To Main Proceedings D:
International System Dynamics Conference
Bergen, Norway
August 7-10, 2000
The Dynamics of Water Allocation in Semi-Arid Regions
Jennifer M. Andersen’ and David N. Ford’
ABSTRACT
The critical role of water for sustainable development and limitations of supply management have
increased the importance of demand management in meeting water needs. As an integral part of demand
management in water-stressed regions water allocation policies address the competition between different
user groups for scarce water resources. This paper presents a dynamic simulation model of a water system
in semi-arid regions for the purpose of analyzing the effectiveness of allocation policies in meeting two
objectives: 1) fill current demand and 2) provide adequate supply for future use. The model was calibrated
and tested with data and policies from the Mediterranean island of Cyprus. Analysis of water allocation
policies reveal that locally rational but overly risk adverse policies degrade performance and that
counterintuitive water allocation policies can be more effective in satisfying both current demands with
future water supply needs than current policies.
INTRODUCTION
One-fifth of the world’s population lack access to adequate, clean water supplies. This
threatens national security as well as prosperity, prompting Wally N’Dow, Secretary-
General of the United Nations Conference on Human Settlements to predict “...a shift
from oil to water as the cause of great conflicts between nations and peoples.” (U.S.
Water News Online 1996). Increases in supplies of water are limited because easily
accessible sources are invariably exploited first (Brooks, 1997), causing underutilized
water sources to grow increasingly difficult and expensive to exploit. While supply
management approaches to insufficient water can help in some areas of the globe it cannot
indefinitely relieve the pressure on the world’s water supply in the future (Postel, 1992).
This is particularly true in areas of water scarcity (Al-Ibrahim, 1990). For example Hamdy
et al. (1995) classified Mediterranean countries into three groups according to future
water problems: 1) countries where water supplies are currently sufficient, 2) semi-arid
countries with currently sufficient but declining resources and 3) arid countries already
facing water shortage crises. Semi-arid regions are characterized by long, hot, dry
summers and short, mild, wet winters. Tourism is also highest in the summer, in some
1 -
Consultant, Modeldata, Bergen, Norway <Jennifer.Andersen@ powersim.no>
cases increasing the population by 50 to 100%. Although these countries can currently
meet their overall water needs they face periods of shortages due to high demand and
inconsistent supply. Total demand can only be met by means such as over pumping
aquifers, which allows salt water intrusion and pollution of the aquifer (Brooks, 19xx).
These countries cannot sustain any significant increase in per capita withdrawals or
economic growth with their current water management and can only partially meet their
current water needs.
We focus on semi-arid regions due to their combined critical need for improved water
resource management and opportunities to avoid crisis conditions. Having done what they
can to increase available supply through water development projects, water managers in
these regions must manage demand in addition to supply. Three forms of demand
Management are used to match demand and supply: total demand management, load
management and allocation. Total demand management reduces water needs. Semi-arid
regions are often dominated by water uses (e.g. agriculture) and economic forces (e.g.
growth) that severely limit the reduction of total demand. Load management changes the
pattern of supply, use, or both over time to match periods of high demand with periods of
high supply. Allocation of available supply occurs when water managers cannot or choose
not to meet all demands and distribute available supply among users. Allocation decisions
divide scarce resources among competing uses to meet various social, economic or
political goals (Stiles, 1997). Allocation is often the primary tool of water managers in
semi-arid regions (Haten-Moussallen, Gaffney, Cox, and Batho, 1999). Due to its critical
tole we focus here on the impacts of allocation policies on filling water needs. In contrast
to water resource management models that focus on supply characteristics such as
variability (e.g. Wilchfort and Lund, 1997) we focus our investigation on management
policies as they are used by practitioners. Understanding these policies is critical to
understanding system performance. By modeling the actual policies of practitioners we
investigate their impacts on performance and changes which can improve system
performance.
® “aveierarPavtesen Deparment af Ciil Rugieedug: “Toes Ana Univenany, College Staten TX
77843-3136 <DavidFord@tamu.edu>
2
ALLOCATION POLICIES
Water managers in semi-arid regions face the difficult task of developing and
implementing allocation policies that will simultaneously fulfill current demand and save
adequate supplies to provide continuity of supply during droughts. The underlying
objective of this second allocation objective is to translate an inherently uncertain supply
into a predictable and dependable series of releases that users need for prosperous and
sustainable lifestyles and work. To meet these goals allocation policies must strike a
balance between how much water should be released in years of plentiful supply and how
much should be saved for drier years. Understanding how the allocation policies used by
Managers impact current demand satisfaction and continuity of supply requires
understanding of how managers use the information available to them to develop
expectations of future supply and set allocations and an understanding of how those
decisions impact current and future fulfillment of demand. Access to managers of the
water system studied provided us rich data concerning the information, parameters and
processes used to allocate limited water resources. This allowed us to model and analyze
an important aspect of water resource management with a depth and richness rarely
possible.
