th percentile of fish lengths (juveniles) changed among years.

We used acoustic telemetry to investigate our hypotheses regarding the effects of CWP on emigration and spawning migration behaviour ( Table 1 ). In May 2013, ten acoustic receiver stations (Model VR2W; Vemco, Nova Scotia, Canada) were installed along the study reach. This was immediately followed by the 2013 electrofishing surveys, during which 25 adult Murray cod (486–1055 mm TL ( Supplementary Material 2 ) were implanted with Vemco Model V16-4x, 69-kHz acoustic transmitters (estimated battery life of 2356 days). While fish were distributed throughout the study reach, a greater proportion of fish was tagged in the most downstream sections ( Supplementary Material 3 ). Details of surgical procedures are described in (Tonkin et al., 2022). Acoustic receivers were visited annually between March and May from 2014–2018, for battery replacement and data download.

We analysed movement as changes in location (defined as the reach between listening stations) and direction. A change in location was defined as a fish being detected at a station not associated with its previous location, with the direction a fish had moved (upstream or downstream) also determined (change relative to its previous position). We then investigated how flow, temperature and time of year influenced these movements, especially the effect of CWP on emigration and spawning migration behaviour. We hypothesised that the influence of these factors would be greater prior to June 2016 (characterised by unseasonal high discharge and low water temperatures during spring and summer) and less in the subsequent years that had low discharge and more seasonal water temperatures during spring and summer.

For these two periods, the effects of flow, temperature and time of year on the likelihood of detection, movement and direction of movement were analysed using general additive mixed models (GAMM). In particular, we considered fixed effects related to flow and temperature, their variability, and change from the previous measure. The time of year was considered in two separate ways, either as a smoothed time-of-year or as a factor with levels of spawning, return from spawning and all other times.

The direction of fish movement, upstream or downstream, was first analysed by determining the independence of direction of movement and the time-of-year as a factor using a χ² test. Then, the effect of two periods, discharge, temperature and time of year (as fixed effects) and individual fish (as a random effect) on the likelihood of a movement being upstream (as opposed to downstream) was analysed using a GAMM. The effect of time of year was done using a cyclic smoother over the days of the year for each period separately. The analysis was conducted using the statistical program R 3.4.4 (R Core Team 2021) and the package gamm4 (Wood and Scheipl, 2014). Final model selection was achieved using Akaike information criteria (AICc), corrected for small samples to determine the model with the most evidence (Burnham and Anderson, 2010).

Distinguishing fish as either stocked or naturally spawned was integral to the objectives of this project and we used two methods to determine this. Firstly, all Murray cod stocked between 2013 and 2015 (inclusive) were chemically marked (calcein) allowing their identification as hatchery-produced fish. Detection of these marked fish required otoliths to be removed, thin-sectioned and viewed microscopically for the presence of the calcein mark using fluorescence backscatter (e.g. Crook et al., 2016). The second method used the fish’s otolith chemical composition to identify its birthplace. Specifically, we used laser ablation inductively coupled plasma mass spectrometry (LA-ICPMS, strontium isotope ratios) to discriminate between recently stocked and naturally spawned fish based on otolith strontium isotope ratios (87Sr/86Sr) as this partially reflects the chemistry of ambient water at the time of deposition (Campana and Thorrold, 2001). In rivers and streams, dissolved 87Sr/86Sr are an artefact of catchment geology and are increasingly being used to discern the natal habitats and movements of numerous diadromous and freshwater fishes in the southern MDB (e.g. Thiem et al., 2022; Zampatti et al., 2022).

In autumn 2017, 2018 and 2021, otoliths were collected from 35 juvenile (<150 mm TL, age <2 years) Murray cod in the Mitta Mitta River. We then sampled the 87Sr/86Sr at the core and edge of otolith transverse sections which represent the fish’s place of birth and capture, respectively. We also compared 87Sr/86Sr of the samples from the Mitta Mitta River with the otolith core values generated from 47 Murray cod fingerlings sourced from each of the hatcheries used to stock the Mitta Mitta River during from 2016–2021.

