You are on page 1of 10

WATER RESOURCES BULLETIN

VOL. 1 3 , NO. 3 AMERICAN WATER RESOURCES ASSOCIATION JUNE 1977

PREDICTING INSTANTANEOUS PEAK DEMAND IN


RURAL DOMESTIC WATER SUPPLY SYSTEMS'

Trevor C. Hughes and Ronald V. Canfield2

ABSTRACT: Many rural areas of the United States still have no public domestic water systems.
Typical land use patterns in these areas may require 1/2 mile or more of pipe per farm
connection. Public systems serving these areas are economically feasible only if realistic
short-term peak demand standards are available for their design. The lack of reliable data upon
which to establish such criteria has resulted in a large variation in criteria among state and
federal agencies involved in financing and in approving construction of these systems. During
the summer of 1975 three distribution laterals of a rural system in Utah were master metered
and instantaneous peak flows were recorded for 4 months. The metered lines served 4,12, and
22 farm houses each. The frequency distribution of peak flows has been analyzed and
compared with that developed during similar research in Mississippi and with the existing design
standards of the Farmers Home Administration and the State of Utah.
(KEY TERMS: water supply; rural; design criteria; water demand; maximum flow rates.)

INTRODUCTION TO THE PROBLEM


Traditional design criteria for municipal water distribution systems typically include
generous factors of safety for the following reasons which are either factual or at least are
assumed by engineers or approval authorities to be factual:
1. Some of the line capacities are governed by fire flow and domestic demands are
therefore irrelevant.
2. In lines which are not governed by fire criteria large safety factors provide a hedge
against unforeseen future growth.
3. Failure of a pipeline to deliver even infrequent short-term peak demands causes
very negative feedback t o utility managers and engineers but no one complains about
over investment in unused capacity once the investment has been approved.
4. Public health approval authorities tend t o be extremely cautious about avoiding
possible negative line pressures which could draw contaminated ground water into the
system.

'Paper No. 76087 of the Water Resources Bulletin. Discussions are open until February 1.1978.
'Respectively, Associate Professor of Civil and Environmental Engineering at the Utah Water Re-
search Laboratory, and Associate Professor of Applied Statistics and Computer Science, Utah State
University, Logan. Utah 84322.

479
480 Hughes and Canfield

5 . There is a serious lack of data on short-term peak demands in distribution lines


serving various numbers of customers. Flows from wells and treatment plants (above
equalizing reservoirs) which are related to daily peaks are usually metered and are
well known. This is not the case with instantaneous peaks because few distribution
laterals are master metered. The tendency then is to use whatever has worked in the past
as a design standard.
The situation described above undoubtedly results in substantial over investment in
municipal systems. This may not be a serious problem in large systems where economies
of scale result in relatively low water costs per family, however, it can be a crucial
problem for low density rural domestic systems.
Many rural areas of the United States still have no public domestic water systems.
Typical land use patterns in such areas may require 1/2 mile or more of pipe per farm
connection. During the last 20 years domestic systems (excluding conventional fire
protection in outlying areas) have become economically feasible due to the availability of
high quality, low-cost materials for small diameter distribution mains such as plastic pipe.
The Fanners Home Administration water and sewer loan and grant program which is
being funded at approximately $500 million annually has enabled thousands of these
types of water systems to be constructed. The rural water supply program, however, is
now facing very serious problems for several reasons: (1) Dramatic increases in energy and
other costs have inflated the costs of all waterworks materials, but particularly that of
plastic pipe, which is a petroleum derivative; (2) Enforcement of the recently enacted
Safe Drinking Water Act will cause very striking increases in the costs of both existing and
planned rural systems (much greater costs per customer than in urban systems); ( 3 )
Substantial differences of opinion exist between public water supply approval agencies
over what constitutes adequate peak flow criteria for rural systems. Farmers Home
Administration design criteria for example requires 1 to 3 gallons per minute per
connection on distribution lines serving small numbers of connections as compared to
some state criteria which requires 5 gpm or more (Utah, 1974, and Butler, 1955). As a
result many rural systems which could be constructed in some states are not economically
feasible in other states because of higher local criteria (which overrides the federal
criteria).