The Mediterranean island of Cyprus is an example of a semi-arid country where water
allocation policies have important impacts. Cyprus experience water shortages but is
expected to meet their water demand in the near future through water supply management
(Hamdy et al., 1995). In years of above-average rainfall existing and new supplies are
expected to provide enough water to meet demand from all sources. However droughts
every two to four years are common on Cyprus (Haten-Moussallen et al. 1999). Being
aware of this behavior pattern, Cypriot water managers plan for droughts when allocating
water. They acknowledge that they cannot sustain any significant increase in demand and
one or two years of lower-than-average rainfall will force stricter demand management
and allocation polices (Grimble and Archimandritou 1982c). Therefore effective allocation
policies for Cyprus must use water storage to de-couple the highly variable and relatively
unpredictable supply from the desired reliable and consistent outflows without sacrificing
fulfilling user needs.
Means of improving water allocation policies in semi-arid regions are not obvious. The
interactions of water manager decisions, political and social objectives and priorities, the
water supply system and demand centers are dynamic, delayed, nonlinear and closed in
that system conditions and information are fed back to managers to contro] the system.
These factors create a very complex decision environment for water managers. They have
a central, unknown variable (water supply) upon which they must base allocation decisions
which will in turn create a multitude of economic, political and social effects. For example
allocation decisions for some crops must be made before most of the rain has fallen. Water
managers are forced to build expectations of water supplies across future months for
annual allocations and years to meet long term supply needs. Therefore farmers must plant
some crops under allocation decisions that were made based on projected, not actual,
rainfall amounts. In addition, the year following a drought year is often a time of restricted
water supply as managers replenish storage. Therefore from the users perspective drought
conditions persist after the rainfall has returned to normal. This complexity has caused
purely economic approaches to the design of water allocation policies at the same location
studied here to generate results that are inconsistent with actual behavior and confounding
to researchers (Haten-Moussallen et al. 1999). An approach is needed that explicitly
addresses the dynamic complexity of water allocation decision-making and its impacts on
water resource system performance.
Water managers in semi-arid regions need an improved understanding of how allocation
policies impact multiple objectives. Explicit descriptions of policies as used in practice
with the information used, expectations developed, use options considered and priorities
of uses are needed to understand current practice. This can form a basis for improvement.
In addition to improved policy descriptions a means of predicting the impacts of current
policies and policy alternatives on water resource system performance from both user and
managerial perspectives is needed to evaluate policies. Finally, a means of analyzing the
structure through which allocation policies influence performance is needed to design
improved policies and transfer lessons from analysis cases to other systems. We address
these needs by developing a dynamic simulation model of the interactions of water
allocation policies and a water resource system including managerial agents and decision
making, information flows and physical system responses to allocation policies. We
describe the model structure in the next section. Then the model’s calibration and testing
4
with a specific water resource system in Cyprus and the allocation policies used to manage
it are described. We illustrate our model analysis approach and explain how analysis
results are used to identify weaknesses in current policies and direct the design of
improved policies. Finally, we draw conclusions concerning our work and make
recommendations for future research.
THE MODEL
Our model is a system of nonlinear differential equations. Model components and their
interactions are based on existing water resource theories and our field studies. For
example the structure of the water storage sector is based on the conservation of mass,
decision-making structures on the theory of bounded rationality (e.g. Simon 1995),
allocation policies on resource management theories (e.g. Jacobs and Vogel 1998), and
existing water resource models (e.g. Belaineh, Peralta, Hughes 1999) as well as previous
dynamic water resource models. Consistent with previous research, realistic storage
conditions are modeled, including the preservation of dead storage volume and flood
conditions (Jacobs and Vogel 1998, Haten-Moussallen et al. 1999, Sheer, Ulrich, and
Houck 1992). We focus here on the model structures, behavior and policies that reflect
the inadequate supply conditions that dominate water-stressed regions and water
allocation policies. Andersen (1998) provides complete model description and
documentation. Because no closed-form solutions are known we simulate the system's
behavior.
Our model includes water uses that differ in their volume, efficiency and timing of use and
contributions to economic performance. These differences can have significant impacts on
performance and managerial decision-making and are therefore important in describing
and analyzing allocation policies. Figure 1 shows the interactions among the three demand
sectors (agricultural use, residential use and tourism use), the water storage sector and the
water allocation sector. The Agricultural Use, Residential Use and Tourism Use sectors
accept water from the Water Storage sector and provide information on their respective
demand levels to the Water Allocation sector. Within the Water Allocation sector,
information concerning the supply from the Water Storage sector and demand from the
three use sectors are used to predict available supply. Allocation policies are then applied
to determine releases for specific uses, which reduces supplies in the Water Storage
sector.
