Calibration of On-Demand Irrigation Network Models

Miguel Ángel Moreno Hidalgo, Ph.D.1; Patricio Planells Alandi, Ph.D.2;

José Fernando Ortega Álvarez, Ph.D.3; and José María Tarjuelo Martín-Benito, Ph.D.4

Abstract: In this study, a new procedure for calibrating on-demand irrigation network models was developed. This procedure used a new
objective function called maximum data with a reasonable error 共MDRE兲 for calibrating the network. It was compared with the two more
commonly used objective functions in calibration procedures that are the simple least squares 共SLS兲 and the maximum likelihood
Downloaded from ascelibrary.org by Selcuk Universitesi on 12/24/14. Copyright ASCE. For personal use only; all rights reserved.

estimator for the heteroscedastic error case 共HMLE兲. In order to carry out the calibration, a quasi-Newton optimization method was used
having as variable the Hazen-Williams head losses coefficient 共C兲. This procedure was applied to an on-demand irrigation network located
in Tarazona de La Mancha 共Albacete, Spain兲 where flow and pressure at hydrant level was measured. The calibration procedure using the
MDRE objective function was applied considering all the pressure control points simultaneously and the obtained results were compared
with the results of considering the pressure control points independently. Therefore, the effect of the location of the pressure control point
was studied. Results showed that, when the proposed objective function was used, the root mean squared error 共RMSE兲 comparing the
measured and simulated data after calibration was lower than when the SLS or HMLE objective functions were used. The location of the
pressure control points throughout the irrigation network could affect the results; therefore, it was more accurate to use all the control
points simultaneously than independently in the calibration process.
DOI: 10.1061/共ASCE兲0733-9437共2008兲134:1共36兲
CE Database subject headings: Simulation models; Calibration; Head loss; Pipe networks; Irrigation.

Introduction A calibrated hydraulic model can be used to provide informa-

tion about the hydraulic behavior of a network in an identified
Hydraulic systems analysis consists of the study of the main hy- scenario, and can be a useful tool for solving network manage-
draulic parameters 共flow and pressure head兲 for networks working ment problems. To develop a calibrated hydraulic network, sev-
under several conditions. In order to develop an exhaustive study eral methodologies can be used. The test-error methodology is
of a network, it is necessary to have computational tools that can mainly used, but it is labor intensive and often yields unsatisfac-
carry out complete calculations for hydraulic conditions of the tory results. Optimization techniques have been developed 共Dias
network 共Lamaddalena and Sagardoy 2000兲. Several user friendly et al. 2000; Wu et al. 2004兲 that offer more accurate results. These
computer models have been developed for analyzing hydraulic optimization techniques generally involve the maximization or
networks. Among all those models, the three most commonly minimization of a user-defined objective function, which com-
used for irrigation networks are: EPANET 共Rossman 2000兲, pares the model output with measured values. The most com-
COPAM 共Lamaddalena and Sagardoy 2000兲, and GESTAR monly used objective function in optimization procedures is the
共Aliod et al. 1997兲. In this paper, EPANET was used because it is simple least squares 共SLS兲. However, many other objective func-
a solid model that has been tested worldwide for different types of tions can also be used, depending on the aim and characteristics
water distribution networks. of the optimization process. In this paper, a new objective func-
tion was developed that improves the results obtained in the cali-
1
Researcher, Agricultural Engineer, Centro Regional de Estudios del bration process.
Agua, Castilla-La Mancha Univ., Campus Universitario, E02071 Albac- The optimization problem for hydraulic network calibration is
ete, Spain. E-mail: MiguelAngel.Moreno@uclm.es highly nonlinear, but is made simpler because the response sur-
2
Professor, Industrial Engineer, Centro Regional de Estudios del
faces are generally convex without region of attractions, local
Agua, Castilla-La Mancha Univ., Campus Universitario, E02071 Albac-
ete, Spain. E-mail: Patricio.Planells@uclm.es minima, etc. Therefore, classical approaches, such as the quasi-
3
Professor, Agricultural Engineer, Centro Regional de Estudios del Newton method, are sufficient to achieve a calibrated model that
Agua, Castilla-La Mancha Univ., Campus Universitario, E02071 Albac- has good results without being computationally expensive.
ete, Spain. E-mail: Jose.Ortega@uclm.es The main goal of this study was to develop a new objective
4
Professor, Agricultural Engineer, Centro Regional de Estudios del function that improves model fitness in the calibration process. In
Agua, Castilla-La Mancha Univ., Campus Universitario, E02071 Albac- order to carry out the calibration process, a quasi-Newton optimi-
ete, Spain. E-mail: Jose.Tarjuelo@uclm.es zation method was used having as variable the Hazen-Williams
Note. Discussion open until July 1, 2008. Separate discussions must head losses coefficient 共C兲. This procedure was applied to an
be submitted for individual papers. To extend the closing date by one
on-demand irrigation network located in Tarazona de La Mancha
month, a written request must be filed with the ASCE Managing Editor.
The manuscript for this paper was submitted for review and possible
共Albacete, Spain兲 where flow and pressure at hydrant level was
publication on July 3, 2006; approved on May 29, 2007. This paper is measured. The calibration procedure was applied considering all
part of the Journal of Irrigation and Drainage Engineering, Vol. 134, the pressure control points simultaneously and the obtained re-
No. 1, February 1, 2008. ©ASCE, ISSN 0733-9437/2008/1-36–42/ sults were compared with the results of considering the pressure
$25.00. control points independently. Therefore, the effect of the location

