CHAPTER THREE

FREQUENCY ANALYSIS
 Frequency analysis is a procedure for estimating the frequency (or the probability) of
occurrence of extreme events
 Objective of frequency analysis of hydrologic data is to relate the magnitude of extreme
events to their frequency of occurrence using probability distributions.
 Hydrologic data to be analyzed is assumed to be independent and identically distributed
and the hydrologic system is assumed to be stochastic, space-independent and time
independent
 The assumption of identical distribution or homogeneity is achieved by selecting the
observations from same population (i.e., no changes in the watershed and recording gauges
are made)
 The assumption of independence is achieved by selecting the annual maximum of the
variable being analyzed as the successive observations from year to year will be independent.
 CONT…
 The results of flood frequency analysis can be used for many engineering purposes
 flood flow frequency analysis can be used in the design of dams, bridges, culverts, flood
controlling devices,
 Urban flooding : to find design storms
 Drought frequency and magnitude for agricultural planning

RETURN PERIOD
 It can be defined as the average length of time for an event of given magnitude to be equaled or
exceeded in a statistical sense.

 The return period of a given magnitude is defined as the average recurrence interval between events
equaling or exceeding a specified magnitude

 The concept of return period of any hydrologic event (e.g., flood, rainfall, river discharge, landslide,
wind storms) plays a key role in risk and uncertainty analysis in hydro-climatic studies.
 CONT…
• An extreme event is said to be occurred, if a random variable X is greater than (or equal to) a
level 𝑥𝑇 .
• The time between the occurrences of X > 𝑥𝑇 is called the ‘recurrence interval’ (τ).
• The expected value of τ, E(τ), is the average number of years in which the event X > 𝑥𝑇
returns.
• E(τ) is the return period ‘T’ of the event X > 𝑥𝑇 .
• The concept of return period is used to describe the chances of occurrences.
• The probability p = P(X > 𝑥𝑇 ) of occurrence of the event X > 𝑥𝑇 in any observation is related
with return period as follows:
• The probability of occurrence of an event in any observation is the inverse of its return period.
P(X ≥ xT) = 1/T
 CONT…
• Therefore the T year return period event X > 𝑥𝑇 and it occurs on an average once in T years.
• Example – 1: What is the probability that the annual maximum discharge in the river will
exceed or equal 1500 cumec and if it has a return period of 5 year.
• Solution: given the event has 5yr return period =T
1 1
• Probability P(X > 1500) = = = 0.2
𝑇 5
• The complimentary event is that the T year return period event occurs at least once in N year
period; and hence the probability is 1 – (1 – p)N
• Since p = 1/T
1 N
• Required probability is 1- (1- )
𝑇
• Example-2: Obtain the probability that the annual maximum discharge in the river will equal
or exceed 1500 cumec at least once in the next five years
1 N
• Required probability is 1- (1- )
𝑇
1
• = 1- (1- )5 = 0.672
5
 probability plotting and plotting positions formulae

• A probability plot: is a plot of magnitude of a particular event versus its probability of

exceedance.
• It helps to check if a data set fits a particular distribution or not. the plot can be used for
interpolation, extrapolation, and comparison purposes.
• It can be useful for estimating magnitudes with specified return periods
• However, any kind of extrapolation must be attempted only when a reasonable fit is assured
for the distribution.
• The primary step to obtain a probability plot for a given set of data is to determine the
probability of exceedance of each data point.
• this technique of determining exceedance probability of a data point is referred to as
plotting position.
• For sample data, this can be analyzed either by empirical methods or by analytical methods.
 CONT…
• plotting position formulae for some of the common empirical methods are listed
Name of the method Plotting position formula
m
California P=
N
Hazen P=
m − 0.5
N
m
Weibull P=
N+1
Chegodayev P=
m − 0.3
N + 0.4
Blom P=
m − 0.44
N + 0.12
Gringoten P=
m − 0.375
N + 0.25