Model Boundary
ne ee ee ee
| |
| Water Supply |
Crop Residential |
Production Demand
|
| Water Supph |
| ply Demand |
Information |
|
| a |
| Storage Allocation |
| Information |
Demand |
| Information
Allocation . |
Decisions Tourist
| Demand |
| |
| Water Supply |
rn ModelBounday
Inflow to the
Reservoir
Figure 1: Water System Model Sectors
The Water Use Sectors
Each of the three water use sectors models the demand for water and the performance of
the sector. Total Demand (D) is the sum of the agricultural (D,), residential (D,), and
tourism (D,) demands for water:
D=D,+D,+D, (1)
where: D - Total demand for water (m? per month)
D,- Agricultural demand for water (m? per month)
D,- Residential demand for water (m? per month)
D,- Tourism demand for water (m° per month)
Agricultural Use
The Agricultural Use sector models crop irrigation requirements and crop production. A
generic structure simulates each crop type, which is calibrated with parameter values to
represent a specific crop (e.g. citrus trees, potatoes, greenhouse crops). The generic
structure is based on documentation provided by the water development department at the
6
calibration site and interviews with the chief water manager. Differences in growing
seasons among crop types are an important driver of water demand. Therefore the annual
demands for each crop are spread across the year to reflect different growing seasons for
each crop type. We used the same fractional portions for each month which are used by
the water managers in the system we investigated to make allocation decisions. Unit
irrigation requirements are modeled as the product of the annual demand for water per
hectare of the specific crop i (di) and the fraction of annual demand required in specific
months (s;). Consistent with Belaineh, Peralta, and Hughes (1998) water use efficiencies
(e,) are included and the result multiplied by the land area the crop covers (a) to estimate
total irrigation demand. The water demands of individual crop types are aggregated to
estimate the total agricultural demand. Therefore the monthly agricultural water demand
for any number of crop types nis:
Da=dai* ((di*s)/e) iO {1, 2, 3...n} (2)
Where: a; - cultivated land area of crop i (hectacres)
d, - annual water demand of crop i (m? per hectacre per month)
s| - monthly fraction of annual demand for crop i (%)
e, - efficiency of water use by crop i (%)
n- number of crop types
Separately modeling different crop types allows us to analyze the potential use of different
crops and land use plans. To measure the performance of the system from the users
perspective this sector compares the amount of water that each crop needs to the amount
it receives. This ratio drives a nonlinear relationship that was previously developed for
each crop type by water managers and used to estimate the fraction of maximum crop
yield produced. The product of this fraction and the yield possible with optimal water is
the crop produced (Grimble and Archimandritou 1982a).
Residential Use
Residential water demand is water needed for basic household needs such as drinking
water, water for cooking, cleaning, laundry, lawn care and so on. Demand is modeled as
the product of population (p), water required per capita per year (d,) and a multiplier that
adjusts demand for seasonal differences in demand (sy).
D,=p* d,* sa (3)
where: p - population (residents)
d.- unit annual residential water demand (m*/resident/month)
Sq - monthly fraction of annual residential demand (%)
Residential water management performance is measured by comparing the water demand
and water actually supplied to determine the average number of days per month that
supply falls short of demand.
Tourism Use
Water demand for tourism is modeled as the product of the number of tourist arrivals each
month (t), the average monthly water demand per tourist (d,)and the average length of
stay (v).
D,=t* d*v (4)
where: t - tourist arrival rate (tourist per month)
d; - unit tourism water demand (m* per tourist per month)
v - average length of tourist visit (months)
Performance in the tourism sector is measured with the number of months in which
rationing or other measures are required due to releases not completely filling tourism
demand.
The Water Storage Sector
Our storage sector is relatively simple but driven by the inflow data set for our calibration
system (Kypris and Panayiotis 1994). Water stored in the one reservoir is modeled as the
accumulation of actual inflows (I) and losses as measured at the water system (L) and
managed outflows. This is consistent with previous approaches to simulating the impacts
of different allocation policies (Wurbs 1997). In our case the managed outflow are
teleases to users (R) as determined by our model of manager's allocation policies
(described next). Therefore:
68/ét=1-L-R (5)
Where: S - Stored supply of water (m’)
I - net inflows to water storage (m? per month)
L - Water losses from storage (m? per month)
R - Total water releases from water storage (m* per month)
The performance of the storage sector reflects the continuity of supply provided by a
given policy. Due to the long delays in some water resource system feedback loops supply
can change gradually over several years. Therefore we compare the water available for
future use at the end of the simulation period using different policies to assess water
storage management performance.
The Water Allocation Sector
Based on our studies of practicing water managers two critical decisions are made in
distributing available water to users that impact both fulfillment of current needs and long
term supply. First managers decide how much water to release from supply. This form of
supply-side load management allocates water supply between filling demand in the current
year and saving water for future use. Second, managers decide how to allocate the
released volume among users. Our model explicitly separates the policies that describe
these two decision processes, allowing the investigation of their separate and combined
impacts on water resource management performance.
