Journal of Hydrology 517 (2014) 7482

Contents lists available at ScienceDirect

Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

Estimation of the environmental risk of regulated river ow

Kilisimasi Latu a,b,, Hector M. Malano a, Justin F. Costelloe a, Tim J. Peterson a
a
b

Department of Infrastructure Engineering, The University of Melbourne, Vic 3010, Australia

CRC for Irrigation Futures, P.O. Box 56, Darling Downs, Qld 4350, Australia

a r t i c l e

i n f o

Article history:
Received 27 November 2011
Received in revised form 7 May 2014
Accepted 10 May 2014
Available online 17 May 2014
This manuscript was handled by Geoff
Syme, Editor-in-Chief, with the assistance of
Paul Jeffrey, Associate Editor
Keywords:
Environmental ow demand
Environmental risk model
Risk time series
Hydrological ow index
Environmental risk

s u m m a r y
A commonly accepted paradigm in environmental ow management is that a regulated river ow regime
should mimic the natural hydrological regime to sustain the key attributes of freshwater ecosystems.
Estimation of the environmental risk arising from ow regulation needs to consider all aspects of the ow
regime when applied to water allocation decisions. We present a holistic, dynamic and robust approach
that is based on a statistical analysis of the entire ow regime and accounts for ow stress indicators to
produce an environmental risk time series based on the consequence of departures from the optimum
ow range of a river or reach. When applied to a catchment, (Campaspe River, southern Australia) the
model produced a dynamic and robust environmental risk time series that clearly showed that when
the observed river ow is drawn away from the optimum range of environmental ow demand, the environmental risk increased. In addition, the model produced risk time series showing that the Campaspe
River has reversed seasonal patterns of river ow due to water releases during summer periods, which
altered the ow nature of the river. Hence, this resulted in higher environmental risk occurring during
summer but lower in winter periods. Furthermore, we found that the vulnerability and coefcient of variation indices have the highest contributions to consequence in comparison to other indices used to calculate environmental risk.
2014 Elsevier B.V. All rights reserved.

1. Introduction
1.1. Flow regime and environmental risk
The natural ow regime is commonly considered as the main
organising variable of healthy uvial ecosystems (Bunn and
Arthington, 2002; Walker et al., 1995). Environmental risk to rivers
can be dened as a deviation from natural conditions (Horne et al.,
2010) and this change from natural ow affects habitat and biota
of a river system over time (Walker et al., 1995). A commonly
accepted paradigm in environmental ow management is that
the regulated river ow regime should mimic the natural hydrological regimes to sustain the key attributes of freshwater ecosystems (Ladson et al., 1999; Poff and Allan, 1997).
Because the timing and quantity of river ow are critical factors
of water supply, water quality and the ecological integrity of a river
system, the ow regime is strongly correlated with many critical
physicochemical characteristics of rivers, such as water temperature,
channel geomorphology, and habitat diversity (Poff and Allan, 1997).
Corresponding author at: Department of Infrastructure Engineering, The
University of Melbourne, Vic 3010, Australia. Tel.: +61 3 8344 4050; fax: +61 3
8344 4616.
E-mail address: latuk@unimelb.edu.au (K. Latu).
http://dx.doi.org/10.1016/j.jhydrol.2014.05.023
0022-1694/ 2014 Elsevier B.V. All rights reserved.

Flow regulation alters nearly all components of the ow regime with

consequent environmental effects. For example, storages and water
diversions reduce the magnitude and frequency of high ows and
can result in increased deposition of ne sediments in gravels that
reduce macroinvertebrate habitat (Poff and Allan, 1997). Conversely,
irrigation releases during the summer period signicantly change
the seasonality of ow in rivers naturally dominated by winterspring ow and can affect the seasonal cues for aquatic biota, such
as sh (Bunn and Arthington, 2002; Rolls and Arthington, 2014).
The clear relationships and strong linkages between a ow regime
and the environmental assets in both natural and regulated rivers
are increasingly recognized (King et al., 2003; Ladson et al., 1999;
Poff and Allan, 1997; Poff et al., 2010).
1.2. Approaches to estimating environmental risk
Early approaches to estimating environmental risk for regulated
river ow typically did not consider the effects of the entire ow
regime on all environmental assets within a river system but
instead focused on the ecological responses to alterations to a particular ow component (see review by Poff and Zimmerman
(2010)). As a result, environmental risks are often estimated based
on individual or multiple stressors, such as drought indices
that use a combination of one or more hydrological variables

