Statistical Analysis in Hydrology

INTRODUCTION
• Hydrologic processes evolve in space and time in a manner that
is partly predictable, or deterministic, and partly random, or
stochastic process
• In some cases the random variability is so large that the
hydrologist is justified in treating data as purely random

• When there is no correlation between adjacent observations, the

output of the hydrologic system is treated as stochastic, space-
independent, and time-independent

• This type of treatment is appropriate for observations of

extreme hydrologic events, such as floods or droughts
• Extreme events
– Floods
– Droughts
• Magnitude of extreme events is related to their
frequency of occurrence
1
Magnitude 
Frequency of occurence

• The objective of frequency analysis is to relate the

magnitude of events to their frequency of occurrence
through probability distribution
• It is assumed the events (data) are independent and
come from identical distribution 3
 REVIEW OF BASIC CONCEPTS
Probabilistic
Outcome of a hydrologic event (e.g., rainfall amount & duration; flood
peak discharge; wave height, etc.) is random and cannot be predicted with
certainty.

Terminologies
Population
The collection of all possible outcomes relevant to the process of
interest. Example:
(1) Max. 2-hr rainfall depth: all non-negative real numbers;
(2) No. of storm in June: all non-negative integer numbers.

Sample
A measured segment (or subset) of the population.
 REVIEW OF BASIC CONCEPTS
Terminologies
- Random Variable
A random variable is a variable described by a probability distribution
which specifies the chance that the variable will assume a particular
value.
Convention: Capital letter for random variables (say, X) whereas the
lower case letter (say, x) for numerical realization that the random
variable X will take.
Example:
X = rainfall amount in 2 hours (a random variable) ;
x = 100.2 mm/2hr (realization).
- Random variables can be
- discrete (eg., no. of rainy days in June) or
- continuous (eg., max. 2-hr rainfall amount, flood discharge)..
 REVIEW OF BASIC CONCEPTS
Terminologies
Frequency & Relative Frequency
o For discrete random variables:
o Frequency is the number of occurrences of a specific event. Relative frequency is
resulting from dividing frequency by the total number of events. e.g.

n = no. of years having exactly 50 rainy days; N = total no. of years.

Let n=10 years and N=100 years. Then, the frequency of having exactly
50 rainy days is 10 and the relative frequency of having exactly 50 rainy days in
100 years is n/N = 0.1.

o For continuous random variables:

o Frequency needs to be defined for a class interval.
o A plot of frequency or relative frequency versus class intervals is called histogram or
probability polygon.
o As the number of sample gets infinitely large and class interval length approaches to
zero, the histogram will become a smooth curve, called probability density function.
 FREQUENCY ANALYSIS

  General
      Frequency analysis techniques are applied to carry out
flood and drought analysis for the purpose of
 
determining peak and low flow magnitudes required in
the design of hydraulic structures affected by high and
low flows, i.e,
(spillway, bridges, culverts, water intake, irrigation
intakes, flood control structures,etc.)
      Drought analysis refers to analysis of low streamflow
records and drought deficiency volumes
      Proper determination of design values for the hydraulic
structures is important for the safety, economy and
proper functioning of such structures
 Frequency Analysis
• Frequency analysis takes a different approach in
characterising baseflow by deriving the
relationship between the magnitude and
frequency of streamflow discharges from the
hydrographic record.
• In its most common application, a flow duration
curve (FDC) is generated.
• Instead of plotting as a time series, a flow
duration curve shows the percentage of time that
a given flow rate is equalled or exceeded
 Development of Flow Duration Curve
• FDC is constructed from flow data of fixed
time period (eg daily, monthly, annual) by:
– Sorting the flow data in order of decreasing
flow;
– Assigning a unique ranking number m to
each flow, starting with 1 for the maximum
flow to n for the minimum flow, where n is
the number of flow measurements.
 Development of Flow Duration Curve
• The probability P that a given flow will be equalled or
exceeded is defined by

• The flow-probability relationship is typically presented

as a log-normal plot
• Flow duration curves can be constructed for the entire
record of flow measurement, or for specific time
periods such as similar calendar months or seasons.
 Flow distribution curves

Flow distribution curves for examples of (a) high

baseflow and (b) low baseflow streams
 Frequency Analysis
• Hydrologic systems are sometimes impacted by
extreme events such as severe storms, floods, and
droughts.

• The magnitude of an extreme events occurring less

frequently than more moderate events.