The Release Volume Policy
One simple but naive management policy releases water in a pattern that closely mimics
the currently available supply of water, releasing up to demand when the reservoir is full
and less as the reservoir level drops. While this approach fills current demand during
adequate supply it leaves managers vulnerable to exhausting supplies during droughts and
open to criticism for not including droughts in their allocation policies. Worse, one year of
drought can easily cause two or three years of drought-like conditions for users as
managers withhold water from users to allow the storage system to recover before
releases that match demand can be resumed. Since water managers in regions of highly
variable supply are aware of tradeoffs between using water for current needs and saving
water for future needs they attempt to anticipate available supplies and incorporate those
expected supplies into allocation policies. The requirement to make release decisions prior
to the rainfall that provides some of the water to be distributed, the uncertainty of that
rain, and the complexity of the impacts of release decisions preclude defining an optimal
release policy. As one manager we interviewed admitted, “We do not have an algorithm in
helping us to decide on the best possible levels of restrictions per use.” (Andersen, 1998).
Therefore manager’s expectations about supply are paramount to understanding their
release volume policies.
Consistent with the literature on decision-making (e.g. Simon, 1995) we assume that
water managers’ expectations about water storage do not change as abruptly as changes in
the water level of the reservoir. Instead managers are strongly influenced by historical
supplies in formulating their expectations of future water supplies. For example managers
may expect supplies to be inadequate even though current supply is plentiful if the region
has experienced a drought in the previous few years. Therefore expectations lag behind
current conditions. Since recent experiences are given more importance than older
conditions (ref here) we model expected storage (E) as an exponential adjustment toward
the current storage (S)over a period of time (tz) estimated to be 18 months:
6E/ét=(S- E)/te (6)
where: E - Expected water supply (m*)
S - Water supply (m’)
te - Time to adjust expected water supply (months)
In regions prone to drought, water managers often adopt risk adverse policies for
allocating water between present and future needs. In a model of water stressed regions
the adjustment time of expectations about water supply (tz) is one descriptor of the level
of risk aversion or conservativeness incorporated into policies. Longer adjustment times
reflect more conservative policies as managers “remember” times of inadequate supply
longer even as current supplies increase and therefore expect and plan for inadequate
supplies more. Conservative policies for regions that also regularly experience periods of
abundant supply (not addressed here) would be modeled with short adjustment times
when supplies decrease to reflect manager’s quick “forgetting” of plentiful supplies. Such
asymmetric adjustment times have been used to model changes in expectations (Oliva
1999) and can easily be incorporated into our model.
A second descriptor of the managerial level of risk aversion is how managers respond to
the available supply when making decisions about how much water to release. When
anticipating drought managers hold water in storage until they feel an acceptable level of
confidence that adequate supply is available that water can be released more freely. We
10
use the concept of “coverage” to describe how much supply managers want to have
available. We define coverage as the ratio of the expected storage (E) to how much water
is needed to meet the unfilled demand for that particular year or in the next year as the
current year nears it’s end (D,). The water coverage ratio compares the coverage and
desired coverage, a third measure of risk aversion to determine if, and by what margin,
supply will last for the remainder of the dry season or, as current demand approaches zero,
the beginning of the next year. For example a water coverage ratio of 1.2 indicates a 20%
surplus over what is believed to be needed to meet unfilled demand. Values less than 1
indicate a shortage in supply, which would be a waming if occurring in the dry season
since additional supply would not be available until the next rainy season. Releases (R) are
reduced from the levels indicated by purely by demand (D)in response to supply through a
nonlinear relationship between the water coverage ratio and releases (fc). Including the
preservation of dead storage:
R =Min(S - Sq, D * fe(E/ Dy) (7)
Du =D - (Rat (8)
Where: S,- "Dead" (unavailable) storage (m°)
f. - Effect of expected coverage on releases (dimensionless)
D,- Unfilled demand for water (m*)
Our field studies and other management policy research (Ford and Sterman 1998,
Meadows 1970) have found the nonlinear nature of managerial assessments and responses
to system conditions are critical drivers of behavior and essential to understanding how
policies used in practice can be improved. By explicitly modeling these nonlinearities we
capture important system components and relationships missed or inadequately modeled
with purely linear approaches. The shape of this curve is reflective of the conservative
nature of water management on Cyprus. Figure 2 shows an example of the effect of
coverage on releases based on our fieldwork.
dA,
Effect of Coverage on Normal Release
Figure 2: Effect of Coverage on Releases
The Allocation Policy
Water managers often prioritize use objectives to facilitate allocation decisions (Sheer
Ulrich and Houck 1992, Haten-Moussallen et al. 1999). Policies incorporated into the
model reflect the priorities among competing potential uses of available water. In our
fieldwork we observed the following priorities during times of adequate or nearly
adequate supply:
1. Preservation of "dead" (unavailable for use) storage in the reservoir
2. Dictated uses determined by legally binding covenants (e.g. riparian rights of
land owners), recharging aquifers and transfers of water to other reservoirs
3. Residential and tourist uses
4, Agricultural uses (primarily irrigation), with a higher priority given to keeping
long lived production plants such as fruit trees alive
5. Retention of water supply for future use
In times of adequate water supply releases fill user demands (priorities 2, 3, and 4 above).
