You are on page 1of 11

Software for Optimal Integrated Water Resources Management

Richard C. Peralta, :PhD, PE, Member


Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

Water Dynamics Laboratory, Utah State University Research Foundation,


Utah State University, Logan, UT 84322-4105, USA
e-mail: richard.peralta@usurf.usu.edu

Abstract

To best integrate management of available water one should coordinate the


use of groundwater and surface water in time and space. This requires: data
distributed in space and time; the ability to predict system response to stimuli;
clearly formulated management goals, constraints, and optimization problem
scenarios; the ability to compute optimal management strategies; and ways of
implementing the strategies. A strategy is a set of controllable water flows, such as
groundwater extraction and injection rates, surface water diversions, and reservoir
releases. Simulation/optimization (S/O) models link simulators and optimizers to
compute optimal strategies for user-specified management problems. S/O models
are increasingly used for policy development, reconnaissance studies, planning,
design, and management. S/O models can help determine the policies, plans,
systems, and management that can yield the best consequences. ‘Best’ is defined
by the manager/modeler via the formulated optimization problem to be solved.
Best is usually defined in terms of water availability, sustainability, crop
production, economic, social, or environmental criteria, or combinations.
Developing optimal multi-objective solutions and tradeoff surfaces is simple with a
powerful S/O model such as SOMOS. Examples demonstrate S/O model power
for integrated water resources management. Capabilities of Simulation
Optimization MOdeling System (SOMOS) are discussed.

Introduction

Simulation models are used to predict physical system response to stimuli.


Included stimuli can be groundwater pumping, recharge, stream diversion, return
flow, reservoir releases, and flows in canals and drains. Some stimuli can be
managed directly. Simulation/optimization (S/O) models help determine the best
manageable stimuli values (the best management strategy). An S/O model
computes strategies that maximize achievement of user-specified goals while
satisfying specified constraints. An S/O model links: a simulation module that
predicts consequences of management; and an optimization module that calculates
the mathematically best management strategy for a specified problem. An
example integrated water management strategy is a is a set of spatially and
temporally distributed pumping, diversion, and reservoir release flow rates.

S models (such as MODFLOW, MOC and HEC) predict how a physical system
will respond to a user-input management strategy. S models are not designed to
compute optimal management strategies. Using them for that purpose requires

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
trial and error and can yield the mathematically best strategy only for simple
problems. S/O models employ S models (analytical, numerical, or substitutes) to
predict system responses. S/O model prediction accuracy is the same as that of
the employed S model.

For relatively homogeneous or simple systems, S/O models use analytical


equations for simulators, and usually use classical operations research (OR)
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

algorithms as optimizers. Employing analytical equations is simpler than using


numerical (finite difference or finite element) S models. Peralta and Wu (2004)
describe S/O model applications for representative simple integrated water
management problems. S/O models for integrating water management for larger
heterogeneous systems use numerical flow S models as simulators. Peralta and
Shulstad (2004) describe water policy evaluations for such cases. Optimally
managing water quality and quantity (flow and transport optimization) requires
S/O models to use flow and transport simulators. Peralta (2001a) and Peralta et al
(2003) list such examples, using the SOMOS code (SSOL, 2001; Peralta, 2003).
Peralta (2001a) and Peralta et al (2003) describe several direct comparisons
between designs developed by S/O modeling versus designs prepared
simultaneously by trial-and-error S modeling. S/O modeling always produced
superior designs, usually about 20 % better, but sometimes about 50% better.

Both S and S/O models need enough data for predictive simulation. S/O models
also need data to specify management goals and constraints. Examples include
distributed quantitative and qualitative data of soil, geology, existing and potential
water, water use, and limits.

In summary, S/O models are useful for a range of groundwater and conjunctive
water management settings. Here are described three settings: (a) sites having
limited field data, suitable for analytic equation simulation; (b) stream-aquifer
systems needing numerical flow modeling; and (c) reservoir-aquifer-stream
settings needing numerical modeling. Discussed are capabilities of the SOMOA,
SOMO1, and SOMO3 modules of Simulation/Optimization Modeling System
(SOMOS), (SS/OL and HGS, 2001; Peralta, 2003), and a developmental model.

