Professional Documents
Culture Documents
net/publication/318702726
CITATIONS READS
11 855
6 authors, including:
Some of the authors of this publication are also working on these related projects:
EPA - Climate Change Impacts and Risks Analysis (CIRA) project View project
All content following this page was uploaded by Sarah Marie Fletcher on 09 October 2017.
Abstract: Urban planners face challenges in water infrastructure development decisions due to short-term variation in water availability and
demand, long-term uncertainty in climate and population growth, and differing perspectives on the value of water. This paper classifies these
multiple uncertainties and develops a decision framework that combines simulation for probabilistic uncertainty, scenario analysis for deep
uncertainty, and multistage decision analysis for uncertainties reduced over time with additional information. This framework is applied to a
case from Melbourne, Australia, where a drought from 1997 to 2009 prompted investment in a $5 billion desalination plant completed in 2012
after the drought ended. The results show opportunities for significant reduction in capital investment using flexible design. Building no
infrastructure is best in most simulations. However, in 10% of simulations, building no infrastructure leads to regret of greater than $10 billion
compared with a small, flexible desalination plant. Scenario analysis for deep uncertainties underlines the significant impact of assumptions
about the future and also on value judgments about the cost of water scarcity in evaluating infrastructure performance. DOI: 10.1061/(ASCE)
WR.1943-5452.0000823. © 2017 American Society of Civil Engineers.
uncertainties, assumes equal likelihood of each scenario, and se- up to approximately 40% of the city’s demand (Grant et al. 2013).
lects strategies that meet threshold performance criteria across a However, the drought ended before the desalination plant was
large percentage of scenarios (Lempert et al. 2006; Lempert and completed, leaving the plant unused for years. Several studies dis-
Groves 2010). Info-gap theory, in contrast, develops increasingly cuss the political pressure on Melbourne Water and the Victorian
large multidimensional uncertainty sets and identifies the solutions government, a detailed timeline of actions taken, and the institu-
that meet threshold performance criteria for each uncertainty set tional decision-making process in responding to the drought
(Ben-Haim 2001). Decision scaling links decision analysis with (Ferguson et al. 2013; Grant et al. 2013; Porter 2013; Low et al.
bottom-up climate vulnerability analysis, identifying climate-driven 2015).
action thresholds without relying on general circulation models This paper addresses the following question: given similarly un-
to generate climate scenarios (Brown et al. 2012). Such approaches certain and dire situations like those Melbourne faced in 2007 and
have been widely applied to problems in water resources planning in the inherent trade-off between cost, supply risk, and environmental
various countries (Pallottino et al. 2005; Dessai and Hulme 2007; protection, what approach can water planners use in evaluating
Groves et al. 2008; Kasprzyk et al. 2012; Moody and Brown 2012). infrastructure investments when facing multiple uncertainties of
Climate change and other deep uncertainties should be integrated different natures? A decision framework using classification and
into a water resources modeling framework that accounts for the incorporation of multiple uncertainties is applied in order to
full range of uncertainty planners face (Ray and Brown 2015). (1) evaluate and communicate the cost and water supply risk of
While scenario analysis has been demonstrated to be a powerful proposed infrastructure alternatives, and (2) identify the best infra-
planning tool for urban water managers facing deep uncertainties structure alternative based on a planner’s risk preferences across
and varying stakeholder concerns (Lienert et al. 2006), this does simulated water supply futures. A single best solution is not pre-
not preclude the use of probabilistic approaches for different, more sented because that would require value judgments and preferences
easily modeled uncertainties. Population growth, which is more ap- from stakeholders and planners. Rather, results are presented for a
propriate for statistical treatment, is expected to have a larger im- range of sample preferences to demonstrate the usefulness of the
pact on water resource systems to midcentury than climate change framework. The remainder of the paper is organized as follows.