36 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2008

J. Irrig. Drain Eng. 2008.134:36-42.

 of the pressure control points on the calibration method was measured over two irrigation seasons 共2003–2004兲, while pump-
studied. ing station data could only be collected during 2004.

Methodology for Calibrating Irrigation Network Models

Methodology
In the calibration process of network models, several aspects must
be considered, including changes in the construction of a system,
Case Study such as the introduction of new elements, different positioning of
The case study is an on-demand irrigation network that irrigates the pipes, diameter variation, and other factors, as well as pipeline
550.2 ha in Tarazona de La Mancha 共Albacete, Spain兲. Water is aging, changes in pipe roughness, and improperly closed valves.
obtained from five wells in the hydrogeologic unit 共HU兲 08.29, The steps used in the calibration process are:
located in the Júcar basin. Groundwater is conveyed to a reservoir 1. Data acquisition: The topologic and hydraulic data were ob-
with a capacity of 23,000 m3 from which a pumping station, com- tained from the network design document and by checking
posed of ten pumps with 140 HP each, conducts the water to the the network conditions in the field. Topographic data were
Downloaded from ascelibrary.org by Selcuk Universitesi on 12/24/14. Copyright ASCE. For personal use only; all rights reserved.

irrigation network. To guarantee a pressure head of 45 m in all the measured with a GPS, with a precision of one cm.
hydrants of the network, a head of 62 m 共manometric regulation兲 2. Data input to the model: The input data to the model were
is applied at the pumping station. the internal diameter and length of the pipes, the Hazen-
The irrigation network is composed of asbestos cement pipe- Williams head losses coefficient 共C兲, the topology and topog-
lines of large diameters 共400– 600 mm兲 for the main distribution raphy of the network 共500 nodes, 389 being hydrants兲, and
lines, and polyvinyl chloride pipelines 共63– 315 mm兲 for the re- the average pressure head and flow data for each of the eight
mainder of the system. Each plot has its own hydrant, with 389 daily acquisition data periods. Since it is very difficult to
hydrants in total. Hydrant diameter, which depends on plot size measure minor losses in a network, it was assumed that these
and number of sectors within each plot, varies from 2 – 6 in. with losses were included in the friction losses. In addition, the
a flow range of 2 – 60 l / s. Each hydrant has a flow limiter in order head losses coefficient 共C兲 is assumed to depend only on
to supply a discharge proportional to the plot area. Thus, all farm- pipe material. Once the flow in each pipe is determined for
ers have the same water availability. All hydrants are controlled each demand scenario and using a known head 共based on
using an automatic system that allows technicians to centrally manometric regulation of the pumping station兲, head losses
manage the network. The automation system consists of a central can be calculated to obtain the piezometric height in each
computer connected to 14 field control units. These units are node. To calculate the head losses, the Hazen-Williams equa-
small computers storing data from each of the remote terminal tion was used 关Eq. 共1兲兴
units 共RTUs兲, installed in the hydrants. Each field unit controls up
⌬h = 10.62 ⫻ C−1.85 ⫻ L ⫻ Q1.85 ⫻ D−4.87 共1兲
to 63 RTUs, thus controlling up to 63 hydrants.
Permanent solid-set sprinkler irrigation systems are used for where ⌬h = head losses 共m兲; C = Hazen-Williams head losses
95% of the area, the rest being irrigated by drip irrigation sys- coefficient; L = pipe length 共m兲; Q = pipe flow rate 共m3 s−1兲;
tems. and D = pipe internal diameter 共m兲.
The following data were measured from the on-demand irriga- 3. Provisional running of the model: The model should be run,