Conjunctive Use in Simple Stream-Aquifer System with Salinity and Stream


Depletion Constraints

This example illustrates maximizing total blended groundwater plus surface


water to achieve appropriate salinity for irrigation (Peralta, 1999). A farmer uses
one groundwater well and one stream diversion point. He wishes to maximize total
water delivered to his crop during two months. To protect downstream water use,
his use should not reduce stream flow by more than 11,000 m3 d -1 (385,000 ft3d -1)
at the end day 30, or by more than 11,500 m3 d -1 (402,500 ft3d -1) at the end of day
60. Maximum flows of the well and the diversion are each 8,000 m3 d -1 (280,000
ft3d -1). The most water his crop can reasonably use is 13,000 and 16,000 m3 d -1
(455,000 and 560,000 ft3d –1) in the two respective months.

Pertinent information (including x,y location in meters) is: groundwater well (0.2 m
radius), is at (450, 850); hydraulic conductivity is 80 md -1 ; Ground surface is at 45
m elevation; potentiometric surface is initially at equilibrium at 40 m elevation;

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
aquifer saturated thickness is 40 m; stream runs from Southeast (800, 0) to
Northwest (100,1000); and diversion location is at (200,858). Based on salinity, the
stream should provide at least 60 % of the water used during the first month, and
the stream should provide at least 48% of total two-month water use. These
constraints protect seeds during germination, and prevent root-zone salinity buildup,
respectively.
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

The SOMOA (Peralta and Wu, 2004) module can compute the optimal strategy.
SOMOA, which evolved from CONJUS, uses analytical equations and convolution
integrals for simulation and a simplex procedure for optimization.
One would specify: one extraction well; one diversion; two thirty-day stress
periods; pumping and diversion upper limits of 8,000 m3 d -1 in each period;
water quality ratio {diversion/(diversion + pumping extraction)} lower limit of 0.6
for period 1, and 0.48 for the two-month total; stream flow depletion upper limits
of 11,000 and 11,500 m3 d -1 , for periods 1 and 2; and pumping plus diversion
upper limits of 13,000 and 16,000 m3 d -1, respectively.

The SOMOA-computed optimal conjunctive use strategy for months 1 and 2,


respectively, is 4,774 and 8,000 m3 d –1 groundwater, and 7,001 and 4,573 m3 d –1
surface water. Total delivered water is 11,774 and 12,573 m3 d –1 for periods 1
and 2, respectively, and 12,174 m3 d –1 total. Stream flow depletion rate
constraints are tight at 11,000 and 11,500 m3 d –1, respectively. Water quality
ratios are 0.6 for period 1 (tight), 0.36 for period 2, and 0.48 for the season (tight).
Sustainable Ground-water and Conjunctive Use for Heterogeneous System

The Arkansas Grand Prairie covers part of the Mississippi Alluvial Aquifer
(Figure 1). This important rice, soybean and aquaculture producing area has
historically obtained most of its water from part of the Quaternary Mississippi Plain
alluvial aquifer. Declining potentiometric surface elevations prompted groundwater
modeling and conjunctive use analysis. Figure 2 shows a Grand Prairie finite
difference model grid.

Table 1 compares water use and short-term results of five formulations or scenarios.
The Historic Use scenario involves continuing historic use for 10 years. Scenarios
I-IV used constraints to assure that computed groundwater pumping was
sustainable, transient flow simulation, and economic evaluation.

Scenario I maximizes the common proportion, , of current ground-water


withdrawals that can be sustainably pumped from each cell (the largest for which
a solution to the set of steady state equations does not violate any bounds.) To
implement the strategy, the computed percentage reduction of current pumping
equals 100 times ((1- ). This represents applying the correlative rights water law
doctrine.

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

Fig. 1 Mississippi Alluvial Aquifer and groundwater study areas: (A) Bayou
Bartholomew Basin, and (B) Grand Prairie (Peralta and Shulstad, 2004).

Fig. 2 Grand Prairie groundwater model grid (Peralta and Shulstad, 2004).

Constraints for the four optimization scenarios include: lower bounds on


potentiometric head in each cell sufficient to cause at least 20 feet of saturated
thickness; bounds on recharge in peripheral cells to prevent unacceptable
dewatering of boundary rivers and nearby aquifer; and upper limits on groundwater
extraction to avoid pumping more water than is needed in each cell. Pumping upper
bounds are changed in some scenarios to show conservation measures or
availability of diverted surface water.

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
The goal for Scenarios II-IV is to maximize total groundwater extraction. Optimal
pumping strategies differ because the upper bounds on pumping differ. Scenario II
does not allow new water conservation measures or surface water diversion.
Scenario III allows water conservation but no diversion. Scenario IV allows
conservation and diversion.

Scenarios I and II show that historic groundwater use is not sustainable. Scenario I
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

shows that the smallest global pumping reduction needed to achieve sustainability
is 86 percent. Scenario II, not egalitarian, but would still require significant
reduction in groundwater use. Scenario III shows that without surface water
diversion, about half of the water need will be unsatisfied.