PREVIOUS RESEARCH
The lack of data on short-term peak period demands which was cited previously for
water systems generally applies particularly to low density rural type systems. The John
Hopkms University program (Linaweaver, et. al., 1966, and Howe and Linaweaver, 1967)
has provided good statistical models for peak day and peak hour demands in urban
systems. However, even the 15-minute master meter data recorded by the John Hopkins
study are inadequate for rural low density lines in whch peaks may last less than 3
minutes due to the relatively larger impact of very few services on the total pipeline flow.
What is needed t o define design standards for rural distribution lines is data on the
magnitude and frequency of instantaneous peak flows. A paper which does address this
problem was published by H. W. Ginn, et. al.. (1966). In this research the peak flow at
individual meters on a system in Mississippi were recorded at one minute intervals during
peak hours. These data were aggregated to produce a frequency distribution of peaks at a
Predicting Instantaneous Peak Demand in Rural Domestic Water Supply Systems 481

single typical residence. This distribution was then used in conjunction with a statistical
model. The model is based upon the probability of combinations of various events at
several residences occurring simultaneously. The model constructed by Ginn is a totally
different concept than the approach used herein but the results match surprisingly well,
particularly for the low probability events which are likely to be used as design criteria.
This comparison will be described later.

SAMPLING PROCEDURE
Lapoint Culinary Water Incorporated, a rural system in Utah was selected for the pilot
study for the following reasons: (1) It includes several long small diameter dead end lines
(which minimized the number of master meters required); (2) The system was constructed
nine years ago using PVC pipe for all small lines and appears to have essentially no leakage
(separating leakage from customer use was therefore not required); (3) The capacity of
the small diameter lines in the sample area had been challenged by the State Division of
Health before construction of the system and considerable growth had occurred since
construction; therefore a current study of pipeline flows and minimum pressures was of
considerable practical value to the water utility officers in assessing the limits of their
distribution system capacity.
The approach used to assess peak demands was t o install orifice type meters at three
locations: A 2-1/2” diameter line serving 22 families; a 2” line serving 12 families and a
1-1/2” line serving 4 families. A differential pressure transducer with a 24-hour ink chart
was used t o record flows continuously. The three orifices were installed on June 26,
1975, and flows were recorded continuously until the possibility of frost damage to the
recorder required stopping the operation on October 23, 1975. The summer data is
considered t o represent the peak season on this system. Budget limitations allowed
purchase of only one recorder, so it was rotated at approximately two-week intervals
among the three meter stations in order to obtain an intensive record at each station
under varying climatic conditions.
The connections served by the metered lines are typically modest farm houses. The
outdoor yard irrigation demand is quite small. Most of the houses have some landscaping
but others have none. The largest outdoor demand is probably stock watering; however,
only small family type beef and dairy operations were involved. No attempt was made to
separate domestic and outdoor demand components. The total demands encountered in
the sample areas are considered t o be representative of rural low density farm type
customers in a semi-arid climate.

ANALYSIS O F THE DATA


The daily hydrographs recorded for the small numbers of services involved do not
show the pronounced morning and afternoon peaks that are typical of large system
hydrographs. The relatively larger impact on total flow of peak demands by a few services
produces an almost random distribution of events during the day and almost no flow
occurs at night. A typical 24-hour hydrograph for the 12 service meter is shown in Figure
1.
482 Hughes and Canfield

Figure 1. Typical 24-hour Hydrograph for 12 Service Meter.

