Identifying effective water-management strategies in variable climates using population dynamics models


Correspondence author. E-mail:


  1.  Water-resource management should maintain ecological condition, including population viabilities of aquatic taxa. Many arid and semi-arid regions have experienced elevated water regulation and face drying and warming climates.
  2.  We combined stochastic, population dynamics models for four fish species with differing life histories with simulated regulated and unregulated flow regimes to assess the relative robustness of fish population persistence to different scenarios of climate change and water management.
  3.  Water regulation had a larger effect than differences in climate, negatively affecting one species through increased summer flows, and stabilizing population trajectories for two species that were sensitive to cease-to-flow events; the other species was insensitive to regulation or climate.
  4.  The greater importance of water regulation suggests that management of water regulation and human use can be used to insulate fish, to some degree, from the effects of future climate change.
  5.  General deductions from our results, such as the importance of inter-annual variability and the application of demographic modelling tools, are readily transferable to other systems.
  6.  Synthesis and applications. Our scenario-based approach was able to assess the population-level effects of multiple concurrent stressors and represents an effective framework for identifying management strategies that are robust to uncertainty in future environments.


Habitat loss, fragmentation and degradation and the loss of resources needed by the biota are important determinants of biotic decline (Millennium Ecosystem Assessment 2005). Two problems that are difficult to deal with are multiple stressors (Darling & Cote 2008) and differing responses of sympatric species to the same or alternative stressors (Mac Nally et al. 2011). These problems are common to freshwater, terrestrial and marine systems, differing mainly in the nature of the natural resources that are exploited by humans.

Human reliance on freshwater to support agriculture, industry and domestic use means freshwater ecosystems are now among the most heavily altered ecosystem types around the world (Palmer et al. 2008). The ecological and geomorphic effects of flow regulation and water extraction (collectively ‘alteration’) have been documented extensively (Palmer et al. 2008). Attention for the past several decades has turned to the reduction, mitigation and restoration from these impacts through the prescription and management of ‘environmental flows’ (Poff & Zimmerman 2010). The development of robust, defensible recommendations for environmental-flow delivery is difficult because of the uncertainties in the relationships between ecological behaviour and flow (e.g. Poff & Zimmerman 2010). Prediction of future ecological responses to different flow scenarios is hindered by the naturally high variability in many river ecosystems (Webb, Stewardson & Koster 2010) and a reliance on predictive tools that are poorly conceived and unsuited to dealing with variability and uncertainty (Shenton et al. 2012).

As profound as changes in water management have been, future effects on riverine ecosystems will be exacerbated by climate change (Xenopoulos et al. 2005). Many regions will experience less rainfall and increased temperatures over the next century, leading to reduced run-off (Seager et al. 2007) and exacerbating existing tensions between agricultural, social and environmental water needs (Barnett et al. 2008). A challenge for freshwater ecologists, and for water managers, is to assess more accurately the effects of water management on riverine ecosystems, especially given the predictions of climate change.

Climate-change and water-management scenarios have been coupled with population dynamics models (PDMs) using matrix selection, which involves a set of constant matrices of vital rates (fecundity, growth, survival) to represent vital rates under different climatic or flow conditions (Keith et al. 2008). Matrix selection is useful for comparing outcomes among scenarios; it is not intended to make specific predictions per se (McCarthy, Andelman & Possingham 2003) but instead as an heuristic framework (Boyce et al. 2006).

The inclusion of non-stationary climate effects is a relatively straightforward extension to matrix selection (van de Pol et al. 2010), although there are uncertainties in climate-change predictions arising from different general circulation models, CO2 emission scenarios and downscaling to regional predictions (Chiew et al. 2011). The matrix-selection approach does not explicitly consider causal links between climate and vital rates and restricts vital rates to the range of historically observed values assuming that data during historical extremes (e.g. ‘warm’ periods) apply to vital rates under future, warmer climates (Jenouvrier et al. 2009).

The matrix-selection method can be improved by using empirical knowledge of species’ biology and plausible responses to changes in habitat and climate (e.g. Lytle & Merritt 2004). Spring floods have been linked to breeding success of fishes (Mallen-Cooper & Stuart 2003), a response that could be included explicitly in PDMs. Estimating responses to environmental change using explicit causal links allows greater scrutiny of outcomes generated by a model. Studies linking climate-change effects with population dynamics (e.g. Lytle & Merritt 2004) are very uncommon compared to, say, climate-envelope models (Lavergne et al. 2010).

