Journal of Environmental Research And Development Vol. 3 No.

2, October-December 2008

SEASONAL ARIMA MODEL FOR FORECASTING OF

MONTHLY RAINFALL AND TEMPERATURE
Inderjeet Kaushik1 and Sabita Madhvi Singh*2

1. Department of Civil Engineering, Institute of Technology, B.H.U., Varanasi, (INDIA)

2. Department of Civil Engineering, Haldia Institute of Technology, Haldia, (INDIA)

Received August 13, 2008 Accepted November 14, 2008

ABSTRACT

The prediction of temperature and rainfall on a seasonal time scales has been
attempted by various research groups using different techniques. The prediction of
these parameters on monthly and seasonal time scales is not only scientifically
challenging but is also important for planning and devising agricultural strategies.
This paper describes the Box-Jenkins time series seasonal ARIMA (Auto
Regression Integrated Moving Average) approach for prediction of temperature and
rainfall on monthly scales. The Box Jenkins technique is applied to predict
temperature and rainfall of next five years by analyzing last twelve years data
(1994-2006). Previous years data is used to formulate the seasonal ARIMA model
and in determination of model parameters. The performance evaluations of the
adopted models are carried out on the basis of correlation coefficient (R2) and root
mean square error (RMSE). The study conducted at Mirzapur, Uttar Pradesh (India).
The results indicate that the seasonal ARIMA model provide reliable and
satisfactory predictions for rainfall and temperature parameters on monthly scale.
Key Words : ARIMA, Meteorology, Rain fall, Hydroelectric power

INTRODUCTION fact a substantial part of the annual rainfall

The agricultural practices and crop yields
of India are heavily dependent on the
climatic factors like rainfall. Out of 142
million ha cultivated land in India, 92 million
ha (i.e. about 65%) are under the influence of
rain fed agriculture. Unlike irrigated
agriculture, rain fed farming is usually
diverse and risk prone. The monsoon season
* Author for correspondence
is the principal rain bearing season and in
over a large part of the country occurs in this
season. Small variations in the timing and the
quantity of monsoon rainfall have the
potential to impact on agricultural output.
Prior knowledge of monsoon behavior will
help Indian farmers and also policy makers,
to take advantage of good monsoons and also
to minimize crop damage and human
hardship during adverse monsoons. In
addition to the importance of the mean
monsoon seasonal rain-fall of a particular

506
Journal of Environmental Research And Development
year, the intra seasonal variability is also
crucial. Even if the mean annual rainfall is
normal, the delay in the monsoon onset and
the unusual monsoon lulls or breaks in
between may cause severe disruption of
agricultural activities, hydroelectric power
generation and even drinking water supply,
as is observed in many years in the past.
Therefore, farmers need the location and
time-specific information about the monsoon
behavior. As a result, forecasting of monsoon
on time scales of daily, weekly and monthly
is a major scientific issue in the field of
monsoon meteorology.
The forecasting of rainfall and tempe-
rature started more than a century ago17.
After the MONEX program, with the
increased understanding of the monsoon
phenomenon, more sophisticated empirical
statistical models5,7,14,15 were developed
which have been used with varying degrees
of success. However, successful long-range
prediction on shorter than seasonal/monthly
time scale and regional space scales still
remains elusive.
Study area
Fig. 1 : Location of study area Mirzapur city (India)
507
Vol. 3 No. 2, October-December 2008
The selected study area Mirzapur, Uttar
Pradesh (India) is located in the sub-tropical
belt of the river Ganga plain and experiences
a dry tropical climate with marked seasonal
variations. Mirzapur is located between the
forest area of Barkachha reserve forest on the
east and the Dauti Reserved forest on the
west and has a seasonally dry climate
dominated by dry deciduous and dry
savannah grass lands. Khajuri River flowing
along the western end of sub water shed. The
study area has a complex alignment in
orientation and distribution, structure, relief,
slope, estimate and vegetation. Here
monsoon is divided into three seasons
consisting of a cool dry season of northerly
winds during the months of Oct-Feb, a hot
dry season from March-June and hot wet
season from July to September. Monsoon
rain comes in the third week of June with on
set of hot wet season with southeast winds
(mid June – mid October). It is observed that
86 % of the total annual rainfall comes in the
rainy season from July to September.
Journal of Environmental Research And Development Vol. 3 No. 2, October-December 2008

