12/12/2024

4. Autocorrelation analysis 4.1 Correlation and regression

Correlation
Correlation determines the degree of association between two
variables. The association between two variables is called cross-
correlation. Correlation measures the degree of linear dependence
between two variables.

Higher value of one variable associated with higher value of other or

lower value of one variable associated with lower value of other gives
positive correlation. Otherwise, the correlation is negative.
If two variables are independent, then their correlation will be 0.

Regression
Correlation describes the strength of an association between two
variables, and is completely symmetrical, the correlation between A
and B is the same as the correlation between B and A
Regression analysis produces a regression equation where the
coefficients represent the relationship between each independent
variable and the dependent variable.
Regression is the relationship between dependent and independent
variables. It represents the cause and effect relationship, which is
described by polynomial equation.
Multiple regression consists of more than one independent variables.

1
 12/12/2024

Autocorrelation
Autocorrelation is the correlation of a variable with its past and
future values (correlation in itself).
The correlation is between two values of the same variable at
different times Xi and Xi+k. is called the Autocorrelation.
When the autocorrelation is used to detect non-randomness, it is
usually only the first (lag 1) autocorrelation that is of interest. When
the autocorrelation is used to identify an appropriate time series
model, the autocorrelations are usually many lags according to
requirements.
Autocorrelation
coefficient or Serial
correlation coefficient
is given by

Autocorrelation

The plot of serial correlation

coefficient with time lag is
called Correlogram or
Autocorrelation function
(ACF).
Partial Autocorrelation
function (PACF) is the plot
of serial correlation by
neglecting some values.

2
 12/12/2024

4.2 Classification of time series considering deterministic and stochastic process

A time series is an observed realization of the stochastic process, and

is the basis for analysis, identification and description of the process.
Any time series of a random phenomenon can generally be
considered to consist of a deterministic component which is the one
that can be determined uniquely and a stochastic component
consisting of chance and chance dependent effects of event.

a. Sinusoidal series
Sinusoidal data can be represented by

3
 12/12/2024

b. Complex periodic series

Series which repeats itself at regular interval is the periodic series.

II. Non-periodic
a. Almost periodic series
In almost periodic series, there is no common factor for frequency and
no period (Tp).
Physical phenomena producing almost period series frequently occur
in practice when the effect of two or more unrelated periodic
phenomena are mixed.
A typical example of the series composed of the almost periodic and
stochastic elements are tidal levels along coasts and bays. The three
cycles which affect the tides namely the earth’s daily rotation, the
moon’s revolution around the earth and the earth’s revolution around
the sun don’t have equal frequency.

4
 12/12/2024

b. Transient series
All non-periodic data other than almost periodic is called transient
series. Such series is represented by time varying function. The
physical phenomena which produce transient series have numerous
process and drivers.
Such series are formed causally in the case of hydrological series,
superimposed on the other such as periodic and stochastic
components. Three simple examples of the transient series are as

: Damped oscillatory

b. Transient series
Another simple examples are linear or nonlinear trends and jumps in
the hydrological series.
Trends are linear or non-linear and created by the slow changes.
Jumps are created by the sudden changes in the basin

5
 12/12/2024

2. Stochastic series
Stochastic series is described by the probability distribution. The
most widely used distribution in time series analysis for describing
random variable is the normal distribution. The observed series are
fit in the normal distribution and analyzed it.
Stationary Series:-
Stationary stochastic series are time invariant (does not depends upon
time change).
In Hydrology, the mathematical techniques for analyzing stationary
series are well developed.
After identifying and removing deterministic component, the
resulting series is considered to be approximately stationary.
The stationary series is classified into dependent and independent
series.

2. Stochastic series
The independent stationary series is called the pure random series
and easily described through the single probability distribution
function.
Dependent series is again divided into linear dependent and non-
linear dependent.
The linear dependent stationary stochastic series are analysis based
on moving average and autoregressive series.
Non-stationary Series:- It is time variant series for a long duration.
So the statistical parameters are changed with the long run like river
runoff, climate change etc. events.
Autoregressive integrated moving average series represents a kind of
non stationary series for which the analysis techniques have been
developed.

6
 12/12/2024

2. Stochastic series
Components of time series
Observed time series =
stochastic component +
trend component +
periodic component +
catastrophic event

Autocorrelation analysis for the investigation of independent process

It is very extensively used to determine the linear dependence in the
existing series.
The method consists of estimating the autocorrelation coefficients
for various lags and plotting them against the respective lags
resulting in what is known as an empirical correlogram.
The empirical correlogram thus obtained is then compared with the
theoretical correlogram of known process to identify a suitable
mathematical model to describe the process under consideration.

a. The theoretical Autocorrelation function (population data) rk for

an independent and random process is zero for all lags except
zero lag. (for zero lag, it is 1).
b. The Autocorrelation function (ACF) for observed data (sample
data) will not be exactly equal to zero due to sampling effects.
The ACF fluctuates around zero for independent sample data.

7
 12/12/2024

Autocorrelation analysis for the investigation of independent process

To verify this fact the sampling distribution of rk is required.
Anderson has proposed a test of significance for the autocorrelation
coefficients of the time series.
In a random, normally distributed series of length n, if n is large in
comparison to k.
rk is found to be normally distributed with a mean of -(n-k)-1 and a
variance of (n-k-1)/(n-k)2.
The 95% confidence limits (+ for upper limit and – for lower limit)

If all the autocorrelation coefficients for various lags fall within the
tolerance band, the sample data is considered to be independent

The autocorrelation coefficients are falling outside the tolerance band

the annual run off series being analyzed is not independent.

8
 12/12/2024

Correlogram of different types of series

Complex periodic with a period Tp: The correlogram is periodic with
same Tp and unit amplitude.

Linear trend: The correlogram is a horizontal line.

Combination of periodic sine component and independent random

component: The correlogram is periodic with same frequency as sine
component, but the amplitude is not unity.

Combination of complex periodic and independent random

component: The correlogram is periodic with frequency equal to
fundamental frequency of complex periodic series.