Drainage Design Manual
Section 6
TABLE OF CONTENTS
SECTION 6 : FREQUENCY ANALYSIS OF GAUGED DATA...............................................6.1
6.1 General........................................................................................................................ 6.1
6.2 Analysis of Annual and Partial-Duration Series............................................................6.1
6.3 Frequency Analysis Concepts......................................................................................6.2
6.3.1 Plotting Position Formulas................................................................................6.2
6.3.2 Log-Pearson Type III Distribution.....................................................................6.3
6.3.3 Outliers............................................................................................................. 6.7
6.3.4 Incomplete Records and Zero Flows................................................................6.8
Ministry of Works, Housing & Communications
Drainage Design Manual
Section 6
SECTION 6
FREQUENCY ANALYSIS OF GAUGED DATA
6.1 General
Analysis of gauged data permits an estimate of the peak discharge in terms of its
probability or frequency of exceedance at a given site. This is done by statistical
methods provided sufficient data are available at the site to permit a meaningful
statistical analysis to be made. It is suggested that at least 10 years of record are
necessary to warrant a statistical analysis by method presented therein.
At some sites, historical data may exist on large floods prior to or after the period over
which stream flow data were collected. Whenever possible, these data should be
compiled and documented to improve frequency estimates.
6.2 Analysis of Annual and Partial-Duration Series
Before analyzing data, it is necessary to arrange it in a systematic manner. Data can be
arranged in a number of ways depending on the specific characteristics that are to be
examined. An arrangement of data by a specific characteristic is called a distribution or
a series.
The most common arrangement of hydrologic data is by magnitude of the annual peak
discharge. This arrangement is called an annual series. Another method used in flood
data arrangement is the partial-duration series. This procedure uses all peak flows (for
instance all flows above the discharge of approximately bank-full stage) above some
base value.
Partial-duration series are used primarily in defining annual flood damages when more
than one event that causes flood damages can occur in any year. The partial-duration
series avoids a problem with the annual-maximum series. Annual maximum series
analyses ignore floods that are not the highest flood of that year even though they are
larger than the highest floods of other years. While partial-duration series produce
larger sample sizes than annual maximum series, they require a criterion that defines
independence of the discharges to be considered for the frequency analysis.
The difference between the results of the two methods is large at the lower flows and
becomes very small at the higher peak discharges. If the recurrence interval of these
peak flows is computed as the order divided by the number of events (not years), the
recurrence interval of the partial-duration series can be computed in terms of the annual
series by the equation:
TB
1
ln T A ln(T A 1)
Eqn
(6.1)
where:
TB and T A are the recurrence intervals of the partial-duration series and annual
series, respectively.
Comparison between analyses results of the two methods shows that the maximum
deviation between the two series occurs for flows with recurrence intervals less than 10
years. At this interval the deviation is about 5 percent and for the 5-year discharge, the
deviation is about 10 percent. For the less frequent floods, the two series approach one
another.
6.1
Ministry of Works, Housing & Communications
Drainage Design Manual
Section 6
When using the partial-duration series, one must be especially careful that the selected
flood peaks are independent events. This is a tough practical problem since secondary
flood peaks may occur during the same flood as a result of high antecedent moisture
conditions. In this case, the secondary flood is not an independent event. One should
also be cautious with the choice of the lower limit or base flood since it directly affects
the computation of the properties of the distribution (i.e., the mean, the variance and
standard deviation, and the coefficient of skew) all of which may change the peak flow
determinations.
For this reason, it is probably best to utilize the annual series and convert the results to
a partial-duration series through use of Equation 6.1. For the less frequent events
(greater than 5 to 10 years), the annual series is entirely appropriate and no other
analysis is required.
6.3 Frequency Analysis Concepts
Future floods cannot be predicted with certainty. Therefore, their magnitude and
frequency are treated using probability concepts. To do this, a sample of flood
magnitudes are obtained and analyzed for the purpose of estimating a population that
can be used to represent flooding at that location. The assumed population is then used
in making projections of the magnitude and frequency of floods. It is important to
recognize that the population is estimated from sample information and that the
assumed population, not the sample, is then used for making statements about the
likelihood of future flooding. The purpose of this section is to introduce concepts that
are important in analyzing sample flood data in order to identify a probability distribution
that can represent the occurrence of flooding.