Sagittal otoliths from juvenile fish were removed and prepared for ageing and LA-ICPMS using transverse sections. The age (and therefore year class) of each fish was determined by macroscopic age analysis of each otolith section as described in (Anderson et al., 1992). We then used microchemistry LA-ICPMS analysis of 87Sr/86Sr ratios from the otolith edge (from wild fish) and core (hatchery and wild fish) using an experimental system consisting of a ‘Nu Plasma’ multi-collector ICPMS (Nu Instruments, Wrexham, UK), coupled to a ‘RESOlution’ 193-nm excimer laser ablation system (Applied Spectra, USA). Each otolith was ablated on the primordium (core) and the dorsal margin at the widest radius using a 60-µm laser spot. Ablation was performed under a pure helium (He) atmosphere followed by rapid transport of the ablated products to the MC-ICPMS in the argon carrier gas. Methods for online correction for isobaric interferences and mass bias are described in (Woodhead et al., 2005), while accuracy and reproducibility was checked after every 10 otolith ablations using methods described in (Thiem et al., 2022).

We first tested whether the hatchery and Mitta Mitta River fish had different 87Sr/86Sr values and whether these differences varied between years. To do this, we used an Analysis of Variance (ANOVA), comparing 87Sr/86Sr values at the otolith edge from fish captured in the Mitta Mitta River and the otolith core from hatchery fish. We then ran a simulation to determine whether 87Sr/86Sr measured from the otolith core (natal origin) of each juvenile Murray cod captured in the Mitta Mitta River was more closely matched with the otolith edge chemical signatures of the Mitta Mitta River (wild fish) or the hatchery (stocked fish) for each year. Natal origin as either a wild or stocked recruit was only assigned if it was a ≥95% match with the chemical signature to either the Mitta Mitta River or the hatchery.

We used a sample of sagittal otoliths from 39 Murray cod (49–790 mm TL) collected and euthanized during electrofishing throughout the project. As well as enabling the natal origin of juvenile fish to be determined, it also provided a broad assessment of length-age relationships to determine year-class strength. Based on the ages of fish generated from the otolith assessment, it was determined that the length-age relationship of Murray cod in the Mitta Mitta River was similar to other populations in the region ( Supplementary Material 4 ). As such, a length dataset from all Murray cod captured during electrofishing surveys in the Mitta Mitta River from 2011–2018 was used to estimate ages of individuals using the species growth model reported in (Todd and Koehn, 2008; see also description in Supplementary Material 4 ).

We used our estimates of age to assess how flows and temperature in the lower Mitta Mitta River influenced YCS. It was determined that all juvenile fish captured between 2016 and 2018 were of stocked origin (see results Natal origin). Therefore, YCS was estimated by calculating a relative offspring survival probability using the ratio of catch of each year class vs the number of fingerlings stocked into the Mitta Mitta River for the same year (source https://vfa.vic.gov.au/db/native-fish-release-summary). We focussed our analysis on age classes 1 – 3 years due to the low confidence of age estimates for larger fish, and age 0 (YOY) fingerlings sometimes released just weeks before sampling. Offspring of the three age classes was modelled separately as a binomial variable, with a logit link function using generalised linear models. In these models, fixed-effect predictors were electrofishing seconds (to account for effort in detection; null model) and various metrics representing the temperature and discharge dynamics (mean, minimum, extreme events and variation) during the stocking period (day of and 30 days following fish release) that may influence fingerling survival ( Table 2 ). As expected, several of the flow and temperature metrics were correlated so only four of these fixed effects were tested (see Table 2 ). Each age-based model had a stocking window metric compared to the null model using the Akaike information criteria corrected for small samples (AICc) to identify the model which best fit the data.

A Murray cod population model previously developed for the MDB (Todd and Koehn, 2008; Koehn and Todd, 2012) was refined (see below) and used to examine the likely influence of past temperature impacts on survival for two vital rates (see below) and subsequent population dynamics; and use these outputs to explore different management scenarios, specifically altered flow delivery and stocking. The model summarises the life history of Murray cod by explicitly representing 50 age classes, where entry into the first age class combines spawning, egg, larval and early juvenile survival, and where early juvenile fish are defined as those less than one year old.