We built PDMs for four fish species as exemplars of a broad range of life histories: the Murray cod Maccullochella peelii peelii Mitchell, golden perch Macquaria ambigua Richardson, carp gudgeon Hypseleotris klunzingeri Ogilby and Australian smelt Retropinna semoni Weber (Fig. 1). A Winemiller-Rose guild diagram (Winemiller & Rose 1992) classifies species into life-history guilds and our focal species are in three distinct guilds: ‘equilibrium’ (Murray cod), ‘periodic’ (golden perch) and ‘opportunistic’ (carp gudgeon and Australian smelt) (Fig. 1). We used flow-response functions, based on species’ biology, to link PDMs to simulated flows (regulated and unregulated) under several climate scenarios. We sought to characterize the relative robustness, in terms of population persistence, of various approaches to water management given the uncertainties in climate and weather predictions. In doing so, we employ a new approach to environmental flows modelling, providing an alternative to the habitat-based environmental flows approaches commonly used (Shenton et al. 2012). While our focus is on fish responses to water management and climate change in drier areas of south-eastern Australia, the drying and warming futures have wide relevance to other parts of the world that are experiencing similar management and climate-change interactions, such as the south-western USA (Barnett et al. 2008).

Figure 1.

Position of each fish species on the Winemiller—Rose three-dimensional guild diagram (Winemiller & Rose 1992). Variation in three major life-history features (investment in juveniles, fecundity, time to maturation) was used to classify fish species into general guilds representing periodic, opportunistic and equilibrium species.

Materials and methods

Study Region

Flow scenarios and fish responses were based on measurements in lowland river systems in south-eastern Australia. Most data were from the Broken River, a 5th-order river (Strahler 1957) with a basin area of 7100 km2 and a mean annual run-off of 326 GL (10·34 m3 s−1).

Study Species

The four fish species are thought to respond differently to flow variation (Humphries, King & Koehn 1999). While the golden perch does not spawn in the Broken River per se, it spawns in similar, nearby lowland rivers and was included because of its distinct life history and predicted increase in spawning opportunities in the Broken river under a warmer climate (Bond et al. 2011).

Model Framework

We used a hierarchical model structure to link climate and management scenarios to population dynamics. Models were run for each species (Murray cod, golden perch, carp gudgeon and Australian smelt) under a range of climate scenarios (C = historic climate, mild increase in drought risk, large increase in drought risk, increasing drought length or increasing climatic volatility) and management schemes (M = unregulated or regulated). The overall structure is described here, with approaches to parameter specification and estimation described in the following section.

At the top level, climate states (CS = drought, normal or wet) were modelled, conditional on climate scenario, as a first-order Markov process

display math

where j and j′ are the current and previous climate state, respectively, and the πj′→j are the probabilities of transitions between climate states under a given climate scenario (parameter estimates are given in Table 1). Climate states were generated annually for 100 years (= 1, …, 100), with the climate state at = 1 selected randomly.

Table 1. Markov transition probabilities used to generate flow scenarios. The five scenarios represented potential future climates. Values are the probabilities of changing from the state in the row to the state in the column. For example, the number in the top right cell (0·2) is the probability that a drought will break with a wet spell under the current climate
ClimateClimatic statesDroughtNormalWet
Historic climateDrought0·750·050·2
Mild increase in drought riskDrought0·750·050·2
Large increase in drought riskDrought0·750·050·2
Increase in drought lengthDrought0·950·010·04
Increasing climatic volatilityDrought0·50·10·4

Conditional on climate state and management scheme, flow event levels (SprFE = spring flow events, SumFE = summer flow events, and CTFE = cease-to-flow events) were generated according to a categorical probability distribution

display math
display math
display math

where the π(∙) are the probabilities of a particular flow event (parameter estimates are given in Table 2).