(Vörösmarty et al. 2000; Schlosser et al. 2014; Sapkota et al. 2015; “Methods” describes the methods used: the uncertainty framework
Fant et al. 2016). The stationary statistical variation in rainfall, is presented first, followed by a description of its application to
which can be modeled stochastically, often drives uncertainty in the Melbourne. Key results and conclusions are presented in “Results”
short to medium term more than climate change (Lins and Stakhiv and “Conclusions”, respectively. More detailed descriptions of the
1998; Doczi 2014). methods and additional results are provided in the Supplemental
This paper builds on traditional decision analysis and deep Data.
uncertainty methods for water supply planning by developing a
framework to classify and model multiple uncertainties of different
natures. It enables water supply planners to quantify statistical un- Methods
certainties when appropriate, while the impact of different, deep
uncertainties on a risk profile can be assessed using scenario analy-
Decision Framework for Multiple Uncertainties
sis. It also identifies which uncertainties enable learning over time
and incorporates them into a multistage decision analysis model. Many uncertainties simultaneously impact water supply infra-
This model allows the evaluation of real options such as staged structure planning. These include annual reservoir inflows, popu-
deployment where planners learn as uncertainties unfold and utilize lation growth, energy prices, the cost of water supply shortages, and
flexible options if and when they are needed. climate change. A framework for decision making under multiple
The usefulness of the method to water planners is demonstrated uncertainties was developed, illustrated in Fig. 1.
through an application in Melbourne, Australia. The Millennium The first step in Fig. 1 is to classify uncertainties along two
Drought in Southeast Australia from 1997 to 2009 motivated this dimensions: whether they are deep or statistical, and whether there
work. Melbourne’s water supply system comprises a network of 10 is high or low potential for learning over time. Deep uncertainties
storage reservoirs with a total of 1,812 million m3 (MCM) of stor- are those for which the likelihood of the outcomes cannot be de-
age. Net average annual inflows (evaporation and losses subtracted) termined; they are modeled using scenario analysis in order to
were 571 MCM between 1926 and 2014 and demand was avoid estimating likelihood. In this context, scenario analysis refers
401 MCM in 2014–2015 (Melbourne Water 2015b). This results to varying the value of uncertain parameters in order to understand
in a storage to annual runoff ratio of 3.2, which indicates a highly the impact of their uncertainty on the results. While scenarios are
managed system. Catchment-level streamflow management plans not explicitly probabilistic, the choice of values does assign some
uncertainties are realized over time; Step 2: water system and decision risk profile for each alternative based on how it performs under
models evaluate infrastructure alternatives under these uncertainties; many different conditions drawn from simulation and scenarios.
Step 3: performance indicators are displayed in a risk profile for de- Because stakeholder values are not elicited and incorporated in this
cision support study, the identification of a single solution is not feasible. This is
consistent with decision support tools that offer technical analysis
and identify the effects of value judgments on the results but do not
implicit probability to those outcomes; this is important to recog- recommend a single choice to planners (Benayoun et al. 1971;
nize in all deep uncertainty methods. Statistical uncertainties are Loucks 2000; Brugnach et al. 2007).
those that can be estimated using data-driven probability distribu-
tions. Statistical uncertainties are modeled using Monte Carlo sim- Application to Urban Water Supply Planning in
ulation in order to obtain the most precise risk profile. The high Melbourne, Australia
versus low learning potential indicates whether, as the uncertainty
is realized over time, additional observations meaningfully update This analysis framework is applied to an illustrative example from
the planner’s expectations of the future. Based on updated expect- Melbourne, Australia, where planners in 2007 decide what, if any,
ations, planners decide to exercise or ignore flexible options. These additional supply investments should be made over a 30-year plan-
uncertainties are therefore incorporated directly into a multistage ning period (the approximate lifetime of a RO plant). Key uncer-
decision analysis model (de Neufville and Scholtes 2011), in which tainties are identified and classified. Infrastructure alternatives are
flexible options are optimally exercised or ignored. This allows the chosen based on those considered by Melbourne’s water planners
full value of flexible options to be modeled. Uncertainties that do in 2007, with additional flexible alternatives designed for com-
not yield meaningful updates on expected payoffs as observations parison. A model of Melbourne’s water system is developed using
are made over time can be analyzed more efficiently by varying the a simple hydrological model and demand forecasts. A decision
inputs across model runs. analysis model is then used to evaluate the infrastructure alterna-
This uncertainty classification explicitly incorporates different tives, including some with flexible expansion options, based on an
objective function that considers lifetime costs and water shortages.