tion network: as a first step, using the default calibration parameter values
1. Discharge and pressure head data at the hydrant level: Dis- 共Table 1兲. It is useful for locating the pressure transducers
charge data were measured using the automation system, throughout the network in the experiment design process.
which records the output of the hydraulic valve flow meter. 4. Hydraulic data measurement: All authors consider that it is
The automation system records the total volume output of necessary to measure pressure head for different network
each hydrant eight times per day. Since in this case, irrigation working conditions 共low, medium, and high demand兲. La-
does not occur between 9:00 am and 7:00 pm, the periods of maddalena 共1997兲 considers measurements to be sufficient
data acquisition were concentrated during the time of maxi- taken at just one point, but for several demand conditions
mum probability of irrigation, in this case, at night. There- 共seven different working conditions兲. Allen 共1987兲 favors
fore, the periods of cumulative volume measurement at each minimum pressure control points at 20% of the total number
hydrant were from 7:00 pm to 10:00 pm, 10:00 pm to 11:50 of hydrants. Ferreri and Gangitano 共2001兲 suggest that it is
pm, 11:50 pm to 2:00 am, 2:00 am to 4.00 am, 4:00 am to better to increase the number of control points for a working
6:00 am, 6:00 am to 8:00 am, and 8:00 am to 9:00 am. The condition than to measure few points for several working
difference of volume between periods divided by the elapsed conditions. In this work, 13 strategic control points in the
time of the period is the average discharge of each hydrant. network were installed 共Fig. 1兲 after analyzing the results of
2. Pressure head data at the hydrant level: Pressure head data the provisional running of the model with standard values of
were measured at 13 selected points distributed throughout the head losses coefficient. Discharge data were obtained
the irrigation network. Location of these pressure head con- from the flow meter installed in each hydrant by using the
trol points was carried out by simulating the network with automation control system. Frequency of pressure head data
EPANET. Standard roughness values and minor losses equal measurement was 1 min, but average data of pressure head
to 20% of friction losses were used. Pressure transducers and flow were used for each of the eight periods of flow
were placed at those points with variable conditions in the measurement. Since the irrigation time in each plot was
pipes upstream, such as different material, different flows, at around 4 – 6 h 共much longer than the time between each flow
the end of pipes, at hydrants far from the pumping station, at measurement period兲, flow and pressure head variation dur-
hydrants with high elevation, etc. Pressure head data were ing the period was very low. If a high variation of flow or
stored in data loggers every minute. pressure head happened during the period, the model did not
Discharge and pressure head data at the hydrant level were consider this measurement, as will be shown herein.

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2008 / 37

J. Irrig. Drain Eng. 2008.134:36-42.

 Table 1. Standard C Values, Calibrated C Values Considering the Proposed Objective Function, and Average, Maximum, Minimum, and Standard
Deviation of the C Values for the Different Pipe Materials
“Cstandard” “Cstandard” “Ccalculated” “Ccalculated”
Year Month PVC asbestos PVC asbestos
2003 July 140–150 120–140 128.30 98.22
2003 August 140–150 120–140 126.92 96.57
2004 July 140–150 120–140 118.60 81.90
2004 August 140–150 120-140 123.57 98.11
Average 124.35 93.70
Maximum 128.30 98.22
Minimum 118.60 81.90
SD 4.32 7.90
Downloaded from ascelibrary.org by Selcuk Universitesi on 12/24/14. Copyright ASCE. For personal use only; all rights reserved.