Murray cod are assumed to become sexually mature at five years of age, with fecundity increasing depending on the maximum size of fish in the population. For the Mitta Mitta River population, the average maximum size is approximately 110 cm. The model only accounts for females, as males are not considered to limit the population in any way (Todd and Koehn, 2008; Koehn and Todd, 2012), although it is recognised that males are likely to play an integral role in the successful development of eggs hatching into larvae (Rowland, 1998). A 1:1 sex ratio was assumed. The model includes stochastic processes for modelling variation in fecundity rates in all adult age classes and variation in survival in all age and stage classes each year. Variability in fecundity and survival rates was incorporated to simulate the process of natural variation in the environment by modelling demographic and environmental stochasticity (Akçakaya, 1991; Todd and Ng, 2001). Age-specific variation in growth was included as were processes for modelling the effects of density-dependence, with the choice of carrying capacity set by the user. For the purposes of this study, a long-term average of 10,000 female adults was used as the level above which density-dependence acts on the adult population size (for details see Todd and Koehn, 2008; Koehn and Todd, 2012), based on an assumed maximum density of one female adult Murray Cod every 10 m. This assumption was supported by detailed catch–mark–recapture data from the Murray River at Yarrawonga, which supports the estimates of population size (Lyon et al., 2019). We used an initial population size of 1500 female adults (an assumed density of one female adult Murray Cod every 66m) with all scenarios executed over 45 timesteps, equivalent to 45 years. Given the results of the movement component that found fish were largely sedentary in the reach (see Migration results below), immigration and emigration were not included in the model.

A time series of temperature from the Tallandoon gauge on the Mitta Mitta River was collected from 2004 to 2018 (15 years of daily water temperature records). This datum was analysed for the thermal effects on survival of:

The scenarios, generated from the past 15 years and used to model the trajectory of the population and its risk of falling below 500 adult females were:

Scenario 1: Actual natural recruitment conditions without stocking using the observed thermal time series and subsequent likely natural recruitment success (see Supplementary Material 5 ; equivalent to 2 in 15 years of successful recruitment).

Scenario 2: Modified natural recruitment conditions without stocking using a modified thermal time series to test variations in the frequency of natural recruitment above and below the existing recruitment frequency (equivalent to 1 in 15 years, 3 in 15 years, and 4 in 15 years).

Scenario 3: Scenario 1 with the observed stocking (timing and numbers of fish released; see Supplementary Material 6 ) which accounted for the river temperature at the time of release. This scenario represents what has occurred in the Mitta Mitta River the past 15 years.

Scenario 4: Optimal summer flow and temperature conditions and stocked fish release times using the same stocking time series together with temperature records, releasing the fish at the optimal thermal level for that season. This was calculated by firstly using an alternative summer water temperature in a retrospective manner where water could have been released earlier in the season rather than the actual summer release dates (Murray-Darling Basin Authority unpublished data). Secondly, the date that fish were released was altered so as to align with the optimal summer thermal period in each season (warmest 30-day period), thus giving the optimal scenario for the survival of stocked fish.

The number of stocked fish was halved to represent the female proportion of fish when included in the population model, assuming an even sex ratio of stocked fish. The thermal time series and stocking time series were replicated three times to derive an overall time series equivalent to 45 years to capture consequences for the population dynamics over the longer term. All scenarios have a background angler capture rate of 10% (chance of capture) of fish in the legal-size range 55 cm to 75 cm.

Trajectory summaries, including the average (± 1 standard deviation), maximum and minimum female adult population size from each scenario were used to provide a scenario overview. To compare the four thermal time series, the average female adult population size only was used; similarly, for the stocking time series. In addition to using trajectory summary information, risk curves and expected minimum population sizes were used for comparing scenarios. Population risk is characterised by changes in the cumulative frequency distribution of the minimum population size, also known as a risk curve. The risk curve represents the probability of falling to or below a given population size or threshold population size over the period for which the model is run and is commonly used in conservation management (Akçakaya, 1991; Todd and Ng, 2001). To quantify differences between risk curves, the expected minimum population size was calculated as the expected value of the cumulative frequency distribution and may be compared among scenarios.

The size structure of Murray cod captured in the study reach varied considerably in each survey year. This was driven largely by changes in the abundance of juvenile fish in the system ( Figure 3 ). The median length of Murray cod captured in the Mitta Mitta River significantly increased from 204 mm in 2011 to 288 mm and 386 mm in 2013 and 2015, respectively ( Supplementary Material 7 ). Conversely, the median length of Murray cod captured in 2017 was significantly lower than in 2011, 2013 and 2015. This change was driven by variation in the abundance of juvenile fish with the analysis of fish in the lower 20th percentile of size also varying across the survey years ( Figure 3 ). The median sizes of juvenile fish captured in 2011, 2017 and 2018 were 161 mm, 98 mm and 196 mm, respectively (compared to 279 mm and 306 mm in 2013 and 2015). In 2021 (our opportunistic monitoring), the median fish size was 137 mm, within the range of 2011, 2017 and 2018 values.