K. Latu et al. / Journal of Hydrology 517 (2014) 7482

(Steinemann and Cavalcanti, 2006; Sun et al., 2012). These methods used to estimate environmental ow requirements for rivers
focus primary on one or few species that live in the wetted river
channel (Poff and Allan, 1997). The use of arbitrary minimum environmental ows is inadequate, because the structure and function
of a riverine ecosystem and the many adaptations of its biota are
dictated by the pattern of temporal variation in river ows, reecting the natural ow regime paradigm (Arthington et al., 2006;
Richter, 2010; Tharme, 2003). Single stressor focused approaches
have limited application for estimating environmental risk in a
water allocation model, which must consider the entire ow
regime of a river basin, e.g. from cease to ow to overbank ow
(Ladson et al., 1999; Poff and Allan, 1997).
Reviews of recent approaches to estimating environmental risk
show a paradigm shift towards considering changes across the
entire ow regime (Arthington et al., 2006; Poff and Allan, 1997;
Poff and Zimmerman, 2010). Arthington et al. (2006) proposed an
empirical approach that holistically incorporates essential aspects
of natural ow variability across the ow regime and shared across
particular classes of rivers, with these aspects being calibrated
against empirical biological and hydrological data. That approach
has the advantage of addressing ow requirements of many ecosystem components and assessing the ecological consequences of
change in each relevant ow variable. However, the approach is
only applicable in a regional context where a large number of
impacted rivers are considered in order to produce the reference
condition for the same class of rivers and it integrates all past
behaviour into a single score rather than a historical time series
of risk. As such, it has limited application to the water allocation
of a single river. Richter (2010) proposed the notion of a sustainability boundary approach (SBA) as a means for setting measurable goals relevant to environmental ow protection that address
the water quality and environmental ows required to support
water-dependent ecosystem benets. This approach denes the
degree to which human uses of water and land within a catchment
can alter natural or baseline hydrologic conditions without impairing ow-dependent ecosystem benets. The SBA method fosters a
precautionary approach that requires only the determination of
the magnitude of ow rather than relying on current scientic
knowledge to dene every aspect of the ow regime and the associated characteristics of the natural ow paradigm, therefore
greatly reducing scientic uncertainties (Richter, 2010). However,
the SBA approach does not address the cumulative risk due to
important ow variables being outside the optimum range for long
periods of time.
1.3. Aims
We further develop the approach advocated by Richter (2010)
by positing that if the optimum range of the environmental ow
demand of a river or reach is dened, then risk can be measured
by the magnitude of the deviation from that range. While river
health has a number of other aspects besides river ow, the use
of river ow to estimate consequences of water allocation policy
allows the direct linkage between river ow and the potential negative effects of water allocation. We use statistical analysis to estimate environmental risk based on the hydrological indices from
the index of stream condition (Ladson et al., 1999; Merabtene
et al., 2002). Our approach makes no attempt to link the magnitude
of ow alterations and ecological responses.
2. Study area
The study area is the lower reaches of the Campaspe River
catchment in southern Australia (Fig. 1). The study area has a total
catchment area of 2124 km2 and extends from the downstream

75

end of Lake Eppalock (the main storage in the catchment) to the

junction with the Murray River at Echuca. The catchment is relatively at in the downstream northern half with increased higher
terrain towards Lake Eppalock (Chiew et al., 1995). The climate is
fairly uniform with hot summers experienced particularly in the
north. The annual average rainfall is 450 mm (Chiew et al., 1995)
and 69 mm average annual runoff (CSIRO, 2008). The rainfall
occurs throughout the year, with the winter and early spring being
the wettest period and JanuaryFebruary being the driest months
(Chiew et al., 1995). The areal potential evapotranspiration is estimated to be 1211 mm (Potter and Chiew, 2009).
The study area has been divided into three reaches dened by
major demand centres and four major hydrological structures:
Lake Eppalock, Campaspe Weir, Campaspe Siphon and the outlet
to Murray River in Echuca (Fig. 1). In Reach 1, the major water
demand is from private diversion with little irrigation compared
to Reach 2 and 3. In Reach 2, demand consists of the Campaspe Irrigation Areas (East and West). The Rochester Irrigation Areas (East
and West) are located in Reach 3, although they divert water from
the Waranga Western Channel (marked by the boundary of Reach
2 and 3) which carries water from the Goulburn River, a neighbouring catchment to the east. Within Reach 3, the main demand from
the river is by private diversions. Each reach contains a gauging
station at the downstream end where the observed ows were
obtained: Reach 1 406201, Reach 2 406202 and Reach 3
406265. Because the levels of water demand are different for each
reach, the environmental risk must be individually determined.
3. Method
The estimation of Environmental Risk (ER) is based on two key
objectives. Firstly, the approach should produce quantitative,
dynamic and robust time series of environmental risk that take
into account the cumulative effects of past environmental risk. Secondly, it must accounts for risks that arise over the entire ow
regime (cease to ow to overbank ow), thus accounting for most
environmental ow related consequences. In order to satisfy these
two objectives, we propose a method that involves ve key steps as
summarised in Fig. 2.
3.1. Estimation of the EFD from natural ow
The Environmental Flow Demand (EFD) of a regulated river can
be dened as the specied ow magnitude required to maintain
the health of the river (Ladson et al., 1999). In this study, it has
been estimated using recommendations by ecologists and expert
panels based on a ow method assessment (Sharpe, 2006). Typically, these recommended values are in aggregate form and the
ow requirement does not specify the timing when the ow
should be applied in relation to the level of the Regulated Flow
(RF). Instead, only the frequency, duration and magnitude of ow
required within a time frame is dened without a clear link to
the level of the RF in the river. For this reason, the ow requirement must be disaggregated into a time series of EFD to enable
water authorities to adequately provide environmental ow to
the river. This disaggregation has been done using the method
developed by Neal et al. (2005) that involves the derivation of environmental ow requirements from modelled daily Natural Flow
(NF) at a site. This approach assumes that most environmental
ows are only provided if these ows would have occurred naturally (Neal et al., 2005). It allows automated decision making for
the provision of seasonal ows of a given magnitude at a given
annual frequency, and it assumes independence between events
and rates of rising and falling hydrograph limbs. Each component
of the recommended ow is progressively added until a time series