Table 2. Probability of flow events in a given state (drought, normal or wet). Values are the probabilities that a flow event occurs with a given magnitude in drought, normal and wet conditions. Regulated and unregulated flow conditions were given different parameter sets. Estimates were based on the frequency of events in gauge flow data in the Broken River in south-eastern Australia
Spring flows (ML day−1)>81000·0010·0010·03000·005
Summer flows (ML day−1)>2600·10·20·30·20·450·5
Cease-to-flow events (days)0–100·150·750·950·550·950·99

The third level of our hierarchical model mapped flow events (SprFEt, SumFEt and CTFEt) deterministically to scalar multipliers of birth rates and survival probabilities, using the mappings shown in Figs S1–S3 (Supporting information). For each age class i, math formula, where the flow events depended on species and the life-history stages encompassed (see Table 3 and Appendix S1, Supporting information). The λi,t values were restricted to [0,1], so that the default demographic parameters represented an ideal flow scenario.

Table 3. Summary of qualitative flow requirements for the four species used in this study. A dash indicates a lack of knowledge regarding a species’ response to flow and does not mean that no relationship exists between a species’ vital rates and flow. This information was gathered from the literature and the rationale for each flow requirement is presented in Appendix S3 (Supporting information)
Common nameScientific nameSpawningEgg survivalLarval survivalJuvenile survivalAdult survival
  1. a

    Cease-to-flow events.

Murray codMaccullochella peelii MitchellPreference for CTFa <10 days Sep–NovPreference for floods Sep–NovPreference for CTF <10 days Dec–Feb
Golden perchMacquaria ambigua RichardsonPreference for floods Sep–FebPreference for CTF <30 days Dec–Feb
Carp gudgeonHypseleotris klunzingeri OgilbyPreference for low flows Dec–FebPreference for low flows Dec–FebPreference for low flows Dec–Feb
Australian smeltRetropinna semoni Weber

The lowest level of our hierarchical model was a PDM; see Caswell (2001) for a detailed overview of matrix population models. The λi,t were multiplied by birth rates and survival probabilities, which then were used for Poisson and Binomial random variable generation of abundances by age class (described in Appendix S1 in Supporting Information). The deterministic structure of our model can be approximated by E[nt+1] ≈ Λt(flow) × Lt × nt, where nt is a vector of abundances by age class (prior to reproduction), Λt is a diagonal matrix containing the λi,t values described previously and Lt is a Leslie matrix containing birth rates and survival probabilities for each age class. The model structure incorporated density dependence using the simplest empirically justified models: a top-down biomass model was used for Murray cod (Todd et al. 2005) and a Ricker model (Ricker 1975) was used for golden perch, carp gudgeon and Australian smelt (for details, see Appendix S1, Supporting information).

Parameter Specification and Estimation: Climate and Flow Scenarios

Five climate scenarios were considered with both regulated and unregulated flow regimes: historic climate, ‘mild’ increase in drought risk, ‘large’ increase in drought risk, increasing drought length and increasing climatic volatility. A ‘mixed’ scenario was generated by combining both management schemes (i.e. each year had equal probability of being regulated or unregulated) under the ‘historic’ and the ‘large increase in drought risk’ climate scenarios; these scenarios were used to determine whether inter-annual variability in water-management strategies (i.e. regulated years interspersed with completely unregulated years) might moderate the effects of water regulation. Such a situation has occurred in south-eastern Australia in recent years (2011 and 2012), where heavy flooding has overwhelmed the influence of flow regulation. For each of our 12 flow scenarios, 250 realizations were generated.

Climate scenarios were represented by their Markov transition probabilities (Table 1), that is, the scenarios differed in the frequency and length of dry, normal and wet periods. Markov transition probabilities were chosen to generate distinct, plausible future climate scenarios but were not estimated statistically or based on global circulation models per se. Plausibility was based on return times of events such as drought and flood relative to the historical occurrence of these events in gauge flows from the Broken River. Probability distributions for flow events differed between regulated and unregulated flows and were estimated using gauge flows from the Broken River at Casey's Weir, Benalla (Gauge No. 404216; Table 2). Further details and R (R Development Core Team 2009) code are in Appendix S2 (Supporting information).