levels of uncertainty (Walker et al. 2003) and varying potentials
for learning over time. The level dimension can also incorporate
ambiguity, or uncertainties that arise through differences in stake- Uncertainty Classification
holder perspectives (Brugnach et al. 2008; Kwakkel et al. 2010), Five uncertainties, listed in Table 1, are included in this analysis.
such as value judgments about the cost of water shortages. Uncer- They were chosen because of their high degree of uncertainty and
tainties arising from ambiguity are similar to deep uncertainties in potential for impact on Melbourne’s planning decisions; however,
that they are not appropriate for probabilistic analysis, so scenario they are not comprehensive. Certain uncertainties were excluded,
analysis is used instead. Different objective functions can be used to such as the potential for desalination technology costs to decline over
represent different preferences or risk profiles. For instance, a risk- time, because initial analysis of historical data suggested they were
averse planner may prefer to invest in water supply infrastructure unlikely to impact planning decisions on a 30-year timeline. Other
that often sits idle as an insurance policy against drought that min- uncertainties, such as climate change, future agricultural production,
imizes shortage risk. While there is a large amount of literature on policy changes, and structural and observation uncertainty, could be
eliciting and incorporating stakeholder values into decision analysis included in future work. The included uncertainties are classified as
(Renn et al. 1993; Gregory 2000; Hämäläinen et al. 2001), such indicated in Table 1 with brief justifications. Decisions to classify
analysis is not included in this paper. Rather, a sample objective uncertainties using the framework presented in Fig. 1 ultimately rely
function and a few scenarios comparing alternative objective on analyst judgment but are informed by analyses of available histori-
weights are used to demonstrate the method. Additionally, other di- cal data and forecasts. The analysis was based on historical inflow data
mensions of uncertainty, such as the location of uncertainty within a (Melbourne Water 2015b), historical and forecasted population
decision-making process, structural uncertainty, and observational growth (Australian Bureau of Statistics 2014), and historical and fore-
uncertainty, are not addressed here. casted electricity price (ACILTasman 2011). Further details are avail-
In the second step in Fig. 1, a model of the water system is com- able in methodology sections A–C of the Supplemental Data.
bined with the decision model to evaluate and compare infrastruc-
ture alternatives. These infrastructure alternatives can be generated
Infrastructure Alternatives
through a screening model or planner input. The water system
model is simulated to generate the distribution of cost and water Six infrastructure alternatives comprised of combinations of three
supply risk for each infrastructure alterative over all statistical projects—a 150 MCM=year RO plant based on the Wonthaggi
Table 2. Definitions for the Six Infrastructure Alternatives Evaluated and Compared in the Decision-Modeling Framework
Capital expenditure
Infrastructure alternatives (millions of dollars) Capacity (MCM=year)
Downloaded from ascelibrary.org by Sarah Fletcher on 07/26/17. Copyright ASCE. For personal use only; all rights reserved.
S1: No build 0 0
S2: Pipeline and irrigation upgrade 1,002 Variable: maximum 100
S3: Small RO plant with expansion option 2,045 [þ1,095] Firm: 75 [þ75]
S4: Large RO plant 2,900 Firm: 150
S5: Small RO plant with expansion option; pipeline and irrigation upgrade 3,047 [þ1,095] Firm: 75 [þ75] variable: maximum 80
S6: Large RO plant; pipeline and irrigation upgrade 3,902 Firm: 150 variable: maximum 80
Note: S3 and S5 were developed to compare a staged deployment approach to the full, upfront deployment approach used in S4 and S6.
plant, a 100 MCM=year capacity pipeline and accompanying irri- both the existing reservoir system and a new supply infrastructure.
gation system upgrade based on the Sugarloaf pipeline and accom- The model estimates the annual cost and water shortages (i.e., un-
panying upgrades, and a 75 MCM=year RO plant designed with a served demand) over a 30-year period starting in 2007 for each in-
flexible option to expand to 150 MCM=year if desired—are de- frastructure alternative. Monte Carlo simulation is used to develop
signed and evaluated. The smaller (75 MCM=year) desalination distributions for these estimates based on uncertainty in inflows and
plant was not considered by Melbourne Water in 2007; it was in- population growth.