76

K. Latu et al. / Journal of Hydrology 517 (2014) 7482

Fig. 1. Presents the Lower Campaspe Valley as the study area and the demand nodes in relation to the Campaspe River and locations of the three reaches.

of EFD is constructed over the period of modelling. The daily NF

data are required as a reference series to determine whether environmental ows should be provided to the RF (Neal et al., 2005).
We use natural ow obtained from the Victorian Department of
Sustainability and environment for this study based on the
approach by Neal et al. (2005).
The emphasis of this paper is on how an ER time series is developed rather than how the EFD is determined. The latter can be generated with any of the methods discussed in the literature
(Anderies, 2005; Arthington et al., 2006; Tharme, 2003). Because
the model developed by Neal et al. (2005) includes the key hydrological factors, we considered this method appropriate to provide
an acceptable EFD series, as shown in Fig. 3. However, we recognise
that relying on modelled daily natural ow to produce monthly
EFD introduces uncertainty due to assumptions used in modelling
the natural ow, which may affect the EFD and subsequently the
environmental risk.

3.2. EFD optimum range

The optimum risk prole for an EFD is that where the regulated
monthly ow satises the EFD, and therefore would present minimum risk (Richter, 2010). The monthly median, 25th and 75th percentiles are used to dene the optimum range (see Fig. 4 for
example of EFD for Reach 1). The selection of 25th and 75th percentiles provide an arbitrary range so that if a monthly regulated
ow falls within this range then EFD is satised. The values are
changeable which allows the user to narrow or expand the risk
band to investigate their effects on risk sensitivity. If a river ow
magnitude is outside this optimum range then there is considered
to be environmental risk, and the level of risk is related to the magnitude of departure from the risk band. In addition, we assume that
the percentile range is time invariant. The use of percentiles avoids
making assumptions that the data are normally distributed from
one month to another.

77

K. Latu et al. / Journal of Hydrology 517 (2014) 7482

EFD Estimations

Flow Band Specifications

Step 2

Step 1
EFD LL
Daily Natural
Flow

EFD UL
EFD Optimum
Range

Env Flow
Requirements
High Flow Index

Step 3

Step 4
Low Flow Index

Total Consequence
(C)/5

Zero Flow Index

Step 5

Variation Index

Environmental Risk Time-Series

Vulnerability Index

Fig. 2. Five key steps of estimating the environmental risk time series of a regulated river basin.

10000
Natural Flow
Environmental Flow Demand

Flow (ML/m)

Regulated Flow

5000

0
Apr-02 May-02 Jun-02 Jul-02 Aug-02 Sep-02 Oct-02 Nov-02 Dec-02 Jan-03 Feb-03 Mar-03

Month
Fig. 3. Environmental ow demand, natural ow and regulated ow data for Reach 1 for 2002.

18000

EFD-Upper Limit
Median
EFD-Lower Limit

Flow (ML/m)

15000
12000
9000
6000
3000
0
Jan

Feb Mar Apr May Jun

Jul

Aug Sep Oct Nov Dec

Month
Fig. 4. The annual optimum range of the EFD for Reach 1.