Parameter Estimation: Scaling Values

The diagonal matrix Λt contained the λi,t scaling values. These values were based on ‘flow-response functions’ that represented the relationship between flow events and demographic rates; for similar mechanistic approaches see Lytle & Merritt (2004) and Thompson et al. (2012). We focused on flow because flow is linked implicitly to changes in many environmental factors (e.g. channel morphology, water temperature; Lytle & Merritt 2004) and is one of the main management levers. This approach is supported by evidence of a relationship between hydrologic variability and among-river variation in life-history success in fish (Mims & Olden 2012).

The importance of flow events (summer flows, spring flows, cease-to-flows) depends on species and life stages within species. We used case studies in the literature to hypothesize qualitative responses to flow for each species (outlined in Table 3). Cease-to-flow events were characterized by their duration, with extended CTF events predicted to affect Murray cod and golden perch negatively through habitat alteration and associated physicochemical changes (Figs S1–S2, Supporting information). Spring floods were predicted to have a positive effect on egg production in golden perch (Fig. S2, Supporting information) (Mallen-Cooper & Stuart 2003) and a positive effect on larval survival in Murray cod (Fig. S1, Supporting information) (Rowland 2004). These effects might result from increased spawning and recruitment opportunities, or from increased food availability (King, Tonkin & Mahoney 2009). We simplified our model by accounting for the occurrence of spring floods only and did not account for flood duration. Carp gudgeon breed during the summer low-flow period and are thought to be sensitive to elevated summer flows (Humphries, Serafini & King 2002). The effects of elevated summer flows depend on duration, so we set a threshold of 30 days, below which summer flows did not affect carp gudgeon (Fig. S3, Supporting information). Elevated flows persisting for >30 days were predicted to have a more negative effect as flow volume increased (Fig. S3, Supporting information). Further details are provided in Appendix S3 (Supporting information).

We used step functions to generate quantitative functions from the qualitative patterns (Table 3) following discussion with several experts with experience in Australian freshwater fishes (see 'Acknowledgements'). Expert opinion was used to estimate the proportional reduction in species’ vital rates at each level of the flow characteristics (i.e. a value of 1 represents ideal conditions, while a value of 0 represents completely inappropriate conditions). Where insufficient information was available, a conservative approach was taken in which vital rates were assumed to be insensitive to variations in the corresponding flow characteristic.

We developed several alternative functions for Murray cod that represented the same qualitative flow responses with different quantitative forms to assess the sensitivity of our models to the form of the step functions (see Figs S4–S6, Supporting information). Consistency among model outputs using alternative forms indicates that our qualitative estimates of flow response were translated reliably into a quantitative form.

Parameter Estimation: Population Dynamics Models

Parameter estimates were required for the Leslie matrix Lt, which contained the transition probabilities among age classes. The elements of Lt were drawn from probability distributions for species’ vital rates (e.g. survival and fecundity; Table 4), so that estimates of both average vital rates and variation around these rates were required.

Table 4. Parameter means, standard deviations and probability distributions used in stochastic demographic models. Model structures are described in Appendix S1 (Supporting information)
SpeciesParameterParameter codeDistributionMeanStandard deviationCarrying capacityStarting adult population sizeQuasi-extinction thresholdReference
  1. a

    Based on values reported in (Todd et al. 2005).

  2. b

    An adult carrying capacity close to starting population was used because the Ricker model allowed populations to grow larger than the stated carrying capacity (up to five times the stated value in this case).

  3. c

    Values were selected to reflect the life histories of these fish species. Variation in these values altered raw abundances but did not alter population trends.

  4. d

    Quasi-extinction thresholds were set between 10% and 75% of the initial population size. Higher thresholds were used to capture variation in population sizes (i.e. some populations never reached 10% of initial sizes and so a threshold at that level would not distinguish among scenarios).