cluded here as an alternative to assess the value of staged deploy- The main components of the water balance are net reservoir in-
ment. The flexible design requires the plant site to be sized to fit flows (inflows minus evaporation), demand from end users, water
twice as many membrane modules as needed before the expansion, imported from new infrastructure, and environmental outflows. To
incurring additional capital costs upfront in exchange for the option model future inflows, 100,000 synthetic annual 30-year inflows are
to expand cheaply and quickly later. The no-build alternative (S1 in generated using an autoregressive moving average (ARMA) time-
Table 2) is included as a baseline option where no new infrastruc- series model fit to historical inflow data dating back to 1926. This
ture is developed. approach captures the autocorrelation observed in runoff. It as-
The desalination plants incur both high capital costs and high sumes inflows to be a stationary stochastic process over the 30-year
operating costs. The operating costs are high due to the energy planning period. Scenario analysis is used to vary the mean and
intensity of seawater desalination and membrane replacement variance of this process based on estimates of climate change im-
requirements. In addition to substantially lower capital costs, the pacts on runoff in Australia by Strzepek et al. (2013). This is a sim-
pipeline and irrigation upgrade also incurs much lower operating ple approach; climate change is not a focus of this paper. Future
costs; pumping is the largest component. The desalination plants, work evaluating longer planning periods could use more sophisti-
however, provide firm capacity; they can reliably provide the maxi- cated statistical approaches such as downscaling from general cir-
mum design capacity during a dry year. The pipeline system is culation models. Annual water demand is modeled as the product
market-based: farmers whose irrigation systems have been up- of population and demand per capita. Population projections from
graded to be more efficient sell excess water to the utility at their the Australian Bureau of Statistics for the city of Melbourne are
discretion. Farmers are less likely to provide water during dry used, ranging from 50,000 to 150,000 people per year (Australian
years, so it is unlikely that the full 100 MCM=year is available each Bureau of Statistics 2014). The base case demand per capita of
year. This dynamic is approximated by assuming a correlation be- 100 kL=person=year is based on historical demand data and is var-
tween inflows to Melbourne’s reservoir system and water available ied using scenario analysis to assess the effect of demand-reduction
from the pipeline, with 80 MCM available during average wet years measures; future work could assess and compare demand-side al-
and no water available during dry years. Cost estimates (capital ternatives directly. Water is imported from any new infrastructure
expenditures, fixed operating costs, and variable operating costs) capacity if reservoir storage levels go below one of two thresh-
are developed for each infrastructure alternative using cost data olds. Pipeline water is imported below an intervention threshold of
from the Sugarloaf pipeline project, cost information on the Won- 980 MCM and desalination water is imported below an emergency
thaggi plant, cost data from comparable RO plants, and input from threshold of 580 MCM, the maximum allowable drawdown. These
desalination experts. Details are available in “Infrastructure Cost thresholds are set by Melbourne Water (2015c) and used in other
Estimates” of the Supplemental Data. studies of Melbourne’s water system (Turner et al. 2014). In this
paper, water shortage is defined as water demand that cannot be
met without reducing demand or withdrawing below the emergency
Water System Model
threshold. The operational rules used to model the import and out-
A water system model is developed using a water balance approach flow decisions are illustrated in Fig. S2. The water system model
to model reservoir storage and water supplied to end users from aggregates the individual reservoirs and uses an annual time step.
Results
Decision Analysis Model
The decision model uses the estimates for cost and water shortages First results showing the impact of the statistical uncertainties, res-
from the water system model to evaluate the six infrastructure al- ervoir inflows and population growth, are presented in “Statistical
ternatives (Table 1) over a 30-year planning period. For each syn- Uncertainties.” This includes results from the water system model
thetic 30-year inflow series, the model ranks the six infrastructure showing the cost and water shortage risk of each infrastructure op-
alternatives using multistage decision analysis, which is frequently tion (Fig. 2) as well as results of the decision model (Fig. 3). These
used to evaluate real options (de Neufville and Scholtes 2011). The results are shown as distributions because Monte Carlo simulation
model can be understood as a decision tree, in which the population is used for statistical uncertainties. Then, results showing the effect
of deep uncertainties using scenario analysis are presented in “Deep
Downloaded from ascelibrary.org by Sarah Fletcher on 07/26/17. Copyright ASCE. For personal use only; all rights reserved.