3.3. Estimation of the environmental consequences

Risk is commonly dened as the consequence of an event (i.e.
ow occurring outside of the dened EFD) multiplied by the probability of that event occurring (Cha and Ellingwood, 2012; Hart

et al., 2005). In this study, we calculate the environmental consequences based on the difference between the EFD and RF over
the entire ow regime using a number of indices. As we use the
actual RF (either observed or predicted in a scenario) the probability of that RF occurring is considered to be one and so only the consequences are used in the risk estimation.
The Index of Stream Condition (ISC), developed for southern
Australian rivers, provides a set of multiple indicators to estimate
consequence to river health (Ladson et al., 1999). Aspects of the
ISC have been adapted to produce consequence scores that are ow
related (Ladson et al., 1999) using selected measures of the hydrology index (one of the ve sub-indices of ISC) and readily available
ow data. The components of the Hydrological Index of the ISC
allow a comparison between the RF and EFD over most of the ow
regime (e.g. up to bankfull). The Consequence Index (C) for this
study consists of ve sub-indices; Low Flow Index (LF), High Flow
Index (HF), Zero Flow Index (ZF), Coefcient of Variation Index (CV)
and Vulnerability Index (VI). Each index is outlined below. In addition, the sub-indices can be calculated over varying Cumulative
Periods (CP) allowing for the impact of previous deviations from
the EFD to be taken into account in the current time step. The
monthly ow data used for the risk estimations ranged from Oct
1977 to Dec 2005. For a 10 year CP to be used for calculating the
value of Jan 1995, the data should at least start from January 1985.

78

K. Latu et al. / Journal of Hydrology 517 (2014) 7482

Alteration of the magnitude of low ows changes the availability of in-stream habitats, which in turn can lead to a long-term
reduction in the viability of the populations of ora and fauna
(Lu et al., 2011; Sharpe, 2006). A Low Flow Index has been developed (LF, Eq. (1) that measures the changes in low ow magnitude
between the RF and EFD. A review by Nathan et al. (2009) found
that the low ow requirements often correspond to the daily 90%
exceedance ow. However, at a monthly time step, the index is calculated using two ow thresholds: one based on the ow exceeded
91.7% of the time (i.e. 11 months out of 12) and ow exceeded
83.3% of the time (10 months out of 12) (Nathan et al., 2009).



LF 91:7;t jPQ 91:7 EFD  PQ 91:7 R j

where LF91.7,t is the range-standardised Low Flow Index based on

the 91.7% exceedance ow for specied CP of years (j) (e.g.
110 years). The variable j is the number of years counting back
from the current month t. It represents the period of data used to
determine the PEFD and PR. P(Q91.7)R is the proportion of months that
the 91.7th percentile ow of RF (Q91.7)R is exceeded by the annual
91.7th percentile EFD ow (Q91.7)EFD. P(Q91.7)EFD is the proportion
of years of the EFD that the 91.7th percentile ow over j years is
exceeded by the annual 91.7th percentile of the EFD. The above process is repeated for the 83.3% exceedance ow. Then the rangestandardised Low Flow Index (LFt) is calculated as the average
(LF91.7) and (LF83.3). The advantage of using two range-standardised
low ow indices to calculate the LFt, is that it reduces the level of
bias and dominance of either the RF or EFD in determining of LFt.
High ows act as a natural disturbance in river systems, as they
can remove vegetation and organic material while resetting successional processes, especially during winter periods (Lu et al.,
2011; Nathan et al., 2009), and may provide adverse consequences
if they occur during the summer period. The High Flow Index (HF)
measures the change in high ows for a set of RF and EFD conditions. A similar approach is used for estimating LF but with 8.3%
and 16.7% exceedance ows used in Eq. (1).
The zero ow index compares the proportion of zero ow spells
occurring under EFD and RF conditions (Eq. (2)). If the number and
length of cease to ow spells remains unchanged between EFD and
current ow conditions, then the value of the index is 0, indicating
no consequence. The zero ow index is expressed as:

ZF t maxPZ EFD ; PZ R  minPZ EFD ; PZ R j

where ZFt is the zero ow index for month t and PZEFD is the proportion of zero ow events over the whole monthly data set for a given
EFD condition. PZR is the proportion of zero ow events over the
period j for the regulated ow. The max and min ensure that the
higher value between PZEFD and PZR is the value where the lesser
of the two is taken as reference.
The coefcient of variation of monthly ows (CVt) between the
RF (CVR) and EFD (CVEFD) for the CP j measures the variability
between the two ows (Eq. (3)). Seasonal variation in ow acts
as an important hydrological driver of aquatic ecosystems and is
relatively easy to predict. Due to irrigation releases and private
diversions, regulated systems have lower CV values than unregulated rivers with natural ows. The CV is applied according to the
following criteria:

if CV EFD > CV R ; then CV t CV R =CV EFD j

if CV EFD > CV R ; then CV t 1

We propose a new index called the Vulnerability Index (VI) (Eq.