Murray codFecundity (5–6-year-olds) f 1 Lognormal40002000400a100a10dAverage of 5–6-year-olds (Todd et al. 2005)
 Fecundity (7+-year-olds) f 2 Lognormal130004000   Average of 7–10-year-olds (Todd et al. 2005)
 Egg survival s E Beta0·50·075   (Todd et al. 2005)
 Larval survival s L Beta0·00120·001   As for sE
 1-year-old survival s 1 Beta0·450·094   Rounded value (Todd et al. 2005)
 2-year-old survival s 2 Beta0·600·12   As for s1
 3-year-old survival s 3 Beta0·70·14   As for s1
 4-year-old survival s 4 Beta0·750·11   As for s1
 5-year-old survival s 5 Beta0·80·12   As for s1
 6-year-old survival s 6 Beta0·80·12   As for s1
 7 + -year-old survival s 7 Beta0·850·09   Geometric mean over 7–10 year-olds (Todd et al. 2005)
Golden perchFecundity (3-year-olds) f 1 Lognormal100000500002000b2000c1500dBased on the value reported in Growns (2004), but reduced to reflect the lower fecundity of 3-year-old fish, which were considered to be not fully mature
 Fecundity (4 + -year-olds) f 2 Lognormal300000100000   4-year-olds were considered fully mature (Mallen-Cooper & Stuart 2003) and this value was based on that reported in Growns (2004)
 Egg survival s E Beta0·00330·001   Selected arbitrarily to avoid instability generated by high fecundities of golden perch
 Larval survival s L Beta0·150·05   As for sE
 1-year-old survival s 1 Beta0·530·1   Based on multi-year cohort data for the Moonie river (N. Bond, unpublished data)
 2-year-old survival s 2 Beta0·550·1   Based on s1, but increased to reflect the longevity of golden perch. Survival was assumed to increase as golden perch aged
 3-year-old survival s 3 Beta0·650·1   As for s2
 4 + -year-old survival s 4 Beta0·80·1   As for s2. This value was consistent with an average age of 10 years, with the potential for some fish to grow to >20 years of age (Mallen-Cooper & Stuart 2003)
Carp gudgeonFecundity f 1 Lognormal5002504000b4000c400dBased on Perry & Bond's (2009) value of 100 (updated to incorporate egg and larval survival) (Perry & Bond 2009).
 Combined egg and larval survival s 0 Beta0·20·05   Estimated from cohort data from broken river (P. Humphries, unpublished data)
 1-year-old survival s 1 Beta0·150·05   Selected to be consistent with values for s0 and s2
 2-year-old survival s 2 Beta0·10·04   Based on (Perry & Bond 2009)
Australian smeltFecundity f 1 Lognormal20010040000b40000c15000dBased on estimate provided in (Humphries, King & Koehn 1999) and the estimates provided in (Milton & Arthington 1985), both of which suggested 100-1000 eggs
 Egg survival s E Beta0·20·05   Selected to be consistent with life-history characteristics. The high relative reproductive effort of Australian smelt (Milton & Arthington 1985) was assumed to result in relatively low survival of early life stages
 Larval survival s L Beta0·20·05   As for sE
 1-year-old survival s 1 Beta0·30·1   As for sE.
 2 + -year-old survival s 2 Beta0·050·01   Selected to match the expected life span of Australian smelt (Milton & Arthington 1985), which is thought to be 1–2 years (i.e. survival of 2-year-olds is very low)

Vital-rate estimates for each species were gathered from the literature (Table 4). If estimates were unavailable, values from related species were used. Population trajectories were simulated in R and 1000 population trajectories were simulated for each of the 250 realizations of a given flow scenario.

Model Outputs

Our PDMs produced a probability distribution of population size at each time step for each of the 250 realizations of a given flow scenario. From this, we calculated mean and median population trajectories and associated quantiles. Quasi-extinction risk (Holmes et al. 2007), which we defined as the probability of declining below a specific threshold (see Table 4) at any time step, was calculated for each realization of each flow scenario, generating a distribution of 250 quasi-extinction probabilities for each flow scenario. We used ‘persistence probabilities’ as our final model output, calculated as one minus the quasi-extinction risk.


Effects of Regulation and Climate Change

Changes in climate-altered persistence probabilities for three of the four species (Murray cod, golden perch, carp gudgeon; Fig. 2), but water regulation had an equal or greater effect. Water regulation increased the likelihood of negative outcomes for carp gudgeon, but stabilized population trajectories and reduced the likelihood of negative outcomes for golden perch and Murray cod (Fig. 2). Changes in climate and water regulation affected the variability of outcomes but only slightly influenced median persistence probabilities (Fig. 2). Australian smelt did not respond to water regulation in the modelling (but see Table 3). Increasing drought risk or drought length negatively affected populations of Murray cod and golden perch, but had little effect on the carp gudgeon (Fig. 2). Under the mixed scenario, there was a reduction in the negative effects of regulation on carp gudgeon and a reduction in the negative effects of unregulated flows on Murray cod and golden perch (Fig. 2); recall that the mixed scenario used the ‘historic’ and ‘large increase in drought risk’ climatic scenarios.