However this is often not possible because total releases are less than total demand
(equation 7). In those years total releases must be distributed among users. To do this
managers first preserve dead storage and fill all dictated uses (priorities 1 and 2). The
remaining water released is distributed proportionately among agricultural, residential and
tourism uses based on their contribution to total demand. Agricultural releases are used
first to fill minimum long-lived production plant needs as suggested by Keshari (2000).
12
The remainder of the agricultural releases is distributed among the different crop types
proportionately according to each crop type’s contribution to agricultural demand.
The policy above captures the fundamental drivers of water release and allocation but
remains a generalization of the actual policies used by water managers at our calibration
site. For example, actual policies varied from the model description during the first few
years of operation as managers initially filled the reservoir. Managers have also applied a
trial-and-error approach in search of an optimal policy. For example during one period
after dead storage and dictated demands were met 80% of domestic demand was provided
first, then a percentage of the demand for long-lived production crops, then the remaining
20% of domestic demand, then other crops’ demand, etc. However these managers
consistently used the rank order of priorities above and a policy of allocation
proportionate to contribution to total demand throughout the simulation period. Therefore
the model is considered useful for policy analysis.
Model Testing and Calibration
To test our model and provide a real-world basis for policy analysis we calibrated it to the
Kouris Dam water district on the island of Cyprus. The Kouris Dam is the largest reservoir
on Cyprus with a 92 MCM reservoir and is part of Cyprus's Southern Conveyor Project
(Cyprus— Southern Conveyor Project Case Study, 1986; Water Development in Cyprus,
1996). Parameter estimates are based on research literature, data from the Southen
Conveyor Project and our field studies of water management in the region. Actual records
of losses and uses due to non-agricultural, residential or tourism causes were used to
model dictated uses. Over the eight-year records available for the case study these flows
averaged approximately 10% of total releases with relatively low variability. They
therefore are assumed to not influence our release policy conclusions. Through extensive
field work we gathered reliable data for calibrating the majority of the model’s exogenous
variables that describe a particular water resource system, including time series data on
rainfall, population, tourist arrivals as well as the nonlinear crop water to yield
relationships, maximum yields for different crops throughout the year and average length
of tourist visits. Eight years of monthly historical data (1988 - 1996) for total releases
from the reservoir were collected and reservoir storage volumes were calculated from
13
actual inflow, losses and releases to test our model's ability to replicate actual system
behavior. Figure 3 shows the actual and simulated reservoir storage. Simulations used the
Euler integration method and weekly time step.
Reservoir Storage
70 000 000
ra
§ 60.000 000 + 5 4 2
& 50000 000
‘Y 40.000 000 Model
g [~ a
Real
30.000 000 | —7- Real
20 000 000
10 000 000 + ' + ™ + ' t +
0 12 24 36 48 60 R 84
Months
Figure 3: Actual and Simulated Reservoir Storage
The simulated behavior closely matches the behavior pattern (shapes, timing and
amplitudes) of the actual system behavior with acceptable error (R?=96%). Disaggregation
of the error using Theil statistics (Sterman, 1984) reveal the majority (92%) of the error is
due to covariation and not variation or bias. This indicates that differences between
simulated and actual behaviors are due primarily to mismatches between individual
simulated and data points and not due to a systemic bias (vertical translation)or
exaggeration of amplitudes. Comparison of model and actual behavior for total releases
shows more but still acceptable error (R?=53%) and high (89%) covariation. Andersen
(1998) describes additional tests of model structure and behavior used to develop
confidence in the model's ability to simulate water system behavior from the underlying
structural drivers and for analyzing water allocation policies.
MODEL ANALYSIS
Sensitivity Analysis
Sensitivity analyses were performed to identify system components which most influence
system behavior. Tests were limited to components which managers can reasonably
influence. For example managerial expectations were tested but population was not. Water
storage was simulated for a base case and pessimistic and optimistic values of 19
parameters, reflecting the modelers’s estimate of the 90% confidence band. Performance
14
ranges relative to the base case performance when test values were individually set to the
pessimistic and optimistic values were used to identify influential parameters. See Ford
(1995) Mahieu (1998) and Andersen (1998) for detailed examples of our approach to
sensitivity analysis. water storage was found to be most sensitive to the total demand
expected by managers, managerial response to coverage in determining releases, crop
efficiency of water use and manager's desired supply coverage. The high sensitivity of
performance to managerial expectations and beliefs supports our hypothesis that these
policies are important in improving water resource system management. In particular, two
of the sensitive parameters (response to coverage and desired coverage) describe the
degree of risk aversion in managerial decision-making.