Of the 25 tagged Murray cod, three fish were not detected again after their release, although one of these fish was caught and released by an angler close to its original tagging location. More than 1 800 000 detections (condensed to 4 850 daily detections) from 22 of the 25 tagged fish were investigated for movement. In general, the 23 tagged fish displayed strong site fidelity with 10 fish detected on fewer than 20 days of the study period and a very low probability of movement beyond their tagging location (estimated 1-5% on any given day, Figure 4 ; Supplementary Material 2 ). Nevertheless, 21 fish (84% of tagged fish) moved beyond their tagging location in the form of either a seasonal spawning-type behaviour (with return movements) or a change in location ( Supplementary Material 2 ). There were some large movements, with one fish moving more than 50 km upstream (in 26 days) and returning the same distance downstream (taking just 16 days) in each of the three years during the study. No fish were recorded at the most upstream receiver during the first two years of the study.

Detections were analysed over a fortnightly (14-day) time-step because of the sparseness of daily and weekly data. There is little evidence that temperature, discharge and time of year affect the direction of Murray cod movement in this study ( Supplementary Materials 8A, B ). Of the fish that moved, most were downstream, however, the χ² test for independence did not show any significant departure from independence (p-value = 0.33). Furthermore, when other explanatory variables were used to analyse the data, the null GAMM model had the smallest AICc ( Supplementary Material 8A ).

The best model explaining movement (location change) included temperature, time-frame (before or after June 2016) and the cyclic smoothed day of the year as explanatory variables ( Supplementary Materials 8C–E ). The model explained only a small amount of the variation (adjusted R² = 3.0%). Peak movement times differed before and after July 2016 ( Figure 4 ). Prior to July 2016, movement rates peaked during February while after July 2016, the peak movement period was much earlier in the season, from October to November ( Figure 4 ). Movements were very unlikely (<1%) from May to August ( Figure 4 ). The model also showed that at higher temperatures, the likelihood of a Murray cod moving increases. The coefficient (on the logit scale) for the standardised temperature is 0.55 (95% CI is 0.04 to 1.05) with the odds of a fish moving increasing by a factor of 4.3 when the temperature increases from 12°C to 23°C. In general, there was less fish movement in the later period, with the odds of movement being reduced by an average of 86% (95% CI from 61% to 95%).

Just three fish captured in 2013 and 2015 were of a size estimated to represent fish stocked or spawned in the 2013–2015 period inclusive (112–161 mm TL). All three fish were determined to be of stocked origin, based on the presence of a calcein mark on their otoliths. For latter years, our comparison of otolith edge 87Sr/86Sr ratios of fish caught from the Mitta Mitta River with otolith core values from hatchery-reared fish indicated that all but one hatchery (used in a single stocking event) had distinguishable otolith 87Sr/86Sr ratios for their respective stocking years. Specifically, despite significant differences in the otolith 87Sr/86Sr ratios within the Mitta Mitta River amongst years (ANOVA, p<0.05), these values were significantly different from the Murray Darling Fisheries (MDF), Snobs Creek and Murrumbidgee hatcheries in their respective years ( Figure 5 ; all p<0.001, Supplementary Materials 9, 10 ). Only at the Maintongoon hatchery, that was used to stock some fish in 2017, were the otolith 87Sr/86Sr ratios indistinguishable from the Mitta Mitta River (ANOVA, p>0.05, Supplementary Materials 9, 10 ).