Figure 2.

Probability of persistence under natural (black) and regulated (light grey) flow regimes under each climate scenario. Persistence probabilities also are shown for ‘historic’ and ‘large increase in drought risk’ climate scenarios under mixed (dark grey) flow regimes. Results are shown for (a) Murray cod, (b) golden perch, (c) carp gudgeon and (d) Australian smelt. Only two scenarios only are shown for Australian smelt because models for this species did not incorporate an explicit flow response (Scenarios A and B are 250 realizations each and represent any given climate scenario). The middle line of each box is the median, box edges are the 25th and 75th percentiles, and whiskers are the 2·5th and 97·5th percentiles.

Sensitivity Analysis

Changes to flow-response functions had little effect (Fig. S7, Supporting information). Although absolute persistence probabilities changed slightly, the overall trends in our results were unchanged (Appendix S3, Fig. S7, Supporting information).


Identifying the optimal (in the sense of operations research) management strategies for animal populations living in variable and changing environments requires good predictions of future climates, which are unavailable and are unlikely to become available in the near to medium future. Our scenario-based approach explicitly incorporated uncertainty in future climates, identifying trends in persistence probabilities that were robust to uncertainty in climate predictions. According to our models, the effects of water regulation are likely to have a larger effect on populations than will climate change. Even under extreme climate-change scenarios (e.g. 20 years of continuous drought), the effects of regulation were at least as large as the effects of climate change (Fig. 2). These deductions highlight the effects that flow regulation already has wrought on many fish populations, while emphasizing the need for future management strategies to consider the risks associated with alterations to hydrology stemming from climate change (Barnett et al. 2008). The system we considered is highly variable and these results may not apply in qualitatively different systems (e.g. hydroclimatically stable systems). Regardless, scenario-based approaches using quantitative models are useful for determining the effectiveness of alternative management strategies in uncertain futures (Jenouvrier et al. 2009).

Existing methods for assessing environmental flows typically rely on static flow rules, which do not explicitly model ecological outcomes or dynamics (Shenton et al. 2012). By directly linking flow scenarios to ecological response variables (e.g. persistence of biota), which are the foci of the management objectives, rather than to hydrologic surrogates, our approach allows model predictions to be validated and updated as new data become available. Moreover, monitoring of population trajectories in response to extreme events becomes the focus, which may not have been the case when flow rules were developed (Shenton et al. 2012). Such validation is crucial to assess the effectiveness of implementing environmental flows and forms the basis of effective adaptive management (Williams, Szaro & Shapiro 2009).

Species’ responses in our models differed and were not always consistent with the current understanding of these species’ responses to flow. Regulation decreased the persistence probability of carp gudgeon by increasing summer flows, a result that is consistent with observation (Humphries, Serafini & King 2002). However, regulation increased the persistence probability of Murray cod and golden perch, which is inconsistent with observations. Population sizes of the Murray cod are much reduced in many regulated rivers from habitat loss, pollution (including thermal), over-fishing, declines in primary production and interactions with invasive fish species (Rowland 2004). Our models did not include these factors, considering only the effects of flow regulation and climate change. Importantly, the population trajectories output from demographic models such as this can be compared directly with monitoring data and hence fit into a rigorous adaptive management framework (Williams, Szaro & Shapiro 2009). This is in contrast with the habitat suitability derived from hydrologic transformations, which cannot easily be translated into population responses, making anomalies such as this much harder to detect (Shenton et al. 2012).

The development of management schemes is complicated by conflicting outcomes among species (e.g. Murray cod and carp gudgeon had opposite responses to water regulation, Fig. 2). Our models suggested that higher levels of inter-annual variability improved multiple-species persistence (‘mixed’ models in Fig. 2). Although the mechanisms underlying multiple-species coexistence are debated (Chase & Myers 2011), inter-annual variability probably is important for allowing multiple species to persist. This is consistent with the high interannual variability in river flows that characterizes Australia's rivers (McMahon & Finlayson 2003) and rivers in many other parts of the world. There is no ‘one-size-fits-all’ approach to water management for population persistence, with static prescriptions being unable to accommodate the conflicting requirements of multiple species (Poff et al. 1997). Flexible approaches to water management will need to allow greater interannual variability in flow regimes, which may result in improved species persistence irrespective of changes in climate because different species benefit under different conditions (Poff et al. 1997).