The same analysis procedure was used to analyze the sensitivity of crop production to
those parameters most directly related to them. All three crop types are most numerically
sensitive to maximum yield, cultivated area of the specific crop type, and the crop’s
efficiency of water use. This is consistent with the findings of authors who have promoted
the use of more efficient conveyance and distribution systems in combating water
shortages (Makin, 1982; Mill, 1995; Mishalani, 1988; Postel, 1992, 1989; Roodman,
1996; Van Tuijl, 1993; Xie, 1993). These factors are all impossible or very difficult for
water managers to influence.
Base Case Analysis
A base run was generated using the calibration conditions of the model except that
dictated releases were held constant to focus on policies for residential, tourist and
agricultural use described above. Therefore the policy in the base case reflects the policies
used in Cyprus during the simulation period. The behavior of the storage of the base case
is close to the calibration and testing case (Figure 3) in both shape and numerical values.
The optimum yield for any crop requires the proper mix of soil conditions, sunlight,
evapotranspiration, etc. as well as adequate water. Similarly, families, businesses and
tourist areas respond to water availability in ways not included in the model such as the
common Cypriot practice of filling rooftop water tanks with rain or tap water during wet
times to provide short-term relief in times of severe water shortages. Because these factors
impact actual optimal performance and are beyond the scope of our model simulated
15
performance values are not predictions of performance in real systems. To evaluate
different policies we compare the performance of the system when managed using specific
policies to the optimal performance possible in the model. By simulating and measuring
performance over an eight year period that includes both times of adequate and inadequate
supply we capture both the ability of policies to fill current demands in different naturally
occurring supply conditions. We measure the ability of policies to reserve adequate supply
for future use with the stored supply at the end of the eight year simulated period using the
policy.
The performance of the system using the base case policy is shown in Table 1. Of the three
crop types modeled citrus performed best in the base run policy. Citrus farmers lost 25.6%
of maximum yield because of water shortages, compared to 67.0% and 65.6% for
greenhouses and potatoes, respectively. Residential users suffered an average monthly
deficit of 2.64 cubic meters per month per capita over the 8 years, while tourism was
particularly hard-hit with 90.6% of the months (87 of 96) experiencing a water shortage.
Performance Measure Units Base Case Optimum Variance
Performance: Performanc«
Agricultural Water Use
Citrus yield (8-year average) tons/ha 37.2 50 -25.6%
Citrus yield range tons/ha 31.0 0 31.0 tons/ha
Seasons with zero citrus yield each 0 0 0
Greenhouse yield (8-year average) tons/ha 12.2 37 -67.0%
Greenhouse crop yield range tons/ha 18.1 0 18.1 tons/ha
Seasons with zero greenhouse crop yield 0 0 0
Potato yield (8-year average) tons/ha 12.1 35 -65.6%
Potato yield range tons/ha 20.9 0 20.9 tons/ha
Seasons with zero potato yield each 0 0 0
Residential W ater Use
Average residential supply shortfall m3/month/capite 2.64 0 2.64 m3/montk
Tourism Water Use
Months tourist supply shortfall each 87 0 90.6%
Table 1: Performance using Base Case Policy
Reservoir Storage began to recover after 1983 but neither domestic or irrigation supply
matched or exceeded demand thereafter. We hypothesize that a primary cause of these
consistent shortages over the eight years is the risk averse nature of the manager's policy.
Cypriot water managers showed a tenacious adherence to risk averse policies. During
interviews in 1997 they repeatedly mentioned the 1990-1991 drought when inflows to the
16
Kouris Dam and to all the dams on Cyprus were unusually low and stressed the need to
“assume the worst” about inflows and to consider the next few years even in “good”
years, in case such a severe shortage happens again.
We tested our hypothesis and the effects of the risk adverse policy used by managers by
simulating performance using a less risk averse policy. We modeled a less risk averse
policy by decreasing the expected total demand and the desired water coverage and a
steeper slope of the effect of coverage on releases. The performance of this less
conservative policy is better than performance using the current, risk averse policy (table
1). This shows that less risk adverse water allocation policies than are currently used in
semi-arid regions can perform better. Performance using other parameter values to
describe more or less risk averse policies provides consistent results. These results support
our hypothesis. The seemingly reasonable practice of larger releases in times of plentiful
supply and restricted releases in times of inadequate supply contributes to the
inconsistency in releases that Cypriot water managers want to avoid. Overly conservative
policies can fill current demands fairly well but increase vulnerability to future droughts. A
policy that reflects the variability of natural supply (and therefore vulnerability to drought)
is needed.