(4)) that accounts for the shifts in the seasonality of ow. The index
is based on a vulnerability criterion used to quantify the severity of
failure in reservoir operation and to estimate the risk in water supply and demand decit (Evans et al., 2005). For this study, we
rened the approach of Evans et al. (2005) by using the proportion

of time when the two ow magnitudes are different. The advantage of using proportion of time is that each month represents a
magnitude of one, regardless of its ow magnitude.


!,
!

T
T
 X
X


VIt 
P  PR
PEFD 

 t1 EFD
t1

where VIt is the Vulnerability Index at month t using a CP (j), PEFD is

the proportion of time the RF falls inside the EFD range, while PR is
the proportion of time the RF occurs outside the EFD range.
Each of the ve indices range between 0 (no consequence) and 1
(negative consequence). The total consequence score is calculated
out of 1 (Eq. (5)) based on a uniform weighting of the individual
indices. The weighting, and indeed the use of other or additional
ow indices, can be easily modied according to the characteristics
of the catchment under consideration. The model is implemented
using code written in Matlab and we refer to it as the Environmental Risk Model (ERM).

C t 0:2LF HF ZF CV VIt

4. Results
We present results obtained from the application of the ERM
model for Reach 1 of the Campaspe River (Fig. 1) in order to illustrate the capability of the model in producing ER time series. We
rst discuss the consequence indices and their implications followed by the environmental risk time series.
4.1. Consequence indices
The consequence score is the mean of all the consequence indices for each month. The results for CPs ranging between 1 and
10 years are presented using a 10 year period of data from a model
data between Oct 1977 and Dec 2008 (Fig. 5). In general, the
shorter CPs produced higher variation in the magnitude of the consequence score. For longer CPs (e.g. 510 years), the variation was
greatly reduced due to the smoothing effect of using a longer
cumulative period to calculate a single value at a particular point
of time. While the use of a CP enables the past effects of not meeting the environmental ow demands to be included in the investigated month, the longer the CP used, the less sensitive is the
consequence score to immediate past conditions.
The study period of 19952005 consists of a mix of wet, normal
and dry years and the consequence score illustrates negative
effects where the river ows have frequently not met the optimum
range of the environmental ow demands. The step jumps in Reach
1 for the rst three CPs are the response to the large 1996 ood and
the higher ows in 2001, in comparison to the EFD optimum ow
range. In addition, the rst three CPs for Reach 1 shows variable
but declining trends in the consequence scores, which is caused
by the similarity of the regulated ows to the EFD. Fig. 5 suggests
that a 13 years CP may be adequate for calculation of consequence, because for up to 10 years, the consequence effects have
been signicantly smoothed out.
The mean monthly values of each consequence index for a
1 year CP were calculated and are presented in Fig. 6 for between
1995 and 2005. For Reach 1, the coefcient of variation and vulnerability indices have the highest consequence score, indicating that
the regulated river ow is often outside the optimum range of
environmental ow demand for a longer period of time, reecting
the changes in ow seasonality in the river due to system regulation. The low values of the ZF index reect the insignicant number
of zero ows in both the environmental ow demand and the regulated river ow. Interestingly, the high and low ow indices are
much lower in comparison to other indices. This reects the shift

K. Latu et al. / Journal of Hydrology 517 (2014) 7482

79

Fig. 5. Comparison of different consequences produced by environmental risk model for different cumulative periods of the regulated ow (represented by ow) for Reach 1.
The top section shows the ow, upper and lower EFD bands on a log10 y-axis, and the lower section shows the consequences indices between Jan 1995 and Dec 2005.

Fig. 6. Shows the average levels of consequence indices for each month of Reach 1 using a 1 year CP for the period of Jan 19952005.

in timing of ow relative to natural ow pattern (Fig. 3) rather than

large changes in the high ows or low ows.
The historical time series of the consequence indices of Reach 1
are shown in Fig. 7 using a CP of 1 year, chosen to minimize the
smoothing effects resulting from longer CP values. In general, the
vulnerability index is higher than the other indices in all reaches.
All indices reacted differently, with Reach 1 showing a high seasonal ow reversal which is reected in the consistently higher

value of its vulnerability index. The higher values of the coefcient

of variation and the vulnerability indices reect the shifting in timing of the ows and the differences in magnitude between the regulated and EFD optimum ows. This is in contrast to the other
indices with lower consequence values, which mainly focus on
the magnitude differences between the regulated ow and the
EFD optimum ow and place little attention to shifting in timing
of the ow.