Population dynamics models require estimates of birth rates and survival probabilities, which can be difficult to obtain for many species. One way to address this limitation is to use existing knowledge of related species (Shenton et al. 2012). Parameter estimates were available for our four study species, but even crude estimates are unlikely to be available for many other species. Assessing whether our inferences are transferable would be a test of whether PDMs could be applied broadly in the absence of detailed demographic knowledge (but see Caro & O'Doherty 1999 for potential pitfalls of this approach). Our study species were classified into three life-history guilds (Fig. 1) and this classification might apply to other fish species elsewhere in the world (Olden, Poff & Bestgen 2006; Mims & Olden 2012). If our results were transferrable, this would improve our ability to parameterize PDMs. The potential transferability of guild-based results is supported by consistent responses to flow within life-history guilds from different rivers throughout continental United States (Mims & Olden 2012).

Uncertainties in the specification of our scaling functions would be expected to reduce the reliability of our results. Our assessment of alternative step functions suggested that these uncertainties had a minimal effect on overall trends in results (Appendix S3, Supporting information). While we had little data with which to parameterize and assess our flow-response functions, the robustness of model predictions to changes in flow-response functions increases our confidence that the models generate reasonable outputs, consistent with the heuristic nature of our framework.

Concluding Remarks

Many freshwater fish faunas around the world have been declining for decades, usually due to complex mixes of multiple stressors (Warren & Burr 1994). Most assessments have been qualitative and management responses to declines usually are tailored to individual species. With on-going changes in hydrology arising from human management of freshwater resources coupled with longer-term changes in climate, better approaches are needed for dealing with the provision of flow regimes that accommodate demographic information and feedbacks (Lancaster & Downes 2010) and that allow the concurrent modelling of multiple species (Shenton et al. 2012). We demonstrated the feasibility of such a framework for simultaneously examining the effects of different climate sequences and water-management modes on population size and persistence probabilities of multiple species.

The results from our models are primarily heuristic, but our framework provides a potentially insightful method for quantifying population responses to altered water regimes and is readily adaptable to empirical prediction. Traditional approaches to environmental flows modelling focus on hydrologic surrogates, which often will not reflect population dynamics well (Anderson et al. 2006). Although population dynamics are not always of interest (see Arthington et al. 2006), they provide a measure of ecological response that accounts for temporal sequencing and time-lags (Shenton et al. 2012).

A remaining challenge is to convert heuristic models into predictive models. Demographic models have been perceived to be unreliable for generating predictions because vital rates are hard to estimate (Beissinger & Westphal 1998). The consequent uncertainties are problematic for forecasting population changes, but building demographic models forces one to identify and to address quantities (and their uncertainties) that otherwise might be ignored. The increased data requirements of PDMs might be addressed through the use of life-history guilds, which could allow transfer of results among species and regions. Irrespective of data limitations, if population responses to environmental change are of interest, then the PDM approach is a useful one.

We were struck by the common problems that arise in conservation management in terrestrial systems. The strong push for focusing on population and metapopulation viability in terrestrial conservation management (Reed et al. 2002) is equally relevant to the freshwater species that we consider here. The use of multiple, parallel PDMs provides a framework in which compromises between multiple-stressor management and effects on species’ persistence for freshwater animals are handled effectively.


The National Water Commission and the Victorian Water Trust funded the work undertaken in the paper through the project: Farms, Rivers and Markets (Lead CI: John Langford). RM was partly supported by ARC DP120100797. We thank Andrew Western, Ben Gawne, Mike Stewardson, Darren Neilson, Nathan Ning, Geoff Vietz, Sam Lake, Paul Reich, Jim Thomson, Paul Humphries, Nicole McCasker and Shaun Cunningham for helpful insights into the work. We thank two reviewers, N. LeRoy Poff and Ken Newman, for their insightful comments. This is contribution 257 from the Australian Centre for Biodiversity, Monash University.