6.3.1
Plotting Position Formulas
When making a flood frequency analysis, it is common to plot both the assumed
population and the peak discharges of the sample. To plot the sample values on
frequency paper, it is necessary to assign an exceedence probability to each
magnitude. A plotting position formula is used for this purpose. A number of different
formulas have been proposed for computing plotting position probabilities, with no
unanimity on the preferred method. A general formula for computing plotting positions
is:
ia
n a b 1
Eqn. 6.2
where:
i = the rank of the ordered flood magnitudes, with the largest flood having a rank of
1
n
a
= the record length
and b = constants for a particular plotting position formula
The Weibull, Pw ( a = b =0), Hazen, Ph ( a = b =0.5), and Cunnane, Pc ( a = b =0.4)
are three possible plotting position formulas:
i
n 1
i 0.5
Ph
n
i 0.4
Pc
n 0 .2
PW
Eqn 6.3 (a)
Eqn 6.3 (b)
Eqn 6.3 (c)
6.2
Ministry of Works, Housing & Communications
Drainage Design Manual
Section 6
The data are plotted by placing a point for each value of the flood series at the
intersection of the flood magnitude and the exceedance probability computed with the
plotting position formula. The plotted data should approximate the population line if the
assumed population model is a reasonable assumption.
6.3.2
Log-Pearson Type III Distribution
Flood frequency analysis uses sample information to fit a population, which is a
probability distribution. These distributions have parameters that must be estimated in
order to make probability statements about the likelihood of future flood magnitudes. A
number of methods for estimating the parameters are available. The method of
moments, which is just one of the parameter-estimation methods is applied in this
manual.
Several cumulative frequency distributions are commonly used in the analysis of
hydrologic data, and as a result they have been studied extensively and are now
standardized. The frequency distributions that have been found most useful in
hydrologic data analysis are the normal distribution, the log-normal distribution, the
Gumbel extreme value distribution, and the log-Pearson Type III distribution.
Log-Pearson Type III Distribution has found wide application in hydrologic analysis. It is
a three-parameter gamma distribution with a logarithmic transform of the variable. It is
widely used for flood analyses because the data quite frequently fit the assumed
population.
The log-Pearson Type III distribution differs from most of other distributions in that three
parameters (mean, standard deviation, and coefficient of skew) are necessary to
describe the distribution. By judicious selection of these three parameters, it is possible
to fit just about any shape of distribution.
An abbreviated table of the log-Pearson III distribution function is given in 6.1. Using the
mean, standard deviation, and skew coefficient for any set of log-transformed annual
peak flow data, in conjunction with 6.1, the flood with any exceedence frequency can be
computed from the equation:
Y log X Y KS y
Eqn. 6.4
where:
X is value of the flood series
Y is the predicted value of log X
Y is the average of the logarithms of X and
S y is the standard deviation of the logarithms.
K is a function of the exceedance probability and the coefficient of skew.
A) Procedure
The specific steps for making a basic log-Pearson type III LP3 analysis without any of
the optional adjustments are as follows:
1.Make a logarithmic transform of all flows in the series ( Yi log X i ).
2.Compute the mean ( Q ), standard deviation ( S ), and station skew (G) of the
logarithms using Equation 6.5, Equation 6.6, and Equation 6.7, respectively of both
the flood data and its log-transformed set. Round the skew to the nearest tenth (e.g.,
0.32 is rounded to 0.3). Discharges could also be estimated using Generalized skew
and Weighted skew. For this the user could refer to Highway Hydrology, HDS 2,
September 1996 (Metric Version).
6.3
Ministry of Works, Housing & Communications
Drainage Design Manual
Section 6
i 1
Eqn. 6.5
i 1
0.5
2
(Qi Q )
n 1
Qi
2
( Q 1)
i 1
= Q
n 1
0.5
Eqn
6.6
n
n (Qi Q )
i 1
( n 1)(n 2)V
where: V
n (
Qi
1) 3
Q
( n 1)(n 2)V 3
i 1
Eqn. 6.7
S
X
3. Since the LP3 curve with a nonzero skew does not plot as a straight line, it is
necessary to use more than two points to draw the curve. The curvature of the line
will increase as the absolute value of the skew increases, so more points will be
needed for larger skew magnitudes.