80

K. Latu et al. / Journal of Hydrology 517 (2014) 7482

Fig. 7. The top section shows the regulated ow, upper and lower EFD bands on a log10 y-axis. Historical consequence indices over the investigated period for a 1 year CP are
shown on the middle section. The lower section of the gure shows the environmental risk time series for Reach 1 over a 1 year CP regulated ow.

The ER time series produced by the ERM are shown in Fig. 7. The
time series show the expected patterns of risk being higher when
RF is outside of the optimum range of EFD. The results show that
the model produces robust and dynamic ER proles reecting the
differences between RF and EFD. When the RF is within the optimum EFD range then there is minimum risk, rather than no risk,
due to the model accounting for the cumulative risk effects of
unsatised environmental water demand in previous years. When
RF is outside of the optimum range of EFD, the risk is higher. For
instance when a high ow occurs in summer rather than winter,
it results in a higher risk. This reects the highly regulated nature
of this system, where water is released during summer periods to
satisfy irrigation water demands, causing a signicant departure
from the rivers natural low ow status during summer months.
5. Discussion
The ERM cannot be strictly calibrated or evaluated as there are
no observed ecological/environmental risk data with which to
compare hydrological risk indices. However, it provides a relative
measure of risk and the accuracy of the model can be evaluated
qualitatively by comparing its behaviour relative to expectations.
The environmental risk is expected to increase when the environmental ow demands are not being satised and the model results
demonstrate the models consistency in this regard. The model was
able to capture the dynamics of risk arising from water diversions
within the catchment. For example, the mean monthly risk time
series (Fig. 7) shows the risk response to the regulated ow regime
within Reach 1 that is seasonally reversed compared to the natural
ow regime (Nathan et al., 2009; Sharpe, 2006).

The ERM has been developed to be part of a water allocation

model, where risk will be used as an allocation driver. Therefore,
when a water allocation model is calibrated the ERM will provide
an estimate of environmental risk for the period of calibration
and also for any future scenarios. The use of risk as a major driver
for a water allocation model allows the satisfaction of the EFD to be
optimised while taking into account competition with other
demands in the catchment. This approach has the advantage of
producing a time series of risk rather than a risk value for a single
point of time (Arthington et al., 2006; Ladson et al., 1999). As the
consequence index of ERM addresses the entire ow regime, meeting the environmental ow demands should account for most of
the environmental assets of a river.
The hydrological approach of ERM could also be incorporated
into existing frameworks that utilize some hydrological data, such
as those proposed by Arthington et al. (2006) or Poff et al. (2010).
For example, the holistic environmental risk assessment indices
(individually and in aggregated form) and time series of hydrological risk could be calibrated against ecological response data from
regulated rivers to test their utility and potential in frameworks
such as ecological limits of hydrological alteration (ELOHA) developed by Poff et al. (2010). As a result, ERM provides a valuable tool
to assist in meeting the paradigm that the environmental ow
regime should mimic the natural ow of the river in order to minimise environmental risk (Arthington et al., 2006; King et al., 2003;
Poff and Allan, 1997; Richter, 2010; Tharme, 2003).
Only ve risk (consequence) indices were used for this model.
However, ERM has the ability to incorporate more consequence
indices that capture different aspects of the ow regime, such as
an index that reects the frequency and duration of ood events