Data Collection
Mirzapur district’s rainfall and temperature data (1994-2006) is used for analysis and model
formulation as shown in Figs. 2-4.
observed min. Temperature
35.00
Min.Temperature (C)

30.00
25.00
20.00
15.00
10.00
5.00
0.00
Aug- Jan- May- Sep- Feb- Jun- Nov- Mar- Aug- Dec- Apr-
93 95 96 97 99 00 01 03 04 05 07
Months
Fig. 2 : Observed minimum temperature in Mirzapur district (Jan. 1994-Dec. 06)

Fig. 3 : Observed maximum temperature in Mirzapur district (Jan. 1995-Dec. 06)

METHODOLOGY casual mechanism of different components.

A time series is defined as a set of
observations arranged chronologically i.e. a
sequence of observations usually ordered in
time. The principal aim of a time series
analysis is to describe the history of
movements in time of some variable at a
particular site. The objective is to generate
data having properties of the observed
historical record. To compute properties of a
historical record, the historical record or time
series is broken into separate components
and analyzed individually to understand the
508
Once properties of these components are
understood, these can be generated with
similar properties and combined together to
give a generated future time series. Analysis
of a continuously recorded rainfall and
temperature data time series is performed by
transforming the continuous series into a
discrete time series of finite time interval.
Mathematical modeling of rainfall data is a
stochastic process. Several mathematical
models based on the probability concept are
Journal of Environmental Research And Development
available. These models help in knowing the
probable weekly, monthly or annually
rainfall. Over the past decade or so, a number
Fig. 4 : Observed rainfall in Mirzapur district (Jan. 1994-Dec. 06)
Monthly rainfall and temperatures were
analyzed using time series analysis. Time
series models have been extensively studied
by Box and Jenkins (1976) and as there
names have frequently been used with
synonymously with general ARIMA process
applied to time series analysis and
forecasting. Auto Regressive (AR) models
were first introduced by Yule (1926) and
later generalized by Walker (1931), while
Moving Average (MA) models were first
introduced by Slutzky (1937). Wold (1938)
provided theoretical foundation for combined
Auto Regressive Moving Average (ARMA)
process.
Box and Jenkins (1976) have effectively
put together in a comprehensive manner, the
relevant information required to understand
and use time series ARIMA models. A
detailed strategy for the construction of
linear stochastic equation describing the
behavior of time series was examined.
Consider the function Zt represents
forecasted rainfall and temperature at time t
month. Yt is series of observed data of
rainfall and temperature at time t. If series is
stationary then a ARIMA process can be
represented as
509
Vol. 3 No. 2, October-December 2008
of models have been developed to generate
rainfall and runoff.
 pZt =  q Yt ….(1)
Where  is a back shift operator.
If series Y is not stationary then it can be
reduced to a stationary series by differencing
a finite number of times.
 pZt =  q (1-B)d Yt ….(2)
Where d is a positive integer, and B is
back shift operator on the index of time
series so that
BYt= Yt-1; B2Yt = Yt-2 and so on. Thus
further equation (2) can be simplified into
following equation.
(1-Ф1B-Ф2B2-…….-ФpBp) Zt = θ0+ (1-
θ1B-θ2B2- ………-θqBq) at ….. (3)
Where at’s a sequence of identically
distributed uncorrelated deviates, referred to
as “white noise”.
Combining equations (2) and (3) yields
the basic Box-Jenkins models for non
stationary time series
(1-Ф1B-Ф2B2-…….-ФpBp) (1-B)d Yt =
θ0+ (1-θ1B-θ2B2- ………-θqBq) at ….(4)
Equation (4) represents an ARIMA
process of order (p,d,q).
 Journal of Environmental Research And Development
Seasonal ARIMA model represented as
follows for a stationary series i.e.
differencing parameters (d &ds = 0) equal to
Zero, used for forecasting rainfall and
temperature.
 ps  pZt =  qs  q Yt
Where ps and qs are the seasonal
parameters corresponding to AR and MA
process.
Model of type of equation (5) was fitted
to given set of data using an approach
consists of mainly three steps (a)
identification (b) estimation (c) application
(forecasting) or diagnostic checking. At the
identification stage tentative values of p,d,q
and ps,ds,qs were chosen. Coefficients of
variables used in model were estimated.
Finally diagnostic checks were made to
determine, whether the model fitted
adequately describes the given time series.
Any inadequacies discovered might suggest
an alternative form of the model, and whole
iterative cycle of identification, estimation
and application was repeated until a
satisfactory model was obtained.
Min. temp
MIN.TEMP
1.0
.5
ACF
0.0
-.5
-1.0
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69
Lag Number
Fig. 5 : ACF of minimum temperature
Vol. 3 No. 2, October-December 2008
RESULTS AND DICUSSION
Data was analyzed using SPSS a
statistics computer package. Auto correlation
function (ACF) and Partial Auto correlation
function (PACF) of time series of
….(5)
temperature and rain fall were formed up to a
leg interval of 72. The ACF and PACF of
temperature and rainfall are shown in
Figs. 5-10.
The figures of ACF and PACF show that
variables (temperature, rainfall) are periodic
in nature. These functions behave similarly
in their period cycles involving seasonal
variations. Therefore given time series is
periodic and involve seasonal variations.
Based on these assumptions a period of 12
months i.e. one year is assumed for given
time series of temperature and rainfall.
Seasonal ARIMA models for the prediction
of Minimum and Maximum Temperature,
and rainfall with their statistics are described
in Table 1.
MIN.TEMP
Min. temp
1.0
.5
Partial ACF
0.0
-.5
-1.0
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69
Lag Number
Fig. 6 : PACF of minimum temperature
510
 Journal of Environmental Research And Development Vol. 3 No. 2, October-December 2008