4. Compute the logarithmic value for each exceedence frequency using Equation 6.4.
5. Transform the computed values of step 4 to discharges using
X 10 Y
Eqn 6.8
in which
X is the computed discharge for the assumed LP3 population.
6. Plot the points of step 5 on logarithmic probability paper and draw a smooth curve
through the points. The sample data can be plotted on the paper using a plotting
position formula to obtain the exceedence probability. The computed curve can then
be verified, and, if acceptable, it can be used to make estimates of either a flood
probability or flood magnitude.
Log-Pearson III distributions are usually plotted on log-normal probability graph
paper even though the plotted frequency distribution may not be a straight line. It is
a straight line only when the skew of the logarithms is zero.
Example 6.1 (Adopted from HDS 2, 1996)
The Log-Pearson Type III distribution will be illustrated using an exemplary river flood
data (Table Example 6-1-1) for the case of station skew. Appendix 6-1 and Equation 6.4
are used to compute values of the log-Pearson III distribution for the 2- to 100- years
flood using the parameters, Y , S y , and G for the flood data with the following steps.
Step 1:
Transform all the flow into their logarithmic values
This is done by taking the logarithm (to the base 10) of the discharge data. In
the example, it is included in the third column of Table Example 6.1 1.
Step 2:
Computation of the mean ( Q ), standard deviation ( S ), and station skew
(G) of the logarithms using Equation 6.5, Equation 6.6, and Equation 6.7,
respectively for both the flow data and the log-transformed data sets. The
means for the flow data and the log transformed version are figured out to be
186.96 and 2.0912 respectively and standard deviations are computed to be
200.33 and 0.3941 respectively. To facilitate the computation of the
6.4
Ministry of Works, Housing & Communications
Drainage Design Manual
Section 6
generalized skew, the term
Qi
i 1
1) 3 is rendered in the fifth column of
Table Example 6.11. The Station Skew is then computed to be 0.236
employing Eqn. 6.7. The value 0.236 is rounded to the nearest tenth (i.e. to
0.2).
Step 3:
For the Frequency value for which the discharge is to be estimated, compute
the exceedence probability. For Frequency of 10 years, for instance, the
exceedence probability is
0.1 = 1/10
With the G value computed in step 2 and the exceedence probability for the
frequency value under consideration, find the K value from Appendix 6.1. For
the rounded of G value of 0.2 and exceedence probability of 0.1, for instance,
the corresponding K value is read to be 1.30105 and the Y value is
2.6039 = 2.0912 + 1.30105(0.3941).
Step 4: Transform the computed values of step 3 to discharges using Eqn. 6.8. The
discharge for the return period of 10 years, thus, become
401.7 m 3 / s 10 2.6039
Similarly the discharges for the different frequency values could be
computed. The results for the frequency values of two to hundred years are
rendered in Table Example 6.1- 2.