All juvenile Murray cod captured in the Mitta Mitta River which had a birth year of 2015/16 and 2017/18 were found to be hatchery-reared ( Figure 5 ). For the 2016/17 year class, 45% (n=5) of fish displayed natal origin (otolith core 87Sr/86Sr ratios) analogous to the Mitta Mitta River. However, closer investigation revealed these fish were captured at, or near (<1 km), a late-season stocking of Murray cod that were produced at the Maintongoon hatchery located on the upper Goulburn River. These fish were 60% smaller in size (length) than other YOY confirmed to have been stocked three months prior (mean TL = 96.33 ± 6.19mm and 57.80 ± 8.58mm, respectively). Taking this into consideration, we expect that the 2016/17 cohort of fish assigned natal origin belonging to the Mitta Mitta River are in fact stocked from Maintongoon hatchery (the only time this hatchery was used for the waterway), which shares a similar chemical signature value with the Mitta Mitta River ( Figure 5 ). Of the 16 Murray cod samples from the 2020/21 cohort, 70% (n=11) were assigned as wild-spawned natal origin whereas five fish (all of which were collected from the stocking location or sites in close proximity downstream) were identified as stocked from the hatchery ( Figure 5 ). This represented the first confirmed natural recruitment of Murray cod in the lower Mitta Mitta River during the entire study period.

As verified by our ageing data, our estimates of age structure (derived from age-length relationships and otolith data) for fish captured from 2011–2018 used for our survival assessment revealed a high number of fish belonging to the 2015/16 and 2016/17 year classes, with earlier surveys showing 2008/09, 2009/10 and 2010/11 year classes were also prominent ( Supplementary Material 11 ). All of these years coincided with relatively high summer water temperatures, usually (but not always) associated with minimal water releases from Lake Dartmouth.

Binomial models catering for fishing effort (electrofishing seconds) showed that the model which included average temperature for the 30 days following stocking best fit the data for predicting capture of a stocked cohort (our proxy for the survival rate of stocked fish) from 1 to 3 years of age ( Supplementary Material 12 ). About 99% of the variation in survival of stocked fish captured at age 1 and age 3 was explained by fishing effort and average stocking window temperature, while it was 61% for catches of age 2 fish ( Supplementary Material 13 ). There was a significant positive relationship between average temperature within the first 30 days of stocking and survival ( Supplementary Material 13 ). The model for catches of the 1-year age-class equated to a 63% reduction in survival for every one-degree reduction in mean daily water temperature ( Figure 6 ). Effectively, survival (and capture) of fish was very unlikely if mean water temperature was less than 20°C for the 30-day period following fish release (during summer or early autumn).

In Scenario 1, the observed thermal time series (2 in 15 years of successful recruitment) indicates that the Murray cod population would be in decline in all but a few trajectories over the 45-year time span ( Figure 7A ). In Scenario 2, which changed the thermal time series and subsequent recruitment frequency to 1 in 15, 3 in 15 and 4 in 15, on average, none of the modelled natural recruitment achieves population growth; however, the 4-in-15 scenario achieves the largest average female adult population size ( Figure 7B ).

In Scenario 3, the observed stocking and natural recruitment time series, there were periods of population increase and population decrease but overall the average trajectory is relatively stable through time ( Figure 7C ). The average female adult population size from this actual scenario is compared with the estimates from the natural recruitment conditions without stocking (Scenario 1), and the average female adult population size from the optimal stocking time series ( Figure 7D ). Our model predicted a 145% increase in the expected average minimum female adult population size under the current stocking program in comparison to the current conditions (2: 15 years recruitment without stocking). This increased to a 336% change for the optimal stocked scenario. The risk curves show the probability of falling to 500 female adults at least once over the 45-year model period was 95% for the actual recruitment scenario (Scenario 1), but this reduced to 69% with stocked as observed scenario, and to 12% for the optimal stocking scenario ( Figure 7E ). The growth rate through time shows that there are periods where the growth rate is less than 1 indicating periods of population decline and years when the growth rate is greater than 1 indicating an increase in the population size ( Figure 7F ). The overall average growth rate was 1.0024 (Scenario 1); 1.0172 (Scenario 3); and 1.0399 (Scenario 4) ( Figure 7F ). Under optimal stocking (Scenario 4) there remains periods where the population growth rate is less than 1, however the growth rate was greater than 1 42% of the time, 31% of the time for Scenario 3, and only 13% of the time for Scenario 1, even though the overall average the growth rate was marginally greater than 1.

Our results show how key population processes of a threatened freshwater fish are influenced by CWP, how this is likely to influence population dynamics and how we can use these insights to inform specific conservation actions and outcomes. For those population processes that contribute to recruitment, we demonstrated how CWP impacts spawning migration behaviour and YCS. Natural recruitment occurred in just one of the study years (our opportunistic sampling to test management outcomes), thereby supporting our hypothesis (Hypothesis 3) and previous work of (Todd et al., 2005) to show that water temperatures that fall below a critical survival threshold during the core egg and larval period will impact recruitment.