Conclusions
Our findings are counterintuitive because they reverse what appears to be a logical
approach to managing water in areas of shortage. Indeed, the ultimate purpose of building
supply-related infrastructure such as reservoirs, pipeline networks, desalination plants, etc.
is to allow people to control their own water supply. When drought sets in, restrictions on
releases are thought of as temporary coping mechanisms that will be withdrawn when
“normal” conditions return. water managers in areas of inadequate and variable supply
institute release policies that supply to demand conditions in wet years and fall back on
restrictions in dry years. The infrastructure gives the impression that supplying to meet
demand will be the normal operating procedure, while droughts will sometimes interfere
and cause releases to be lower that normal.
This research indicates that in areas of severe water shortage, the chances of providing
consistent releases increase if policies are centered around what can actually be supplied,
17
rather than how and when can we finally release enough water to satisfy all demand. Any
policy that moves away from “rewarding” a population who suffered through drought by
giving them more water when possible will probably have to be combined with
information campaigns to inform and gain the support of the public. It will definitely need
the support of the command centers in government so they do not interfere.
Our results should be considered within the context of the limitations of our model, which
teflect a single-reservoir system without water supplied by sources other than rainfall. This
research also ignored allocation policy that is concerned with distributing water fairly
between competing users. Water was always distributed to a particular demand center
according to its percentage of the total demand. Thus, policies that favor one type of
demand over another, for example domestic demand over irrigation demand, were not
investigated. These policies have great impact on society, particularly the income
distribution of a region. They are arguably more politically charged than the policies
investigated in this research. Providing cheap irrigation water via subsidies, for instance, is
a policy often strongly discouraged in academic literature and in reports from such
organizations as the World Bank and the Food and Agriculture Organization of the United
Nations. Subsidies are still in widespread use, however, and not only for water. This
research can be expanded to investigate the long-term impacts of such prioritization of
sectors on the ability of water managers to provide consistent and reliable supply to all
users
Future variations of the model can also investigate the impact of increasing the efficiency
of supply as well as the situation in which water managers work closely with demand
centers to improve efficiency, timing of demand, and other things not directly under water
managers’ control. Similarly, the model could be expanded to include water conservation
measures. The actions of using water more efficiently through better infrastructure and
water-saving devices could be an extra source of supply in water scarce areas. Along the
lines of Amory Lovins’ “nega-dam” revolution, the water saved could help the quest for
consistent supply by providing relief from situations of overstressed water supply. By
improving our understanding of how managerial policies impact water resource system
performance we can improve that performance and provide the water needed for
sustainable development and prosperity.
18
REFERENCES
Al-Ibrahim, A. (1990) "Water Use in Saudi Arabia: Problems and Policy Implications" ASCE Journal of
Water Resources Planning and Management, Vol. 116, No. 3, May/June 1990, pp. 375-388.
Andersen, J. 1998. Water Allocation in Arid and Semi-Arid Regions. master's thesis. Department of
Information Science. University of Bergen. Bergen. Norway.
Annual Report, Cyprus Tourism Organization, 1995.
Belaineh, G. Peralta, R. and Hughes, T. (1999). "Simulation/Optimization Modeling for Water Resources
Management" ASCE Journal of Water Resources Planning and Management, Vol. 125, No. 3, May 1999,
pp. 154-161.
Brooks, David B., (1997) "Water Demand Management: Conceptual Framework and Policy
Implementation" in Management of Water Demand in Africa and the Middle East: Current Practices and
Future Needs, edited by David B. Brooks, Eglal Rached, and Maurice Saade, International Development
Research Centre, .
Cyprus— Southern Conveyor Project Case Study, Food and Agriculture Organization of the United
Nations, Rome, 1986.
Ford, D. (1995). The Dynamics of Project Management: An Investigation of the Impacts of Project
Process and Coordination on Performance. Doctoral dissertation. Massachusetts Institute of Technology.
Cambridge, MA.
Ford, D. and Sterman, J. Modeling Dynamic Development Processes. System Dynamics Review. 14(1),31-
68. Spring, 1998.
Grimble, RJ. and M. Archimandritou, Southern Conveyor Project Feasibility Study: Agriculture
Economic Analysis, Department of Water Development, Vol 16, 1982.
Grimble, RJ. and M. Archimandritou, Southern Conveyor Project Feasibility Study: Domestic Water
Economics, Department of Water Development, Vol 18, 1982.
Grimble, RJ. and M. Archimandritou, Southern Conveyor Project Feasibility Study: Project Economics,
Department of Water Development, Vol 19, 1982.
Hamdy, A., Mahmoud A.F., and C. Lacirignola, (1995). "Water Crisis in the Mediterranean: Agricultural
Water Demand Management", Water International, 20, 176-187.