• Different available plotting position formulae m = rank, N = number of observations, p =

probability of exceedance.
 CONT…
 For application of these plotting position formulae,
 the first task is to arrange the sample data (of size = N) in descending order of magnitude and
to assign an order number or rank (m).
 Thus, for the first member of the arranged data, i.e., for the largest data, m = 1 will be
assigned, for the second largest data m = 2 and so on.
Hence, for the smallest data in the sample, m = N will be assigned.
 Then, using any of the above-mentioned formulae (Table) probability of exceedance (p) is to
be calculated for all data in the series.
 After determining P hence, return period T which is equal to 1/p is calculated.
We can obtain the probability plot for the given data by plotting its different magnitudes with
corresponding probability of exceedance.
Plotting different magnitudes of the events with their corresponding return period (T ) in semi
log or log–log graph is very popular and useful way of presenting the probability plot.
 CONT..
• Example 3: for the following data given in table below find the exceedance probability and
their respective return periods.
 CONT…
exceedence
Discharge in probability (p) = Return
year Dischagre descending rank(m) m/(N+1) period(T) =1/P
1950 854 1642 1 0.05 21.0
1951 1090 1586 2 0.10 10.5
1952 1580
1583 3 0.14 7.0
1953 487
1580 4 0.19 5.3
1954 719
1303 5 0.24 4.2
1955 141
1090 6 0.29 3.5
1956 1583
1957 1642 854 7 0.33 3.0
1958 1586 804 8 0.38 2.6
1959 218 719 9 0.43 2.3
1960 623 623 10 0.48 2.1
1962 1303 583 11 0.52 1.9
1963 197 487 12 0.57 1.8
1964 583 377 13 0.62 1.6
1965 377 348 14 0.67 1.5
1966 348 328 15 0.71 1.4
1967 804 245 16 0.76 1.3
1968 328 218 17 0.81 1.2
1969 245
197 18 0.86 1.2
141 19 0.90 1.1
140 20 0.95 1.1
1970 140
 CONT…
• When such probability plot is prepared for flood events (with magnitude Q), then this kind of
plot is known as flood frequency plot.
• Depending upon the range of parameters Q and T , the scales in Y- and X-axes can be
arithmetic or logarithmic.
• A best- fit curve (trend line) is drawn through the plotted points, and then by suitable
extrapolation of the line, we can either find out the return period of a certain magnitude of
the event (which is absent from the sample data series) or find out the magnitude of the
event for a particular return period.
• This simple procedure yields reasonably good results for small extrapolation, but with
increase in extrapolation error increases.
 CONT…
• Hydrological data series:
• Partial duration series: A series of data which are selected so that their magnitude is greater
than a predefined base value.
• Annual exceedence series: If the base value is selected so that the number of values is equal
to the number of years.
• Extreme value series: Series including the largest or smallest values occurring in each of the
equally long time intervals of the record.
• Annual maximum series: Series with largest annual values.
• Usually annual maximum series is preferred to use in frequency analysis.
 CONT…
• When the magnitude of a hydro-climatic event differs significantly from the average or
usual range of magnitudes, then such events are termed as extreme events.
• This may take place over one day or a period of time, e.g., severe storms, flash floods,
droughts.
• Frequency analysis is done to determine the frequency of occurrence (or probability of
occurrence) of such extreme events.
• Most of the frequency distribution functions in hydro climatic studies can be expressed in the
form of the following equation, known as the general equation of frequency analysis, given
by
• 𝑿𝑻 = 𝑿 + 𝑲𝑺 …………………………(1)
• where
• 𝑋𝑇 = magnitude of the hydrologic variable with a return period of T ;
• 𝑋 = mean of the hydrologic variable;
• S = standard deviation of the hydrologic variable; and
• K = frequency factor, a function of the return period T and the assumed frequency distribution
function.
 CONT…
 Different probability distribution functions are available for the prediction of extreme
events. Some of them are:
 (i) Normal distribution
 (ii) Lognormal distribution
 (iii) Log-Pearson type III distribution
 (iv) Extreme value type I distribution (or Gumbel’s distribution)