5. Adjustment and calibration of the model: For each data ac- applying two common objective functions. The resulting cali-
quisition period, 12 head loss data values were obtained. The brated models from each objective function were compared by
head losses coefficient 共C兲 was adjusted with calibration pro- calculating the root mean squared error 共RMSE兲 for each. These
cedures to fit the measured data, as detailed below. two objective functions used for calibration are:
1. The maximum likelihood estimator for the heteroscedastic
error case 共HMLE兲. This objective function assumes that the
Calibration Procedures
errors are Gaussian with zero mean and variable variance
The analyzed irrigation network was implemented in Microsoft along the magnitude of the measured parameter. This estima-
Excel to use Solver tool in the optimization process. The objec- tor is mainly used when there is a variation of the errors with
tive function developed in this work was the maximum data with the magnitude of the measured parameter 共Sooroshian and
a reasonable error 共MDRE兲. This objective function tries to maxi- Dracup 1980兲. Eqs. 共3兲–共5兲 describe how to formulate this

再兺 冎再 冋兿 册 冎
mize the number of points for which the difference between mea- estimator. The HMLE equation is
sured and simulated head losses data is inside an error interval n n 1/n −1
delimited by the accuracy in the pressure head measurement 共1 m
Min HMLE = wt␧2t n wt 共3兲
in this case兲. The network is calibrated when the number of points t=1 t=1
inside the error range is maximized. It automatically eliminates
the data that are outside of the error interval. where ␧t = difference between measured head losses 共pobs兲
The objective function is expressed as and simulated head losses 共psim兲 values; n = number of mea-
surements; and wt = weight associated with the measured
j
parameter.
兺
k=1
k
Nm苸n The weight is computed by
Max 100 共2兲 wt = f 2共␭−1兲 共4兲
j⫻n t

where k = number of control points; n = total number of measure- where f t = expectation of the true value 共ptrue兲 共in this case it
ments for each control point 共k兲; and Nm苸n
k
= number of valid mea- is assumed that f t is equal to pobs兲; ␭ = unknown transforma-
surements 共m兲 with an error lower than the maximum established tion parameter that stabilizes the variance. The implicit ex-
error 共1 m in this case兲. pression to estimate ␭ is
To verify the validity of the proposed objective function, the
calibration results were compared with the results obtained by 冋兺 册冋兺 册 冋兺
n

t=1
ln共f t兲
n

t=1
wt␧2t −n
n

t=1
wt ln共f t兲␧2t = 0册 共5兲

2. Simple least squares 共SLS兲 is the most common objective

function used in hydraulic network calibration. In the cali-
bration process with this objective function, the main pur-
pose is to minimize the square of the difference between
measured head losses 共pobs兲 and simulated head losses 共psim兲
data 关Eq. 共6兲兴 so that
n

Min SLS = 兺
t=1
共pobs − psim兲2 共6兲

This objective function assumes that errors are distributed fol-

lowing a Gaussian distribution, are independent, and have homo-
geneous variance for the whole rank of measurements. It is a
specific case of the HMLE in which ␭ is equal to 1.
In order to determine when a model is calibrated, the maxi-
mum difference between measured and simulated head losses
Fig. 1. Network topology and distribution of the pressure control should be established. For irrigation network calibration, Walski
points 共T兲 共1983, 1985, 1986兲 recommends that this difference should be

38 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2008

J. Irrig. Drain Eng. 2008.134:36-42.

 Fig. 2. Relation simulated-measured values using HMLE for July Fig. 3. Relation simulated-measured values using SLS for July 2003
2003
Downloaded from ascelibrary.org by Selcuk Universitesi on 12/24/14. Copyright ASCE. For personal use only; all rights reserved.

lower than for the other two objective functions. This is because
lower than 10–20% of the measured data. Allen 共1987兲 suggests the proposed objective function eliminates abnormal data that can
that the model will be calibrated when 85% of the nodes have an affect the results and because it fits the measured and simulated
error lower than 5% and 95% lower than 10%. Garcia-Serra data for the whole parameter range. A RMSE of around 0.3 m
共1988兲 claims a net difference of ±1 m of pressure head marks corresponds to an approximately 0.5% of the piezometric height
successful calibration. In this case, a maximum difference of in a hydrant, which is a good precision for calibration procedures.
±1 m was employed because it is the most restrictive option of Fig. 5 shows the average values of C obtained in the calibra-
those listed above, and because the accuracy in the pressure head tion process, considering the three different objective functions.
measurement was 1 m. Results corresponding to HMLE are those relative to the initial
The parameter to optimize was the Hazen-Williams head value of ␭. HMLE and SLS results were near the standard C
losses coefficient 共C兲 with the only restriction that its value for value because these objective functions fitted the lower values of
each material was higher than 0. The calibration process has the head losses, but not the higher values. For fitting the smaller
standard values of C as its initial seed. values, small changes in the C value were then required. With the
For the irrigation seasons 2003 and 2004, July and August data MDRE objective function, which fitted simulated and measured
were considered, running the calibration process for each of these head loss values for the whole parameter range, more realistic
months, and comparing the results with the main statistical de- values were obtained 共Table 1兲. In addition, this objective func-
scriptors. An analysis of the result considering all the transducers tion automatically eliminated the values with high error 共due to
simultaneously and considering each transducer independently failures in the measurement兲; therefore, a pretreatment of data
was carried out. Thus, the effect of the location of the transducer was not necessary.
was studied. The new values of C indicate an increase in head losses of
22.5% for PVC and 103.6% for asbestos cement 共minor losses
were included in these percentages兲. A lower value of C for July
2004 was obtained, mainly for asbestos cement pipes. This may
Results