K. Latu et al. / Journal of Hydrology 517 (2014) 7482

(Ladson et al., 1999). The vulnerability and coefcient of variation

indices were dominant when using a CP of 1 year, thus reecting
the differences between environmental ow demand and regulated ow over this time scale, in particular, the shift in ow timing
rather than large changes in high and low ows. For larger CPs, the
coefcient of variation index becomes the dominant index, reecting a larger inter-annual variation in data due to a longer data series being used to estimate consequences. However, when the risk
was assessed using 1 and 3 year CPs, it was found that the risk
measure is only marginally greater with the increase of CP. Therefore, using a 1 year CP was sufcient to estimate the environmental
risk for the Campaspe River as it captures some of the effects of
antecedent conditions but is more sensitive to changes in the ow
regime.
The approach used in this model is similar to those used in previous studies (Ladson et al., 1999; Nathan et al., 2009) in which it
assumes that risks related to attributes of each ow variable are
additive. This assumption may not account for the complex relationships between ow attributes as the changes to different
aspects of the ow regime may interact resulting in synergistic
or forms of other relationship. To provide exibility, the ERM is
designed to be modular. This allows the risk indices and their
summation to be easily updated if interactions between ow components are better understood in the future. An additional assumption of the model is that there is a linear relationship between the
difference in natural and regulated ows and the resulting
environmental and ecological consequence effects. In many river
systems this may not be the case and indeed, previous studies have
suggested that with non-linear relationships a small change from
the natural ow regime may reach a threshold level that has severe
ecological consequences (e.g. Poff et al. (2010)). If this occurs then
the methods described herein may underestimate environmental
and ecological consequences to a river system from ow
alterations.
The use of different CPs has a number of signicant advantages.
Firstly, it captures the effects of the historical variation in climate.
The sequences of dry or wet years and their cumulative effects are
considered. This is a signicant advantage over other approaches
that do not provide a time-series of risk and ignore the signicance
of cumulative risk on the environment. Thus, it enables the decision makers to provide better allocation decisions by considering
the effects of previous risk within the river system and allows a
more cautious approach to how the effects of their decisions may
inuence future risks.
The model could also be used by water managers to facilitate
recovery of a damaged environmental system by forcing allocations to provide water to the environment more adequately. The
methodology developed within this study implicitly assumes that
regardless of the magnitude of the cumulative damage, the system
can be restored. This assumption implies that the system has only
one steady state and hence an innite resilience. While the concepts of ecosystem resilience (see Scheffer et al. (2001)) for a
good overview on this concept) have only recently been applied
to hydrology, a number of recent studies into groundwatervegetation (Anderies, 2005; Peterson et al., 2009; Runyan and
DOdorico, 2010) and wetland-vegetation interactions (Heffernan,
2008) have indicated that hydrological systems can have multiple
steady states. For such hydrological systems, the methodology
used in this study would be unable to facilitate recovery following
an event that caused a shift of the system to an alternative steady
state. Furthermore, the methodology would provide little insight
into when the system is near to switching into an alternative
steady state.
The model can simulate high risk in response to natural events
(e.g. large oods) and river managers need to take this into account
when evaluating results. The model results need to be carefully

81

interpreted to distinguish between high risk values generated by

natural events (e.g. large oods) and those that are generated by
ow regulation (e.g. increased ows in summer due to irrigation
releases).
6. Conclusion
The environmental risk model developed in this study was
found to be capable of producing robust and dynamic environmental risk time series on a monthly basis, which provide an indication
of the overall negative effects to environmental assets that may
occur when the environmental ow water demand is not satised.
The risk is higher when the difference between the environmental
ow demand and regulated river ow is higher. In the Campaspe
catchment, the model produced risk time series that indicated that
the Campaspe River seasonal river ow pattern has been reversed
due to water releases during summer periods. It was also found
that the vulnerability index is the main contributor to environmental risk, which reects the highly regulated nature of this Campaspe River system. The method can be enhanced by adding more
consequence indices once the interactions between environmental
attributors within a river system are better understood. Overall,
identifying the cumulative environmental risk of altered ow
regimes would allow better understanding of regulated river systems and how environmental water allocations can be managed
to minimize risk and maximize ecological benets.
Acknowledgement
This study was funded by the Cooperative Research Centre for
Irrigation Futures.
References
Anderies, J., 2005. Minimal models and agroecological policy at the regional scale:
an application to salinity problems in southeastern Australia. Reg. Environ.
Change 5 (1), 117.
Arthington, A.H., Bunn, S.E., Poff, N.L., Naiman, R.J., 2006. The challenge of
providing environmental ow rules to sustain river ecosystems. Ecol. Appl. 16
(4), 13111318.
Bunn, S.E., Arthington, A.H., 2002. Basic principles and ecological consequences of
altered ow regimes for aquatic biodiversity. Environ. Manage. 30 (4), 492507.
Cha, E.J., Ellingwood, B.R., 2012. Risk-averse decision-making for civil infrastructure
exposed to low-probability, high-consequence events. Reliab. Eng. Syst. Saf.
104, 2735.
Chiew, F.H.S., McMahon, T.A., Dudding, M., Brinkley, A.J., 1995. Technical and
economic evaluation of the conjunctive use of surface and groundwater in the
Campaspe Valley, north-central Victoria, Australia. Water Resour. Manage. 9 (4),
251275.
CSIRO, 2008. Water availability in the Campaspe. A report to the Australian
Government from the CSIRO MurrayDarling Basin Sustainable Yields Project,
CSIRO, Australia.
Evans, R., Dudding, M., Holland, G., 2005. Rivergroundwater interaction. In: A
Practical Management Approach, Engineers Australia 29th Hydrology and
Water Resources Symposium, Canberra, Australia.
Hart, B. et al., 2005. Ecological risk management framework for the irrigation
industry. Report to the National Program for Sustainable Irrigation by the Water
Studies Centre, Monash University, Clayton, Australia.
Heffernan, J.B., 2008. Wetlands as an alternative stable state in desert streams.
Ecology 89 (5), 12611271.
Horne, A., Stewardson, M., Freebairn, J., McMahon, T.A., 2010. Using an economic
framework to inform management of environmental entitlements. River Res.
Appl. 26 (6), 779795.
King, J., Brown, C., Sabet, H., 2003. A scenario-based holistic approach to
environmental ow assessments for rivers. River Res. Appl. 19 (56), 619639.
Ladson, A.R. et al., 1999. Development and testing of an index of stream condition
for waterway management in Australia. Freshw. Biol. 41 (2), 453468.
Lu, H., Huang, G., He, L., 2011. An inexact rough-interval fuzzy linear programming
method for generating conjunctive water-allocation strategies to agricultural
irrigation systems. Appl. Math. Model. 35 (9), 43304340.
Merabtene, T., Kawamura, A., Jinno, K., Olsson, J., 2002. Risk assessment for optimal
drought management of an integrated water resources system using a genetic
algorithm. Hydrol. Process. 16 (11), 21892208.
Nathan, R., Morden, R., Lowe, L., Austin, K., 2009. Development and Application of a
Flow Stressed Ranking Procedure. Sinclair Knight Merz, Victoria, Australia.