Scenario IV best satisfies water need. Even with conservation and diverted surface
water, net return would decline by 23 percent. Omitting either action will cause at
least a one third reduction in net return. Table 2 shows a representative derivative
temporally distributed monthly conjunctive use strategy.

Above results were computed using an transferable S/O model that had limited
optimization capabilities (currently one would use SOMOS to develop the optimal
strategies). This example is presented because it involves sustained groundwater
yield and integrated water use reconnaissance and policy evaluation. Partially as a
result of this and related work: the Arkansas State Water Plan included concepts of
sustained groundwater yield planning in special groundwater management districts;
canals were built to divert river water to the Grand Prairie; and integrated water use
now occurs. SOMOS or its immediate precursors have been reported for other
regional planning optimizations (Belaineh and Peralta, 1995; Das, 2002; Das et al,
2004; Peralta and Shulstad, 2004).

Table 1 Optimal strategies and short-term annual economic consequences (Peralta


and Shulstad, 2004)
Scenario
Historic
groundwater use I II III IV
(base strategy)
Water Need 286 286 286 253 253
(353) (353) (353) (312) (312)
Groundwater 286 38 118 115 62
Use (353) (47) (146) (142) (76)

Surface Water 0 0 0 0 160


Use (197)

Unmet Water 0 248 168 138 31


Need (306) (207) (170) (39)

Change in Net NA -6,985 -4,066 -2,634 -1,948


Economic
Return
Water units are 1000 ac-ft and (106 m 3). Economic return units are 1000 dollars.

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
Table 2 Monthly crop water needs and surface water and ground-water use

Month Monthly % Monthly % Monthly % % of monthly water


of annual of annual of annual needs provided
water needs surface ground Surface Ground
water use water use water water
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

Apr. 5 5 1 99 1
May 7 9 1 99 1
Jun. 29 33 4 98 2
Aug. 32 24 82 64 36
Sep. 2 2 1 99 1
Total 100 100 100
annual

Optimizing Reservoir-Stream-Aquifer (RSA) System Use

Many models have been developed for optimizing reservoir management and
for conjunctive water management. However, generally the conjunctive use models
that have detailed stream/aquifer (s/a) system interactions have not considered
reservoir management rules. Typically, reservoir management models have not
simulated s/a interactions in detail.

To address this, Belaineh et al (1999) performed detailed RSA optimization that


formally included reservoir operating rules. Figure 3 shows significant flows
simulated in their work. Figure 4 schematically illustrates their study area. Figure 5
shows the benefit of increasingly detailed representation of stream-aquifer seepage
within a RSA S/O model—Scenario II-3 best used the water released from the
reservoir for the intended purpose of irrigation.

Belaineh et al (1999) included a salinity constraint to assure adequate blended water


quality for irrigation. Ejaz and Peralta (1995a,b) mention others that have employed
detailed transport simulation in surface water or ground water in S/O modeling.

Fayad and Peralta (2004) optimized management of a hydraulically linked detailed


RSA system, assuming that using mathematical optimization could avoid inclusion
of separate reservoir operating rules. This is valid if the optimization is for a
sufficiently long optimization period and includes constraints that replace normal
reservoir operating rules. They used multi-objective GA with ANNs for
optimizing conjunctive use in a hydraulically connected RAS system. ANN use
reduced simulation computer processing time and made it practical to address
more complex optimization problems. They were able to develop trade-off
surfaces for maximizing hydropower versus maximizing water delivery versus
minimizing water delivery cost (Fig. 6).

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

Fig. 3 Flows modeled in detailed RAS model reported by Belaineh et al (1999).

Fig. 4 RAS study area modeled by Belaineh et al (1999).

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5 Effect of detail in RAS simulation on reservoir performance (Belaineh et al


1999).

85000 Potential
80000 Energy
(kWh)
75000
241,920
Total 70000
239,040
236,160
Cost 233,280
($) 65000 230,400
227,520
224,640
60000 221,760
218,880
55000 216,000
213,120
210,240
50000 207,360

45000

40000
40000 45000 50000 55000 60000 65000 70000
3 -1
Total Water Delivered (f d )
Fig 6 Tri-objective (water cost, water delivered, hydropower) trade-off surface
(Fayad and Peralta, 2004).