Worst Payoff
total shortage spanning the 30-year period at 260 MCM. Cumula-
30% 5th Best tive shortages greater than 2,000 MCM (over the 30-year period)
4th Best occur in 20% of simulation runs (Fig. S6). The pipeline alternative
3rd Best (S2) consistently imposes a cost of close to $1 billion and decreases
20% 2nd Best the shortage risk. The four infrastructure alternatives that include a
Best Payoff desalination plant all increase the average cost and cost variability
while decreasing the shortage risk further. The median water short-
10% age for all four alternatives with desalination is 0, with 90th per-
centile shortages all on the order of 1,000 MCM over the 30-year
Downloaded from ascelibrary.org by Sarah Fletcher on 07/26/17. Copyright ASCE. For personal use only; all rights reserved.
analysis period. Within these four alternatives, those that have more
0 built infrastructure capacity see modest reductions in water shortage
risk but significant increases in cost. Most notably, the alternative of
nt rig g
w h O on
n
Irr
t
Irr
p l la n
n
io
hi
&
t&
w at
pt
the small plant with expansion option (S3) has a very similar water
ot
an
rg e
N
io
La arg
r
pt
shortage risk profile to the large plant (S4), as shown by the width of
I
o
it
D
/O
than 30% of simulations they incur cheaper overall costs than the diverse and multifaceted uncertainties water planners face. The
no-build or pipeline alternatives. The small plant (S3) provides a use of simple yet information-dense visualizations of risk profiles
midrange capital cost alternative that significantly mitigates down- can provide decision support for policy makers.
side risk through the use of a flexible option. Fig. S5 gives more The results demonstrate that moderate investment increases, to-
details. gether with flexible infrastructure design, can mitigate water short-
age risk significantly. Using a staged deployment strategy to expand
if necessary when more information is available reduces shortage
Deep Uncertainties risk at lower cost compared with other, less-flexible infrastructure
So far, results have incorporated uncertainties arising from inflow options. The value of this flexibility is dependent on the nature and
variability and population growth, the two statistical uncertainties magnitude of uncertainties in the system. Furthermore, demand-
addressed. Next, scenario analysis is used to show the effects of the reduction strategies may reduce the need for additional capacity.
deep uncertainties addressed: demand per capita, the shortage pen- These results also suggest that supply infrastructure can be regarded
alty value, and electricity price growth (Table 1). Each bar in Fig. 4 as an insurance policy against drought. Although the investments
shows how often the decision model selected each infrastructure may be used infrequently over their lifetime, they mitigate the risk
alternative as the best across a set of 100,000 simulations. In of severe economic consequences due to water shortages. This view
the base case, indicated using an asterisk, the model chooses both can allow for improved risk assessment along with appropriate val-
the no-build alternative and the small plant (S3) approximately 40% uation and expectations of its utility. However, there are trade-offs
of the time each. The results are highly sensitive to the penalty fac- between infrastructure cost and water shortage risk.
tor, shown in Fig. 4(a). The no-build alternative performs best The Victorian government’s decision to build a 150 MCM=year
in more than 95% of simulations when the penalty is reduced to desalination plant, one of the world’s largest, has been the subject
$5 million=MCM. Likewise, increasing the penalty makes alterna- of heated debate and political backlash given the wealth of reservoir
tives with greater capacity more favorable. Increasing the penalty to water available for years after the plant came online in 2012 (Porter
$50 million=MCM makes the no-build alternative best in approx- 2013; Ferguson et al. 2014). However, as drought impacts the re-
imately 25% of simulations, down from 40%, and the small plant gion again and the population continues to grow rapidly, the plant
with option plus pipeline best in 20% of simulations, up from ap- delivered its first water order in March 2017 (Gordon 2017), dem-
proximately 5%. Increasing the penalty by a full order of magni- onstrating that desalination capacity can play a role in mitigating
tude to $250 million=MCM, consistent with the values in Liu et al. supply and demand uncertainty even if it is used infrequently.