• The objective of frequency analysis of hydrologic

data is to relate the magnitude of extreme events to
their frequency of occurrence through the use of
probability distributions.
  Objectives of Frequency Analysis

  
Magnitude of river Recurrence
interval Peak Flows (Q) or
  Return period (T)

   From data at flow gauging stations

 
Recurrence Interval/ Return Period (T) 
• It is an expression of the frequency of occurrence
of a flow magnitude of a certain magnitude
• It is defined as the average time interval which a
flow of specified magnitude is equaled or
exceeded
 Objectives of Frequency Analysis

 The relationship between the flood magnitude (Q T)

and its return period (T) can be estimated from the
 
analysis of observed flood peaks, while the low
flow magnitude (qT) and its return period (T) can
be estimated from the analysis of observed low
flow values
 For the analysis to be valid, the flood peaks and
low flow values, under consideration, must be of
random magnitude and mutually independent.
 Objectives of Frequency Analysis
The results of flood flow frequency analysis can be
used for many engineering purposes;

– For the design of dams, bridges, culvert, and flood

control structures.
– To determine the economic value of flood control
projects.
– To delineate flood plains.
– To determine the effect of encroachments on the
flood plain.
 Return Period
• Suppose that an extreme event is defined to have occurred
if a random variable X is greater than or equal to some
level xTr

X  x Tr

The recurrence interval, is the time between occurrences of X  xTr

 Return Period

Thus , “the return period of an event of a given magnitude may

be defined as the average recurrence interval between events
equalling or exceeding a specified magnitude”.

The probability of occurrence of the event X  xTr in any

observation is

p  P(X  x Tr )
p 1
E()  
2 p
[1  (1  p )]
Hence, E() = Tr = 1/p
 Return Period

The probability of occurrence of an event in any observation is

the inverse of its return period.

1
p  P(X  x Tr ) 
Tr

What is the probability that a Tr-year return

period event will occur at least once in N years?
 Return Period

• P(X < xTr each year for N years) = (1-p)N

• P(X  xTr at least once in N years) = 1-(1-p)N or

• P(X  xTr at least once in N years) = 1-[1-(1/Tr)]N

 Hydrologic Data
  Series

Data Relevant for Frequency

Analysis  Annual Maximum (AMax) series
     Partial Duration (PD) series
   Annual Minimum (AMin) series
 Annual Maximum Drought Volumes (AM-DV)
series
  Data Relevant for FA

Annual Maximum series (AM)

 
  • The series consists of the peak flow of each year and it
is the most frequent used series
• One of the aspects in favour of the AM series is the
reasonable assumption that the data is not serially
correlated
• A disadvantage of AM series is that the second or third,
etc, highest event in a particular year may be higher
than the maximum event in another year and yet these
values are totally disregarded
  Data Relevant for FA

Partial Duration Series (PD)

• Is a series of data which are selected so their magnitude is greater
than a predefined base value
 
• If the base value is selected so that the number of values in the
series is equal to the number of years of the record, the series is
called an annual exceedence series.
• The series consists of peaks exceeding the threshold q0.
• The PD includes all events above a certain base magnitude, i.e., the
disadvantage of disregarding some of the significant high events in
any particular year in AM series is remedied
• Each event that is included in the PD series must be separate and
distinct
 Data Relevant for FA
Partial Duration Series (PD)
• An extreme value series includes the largest and smallest values
occurring in each of the equally-long time intervals of the record. The
time interval length is usually taken as one year, and a series so
selected is called annual series.
• Using largest annual values, it is an annual maximum
series.
• Selecting the smallest annual values produces an
annual minimum series.
 Daily Flows Peaks over Threshold Annual Maximum Values
450

400

350

300
Flow magnitudes (Cumecs)

250

200

150

100

50

0
1/1/1980 12/31/1980 12/31/1981 12/31/1982 12/31/1983 12/30/1984
Time in Days
  Data Relevant for FA

Annual Maximum Drought Volumes (AM-DV) series

• The Annual Maximum drought volumes series is formed by determining
the maximum deficiency of river flow to meet the specified water
demand in each year of record at a given location
 
• The set of observed annual maximum deficiency volumes at any
gauging station is assumed to be a random statistical sample from the
population of all possible annual maximum deficiencies at a given site.
  Estimation of Frequency/Probability of occurrence of an Event

Graphical Approach
•  Frequencies are determined simply by arranging observed AM
  events in the order of magnitude
• The plotting position formulae are applied to compute the
probability of occurrence of observed AM events
 
• Plot of magnitudes of AM events versus computed probabilities of
occurrences provides a frequency curve which can be used to
estimate the probability of occurrence of any specified event