6.5
Ministry of Works, Housing & Communications
Drainage Design Manual
Section 6
Table Example 6.1-1: Frequency Analysis Computations for the LogNormal Distribution: an Exemplary River
Weibull Plotting
Flood X, (m3/s)
Rank
Y = Log (x)
Position
( Y Y 1) 3
903.30
2.96
0.023
0.071
900.50
2.95
0.045
0.070
495.50
2.70
0.068
0.024
492.70
2.69
0.091
0.024
410.60
2.61
0.114
0.016
370.90
2.57
0.136
0.012
342.60
2.53
0.159
0.010
274.10
2.44
0.182
0.005
267.30
2.43
0.205
0.004
261.10
10
2.42
0.227
0.004
231.10
11
2.36
0.250
0.002
212.70
12
2.33
0.273
0.001
195.10
13
2.29
0.295
0.001
180.10
14
2.26
0.318
0.000
160.30
15
2.20
0.341
0.000
155.20
16
2.19
0.364
0.000
153.80
17
2.19
0.386
0.000
146.70
18
2.17
0.409
0.000
140.60
19
2.15
0.432
0.000
134.50
20
2.13
0.455
0.000
130.80
21
2.12
0.477
0.000
116.90
22
2.07
0.500
0.000
112.10
23
2.05
0.523
0.000
100.20
24
2.00
0.545
0.000
95.10
25
1.98
0.568
0.000
94.90
26
1.98
0.591
0.000
90.60
27
1.96
0.614
0.000
86.40
28
1.94
0.636
0.000
83.50
29
1.92
0.659
-0.001
77.30
30
1.89
0.682
-0.001
71.90
31
1.86
0.705
-0.001
61.20
32
1.79
0.727
-0.003
60.90
33
1.78
0.750
-0.003
60.60
34
1.78
0.773
-0.003
58.00
35
1.76
0.795
-0.004
56.60
36
1.75
0.818
-0.004
56.10
37
1.75
0.841
-0.004
49.50
38
1.69
0.864
-0.007
41.60
39
1.62
0.886
-0.012
34.00
40
1.53
0.909
-0.019
25.20
41
1.40
0.932
-0.036
24.50
42
1.39
0.955
-0.038
22.70
43
1.36
0.977
-0.043
6.6
Ministry of Works, Housing & Communications
Drainage Design Manual
Section 6
Table Example 6.1- 2. Calculation of Log Pearson Type III Discharges for
the Exemplary River Using Station Skew
(1)
Return
Period
(Yrs)
2
5
10
25
50
100
(2)
Exceedence
Probability
(3)
K
(4)
Y
5
X
(m3/s)
0.50
0.20
0.10
0.04
0.02
0.01
-0.03325
0.83044
1.30105
1.81756
2.15935
2.47226
2.0781
2.4185
2.6039
2.8075
2.9422
3.0655
119.7
262.1
401.7
641.9
875.3
1162.8
Example 6.2 (Adopted from HDS 2, 1996)
Compute the plotting position values for the discharge data set given for example
6.1using Weibull Plotting Position Formula.
Step 1: The discharge data is arranged in rank. This is done in column 4 of Table
Example 6.1-1
Step 2: Compute the plotting position value employing Weibulls formula. For the data
ranked first, the plotting position value is (for i = 1 and n=43)
1
0.023
43 1
Using similar procedure, the plotting position values could be calculated for the whole
data set.
6.3.3
Outliers
Outliers, which may be found at either or both ends of a frequency distribution, are
measured values that occur, but appear to be from a longer sample or different
population. This is reflected when one or more data points do not follow the trend of the
remaining data. If the station skew is greater than 0.4, tests are applied for high outliers
first; and if less than -0.4, low outliers are considered first. If the station skew is
between 0.4, both high and low outliers are tested before any data are eliminated.
The detection of high and low outliers is obtained with the following equations,
respectively:
YL Y K N S Y
Eqn. 6.9
and
YL Y K N S Y
Eqn. 6.10
where:
YL is the log of the high or low outlier limit
Y
is the deviation of the sample
K N is the critical deviate taken from Appendix 6.2
If the sample is found to contain high outliers, the peak flows should be checked
against historical data and data from nearby stations. It is recommended that high
6.7
Ministry of Works, Housing & Communications
Drainage Design Manual
Section 6
outliers be adjusted for historical information or retained in the sample as a systematic
peak. The high outlier should not be discarded unless the peak flow is shown to be
seriously in error. If a high outlier is adjusted based on historical data, the mean and
standard deviation of the log distribution should be recomputed for the adjusted data
before testing for low outliers.
To test for low outliers, the low outlier threshold YL of Equation 6.10 is computed. The
corresponding discharge X L = 10 YL is then computed. If any discharges in the
flood series are less than X L , then they are considered to be low outliers and should
be deleted from the sample.
Example 6.3 (Adopted from HDS 2, 1996)
To illustrate the criteria for outlier detection, Equation (6.9) and Equation (6.10) are
applied to the 43-year record for the exemplary River data (Table Example 6.1-1), which
has a log mean of 2.0912 and a log standard deviation of 0.3941. From Appendix 6.2 ,
KN = 2.710. Testing first for high outliers
Y L 2.0912 2.710(0.3941) 3.1592
and
X L 10 3.1592 1443 m s
3
No flows in the sample exceed this amount, so there are no high outliers. Now testing
for low outliers, Equation 6.10 gives
Y L 2.0912 2.710(0.3941) 1.0232
and
X L 101.0232 11 m s
3
There are no flows in the Medina River sample that are less than this critical value.