(2016), increases the magnitude of this shift. The no-build alterna- These insights are relevant to other places facing infrastruc-
tive is chosen as best in less than 5% of simulations and the small ture decisions driven by multiyear droughts. For example, several
plant with option plus pipeline (S5) is chosen in approximately 60% municipalities in California are considering desalination invest-
of simulations. The small plant with option plus pipeline dominates ments (Gillis 2015). Although drought in California is often framed
the distribution of best options over the large plant (S4). Both alter- as the new normal because of climate change, it is important for
natives have 150 MCM of capacity without the expansion in S5; both planners and the public to remember that high variability and
however, half that capacity in S5 is variable rather than firm. This extended periods of low rainfall are normal even in the absence
suggests that the value of flexibility from the expansion option out- of climate change. Using a learning-driven staged deployment ap-
weighs the variability in some of the capacity in S5. proach or increasing demand-reduction efforts may reduce the need
The model results are also highly sensitive to the demand per for capital investment. Alternatively, framing any investments as
capita, shown in Fig. 4(b). Lowering the demand to 80 kL per capita drought insurance that may be used infrequently may increase pub-
per year substantially reduces the need for additional infrastructure, lic acceptance. This strategy was successfully employed in the
with the no-build alternative performing best in more than 95% of United Kingdom, where Thames Water built a large desalination
simulations. The high sensitivity to demand per capita suggests that plant in east London in 2010 citing risks of severe water shortages
demand-side conservation measures may be able to significantly re- (Jowit 2010).
duce the need for supply infrastructure investments, especially if The analysis of the Melbourne case shows that much of the
demand reductions are firmly available during dry years. This dem- choice of the best answer is predicated not only on assumptions
onstrates the profound potential of demand management and points about the city’s future but also value judgments about the value of
to the need for more rigorous analysis of infrastructure expansion water during times of scarcity and society’s appetite for risk. The
decisions in conjunction with demand management options. Results sensitivity to the penalty value suggests that working with stake-
are relatively insensitive to the electricity price growth rate, shown holders to choose a value that reflects society’s tolerance to risk
in Fig. 4(c). Additional scenario analysis on model parameters, will be important in applying this method to prospective planning
in different regions. For example, while population growth pro- water management systems.” Water Res., 47(20), 7162–7174.
vides significant learning opportunities for planners in Melbourne, Dessai, S., and Hulme, M. (2007). “Assessing the robustness of adaptation
Australia, there may be different uncertainties that drive staged decisions to climate change uncertainties: A case study on water resour-
deployment decisions in other regions. Future work could also ex- ces management in the East of England.” Global Environ. Change,
17(1), 59–72.
plore the effect of different water conservation strategies on the ne-
Doczi, J. (2014). “Managing climate risk for the water sector with tools and
cessity for water supply capacity expansion or sustainability goals
decision support.” Climate change and water resources, S. Shrestha,
and objectives (Bhaduri et al. 2016). Some demand-reduction strat- M. S. Babel, and V. P. Pandey, eds., CRC Press, Boca Raton, FL,
egies such as fines or public education may yield uncertain and 239–290.
variable reductions rather than the firm capacity provided by desali- Donoho, D., and Gasko, M. (1992). “Breakdown properties of location
nation. Future work could also include more detailed models of estimates based on halfspace depth and projected outlyingness.”
operational reservoir management to assess the impact of improved Ann. Stat., 20(4), 1803–1827.
operations as an alternative to or in conjunction with infrastructure European Commission. (2013). “Adapting infrastructure to climate
additions. Additionally, new strategies such as fit-for-purpose water change.” Brussels, Belgium, 1–37.
supply, in which water of different quality is used for applications Fant, C., Schlosser, C. A., Gao, X., Strzepek, K., and Reilly, J. (2016).
for which the quality level is adequate (Nair et al. 2014), can be “Projections of water stress based on an ensemble of socioeconomic
evaluated for effects on infrastructure scale and design decisions. growth and climate change scenarios: A case study in Asia.” PLoS
One, 11(3), e0150633.