The principal objective of this research was to determine the frequency distribution of
peak flows. The variable that will be analyzed in detail is the daily maximum. Extraction
of this single event from each of the 24-hour records produced the data which is
summarized in Table I . It should be noted that although the time base increment used in
the probability analysis is one day, the actual duration of the measured peaks is typically
less than 3 minutes (Figure 1). An alternate approach would be to analyze all of the
events above a certain flow rate using, for example, 5 minutes as the time base. This
would provide a much greater number of data points and therefore smaller confidence
limits for a given probability level; however, the more conservative daily maximum events
were selected for the analysis.
The daily peak flow rate data are distributed normally as indicated by the relatively
small skew coefficients shown in Table 1. Ninety-five percent confidence limits on these
coefficients indicate that they are not significantly different than zero, except for the 2”
meter station, which was based upon a relatively small sample (N=l7). Further evidence
Predicting Instantaneous Peak Demand in Rural Domestic Water Supply Systems 483

TABLE 1 . Statistical Parameters for DAY Maximum Instantaneous Flows in


Gallons Per Minute (gpm) Per Service.

Number of Services
Parameter 4 12 22

Number of days (N) 30 17 42


Mean daily maximum (5 3.17 1.97 1.73
Standard deviation ( S ) 0.373 0.199 0.2735
Skew coefficient (g) +.37 - .89 +.39
Maximum measured flow 4.0 2.25 2.29

of the normal distribution is that the data were linearized very well by the normal
probability plots shown in Figure 2. Some recorder difficulties caused by high humidity
in the below grade vault at the 2” meter station resulted in less useable data (17 days)
here than at the other two stations. This smaller sample may account for the relatively
flatter slope of the 1 2 service line. One is even tempted to show a best fit graphical
location of this line which crosses the 22 connection line. However, an analytical
approach using the more conservative t distribution (which incorporates the relative
reliability due to differences in number of data points) supports the approximate
relationship shown in Figure 2.
The probability distribution shown in Figure 2 represents the expected value of unit
demand for a given probability level P or recurrence interval Tr (where P = 1/Tr) assuming

PROBABILITY (PERCENT1
s9.99 99 PI PO M) m MI M w M 20 10 5 2 I 0.5 0.2 0.10 0 5 0.01
I I I I I I I I I I 1 I 1 1 1 1 1 1

,
4.

1.7.
2 I.

I - -
I I 1 I I I I 1

Figure 2. Probability Distribution of Daily Maximum Flow Data.


484 Hughes and Canfield

that the computed means and standard deviations of the data equal the true parameters
for these normally distributed populations. Actually we have no assurance that these
point estimates are precisely equal to the true parameters and so the more conservative t
distribution which accounts for randomness in both the means the standard deviations
was also determined as follows.
We desire t o compute the probability at any risk level ( a ) that the normalized flow
variable exceeds the associated value of the t distribution (Kempthorne and Folks. 1971):

The related value of flow (X) is:

The results of this analysis and a comparison with the graphical solutions from Figure
2 are given in Table 2. The excellent agreement between the t distribution and the
graphical solution indicates that the point estimates of the statistical moments computed
from the data sample are very close to the true population parameters.
The graphical solution however, still represents only the expected values of peak
demands during any time interval. In order t o estimate confidence limits on flows not
being exceeded during a particular time interval, an additional analysis is needed. The
accepted relationship for making such assurance level inferences about hydrologic events
(Linsley, et. aL, 1975) is:

J = 1 - (t)(I-P)N-k Pk

in which J is the probability that an event with average probability of occurrence P will
be exceeded exactly k times during N time intervals. The dimension of N commonly used
for hydrologic data is years; however, in this analysis daily events are used so N is
measured in days. The binomial coefficient

N - N!
( k ) - k! (N-k)!

For the special case where k = 0. that is, the largest event,

J = 1-(I-P) N
Predicting Instantaneous Peak Demand in Rural Domestic Water Supply Systems 485

TABLE 2. Comparison of Unit Demands (gpm/conn) From t Distribution and From


Graphical Linearized Normal Distribution.