Comparison of Objective Functions Used

As described in the methodology, the calibration process was ap-
plied to an on-demand irrigation network located in Tarazona de
La Mancha 共Albacete, Spain兲. Fig. 2 shows the relation between
calibrated and simulated head losses after calibration with HMLE.
When calibrating for the initial values of ␭, this objective func-
tion did not fit the values relative to high flow. Therefore, in this
case, implementing this objective function did not offer any ad-
vantage in the calibration process, but was more computationally
expensive. The SLS objective function had the same problem of
underestimation of head losses, mainly for high head losses 共Fig.
3兲. Thus, these objective functions tended to minimize the error
for lower flows without fitting the whole parameter range. The Fig. 4. Relation simulated-measured values using proposed objective
MDRE objective function fitted the measured and simulated head function for July 2003
loss data in a homogeneous manner in the whole parameter range
共Fig. 4兲. In addition, it eliminated the data with a high error that Table 2. RMSE of Measured and Simulated Data for Each Objective
could affect the final result. The percentage of used data was very Function
high 共70–95%兲, but for some control points this percentage was Year Month MDRE SLS HLME
low, such as T4, caused by problems with data acquisition. These 2003 July 0.27 1.04 1.23
points were eliminated in order to obtain a more accurate result.
2003 August 0.24 3.90 4.08
The RMSE of the measured and simulated data for the appli-
2004 July 0.27 1.90 1.90
cation of each objective function are shown in Table 2. It is
2004 August 0.28 1.40 1.37
shown that the RMSE for the proposed objective function is

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2008 / 39

J. Irrig. Drain Eng. 2008.134:36-42.

 Effect of the Number and Location of Pressure Control
Points on Calibration

When the calibration method, with MDRE as the objective func-

tion, was developed independently for each control point, the re-
sults included in Table 3 were obtained. This Table shows that the
control point T4 was not used in the calibration process, because
it was located very close to the pumping station with very small
head losses. Thus, the proportion of the error in the measurement
was very high. Therefore, it can be concluded that the location of
this control point was not properly done. The control point T5 had
problems of data acquisition and was not used.
Fig. 5. Comparison of C value obtained considering the three objec-
Results showed that C values for PVC were very similar to
tive functions for asbestos cement 共AbsCem兲 and PVC for July and
Downloaded from ascelibrary.org by Selcuk Universitesi on 12/24/14. Copyright ASCE. For personal use only; all rights reserved.

standard values. This is because the flow regime in these pipes

August 2003 and 2004
was smooth turbulent regime 共terminal pipes兲; therefore, friction
losses for small flows are similar to the theoretical.
Head losses coefficient 共C兲 for asbestos cement pipes had a
be caused by the higher crop water requirement in that month higher variability between control points than for PVC pipes.
and, therefore, higher flows, especially in the main pipes 共Fig. 6兲. These differing results were caused by differences in flow in the
In the calibration process, it was assumed that C only varied with pipes that composed the path between the pumping station and
material roughness, but it can be observed that it can be affected each control point. This flow depends on the crops established
by other factors such as the flow in pipes. downstream of these pipes, being different for each control point.
In almost all the cases, the C value for July 2004 was lower than
for the rest of the months.
To make the calibration process easier, the average value of all
the control points was used. In addition, the average value results
in a proper representation of the general network behavior.
Other authors such as Uriol Ardanuy and Montserrat Visacarri
共2002兲 argue that changes in material roughness due to aging are
more important than minor losses; therefore, these minor losses
should be included in the friction head losses in the calibration
process. They obtained a roughness of 1 mm 共including minor
losses兲, for a 40-year-old network, which means an increase in
head losses by 76% of head losses. Guillen et al. 共2000兲 observed
a roughness increase of 0.3 mm, or a 34.26% increase in head
losses for a 20-year-old network, Dias et al. 共2000兲 obtained a C
for PVC between 133 and 140, which has a 0–5% variation com-
pared with the standard value for an urban water network.
Fig. 6. Main pipe flow for July and August 2003 and 2004