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
SOMOS Features

Many S/O models have been developed by researchers. Most are for a
specific site or application. SOMOS represents an effort to make a transferable
family of S/O modules that can address a broad range of water management
issues. SOMOS results from experiences with water management optimization
modeling from small field scale to problems having over one thousand decision
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

variables, and thousands of constrained state variables (Peralta, 2001b). Some of


the complicated simulation models have employed many stress periods and
required extensive computer simulation time. SOMOS includes many techniques
for optimizing strategies for complex linear and nonlinear optimization problems.

SOMOS takes advantage of the fact that different S/O approaches are best for
different problem types. SOMOS has modules tailored for different problem
types. For ease of use, data input formats are as similar as possible for all
modules. Modules employ proven numerical flow and transport, and substitute
response surface simulators. SOMOS versions and modules include a variety of
simulators (analytical equations, public domain numerical models, substitutes) or
can run with any simulator that produces text file output. Module SOMO1 also
uses response matrix (RM), polynomial and other response functions as surrogate
simulators. SOMO3 uses artificial neural networks (ANNs) or any user-defined
simulator as surrogates. SOMOS verifies the accuracy of surrogate simulations.

SOMOS also includes 14 diverse optimization algorithms (many of them world-


famous) to allow the user to utilize defaults or hand-select the best solver fora
particular problem. SOMOS lets the water professional best employ his expertise
in the pursuit of optimal water management strategies. SOMOS provides features
to insure that the professional does not need to become an optimization expert.

Standard and Advanced SOMO1 versions are ideal for hydraulic optimization
problems up to huge sizes having thousands of decision and state variables.
SOMO1 can include transport optimization. SOMO1 optimizes using operations
research (OR) algorithms (such as simplex, gradient search, branch and bound,
outer approximation). SOMO1 includes over 10 embedded commercial solvers,
such as MINOS and DICOPT. SOMO1 solveslinear, nonlinear, mixed integer,
mixed integer nonlinear, multi-objective and stochastic optimization problems.
SOMO1 can link with other models to facilitate large-scale economic modeling.

SOMO3 addresses the same deterministic, stochastic, and multi-objective


optimization problem types as SOMO1, but uses heuristic optimization (HO)--
genetic algorithm (GA), simulated annealing (SA), and integrated tabu search
(TS). SOMO3 optimization can employ artificial intelligence to speed
convergence to optimality. It is ideal for water and contamination management.
Addressed problems have had many stress periods and variables, but have been
smaller than those addressed by SOMO1.

All together, SOMOS can: optimize for 90+ distinct management goals (objective
functions) plus user-defined objective functions; and constrain all pertinent
variables (examples include pumping, stream diversion, flows, cell head, head just
outside well casing, concentration, user-defined). SOMOS has unique tools,

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
including stochastic optimization, for increasing strategy robustness and reliability
under uncertainty. SOMOS also easily does multi-objective optimization and aids
trade-off curve preparation.

SOMOS runs on personal computers, with or without a provided interface. It


allows the user to operate other programs while SOMOS is optimizing. SOMOS’
user-friendly interface speeds input preparation, editing I/O files, and
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

optimizations. Automated processing speeds sequential running of multiple


optimization actions, and permits run restarts and result merges. It has detailed
input error-checking and error messages. SOMOS has also been included within
commercial user interface packages.

Summary

Simulation/Optimization models are becoming more flexible and powerful,


leading to their increased use for aiding water policy-making, planning, systems
design, and management. Different types of simulators and optimizers are better
for different situations. SOMOS contains a range of simulators and optimizers in
its several modules. The SOMOA module is best for field settings where
analytical flow equations are appropriate. SOMOA uses analytical and
convolution (superposition) equations as simulators and classical operations
research optimizers (simplex, branch and bound, and gradient search algorithms).

The SOMO1 module is appropriate for optimizing heterogeneous stream-aquifer


systems describable via numerical flow models. SOMO1 uses classical
optimization algorithms. SOMO3 is best for heterogeneous contaminated
systems, where concentrations must be manageable. SOMO3 employs numerical
flow and transport simulators and heuristic optimizers. Together they constitute
the general SOMOS releases for aquifer and stream-aquifer systems and contain
MODFLOW, MT3DMS and 14 optimization algorithms.

SOMO1WEB and SOMO3 WEB are free versions unlimited with respect to study
area size, but restricted with respect to optimization features. These can be
downloaded at http://www.usurf.org/units/wdl .

Standard and advanced SOMOS versions employ more simulators, including


SEAWAT for salt water intrusion, and have more optimization capabilities. The
SOMO4 module employs user-supplied simulators and variables and is awaiting
application to problems combining features addressed by Ejaz and Peralta
(1995a,b), Belaineh et al (1999), and Fayad and Peralta (2004).