The results showing the influence of CWP on spawning migration behaviour and juvenile survival (Hypotheses 2 and 4) are novel and have not previously been investigated. The greatest number of juvenile fish captured and the subsequent strong year-classes of Murray cod occupying the Mitta Mitta River were associated with higher summer water temperatures, and typically not associated with large releases of cold water from Lake Dartmouth from January–March. Despite a relatively constant stocking rate since 2006, there was still high variability in the survival of fingerlings stocked in the Mitta Mitta River. Like its impacts on spawning and larval survival (Todd et al., 2005), our results indicate CWP also impacts juvenile Murray cod survival, with detection of a stocked cohort very unlikely if mean water temperatures were less than 18-20°C for the 30-day period following the release of fingerlings (during summer or early autumn). These results highlight those conditions influencing juvenile survival must also be considered in addition to those associated with spawning conditions (e.g. Tonkin et al., 2019) when assessing important drivers of recruitment strength for Murray cod.

An important first step in the recruitment process of Murray cod is the spawning migration of adult fish (Koehn et al., 2009). Our results provide evidence of unseasonal cold water releases altering the spawning migration behaviour of Murray cod (Hypothesis 2). Water temperature impacts the swimming performance of warm water fishes (e.g. Golden Perch, Macquaria ambigua; Lyon et al., 2008), with cold water and irrigation releases impacting the swimming capacities and energetics of Murray cod (Whiterod, 2013; Whiterod et al., 2018). We found that the likelihood of movement was positively associated with temperature, with the odds of a fish moving increasing by a factor of 4.3 when the temperature increases from 12°C to 23°C. With this, the timing of peak movement occurred in later spring and summer during the first three years of the study (when CWP was most pronounced). This timing is much later compared to the final two years of the study and previous research on migration and spawning which report larger movements of Murray cod (>10 km) typically undertaken in late-winter and spring (Koehn et al., 2009); (Tonkin et al., 2022), with spawning typically occurring during October and November (Humphries, 2005; Koehn and Harrington, 2006). Altered thermal regimes in rivers below large dams have been linked to changes in migration behaviour of fish in other parts of the world (e.g. Goniea et al., 2006; Thorstad et al., 2008; Tornabene et al., 2020; Moore et al., 2021).

Our analysis of direction of movement did not support our hypothesis of fish moving in a downstream direction and vacating the river reach in response to unseasonal spring and summer cold water releases (Hypothesis 1). This was despite spawning movement behaviour of Murray cod being altered by CWP. We found little evidence that temperature, discharge or time of year affected the direction of movement of Murray cod. Whiterod (2013) reports that Murray cod exhibits broad swimming capacity with low energy demands, therefore reflecting their evolution in some of the world’s most hydrologically and thermally variable rivers. Further to this, our results indicate that once established, Murray cod will remain in the reach even during successive years of elevated discharge and unseasonal water temperatures. As such, other management interventions such as riparian and instream habitat interventions that have been and are underway in the reach are likely to continue to provide benefits to these resident fish (e.g. Lyon et al., 2019; Tonkin et al., 2020).

While the study was able to specifically test the frequency and role of flows in influencing emigration of fish out of the study reach, we did not assess the role of immigration in governing the dynamics of Murray cod in the Mitta Mitta River. The contribution of immigrating fish as a key driver of population dynamics (e.g. Lyon et al., 2021) is likely to be minor given that our data showed no major changes in size structure of the population across years beyond that of small juvenile fish that were assigned as stocked or recruited in situ and very unlikely to have migrated large distances upstream into the study reach.

Finally, in support of our final hypothesis (Hypothesis 6), the output from the population modelling revealed the abundance of fish would very likely decline without any interventions over the 45-year model period. Even with minimal influence of CWP on retention, and growth and survival of larger fish, the conditions required for natural recruitment (spawning and juvenile survival) were still too infrequent to yield a stable population. Indeed, this emphasises the need for management interventions in this altered ecosystem, and relevance of reconciliation ecology for altered rivers globally (see below).