Table 3. Results of the Calibration Procedure Considering Each Control Point Independently
Date T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T11 T12
Jul-03 PVC 140 140 120.41 — — 140 140 — — 140 140 124.93
AbsCement 120.86 96.66 104.92 — — 95.55 100.46 — — 85.73 95.59 111.88
Percentage of 98.82 86.76 88.27 — — 83.44 90.96 — — 83.25 84.47 80.19
used data
Aug-03 PVC 140 140 115.53 — — — 140 135.52 — 140 140 109.95
AbsCement 120.65 95.65 93.92 — — — 88.12 129.78 — 93.79 83.63 82.35
Percentage of 85.33 69.23 81.45 — — — 80.88 88.03 — 77.78 94.89 91.03
used data
Jul-04 PVC 140 — 119.41 — — 140 140 138.51 — 116.66 140 140
AbsCement 97.01 — 82.64 — — 77.77 84.06 108.4 — 94.22 74.69 81.88
Percentage of 78.67 — 73.26 — — 67.92 73.71 54.27 — 70.83 78 69.23
used data
Aug-04 PVC 140 140 113.24 — — — 140 126.94 129.18 — 140 140
AbsCement 113.21 96.89 79.46 — — — 87.33 114.68 135 — 73 72.36
Percentage of 79.12 82.76 73.68 — — — 76.7 69.47 78.95 — 75.86 60
used data

40 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2008

J. Irrig. Drain Eng. 2008.134:36-42.

 Fig. 9. Example of pressure lack in head for the irrigation season
2004
Downloaded from ascelibrary.org by Selcuk Universitesi on 12/24/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 7. Pressure distribution map considering standard C values for a process in order to accurately simulate the real behavior of the
demand scenario network. Nowadays, optimization techniques are the main tools
used in calibration procedures. The selection of the proper objec-
tive function in the optimization process will increase the effi-
EPANET Application with the Calibrated Model
ciency and effectiveness of the calibration process. In this study,
Fig. 7 shows the pressure head distribution map for a demand the MDRE objective function is shown to provide more realistic
scenario 共July 17, 2004 at 8.00 am兲 using standard C values results than the SLS and HMLE objective functions, and also
共Table 1兲, and Fig. 8 shows the same simulation using calibrated eliminates automatically those values with an error higher than
C values obtained with the MDRE objective function 共Table 1兲. In the maximum established.
Fig. 7, there is a notable lack of pressure 共less than 45 m兲 in the In a practical sense, it is impossible to distinguish between
last hydrant of the northeast part of the network when standard friction head losses and minor losses in an operating network.
values were used. However, when the calibrated model was used Including the minor losses as friction losses promotes better re-
共Fig. 7兲, the lack of pressure appears not only in one hydrant but sults in the calibration process.
in several hydrants. Therefore, the calibrated model helps network The proper location of the pressure transducers in a hydraulic
managers to predict problems in network performance. network is very important for obtaining an accurate calibrated
During the 2004 irrigation season, in which crop water re- model. They should be located at those points that allow controls
quirements were very high, some low pressure conditions ap- for different pipe material, at different conditions of local head
peared in the network 共Fig. 9兲. If the technicians possessed the losses, and at different demands downstream of the control point.
calibrated model, those events may not have happened, because Using the averaged values for all the control points produced a
the technicians could have carried out an open hydrant redistribu- good representation of the general characteristics of the network,
tion to avoid the problem. and yields an accurate and useful hydraulic model.

Conclusions
Acknowledgments
Simulation models of hydraulic networks are very useful tools for
This research work was funded by the Spanish Ministry through
proper management. However, these models require a calibration
the R&D Plan of Science and Technology, and included within
the project “Manejo eficiente del agua de riego y la energía en
zonas semiaridas 共Ref. No. AGL2001-1180-C02-01兲.”