SOMOS and its user’s manual are marvelous tools for novice and experienced
optimizers. A relatively new modeler can compute optimal management
strategies. An experienced hydrologist can craft magnificent strategies. SOMOS-
designed strategies are commonly 20-40% percent better than those developed
using simulation model and trial and error (Peralta et al, 2002). The greatest
improvement has been 58%. SOMOS utility is being enhanced as it is interfaced
with powerful commercial visualization and modeling systems.

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change
References

Belaineh, G. and R.C. Peralta. 1995. Considering ecological constraints while


optimizing sustained groundwater yield, Pahvant Valley, Utah. Proceedings
AWRA Conference, Water Conservation in the 21st Century: Conservation,
Demand and Supply. Salt Lake City, Utah, p. 101-110.
Belaineh, G., Peralta, R.C. and T.R. Hughes. 1999. Simulation/optimization
Downloaded from ascelibrary.org by LULEA UNIVERSITY OF TECHNOLOGY on 03/28/16. Copyright ASCE. For personal use only; all rights reserved.

modeling for water resources management. Journal of Water Resources


Planning and Management, 125(3):154-161.
Das, R. 2002. Planning sustainable optimal groundwater yield for the Utah part of
Cache Valley. Irrigation Eng. MS thesis. Utah St. Univ., Logan, UT. 115 p.
Das, R., Peralta, R. C., and B. Timani. 2004. Cache Valley: optimizing
sustainable water use and ecosystems while considering water rights. In
Proceedings of EWRI 2004 World Congress.
Ejaz, M.S. and R.C. Peralta. 1995. Maximizing conjunctive use of surface and
ground water under surface water quality constraints. Adv.in Water
Resources. 18(2): 67-75
Ejaz, M.S. and R.C. Peralta. 1995. Modeling for optimal management of
agricultural and domestic wastewater loading to streams. Water Resources
Research. 31(4): 1087-1096
Fayad, H. C., and R. C. Peralta. 2004. Multi-objective conjunctive use
optimization. In Proceedings of EWRI 2004 World Congress.
Peralta, R.C. 1999. Conjunctive Use of Ground Water and Surface Waters for
Sustainable Agricultural Production. FAO of the UN Consult. Rep. 158p
Peralta, R. C. 2001a. Remediation simulation/optimization demonstrations. In
Proceedings of MODFLOW and Other Modeling Odysseys. 2001. Eds, Seo,
Poeter, Zheng and Poeter, Pub. IGWMC. p. 651-657.
Peralta, R. C. 2001b Simulation/optimization applications and software for
optimal ground-water and conjunctive water management. Ibid. p. 691-694.
Peralta, R. C., Kalwij, I. M., and S. Wu. 2002. Optimal P&T designs for Blaine
Naval Ammunition Depot. Proj. completion rep. for Navy. Systems
Simul./Optim.Lab., Bio. & Irrig. Eng. Dept., Utah St. Univ. 36 p.
Peralta, R. C., Kalwij, I. M. and B. Timani. 2004. Optimizing complex plume
pump and treat systems for Blaine Naval Ammunition Depot, Nebraska. In
Proceedings of EWRI 2004 World Congress.
Peralta, R. C. 2003. SOMOS Simulation/Optimization Modeling System. In
Proc., MODFLOW and More 2003, IGWMC, Golden, CO. p 819-823.
Peralta, R. C. and R. Shulstad. 2004. Optimization modeling for groundwater and
conjunctive use water policy development. In Proceedings of FEM-
MODFLOW Int.Conf., IAHS, Karlovy Vary, Czechoslovakia, Sep, 2004.
Peralta, R. C. and S. Wu. 2004. Software for Optimizing International Water
Resources Management. In Proceedings of EWRI 2004 World Congress,
ASCE, Salt Lake City, U.S.A. Jun, 2004.
Peralta, R. C., Kalwij, I. M. and S. Wu. 2003. Practical simulation /optimization
modeling for groundwater quality and quantity man. In MODFLOW & More
2003: Understanding through Modeling, IGWMC, Golden, CO., p 784-788.
Systems Simulation/Optimization Lab. and HydroGeoSystems Group. 2001.
Simulation/Optimization MOdeling System (SOMOS) users manual. SS/OL,
Bio. & Irrig. Eng. Dept., Utah State Univ., Logan, Utah. 457 p.

Copyright ASCE 2005 EWRI 2005


Impacts of Global Climate Change

You might also like