Recurrence Number of Services


interval Parameter 4 12 22

P = .25 t ,683 .689 ,681


Tr = 4 d a y s 3.43 2.1 1 1.91
'0)
'(Fig. 2) 3.38 2.07 1.90

P = .01 t 2.462 2.561 2.421


TI = 100days 4.10 2.49 2.39
X(t)
'(Fig. 2) 4.18 2.49 2.40

P = .0005 t 3.659 3.965 3.541


Tr = 5.5 yrs. 4.55 2.18 2.70
'(0
'(Fig. 2) 4.60 2.71 2.66

Figure 3 shows a comparison of expected values of demand levels and 95 percent


confidence limits on these levels over the probable range of interest for design standards.
This comparison is also given for a few particular values in Table 3.
One could say, for example that the expected peak unit demand by 22 services during
any 5-year period is 2.68 gprn. We could also say with 95 percent assurance that this peak
would not exceed 2.91 gpm during the next 5 years. The relatively small difference
between the expected values and the confidence limit values is further evidence of the
statistical reliability of the samples.

CONCLUSIONS
In order to derive design standards a particular probability level must be selected. This
is a judgment decision which may vary widely between various engineers. If customer
satisfaction is the governing issue certainly a 3-minute pressure below design level once a
year (P = .00?7) or more often should be adequate; however, public health officials tend
to prefer risk probabilities approaching zero. Their reasoning is that if the following
combination of events occur simultaneously then contamination of the system can result:
( I ) The pressure falls not just below the minimum design requirement (usually 20 psi),
but also below zero; (2) At the precise location on the line where this pressure failure
occurs and at this same time, one of two situations exists - there is a leak in the line at a
point where the pipe is below the watertable, or there is an irrigation hose connected to
the system with the hose bib open and the other end submerged in a puddle; (3) The
water which is drawn into the system by this unlikely combination of events is
sufficiently contaminated and the negative pressure exists for a sufficient duration that
the quality and quantity of water entering the system constitutes a public health hazard.
The probability of this combination of events occurring simultaneously in time and space
is the product of the individual event probabilities. It would seem clear that this product
486 Hughes and Canfield

PROBABILITY IPERCENT I
-98 SS 911 90 80 70 SO 50 40 50 20 10 5 2 I 0 5 02010.05 001
I I 1 I I I I I I I 1 I l l l l l l

A w09e

0 I I I I I I l l
1
2 5 10
Wrl I 5 Y m \lZ771.0rd
RECURRENCE INTERVAL IN DAYS

Figure 3. Comparison of Most Probable and Confidence Limit Excedance Levels.

TABLE 3. Ninety-five Percent Confidence Levels for Unit Demands.


(J = .05)

Number of Services
l i m e Period Robabitity
(N) 8) Parameter 4 12 22

100 days 5.13 95%Assurance 4.6 2.74 2.69


Most probable 4.2 2.52 2.42
1 Year 1.405 95%Assurance 4.75 2.81 2.78
Most probable 4.39 2.62 2.54
5 years 2.81 ( l o i s 95%Assurance 4.93 2.93 2.91
Most probable 4.6 2.13 2.68

of probabilities would approach zero even if the probability of the particular peak flow
event analyzed herein does not. It would follow that a rational criteria would be to select
that peak flow probability based upon customer satisfaction (a once a year recurrence
interval for example).
The paper by Ginn cited previously did not recommend a particular probability level
but did use the .0001 level (27-year recurrence interval) for purposes of comparing with
other existing criteria. In order to build upon this comparison in a consistent way, this
same probability level will be presented here. It should be noted that the expected value
Predicting Instantaneous Peak Demand in Rural Domestic Water Supply Systems 487