Our results clearly illustrate that due to the impact of cold water on the key processes governing recruitment, a population of a fish species with considerable cultural, recreational and conservation significance will not persist in the CWP-impacted reach of the Mitta Mitta River. Our results do, however, support several of our management hypotheses that indicate that interventions to bypass or overcome such recruitment bottlenecks can result in the recovery of the species. The otolith microchemistry results together with our population model predictions indicate that without ameliorating the CWP completely (e.g. Gray et al., 2019), stocking is required for this population to persist. Our model predicted a 145% increase of the expected average minimum female adult population size under the current stocking program (which commenced in 2006) and is likely to be a key factor in the apparent resurgence of Murray cod in the study reach since this time. Indeed, stocking has been shown to be of most benefit in systems where the conditions required for a self-sustaining population are not met (Secor et al., 2000; Forbes et al., 2016). Similarly, aquaculture-based restoration used to boost fish numbers and to overcome spawning habitat and recruitment deficiencies was critical in rebuilding and maintaining some stocks of Caspian Sea sturgeons (Secor et al., 2000).

It should also be noted that stocked fish were not always released in our study system under what we considered as the optimal temperatures. Hatchery-reared fish are faced with a variety of factors influencing their survival during the first year of life (Sutton and Ney, 2001). Some studies suggest most mortality occurs in the first few days following release due to predation, competition and biological metabolic thresholds such as temperature (Fielder, 1992; Blaxter, 2000; Svåsand et al., 2000; Brown and Day, 2002). Our results indicated that stocking the same number of fish with a more suitable thermal profile would significantly improve survival. Model predictions of the optimally stocked scenario indicated an increase in female adult population size from 145% (for the current stocking scenario) to 336% in comparison to the non-stocked scenario. Moreover, our scenario analyses showed that the same volume of water could still be released to meet human needs, but the timing of the release of both water and stocked fish would need to be slightly altered to improve the survival of juvenile fish.

We also showed that modifications to water transfers at critical times to reduce CWP impacts on the Murray cod population can also be extended to improve natural recruitment by minimising water transfers during the core spawning period. However, our modelling showed population decline continued even if managers could double the current natural recruitment frequency (to 4 in 15 years). Indeed, our opportunistic flow management and monitoring year provided an example of improvements to both natural recruitment and stocked fish survival that can be achieved through improved communications among researchers, fisheries and water management and operational agencies (Koehn et al., 2019).

A valid question remains regarding whether the reliance on stocking hatchery-reared fish and managing water releases to maintain a population in a severely altered system constitutes conservation management? We argue yes, specifically if viewed beyond an individual population in isolation, and instead considered in a reconciliation ecology and novel ecosystem setting within a landscape management framework (see Hobbs et al., 2014). Collectively, given that Murray cod is found across a spectrum of altered waterways in south-eastern Australia, managing this and other populations across the landscape will provide a greater range of intervention opportunities, use limited resources more effectively, and increase the chances of achieving management goals for the species. For example, Murray cod is an important recreational species, but such harvest pressure has been shown have a significant influence on its populations (Koehn and Todd, 2012). Simply maintaining populations across the landscape will absorb fishing pressure in the region that would otherwise be occurring at a few locations that contain self-sustaining remnant stocks of Murray cod. Also, if the hatchery and stocking practices used to maintain this population is undertaken with genetic diversity and variation considerations (e.g. Lyon et al., 2012), the population can be a source of broodfish for future captive breeding programs for the whole species.

Maintaining a spectrum of novel ecosystems across the landscape should also make species more resilient to larger perturbations such as climate change. Specifically, some populations occupying some rivers may come under increased threat from reduced or more frequent or severe variations in streamflow (e.g. Balcombe et al., 2011). Conversely, the opportunities for modifications to water management to achieve improvements to the processes supporting populations in novel systems such as our study reach may increase under climate change. As such, considering climate change impacts across the landscape and spectrum of novel ecosystems is essential to more effectively achieve outcomes for individual ecosystems and species across the landscape (e.g. Nelson et al., 2009; Hobbs et al., 2014). We therefore suggest an exploration of climate change impacts as well as different management approaches for the species across the landscape would be a valuable exercise. Specifically, an extension of our population modelling approach would help guide prioritisation of management actions across a spectrum of novel ecosystems with varying degrees of habitat modification and population losses under a changing climate.