82

K. Latu et al. / Journal of Hydrology 517 (2014) 7482

Neal, B., Sheedy, T., Hansen, B., Godoy, W., 2005. Modelling of complex daily
environmental ow recommendations with a monthly resource allocation
model. In: Engineers Australia, 29th Hydrology and Water Resources
Symposium, Canberra.
Peterson, T.J., Argent, R.M., Western, A.W., Chiew, F.H.S., 2009. Multiple stable states
in hydrological models: an ecohydrological investigation. Water Resour. Res. 45
(3), W03406.
Poff, N.L., Allan, J.D., 1997. The natural ow regime. Bioscience 47 (11), 769784.
Poff, N.L., Zimmerman, J.K.H., 2010. Ecological responses to altered ow regimes: a
literature review to inform the science and management of environmental
ows. Freshw. Biol. 55 (1), 194205.
Poff, N.L. et al., 2010. The ecological limits of hydrologic alteration (ELOHA): a new
framework for developing regional environmental ow standards. Freshw. Biol.
55 (1), 147170.
Potter, N.J., Chiew, F.H.S., 2009. Statistical characterisation and attributation of
recent rainfall and runoff in the MurraryDarling Basin. In: 18th World IMACS/
MODSIM Congress, Cairns, Australia.
Richter, B.D., 2010. Re-thinking environmental ows: from allocations and reserves
to sustainability boundaries. River Res. Appl. 26 (8), 10521063.
Rolls, R.J., Arthington, A.H., 2014. How do low magnitudes of hydrologic alteration
impact riverine sh populations and assemblage characteristics? Ecol. Ind. 39,
179188.
Runyan, C.W., DOdorico, P., 2010. Ecohydrological feedbacks between salt
accumulation and vegetation dynamics: role of vegetationgroundwater
interactions. Water Resour. Res. 46 (11), W11561.
Scheffer, M., Carpenter, S., Foley, J.A., Folke, C., Walker, B., 2001. Catastrophic shifts
in ecosystems. Nature 413 (6856), 591596.
Sharpe, A., 2006. Campaspe River Environmental FLOWS Assessment. Sinclair
Knight Merz, Victoria, Australia.

Steinemann, A.C., Cavalcanti, L.F.N., 2006. Developing multiple indicators and

triggers for drought plans. J. Water Res. Plann. Manage. 132 (3), 164174.
Sun, L., Mitchell, S.W., Davidson, A., 2012. Multiple drought indices for agricultural
drought risk assessment on the Canadian prairies. Int. J. Climatol. 32 (11),
16281639.
Tharme, R.E., 2003. A global perspective on environmental ow assessment:
emerging trends in the development and application of environmental ow
methodologies for rivers. River Res. Appl. 19 (56), 397441.
Walker, K.F., Sheldon, F., Puckridge, J.T., 1995. A perspective on dryland river
ecosystems. Regul. Rivers: Res. Manage. 11 (1), 85104.

Glossary
CI: Consequence Index
CP: Cumulative Period
CV: Coefcient of Variation Index
EFD: Environmental Flow Demand
ER: Environmental Risk
ERM: Environmental Risk Model
HF: High Flow Index
ISC: Index of Stream Condition
LF: Low Flow Index
NF: Natural Flow
RF: Regulated Flow
ZF: Zero Flow Index
VI: Vulnerability Index