12/3/2024

Characteristics of
Hydrologic Series

2.1 Types of hydrological time series

a. Full time series and partial duration series
If a time series contains all recorded observations over time, the series is known as full time
series, e.g. daily rainfall.
If a time series contains selected observations, the series is known as partial duration series, e.g. 1
day maximum rainfall (for each year).

b. Continuous and discrete time series

If time series data are recorded continuously with time, the series is called continuous time series,
e.g. water level data recorded continuously.
If time series data are recorded at discrete points in time, the series is called discrete time series,
e.g. daily rainfall, daily discharge.
Discrete time series is often derived from a ‘continuous time series. Usually in hydrology, a time
series is of the discrete type.
A continuous plot of a discrete time series should not be confused with a ‘continuous time series’.

1
 12/3/2024

c. Univariate (single) and multivariate (multiple) time series

Univariate time series is a time series of one variable at a given site, e.g. rainfall measured at each
site, mean rainfall of a basin. Multivariate time series is a set of two or more time series, e.g.
discharge data of different stations in a river.

d. Uncorrelated and correlated time series

If value of a variable at time t does not depend on the past values, the series is called uncorrelated
or independent.
Otherwise, the series is called correlated (auto-correlated or serially correlated), e.g.
autocorrelation of stream flow due to storage effects causing the water to remain in the system.

e. Intermittent and counting time series

If a variable has zero and non-zero value throughout the length of the record, the series is called
intermittent time series, e.g. rainfall data.
If a time series is derived by counting the number of occurrences of event, it is called counting
time series, e.g. count of rainy days in a month.

f. Regularly (evenly) and irregularly (unevenly) spaced time series

If data is recorded on regular time interval, the series is known as regularly spaced, e.g. daily
discharge data.
If data is not observed on regular time interval, the series is known as irregularly spaced, e.g.
water quality measurements.

2
 12/3/2024

g. Stationary and non-stationary time series

A stationary time series is a data sequence where statistical properties like mean and variance
remain constant over time, meaning the pattern of the series does not change significantly
throughout its duration, such as annual rainfall, annual discharge.
Otherwise, the time series is non-stationary, e.g. monthly rainfall. Non-stationary time series
exhibits changing statistical properties, often displaying trends, seasonality, or increasing variance
as time progresses, indicating a pattern that evolves over time
First order stationary: only mean remaining constant with time
Second order stationary: mean and variance remaining constant with time
Weak stationary: second order stationary
Strong stationary: stationary in mean, variance and other statistical properties

2.2 General properties of hydrological time series

Homogeneity: If data in series belong to one population or single events of data, the series is
homogeneous.
The non-homogeneity in a time series is due to multiple set of observation like natural or
manmade causes, such as land-use changes, urbanization, and deforestation.
Inconsistency in time series occurs due to systematic error in observation. The longer the time
series, the greater the chances of non-homogeneity.

Persistent: If value of variable depends on past values, the series is persistent. Time dependence
in hydrological series is due to the geophysical process of water storages in a basin, their random
fluctuation from year to year and periodic fluctuation within a year.

3
 12/3/2024

Components of time series

A hydrological time series contains both deterministic and stochastic components. The
deterministic components are: trend, jump and periodicity. Trend is a long term movement in time
series. The trend in time series exists due to natural or manmade causes, such as land-use changes,
urbanization, deforestation, climate change.
Periodicity is regular fluctuations (cyclic behavior) in time series. If the fluctuation in times
series is for a period of less than a year, the property is called seasonality. The seasonality is due
to the rotation of the earth around the sun.
The abrupt change in time series is referred to as jump or shift, which occurs due to sudden
change in basin characteristics.

Non-parametric test:
A non-parametric test in statistics does not assume that the data has been taken from a normal
distribution. A normal distribution belongs to a parametrized family of probability distributions
and includes parameters such as mean, variance, standard deviation, etc. Thus, a non-parametric
test does not make assumptions about the probability distribution's parameters.

In parametric test it is simple, the events follow the normal distribution curve and test like t-test,
z-test, Anova etc.

Level of significance:
Probability of being incorrect (P > 0.05 is the probability
that the null hypothesis is true). A statistically significant
test result (P ≤ 0.05) means that the test hypothesis is
false or should be rejected.

4
 12/3/2024

Detection of trend
A. Regression test for linear trend
Fit a straight line in the form of y = at+b.
y = given variable, t = time, a & b = constant

From linear regression analysis,

If a is not close to zero, then there is presence of linear trend in time series.

B. Spearman’s rank correlation test

It is a statistical measure to show the strength of a relationship between two variables. Minimum
10 numbers is required for the test.
List the data Xi, i = 1,2,………n, in date order, and the rank (ri) in ascending order of magnitude
(with smallest allocated rank 1 and largest rank n)
Compute spearman’s rank correlation (Rsp)

Compute t test statistic

Find tabulated value of t for α/2 level of significance (say α = 5%) and (n-2) degree of freedom.
If computed t < tabulated t, then there is no trend.

5
 12/3/2024

C. Mann -Kendall test

Rank (R) the time series data in ascending order of magnitude (with smallest allocated rank 1 and
largest rank n).
Compute Mann-Kendall test statistic S as

sgn (x) = 1 for x>0, sgn (x) = 0 for x = 0, sgn (x) = -1 for x<0
Standard deviation of S is

Z- test statistic is

Find tabulated value of Z for level of significance (e.g. α = 5%, Z = 1.96).

If computed Z<tabulated Z, then there is no trend.

6
 12/3/2024

Removal of data
1. Moving average method
Moving average is the average of n
successive intervals of data.
Compute moving average and subtract it
from original data. (discussed in subsequent
chapters)
2. Linear trend
Fit a straight line of the form y = at + b. At
time t, subtract the values of the line from
the original data.

Detection of periodicity
By visual inspection of time series: noticing effect of periods in the plots
By performing autocorrelation and spectral analysis (discussed in subsequent chapters)
Detection of jump: By visual inspection of time series

Split record test for checking stationarity

Testing whether mean and variance of subsets of a series are not significantly different.
Approach
Divide the record into two non-overlapping sub-sets, j = 1, 2,……, m and j = m+1, m+2,…..,n.
Compute mean and variance of each sub-set.
Examine the variance (V) first using F test
F=V(a)/V(b) where V(a)>V(b)
Compare the computed F with the tabulated value of F for m-1 and n-m-1 degree of freedom at
α/2 level of significance.
If the computed F < tabulated value of F, then the variances of subsets can be accepted as
estimates of the same population variance.

7
 12/3/2024

Compute pooled estimate of variance (Vp) as

Test statistics for mean is

Compare the computed t with the tabulated value of t for n-2 degree of freedom at α/2 level of
significance (two tailed test).
If the computed t < tabulated value of t, then the mean zero difference between two subsets is
accepted.

Double mass curve analysis for checking consistency

• Compute cumulative sum of data (mostly annual data is sufficient) of base stations (A).
• Compute cumulative sum of mean value of neighbouring stations (B).
• Plot A versus B and fit points by a straight line. If there is no break in the line, the data is
considered to be consistent.