Max. temp
MAX.TEMP Max. temp
MAX.TEMP
1.0 1.0

.5 .5

Partial ACF
ACF

0.0 0.0

-.5 -.5

-1.0 -1.0
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69

Lag Number Lag Number

Fig. 7 : ACF of maximum temperature Fig. 8 : PACF of maximum temperature

Rainfall
RAINFALL Rainfall
RAINFALL
1.0 1.0

.5 .5
Partial ACF
ACF

0.0 0.0

-.5 -.5

-1.0 -1.0
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69

Lag Number Lag Number

Fig. 9 : ACF of rainfall Fig. 10 : PACF of rainfall

Table 1
Seasonal ARIMA model
Parameters R2 RMSE
(pdq, psdsqs)
Minimum Temperature (504,202) 0.94 6.572
Maximum Temperature (303,202) 0.87 5.81
Rainfall (302,201) 0.72 62.40

Observed and predicted values of next fives years are determined and plotted as shown in
Figs. 11-13

511
Journal of Environmental Research And Development Vol. 3 No. 2, October-December 2008

50 Observed predicted
45
Max. Temperature (C)

40
35
30
25
20
15
10
5
0
Mar-94 Feb-96 Jan-98 Dec-99 Nov-01 Oct-03 Sep-05 Aug-07 Jul-09 Jun-11

Months
Fig. 11 : Comparison of observed and predicted values of maximum temperatures
In the analysis of observed data there are good accuracy as statistics of models
many missing values which results into less indicate. The accuracy of predictions made
accuracy of predicted model in case of
rainfall. Higher values of R2 of models for for rainfall by seasonal ARIMA model is less
temperatures and rainfall show better because data is abrupt and contains many
accuracy of prediction and can be used in missing values, which increases white noise
forecasting temperature and rainfall a few in the system; because no value is assigned
months ahead.
to missing values and are assumed zero.
CONCLUSION Accuracy of these predictions can be
Seasonal ARIMA models can predict increased in future by using these predicted
minimum and maximum temperature with values for missing values.

35.00 observed predicted

30.00
Min. Temperature (C)

25.00
20.00
15.00
10.00
5.00
0.00
Aug-93 Sep-95 Sep-97 Oct-99 Nov-01 Nov-03 Dec-05 Jan-08 Jan-10 Feb-12
M onths
Fig. 12 : Comparison of observed and predicted values of minimum temperatures