Therefore, the entire sample should be used in the log-Pearson III analysis. If any
discharges in the flood series are less than XL, then they are considered to be low
outliers and should be deleted from the sample. The moments should be recomputed
with the data set that does not bear the low outlier.
6.3.4
Incomplete Records and Zero Flows
Stream flow records are often interrupted for a variety of reasons. Gages may be
removed for some period of time, there may be periods of zero flow and there may be
periods when a gage is inoperative either because the flow is too low to record or it is
too large and causes a gage malfunction. If the break in the record is not flood related,
such as the removal of a gage, no special adjustments are needed and the segments of
the interrupted record can be combined together to produce a record equal to the sum
of the length of the segments. When a gage malfunctions during a flood, it is usually
possible to estimate the peak discharge from high water marks or slope-area
calculations. The estimate is made a part of the record, and a frequency analysis
performed without further adjustment.
Zero flows or flows that are too low to be recorded present more of a problem since in
the log transform, these flows produce undefined values. In this case, an adjustment
based on conditional probability that is applicable if not more than 25 percent of the
6.8
Ministry of Works, Housing & Communications
Drainage Design Manual
Section 6
sample is eliminated. The adjustment for zero flows also is applied only after all other
data adjustments have been made. The adjustment is made by first calculating the
relative frequency, Pa , that the annual peak will exceed the level below which flows
are zero, or not considered (the truncation level):
Pa
M
n
Eqn. 6.11
where:
M is the number of flows above the truncated level
is the total period of record.
The exceedence probabilities, P , of selected points on the frequency curve are
recomputed as a conditional probability as follows:
P Pa * Pd
Eqn. 6.12
where:
Pd is the selected probability
QS log(Q0.50 ) K 0.50 ( S S )
Eqn. 6.13
Since the frequency curve adjusted by Equation 6.12 has unknown statistics, its
properties, synthetic values, are computed by the equations:
SS
log(Q0.01 / Q0.50 )
K 0.01 K 0.50
Eqn. 6.14
and
log(Q0.01 / Q0.10
log(Q0.10 / Q0.50 )
G S 2.50 3.12
Eqn. 6.15
where:
Q S , Ss , and Gs are the mean, standard deviation, and skew of the synthetic
frequency curve, Q0.01 , Q0.10 and Q0.50 are discharges with exceedence
probabilities of 0.01, 0.10 and 0.50, respectively, K 0.01 and K 0.50 are the logPearson III deviates for exceedence probabilities of 0.01 and 0.50, respectively.
The values of Q0.01 , Q0.10 and Q0.50 must usually be interpolated since
probabilities computed with Equation 6.11 are not normally those needed to compute
the properties of the synthetic or truncated distribution.
The log-Pearson III distribution can then be computed in the conventional manner using
the synthetic statistical properties. It is recommended that the distribution be compared
with the observed flows since data adjusted for conditional probability may not follow a
log-Pearson III distribution.
The procedures for the aforementioned discussions could be summarized as follows:
1. Obtain site information, the systematic station data, and historic information. This
data should be examined for changes in watershed conditions, gage datum, flow
regulation, etc. It is in this initial step that missing data should be estimated if
indicated by the project.
2. Order the flood data, determine the plotting position, and plot the data on selected
probability graph paper (usually log-probability). Examine the data trend to select
6.9
Ministry of Works, Housing & Communications
Drainage Design Manual
3.
4.
5.
6.
Section 6
the standard distribution that best describes the population from which the sample
is taken. Use a mixed-population analysis if indicated by the data trend and the
watershed information.
Compute the sample statistics and the frequency curve for the selected distribution.
Plot the frequency curve with the station data to determine how well the flood data
are distributed according to the selected distribution.
Check for high and low outliers. Adjust for historic data, retain or eliminate outliers,
and recompute the frequency curve.
Adjust data for missing low flows and zero flows and recompute the frequency curve.
Check the resulting frequency curve for reliability.
6.10
Ministry of Works, Housing & Communications