Haten-Moussallen, M., Gaffney, B., Cox, C. and Batho, M. (1999) "Solutions to Water Scarcity in Cyprus
- A Proposal for Water Banking" Water Resources Group. Department of Civil and Environmental
Engineering. Massachusetts Institute of Technology.
Jacobs, J. and Vogel, R. (1998). "The Optimal Allocation of Water Withdrawls in a River Basin" ASCE
Journal of Water Resources Planning and Management, Vol. 124, No. 6, Nov/Dec 1998, pp. 357-363.
Jenkins, M.W. and Lund, J. R. (in press) “Integrated Y ield and Shortage Management Under Multiple
Uncertainties” ASCE Journal of Water Resources Planning and Management
Keshari, A. (2000). Discussion of "The Optimal Allocation of Water Withdrawls in a River Basin" ASCE
Journal of Water Resources Planning and Management, Vol. 126, No. 1, Jan/Feb 2000, pp. 37-38.
19
Kypris, D. C. and Panayiotis N. (1994). Monthly River Flows in Cyprus, Department of Water
Development,
Lovins, Amory, The Negawatt Revolution: Solving the CO2 Problem, Keynote Address, Green Energy
Conference, Montreal, 1989.
Mahieu, Laurent, An Operationalisation of the Resource-Based View (RBV) of the Firm, Masters Thesis,
University of Bergen, Bergen, Norway, 1997.
Makin, MJ., Southern Conveyor Project Feasibility Study: Irrigated Agricultural Development,
Department of Water Development, Vol 4, 1982.
Meadows, Dennis L. The Dynamics of Commodity Production Cycles. Wright-Allen Press. Cambridge,
MA: USA.
Mill, G.A. andJ.A. Theophilou, Effluent Re-Use in a Tourist Resort— Larnaca Sewage Treatment and
Irrigation Project— Cyprus, Water Science Technology, 32(9), 115-124, 1995.
Mishalani, N.R. and R.N. Palmer, Forecast Uncertainty in Water Supply Reservoir Operation, Water
Resources Bulletin, 24(6), 1237-1245, 1988.
Oliva, R. and J. Sterman. (1999). "Cutting Corners and Working Overtime: Quality Erosion in the
Service Industry." Harvard Business School Working Paper #99-138.
Papachristodoulou, S., Papayiannis C. and G.S. Panayiotou, Norm Input-Output Data for the Main Crop
and Livestock Enterprises of Cyprus, Cyprus Agricultural Research Institute, 1992.
Papandreou, C., Southern Conveyor Project: Crop Production Inputs, Department of Water Development,
1981.
Postel, Sandra, Last Oasis: Facing Water Scarcity, W.W. Norton & Company, 1992.
Postel, Sandra, Water for Agriculture: Facing the Limits, Worldwatch Institute, 1989.
Roodman, David Malin, Paying the Piper: Subsidies, Politics and the Environment, Worldwatch Institute,
1996.
Shawwash, Z.K. and $.0.D. Russell, Use of System Dynamics for Managing Water in Jordan,
Proceedings of the System Dynamics Conference, 1996.
Sheer, D. Ulrich, T. and Houck, M. (1992). "Managing the Lower Colorado River. ASCE Journal of
Water Resources Planning and Management, Vol. 118, No. 3, May/June 1992, pp. 324-336.
Simon, H. A. (1995). The Sciences of the Artificial. The MIT Press. Cambridge, MA.
Socratous, G., Hadjipetrou, P. and O. Evripidou, Southern Conveyor Project Progress Report, Department
of Water Development, 1996.
Sterman, John D., Appropriate Summary Statistics for Evaluating the Historical Fit of System Dynamics
Models, Dynamica, Vol. 10, Winter 1984.
Stiles, Geoffrey, Demand-side Management, Conservation, and Efficiency in the Use of Africa's Water
Resources, in Water Management in A frica and the Middle East: Challenges and Opportunities, edited by
Eglal Rached, Eva Rathgeber, and David B. Brooks, International Development Research Centre, 1997.
U.S. Water News Online (1996) "United Nations conference to address global need for adequate supplies
of clean, fresh water" April 1996. http://www.uswatemnews.com/archives/arcglobal/6unwater.html
20
Van Tuijl, Willem, Improving Water Use in Agriculture: Experiences in the Middle East and North
Africa, The World Bank, Washington, D.C., 1993.
Water Development in Cyprus, Water Development Department, 1996.
Wilchfort, O. and Lund, J. (1997). "Shortage Management Modeling for Urban Water Supply systems.
ASCE Journal of Water Resource Planning and Management. 123 (4), July/August 1997, pp. 250-258.
Wurbs, R. "Water Availability under the Texas water rights system" Journal of American Water Works
Association. 89,(5): 55-63. May, 1997.
21