i) Normal distribution
• Normal distribution is the most frequently used continuous probability distribution function.
When mean is zero and variance is 1, the distribution is called as standard normal
distribution.
• If a hydrologic variable (X) follows normal distribution, the frequency factor K equals its
standard normal variate Z.
 CONT…
• From equation 1, we can express K as K = (𝑋𝑇 −𝑋 )/ S, which is the standard normal variate Z.
• So, in order to determine an extreme event with a particular return period, we have to
calculate its exceedance probability (hence non-exceedance probability) and corresponding Z
value using a standard normal table.
• Now using this Z value as frequency factor K , the extreme event can be determined from
Equation 1.
• Example 4 : Consider a 50-year data of annual maximum 24 h rainfall depth at a particular
place follows normal distribution with mean 92.5mm and standard deviation 34mm.
Determine the magnitude of annual maximum rainfall with return period of 20 years.
• Solution: For the given 50-year data, mean 𝑋= 92.5 mm and standard deviation S = 34 mm.
• Now, for 20-year return period, T = 20; P(X > 𝑥20 ) = 1/20 = 0.05
• P(X ≤ 𝑥20 ) = 1 − 0.05 = 0.95
• From a standard normal table (Table B.1), for φ(Z) = 0.95, Z = 1.645
Thus, the frequency factor K = 1.645.
From Eq. 1, 𝑥20 = 𝑋+ K S = 92.5 + (1.645 × 34) = 148.43mm.
 CONT…
ii) Lognormal Distribution
 Lognormal distribution is a continuous probability distribution of a random variable which is such that
its logarithmic transformation follows a normal distribution.
 It is mostly applicable for hydrologic variables like monthly rainfall depth, river discharge
volumes.
 It is used to determine the extremes of variables at monthly and annual scales
 If a hydrologic variable (X) follows lognormal distribution, we have to transform the X values into a
series of Y values where Y = ln (X).
As X follows lognormal distribution, Y will follow normal distribution.
 Then, we have to follow the same procedure explained in normal distribution to determine the
frequency factor for Y series.
 Then using Eq. 2 (same as Eq. 1 but for variable Y), we can determine the magnitude 𝑦𝑇 for a
particular return period T and from 𝑦𝑇 we can compute 𝑋𝑇 by taking antilog(𝒚𝑻 )
 𝒚𝑻 = 𝒚 + 𝑲𝒚 𝑺𝒚 ……………………..(2)
𝟏 σ(𝒚𝒊 −𝒚)𝟐
 where 𝒚= σ 𝒚𝒊 ; 𝑺𝒚 =
𝒏 𝑵−𝟏
 CONT…
• 𝑦𝑇 = magnitude of the variable Y with a return period of T ,
• 𝑦 = mean of the magnitudes of Y,
• 𝑆𝑦 = standard deviation of the magnitudes of Y, and
• 𝐾𝑦 = frequency factor for Y
• The values of 𝑦 and 𝑆𝑦 can also be computed from the mean (𝑋 ) and standard
deviation (𝑆𝑋 ) of the original data. The equations are as follows:
2
1 𝑋
• 𝑦 = ln[ ]
2 𝐶𝑉2 +1
𝑆𝑥
• 𝑆𝑦 = ln(𝐶𝑣2 + 1) , where, 𝐶𝑣 =
𝑋
 CONT..
• Example 5: Consider the annual maximum rainfall data at a chiro gauging station over a time
period of 15 years as shown in Table below:
Year 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001
Annual
max 31 50.4 52 21 36.4 73.5 68.9 73 45 60

• Determine the 10 and 50- year Rainfall using lognormal distribution.