• Plot of the frequency curve on a probability paper (special

designed paper) will plot the curve in a straight line curve.
Probability Plotting

Bloom’s Formula : b=3/8

Probability Plotting

Designed Probability
Graph
 Probability Plotting

m
Simple Formula : P(X  xm ) 
n

m 1
California’s Formula : P(X  xm ) 
n

m  0.5
Hazen’s Formula (1930) : P(X  xm ) 
n

m  0.3
Chegodayev’s Formula : P(X  xm ) 
n  0.4
 Probability Plotting

m
Weibull’s Formula : P(X  xm ) 
n 1

When m = the rank of a value in a list ordered

by descending magnitude
n = the total number of values to be
plotted
xm= the exceedence probability of the
mth largest value
 Probability Plotting

Form of most plotting position formulas

m b
P(X  xm ) 
n  1  2b

When m = the rank of a value in a list ordered

by descending magnitude
n = the total number of values to be
plotted
xm= the exceedence probability of the
mth largest value
b = a parameter
Probability plot of flood flows
Frequency curve plotted on Gumbel probability paper
 Example 1
• The values for annual precipitation in station X
from 1911 to 1979 are shown in Table 1.

What is the probability that the annual

precipitation R in any year will be less than 35
in? Greater than 45 in? Between 35 and 45 in?
Annual precipitation in Station X, 1911-1979 (in)
 Solution:

There are n = 79-11+1 = 69 data. Let A be the event R < 35.0,

B the event R > 45.0. nA = 23, nB = 19.

Therefore,
P( A)  23 / 69  0.333
P( B )  19 / 69  0.275.

The Probability that the annual precipitation is between 35 and 45

can now be calculated

P (35.0  R  45.0)  1  P ( R  35.0)  P ( R  45.0)

= 0.392
• Probabilities estimated from the sample data are
approximate, because they depend on the specific
values of the observations in a sample of limited size

• An alternative approach is to fit a probability

distribution function to the data and then to
determine the probabilities of events from the
distribution function
Many statistical distributions have been proposed for flood
frequency analysis.
The distributions have been proposed because of their ability to
model different shapes of histogram of flood peaks as shown in Fig.2

Fig.2 Typical shapes of histogram for flood peaks

Probability distributions for fitting Hydrological data
 Some Commonly Used Distributions
• NORMAL DISTRIBUTION
 1  x  μx 
2


f N x | x,  2
x = 1
exp -
 2

 σ


 , for  < x < 

2π  x   x  

Standardized Variable:
X 
Z


Z has mean 0 and standard deviation 1.

The corresponding standard normal distribution has
probability density function

1 z2 / 2
f ( z)  e   z  
2
PDF for the standard normal distribution, (μ = 0, σ = 1)

0.5

0.4 1 z2 / 2
f ( z)  e
(z ) 0.3 2
0.2

0.1

-3 -2 -1 0 1 2 3

The standard normal cumulative distribution function

z
1 z2 / 2
F ( z)  
 2
e du

Where u is the dummy variable of integration, has no analytical form.

Its values are tabulated in Table
Table 1-Cumulative probability of the
standard normal distribution
  Statistical Distributions to Estimate Flood Magnitudes

• The statistical distributions provide the essential basic formulae to

model a peak and low flow data aimed at deriving frequency
curves required to determine flood low flow magnitudes.
 
• Many statistical distributions have been proposed to date. In this
course only three commonly used distributions are considered
namely:
        i) Gumbel/Extreme value type 1 (EV1) (Gumbel,1941)
        ii) Lognormal (LN) (Hazen,1914)
iii) Log Pearson type 3 (LP3) (USWRC,1967)
  Risk of Failure
• The decision of what return period is
appropriate for the design of a particular project is
 
not solely a hydrological problem. Constraints
usually considered include: Economic, Political
& Environmental
• Improvement on the safety or reliability of the
scheme has implications on the costs of the
project
• It is however important to consider the
probability, or risk, of the design flood being
exceeded during the expected life of the project
  Estimation of Risk of Failure
Example:
  Suppose the design event has a return period of T years and the
projected life period of the project is L years
   
• The corresponding annual probability of exceedance, p is given
as p = 1/T
• The probability of non-occurrence in any one year q, is given as
q = (1-1/T)
• The probability of non-occurrence in L years is q = (1-1/T)L
• The probability that the flood magnitude Q will occur at least
once in the L years is r =1-(1-1/T) L
• The value of r is the risk of failure, the flood Q will be exceeded
in the L years of the projected life of the project