Ferguson, B. C., Brown, R. R., Frantzeskaki, N., de Haan, F. J., and Deletic,
Acknowledgments A. (2013). “The enabling institutional context for integrated water man-
agement: Lessons from Melbourne.” Water Res., 47(20), 7300–7314.
We thank Professor Hector Malano and Dr. Meenakshi Arora at the Ferguson, B. C., Brown, R. R., Frantzeskaki, N., de Haan, F. J., and Deletic,
University of Melbourne for discussions and assistance in identify- A. (2014). “The enabling institutional context for integrated water man-
ing initial data sources for this work. We are grateful to Professor agement: Lessons from Melbourne.” Water Res., 47(20), 7300–7314.
Richard Larson at Massachusetts Institute of Technology (MIT) for Frost, L., et al. (2016). “Water, history and the Australian city: Urbanism,
suburbanism and watering a dry continent, 1788-2015.” Cooperative
insights that shaped the early stages of this work and to three anon-
Research Centre for Water Sensitive Cities, Melbourne, Australia.
ymous reviewers whose thoughtful comments greatly improved this
Gersonius, B., Ashley, R., Pathirana, A., and Zevenbergen, C. (2013). “Cli-
paper. mate change uncertainty: Building flexibility into water and flood risk
infrastructure.” Clim. Change, 116(2), 411–423.
Gillis, J. (2015). “For drinking water in drought, California looks warily to
Supplemental Data
sea.” The New York Times, Apr. 11, A1.
Gordon, J. (2017). “Desal water will be needed every year, says Victorian
Further description of methods used, Figs. S1–S13, and Tables S1
minister Lisa Neville.” The Age, Mar. 19.
and S2 are available online in the ASCE Library (www
Grant, S. B., et al. (2013). “Adapting urban water systems to a changing
.ascelibrary.org). climate: Lessons from the millennium drought in southeast Australia.”
Environ. Sci. Technol., 47(19), 10727–10734.
Gregory, R. (2000). “Using stakeholder values to make smarter environ-
References
mental decisions.” Environ. Sci. Policy Sustainable Dev., 42(5), 34–44.
ACIL Tasman. (2011). “Wholesale energy cost forecast for serving residen- Groves, D. G., and Lempert, R. J. (2007). “A new analytic method for find-
tial users.” Australian Energy Market Commission, Melbourne, ing policy-relevant scenarios.” Global Environ. Change, 17(1), 73–85.
VIC, Australia. Groves, D. G., Yates, D., and Tebaldi, C. (2008). “Developing and applying
Australian Bureau of Statistics. (2014). “Regional population growth.” uncertain global climate change projections for regional water manage-
〈http://www.abs.gov.au/ausstats/abs@.nsf/products/AC53A071B4B231 ment planning.” Water Resour. Res., 44(12), 1–16.
A6CA257CAE000ECCE5?OpenDocument#PARALINK1〉 (May 1, Hämäläinen, R. P., Kettunen, E., and Ehtamo, H. (2001). “Evaluating a
2016). framework for multi-stakeholder decision support in water resources
Baker, J., Block, P., Strzepek, K., and de Neufville, R. (2014). “Power management.” Group Decis. Negotiation, 10(4), 331–353.
of screening models for developing flexible design strategies in hy- Harrison, M. (2010). “Valuing the future: The social discount rate in cost-
dropower projects: Case study of Ethiopia.” J. Water Resour. Plann. benefit analysis.” Australian Government Productivity Commission,
Manage., 10.1061/(ASCE)WR.1943-5452.0000417, 4014038. Canberra, Australia.
Benayoun, R., de Montgolfier, J., Tergny, J., and Laritchev, O. (1971). Hashimoto, T., Stedinger, J. R., and Loucks, D. P. (1982). “Reliability, resil-
“Linear programming with multiple objective functions: Step method iency, and vulnerability criteria for water resource system performance
(stem).” Math. Program., 1(1), 366–375. evaluation.” Water Resour. Res., 18(1), 14–20.
View publication stats J. Water Resour. Plann. Manage., 2017, 143(10): 04017061