Notation

The following symbols are used in this paper:

C ⫽ Hazen-Williams head losses coefficient;
D ⫽ pipe diameter;
f t ⫽ expectation of the true value;
k ⫽ number of control points;
L ⫽ pipe length;
k
Nm苸n ⫽ number of valid measurements 共m兲 with an error
lower than the maximum established error;
n ⫽ number of measurements;
Q ⫽ pipe flow rate;
pobs ⫽ measured head loss values;
Fig. 8. Pressure distribution map considering calibrated C values for psim ⫽ simulated head loss values;
a demand scenario wt ⫽ weight associated with the measured parameter;

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2008 / 41

J. Irrig. Drain Eng. 2008.134:36-42.

 ␧t ⫽ difference between measured and simulated modelos hidráulicos de redes de distribución a la demanda.” Riegos y
values; and Drenajes XXI, 102 共in Spanish兲.
␭ ⫽ parameter that stabilizes the variance. Lamaddalena, N. 共1997兲. “Integrated simulation modeling for design and
performance analysis of on-demand pressurized irrigation systems.”
Ph.D. thesis, Universidade Técnica de Lisboa, Instituto Superior de
Agronomía, Lisboa, Portugal.
References Lamaddalena, N., and Sagardoy, J. A. 共2000兲. “Performance analysis of
on-demand pressurized irrigation systems.” FAO Irrigation and
Aliod, R., Eizaguerri, A., Estrada, C., and Perna, E. 共1997兲. “Dimen- Drainage Paper No. 59, Rome.
sionado y análisis hidráulico de redes de distribución a presión en Rossman, L. A. 共2000兲. EPANET 2, users’ manual, Water supply and
riego a la demanda: Aplicación del programa GESTAR.” Riegos y Water Resources Division, National Risk Management Research
Drenajes XXI, 92, 22–38 共in Spanish兲. Laboratory, U.S. Enviromental Protection Agency, Cincinati.
Allen, R. 共1987兲. “Network analysis. The real story.” Proc., Third Int. Sorooshian, S., and Dracup, J. A. 共1980兲. “Stochastic parameter estima-
Conf. of Computer Applications for Water Supply and Distribution, tion procedures for hydrologic rainfall-runoff models: Correlated and
Leicester Polytechnic, U.K. heteroscedastic error cases.” Water Resour. Res., 16 共2兲, 430–442.
Downloaded from ascelibrary.org by Selcuk Universitesi on 12/24/14. Copyright ASCE. For personal use only; all rights reserved.

Dias, M. C. B. F., Vieira, J. M. P., Valente, J., and Coelho, S. T. 共2000兲. Uriol Ardanuy, A., and Montserrat Visacarri, J. 共2002兲. “Calibración de
“Calibração de Modelos de Simulação de Quantidade e Qualidade de una red de distribución a presión en una finca de 310 ha mediante el
Água em Redes de Distribuição. O Caso da Zona Oeste da Cidade de programa GESTAR.” Proc., XX Congreso Nacional de Riegos,
Bragança.” Anais do IX Encontro Nacional de Saneamento Básico, Ciudad Real, Spain 共in Spanish兲.
Loures, Portugal 共in Portuguese兲. Walski, T. M. 共1983兲. “Technique for calibrating network models.” J.
Ferreri, G. B., and Gangitano, L. 共2001兲. “Effetto del tipo e del numero di Water Resour. Plann. Manage., 109, 360–372.
misuratori sull’efficacia della calibrazione delle scabrezze di reti dis- Walski, T. M. 共1985兲. “Assuring accurate model calibration.” J. Am.
tributrici in pressione.” Proc., L’Aqua, 2000, 57–70 共in Italian兲. Water Works Assoc., 38–41.
Garcia-Serra, J. 共1988兲. “Estudio y mejora de las técnicas de calibración Walski, T. M. 共1986兲. “Case-study—Pipe network model calibration is-
de modelos de redes hidráulicas.” Ph.D. thesis, Universidad Politéc- sues.” J. Water Resour. Plann. Manage., 112, 238–249.
nica de Valencia 共in Spanish兲. Wu, Z. Y., Arniella, E. F., and Gianella, E. 共2004兲. “Improving project
Guillen, J., Bescós, M., Doz, J. R., Marzal, A., and Aliod, R. 共2000兲. productivity and model quality for large water systems.” J. Am. Water
“Metodologías y resultados para la validación y la calibración de Works Assoc., 27–34.

42 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2008

J. Irrig. Drain Eng. 2008.134:36-42.