of the 27-year recurrence peak approximates the I-year 95 percent confidence level (see
Figure 3 ) and so this may represent a reasonable design criteria.
The maximum peak day instantaneous demands for the Lapoint analysis that would be
expected once in 27 years (and which would last about 3 minutes) are shown in Figure 4.
Also shown in Figure 4 are the Ginn model demands for the same recurrence interval, the
Utah State Division of Health Criteria (Utah, 1974) and the Farmers Home
Administration Criteria (Hawkes, 1976). The latter criteria includes two curves: (1) a
minimum acceptable standard (where little or no outdoor irrigation is required and (2) an
average recommended curve (presumably where outside demand is significant). The
Farmers Home Administration National Office claims that more than 5,000 systems have
been and are now in operation which were designed by the minimum FmHA standard and
that none of them report problems in meeting peak flows (Hawkes, 1976).
Figure 4 shows close agreement between this research and the Ginn model. This is
particularly interesting in view of the totally different approaches used by the two
projects. The FmHA average criteria requires more pipe capacity by 1/2 to 1
gpm/connection than the Lapoint or the Ginn criteria but is clearly the same form of
equation (same scale effect) as the Ginn standard. The FmHA minimum criteria, however,
requires much smaller pipe diameters than any other standard for lines serving more than
10 customers. The fact that this minimum criteria is accepted and widely used in most
states is likely due to the fact that very infrequent, very short-term pressure drops are not
a significant problem rather than the idea that such peaks never occur.
Standards such as the Utah State Criteria that do not recognize any diversification
effect for lines serving less than 60 customers clearly are overly conservative (except in
extremely small lines where the reverse is true). The concept of providing capacity for
future growth should be handled as recommended by Ginn, e l al., as a separate problem.
That is, the design peak flow per connection and the number of future connections for

Figure 4 . Comparison of Measured Peak Flow Distributions to Existing Standards.


488 Hughes and Canfield

w h c h capacity is to be provided should both be clearly defined. The former should be


determined as realistically as possible, and the latter should be as liberal as possible within
the economic constraints of the particular situation. Using very high unit demand criteria
with the supposed justification that this provides added capacity for unforeseen growth is
not good design practice.
It appears that in this context the FmHA average standard is conservative and could be
reduced 1/2 t o 1 gpm depending upon the number of services involved when designing for
the type of Western United States rural service described previously. The FmHA
minimum standard, however, appears to be low for the peak demand measured during
this study. An interesting area of future research would be the simultaneous measurement
of peak flows and line pressures on a sample of rural systems which have been designed
by the FmHA minimum standard. This would help answer the question of whether such
systems actually deliver peak demands or if desired flows are being constrained
hydraulically.

LITERATURE CITED
Butler, H. E., 1955. Design criteria for distribution systems. Journal AWWA. 47: pp. 1148-1152.
Cinn, H. W., M. W. Corey, and E. J . Middlebrooks, 1966. Design parameters for rural water
distribution systems. Journal AWWA, 58: pp. 1595-1602.
Hawkes, E. Lee, 1976. Correspondence with Farmers Home Administration National Office.
Washington, D.C.
Howe, Charles W., and F. P. Linaweaver, Jr., 1967. The impact of price on residential water demand
and its relation to system design and price structure. Water Resources Research, 3(1) pp. 13-32.
Kempthorne, O., and L. Folks, 1971. Probability, statistics, and data analysis. Iowa State University
Press, Ames, Iowa, 555 pp.
Linaweaver, F. P., John C. Geyer, and Jerome B. Wolff, 1966. A study of residential water use.
Federal Housing Administration, Department of Housing and Urban Development, Washington,
D.C. 20402, 79 pp.
Linseley, Ray K., Jr., Max A, Kohler, and Joseph L. H. Paulhus, 1975. Hydrology for engineers.
McCraw-Hill Book Company, New York, N . Y. 482 pp.
Utah State Division of Health, 1974. Unpublished criteria for water supply system peak flows and
friction coefficient minimum design standards. Personal communication with Environmental Health
section personnel. Salt Lake City, Utah.

You might also like