512
Journal of Environmental Research And Development
600
Rainfall(cm) 500
400
300
200
100
0
Aug-93 Sep-95 Sep-97 Oct-99 Nov-01 Nov-03 Dec-05 Jan-08 Jan-10 Feb-12
Fig. 13 : Comparison of observed and predicted rainfall
Box-Jenkins ARIMA technique can be
used to predict the seasonal and monthly
mean rainfall and temperature, using rainfall
and temperature time series as inputs.
Various verification statistics have shown
that prediction skill is good. Previous 12
years mean monthly rainfall and temperature
values are used to predict next year’s values
and can be predicted with sufficient lead-
time and good accuracy. This indicates that it
may be possible to develop a suitably
configured seasonal ARIMA model for
predicting monthly rainfall and temperature
of any region.
REFERENCES
1. Ashan M. O. and Connor K.M., A
reappraisal of the kalman filtering
technique, as applied in river flow
forecasting. Journal of Hydrology 161,
197–226, (1994)
2. Haltiner J. and Salas J.D., Short term
forecasting of snowmelt runoff using
armax models, Water Resources
Bulletin 24, 1083–1089, (1988)
3. Shamseldin A.Y., Application of a
neural network technique to rainfall-
runoff modelling. Journal of Hydrology
199, 272–294, (1997)
Vol. 3 No. 2, October-December 2008
observed predicted
Months
4. Basu S. and Andharia H.I. The chaotic
time-series of Indian monsoon rainfall
and its prediction. Proc Indian Acad Sci
(Earth Planet Sci), 101, 27-34, (1992)
5. Gowariker V., Thapliyal V., Sarker
R.P., Mandal G.S. and Sikka D.R.
Parametric and power regression
models-new approach to long range
forecasting. Mausam 40, 115-122,
(1989)
6. Tang Z., Almeida C. and de Fishwick
P.A., Time series forecasting using
neural networks versus Box-Jenkins
methodology. Simulations 57, 303-310,
(1991)
7. Shukla J. and Paolino D.A., The
Southern Oscillation and long-range
forecasting of the summer monsoon
rainfall over India, Mon Weather Rev
111, 1830-1837, (1983)
8. Sarle W.S., Neural networks and
statistical models. Proc 19th Ann SAS
Users Group International Conference,
1538-1550, (1994)
9. Satyan V., Is there an attractor for Indian
summer monsoon? Proc Ind Acad Sci
(Earth Planet Sci), 97, 49-52, (1988)
513
Journal of Environmental Research And Development Vol. 3 No. 2, October-December 2008

10. Navone H.D. and Ceccatto H.A.,
Predicting Indian monsoon rainfall: a
neural network approach., Clim Dyn,
10, 305-312, (1994)
11. Normand C., Monsoon seasonal
forecasting QJR Meteorol Soc, 79, 463-
473, (1953)
12. Perrone T.J. and Miller R.G.,
Generalised experimental markov and
model output statistics: a comparative
verification, Mon Weather Rev, 113,
1524-1541, (1985)
13. Pathasarathy B., Munot A.A. and
Kothawale D.R. All India monthly and
seasonal rainfall series : 1871±1993.
Theor ApplClimatol, 49, 217±224,
(1994)
14. Thapaliyal V., ARIMA model for long-
range prediction of monsoon rainfall in
Peninsular India. India Meteorol Dept
Monogr Climatology, 12/81, (1981)
15. Thapaliyal V., Stochastic dynamical
model for long-range prediction of
monsoon rainfall in peninsular India.
Mausam 33, 399-404, (1982)
16. Vautard R., Yiou P. and Ghill M.,
Singular spectrum analysis: a toolkit for
short, noisy and chaotic series. Physica D,
58, 95-126, (1992)
17. Walker G.T., Seasonal weather and its
prediction. Brit Assoc Adv Sci, 103, 25-
44, (1933)
18. Webster P.J., Magana V.O., Palmer
T.N., Shukla J., Thomas R.A., Yanai M.
and Yasunari T., Monsoons: Processes,
predict-ability, and prospects for
prediction. J Geo Phys Res., 13, 14451-
14510, (1998)




INSTRUCTIONS FROM PUBLISHERS

It is a condition for publication that the authors must give an undertaking in the
writing at the time of submission of papers that the manuscripts (research papers)
submitted to JERAD have not been published and have not been submitted for
publishing elsewhere, manuscripts are their original work. Furthermore, it should
also be noted that the manuscript will not be returned in any case, whether
accepted or rejected. Acceptance of research article will be communicated to
authors in due course of time.

514