• Solution: For the given maximum rainfall data series (X), convert the X values into a series
of Y values where y = ln(x).
Year 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001
Annual
max (X) 31 50.4 52 21 36.4 73.5 68.9 73 45 60
Y= ln(X) 4.08 4.58 2.18 3.43 3.92 3.95 3.04 3.59 4.30 4.23
• Mean ( 𝒚 )= 3.73 and standard deviation (𝑺𝒚 ) = 0.707
• Now, for a 10-year flood, T = 10; P(Y > 𝒚𝟏𝟎 ) = 1/10 = 0.1
 CONT…
• P(Y ≤ 𝒚𝟏𝟎 ) = 1 - 0.1 = 0.90 = φ(𝒁𝟎.𝟏 )
• From a standard normal table (Table B.1), we got Z10 = 1.282 = the frequency factor K.
• Hence 𝒚𝑻 = 𝒚 + 𝑲𝒚 𝑺𝒚 = 3.73 + (1.282 × 0.77) = 4.636.
• thus, y10 = ln(x10 )= 4.636, hence, x10 = 103.13 cumec.
• Similarly, we can obtain x50 and x100.
iii) Log-Pearson Type III Distribution
 This distribution is popularly used for flood frequency analysis.
 Estimation of frequency factor follows the same procedure as in case of lognormal distribution
 the steps to be followed is as flow:
First, we have to convert the X values into a series of Y values where y = log10 (x).
 Then, three statistical parameters are calculated for this transformed data series Y , namely mean (𝒚),
standard deviation (Sy), and coefficient of skewness (Cs).
𝒏 σ(𝒚−𝒚)𝟑
 Cs =
(𝒏−𝟏)(𝒏−𝟐) 𝑺𝟑𝒚
 CONT..
 frequency factors are obtained from z-Table for a particular return period or exceedance
probability or compute using the following equation.
𝟐 𝟏 𝟑 𝟐 𝟐 𝟑 𝟒 𝟏 𝟓
 𝑲𝑻 = 𝒛 + 𝒛 − 𝟏 𝒌+ 𝒛 − 𝟔𝒛 𝒌 − (𝒛 -1) 𝒌 + 𝒛𝒌 + 𝒌
𝟑 𝟑

𝑪𝒔 𝟐.𝟓𝟏𝟔+𝟎.𝟖𝟎𝟐𝟗𝒘+𝟎.𝟎𝟏𝟎𝟑𝟑𝒘𝟐
Where, 𝒌 = ;𝒛 =𝒘−
𝟔 𝟏+𝟏.𝟒𝟑𝟐𝟖𝒘+𝟎.𝟏𝟖𝟗𝟑𝒘𝟐 +𝟎.𝟎𝟎𝟏𝟑𝟏𝒘𝟑

𝟏ൗ
𝟏 𝟐 𝟏
 𝒘 = 𝒍𝒏 and 𝑷 =
𝒑𝟐 𝑻

 Next, the magnitude yT for a particular return period T can be computed using Eq. 2.
 The value of xT can be computed from yT , using antilog(yT ).
 CONT..
Example 6: Consider the Example 4, and determine the 10-, 50-, 100-year floods using
log-Pearson type III distribution.

Solution: For the given flood data series (X), convert the X values into a series of Y values
where y = log10 (x).
Year 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001
Annual
max (X) 31 50.4 52 21 36.4 73.5 68.9 73 45 60
Y=
lOG10(X) 1.49 1.70 1.72 1.32 1.56 1.87 1.84 1.86 1.65 1.78
 CONT…
• For this Y series calculate; mean (𝒚) = 1.68, std. deviation
(sy) = 0.18, and coefficient of skewness (Cs) = −0.89. Now,
for a 10-year flood, T = 10; P(Y > y10) = 1/10 = 0.1.
• From Table 5.6, we got K10= 1.166 for CS= −0.8 and K10=
1.147 for Cs = −0.9
• So, for Cs = −0.89, by linear interpolation, K10 = 1.148
Hence 𝒚𝑻 = 𝒚 + 𝑲𝑻 𝑺𝒚 = 1.68 + (1.148*0.18) = 1.887
𝒚𝑻
XT = 10 = 101.887 = 77.1
 CONT…
iv) Extreme value type I distribution (or Gumbel’s distribution)
• Extreme value type I distribution is mostly used to analyze extreme events like flood peaks,
maximum rainfall
• For a hydrologic extreme event X, following Gumbel’s distribution, exceedance probability
of X = x0 is given by
−(𝒙−𝜷)ൗ
−𝑒 ά
• P(X ≥ X0 ) = 1- 𝑒
• introducing a dimensionless variable y known as Gumbel’s reduced variate, given by
y = (𝒙−𝜷)ൗά where α and β are scale and location parameter of Gumbel’s distribution,
respectively. So, the above equation rearranged as
−𝑒 𝒚 1
P(X ≥ X ) = 1- 𝑒
0 = OR
𝑇
𝟏
𝒚𝑻 = − 𝐥𝐧 𝒍𝒏 ……………a
𝑻−𝟏
• For Gumbel’s distribution, standard deviation and mean are given by
 CONT..
• Sx = 1.2823ά => ά = Sx /1.2825
• x = β + 0.5772 ά ⇒ β = 𝑥 − 0.5772 ά ⇒ β = 𝑥 − 0.4501Sx
• Using above equations of α and β, we can express y as
𝑥−β 1.2815(𝑥−𝑥 )
• 𝑦= = + 0.5772
α 𝑆𝑥

• Now for a particular return period T , let us designate y as y and x as x , then

T T

1.2815(xT−𝑥)
• yT = + 0.5772 or
𝑆𝑥
(yT − 0.5772)
• xT = 𝑥 + 1.2815
𝑆𝑥 or xT = 𝑥 +K𝑆𝑥
• is the general equation for hydrologic frequency analysis, where the
(yT − 0.5772)
frequency factor KT = 𝟏.𝟐𝟖𝟏𝟓
…………..b
 CONT….
• The equation above constitutes basic Gumbel’s equations and is only
applicable to a sample of infinite size (i.e., sample size N → ∞).
• But in practice, annual data series of extreme hydrological events like
maximum flood, maximum rainfall are of finite sample size.
• Hence, Eq. above is modified to take care of the finite sample size N
as shown below.
(yT − 𝑦𝑛) 𝟏
• xT = 𝑥 +K𝑆𝑥 where 𝑘 = where 𝒚𝑻 = − 𝐥𝐧 𝒍𝒏 is
𝑆𝑛 𝑻−𝟏
reduced variate for return period T and yn is reduced mean, a
function of T and sample size N; as N → ∞,yn → 0.5772. Sn is the
reduced
 CONT..
• standard deviation, a function of T and sample size N; as N → ∞,
Sn → 1.2825.
• Tables are available for determining yn and Sn for a certain sample size (N) and
return period (T ).
• Reduced variate yT can be directly calculated by using equation(a) above
• Then, K can be calculated by using respective equation(b) above.
• Or you can use table which directly gives the values of frequency factors (K )
for different sample size (N) and return period (T ).
• If the given sample size is said to be infinite (practically very large), then we do
not need to use this table, and we can directly use Eq. to calculate K.
 CONT..
• If the given sample size is said to be infinite (practically very large), then we do not need to
use this table, and we can directly use Eq. to calculate K.
• Example 7: The mean annual maximum daily rainfall in a city is 105mm and standard
deviation is 45mm. Determine the depth of daily rainfall with 5-year return period in that city.
Use Gumbel’s method and assume sample size to be very large.
• Solution: If X is a random variable which describes mean annual maximum daily rainfall in
a city, then mean (𝑥) = 105 and standard deviation (Sx) = 45. For T= 5-year return period,
reduced variate will be
𝟏
• 𝒚𝑻 = − 𝐥𝐧 𝒍𝒏 = 1.5
𝑻−𝟏
• As the sample size (N) is very large, we can directly use Eq. (b) to evaluatefrequency factor K
as shown below
(y − 0.5772)
• KT = T 𝟏.𝟐𝟖𝟏𝟓
= 0.72
• Now, the depth of annual maximum daily rainfall with 5-year return period in the city
• xT = 𝑥 +K𝑆𝑥 = 105 + (0.72 × 45) = 137.40 mm.