Print : ISSN 0974-8

Online : ISSN 0976-2485

NAAS Rating : 2.7

Trends
in
Biosciences
A Monthly International Journal

Volume 7 Number 1 January, 2014

Online version available at

www.trendsinbiosciencesjournal.com

Dheerpura Society for Advancement of Science

and Rural Development
 Print : ISSN 0974-8
Online : ISSN 0976-2485
NAAS Rating : 2.7

Trends
in
Biosciences
A Monthly International Journal

Volume 7 Number 1 January, 2014

Online version available at

www.trendsinbiosciencesjournal.com

SPECIAL OFFER
Life Membership
Journal Membership - Rs. 500
Benefit : Online access of Journal for lifetime alongwith certificate of membership
Journal Membership with free authorship - Rs. 4000
Benefits : Authorship charges for publishing research papers in
Trends in Biosciences Journal is nil (No authorship charges) for lifetime.
Award of fellow of DSAS&RD Society

Dheerpura Society for Advancement of Science

and Rural Development
Branch Office : Kanpur (U.P.) 208 018, India
Trends in Biosciences 7(1): 39-41, 2014
Weather Based Pre-harvest Forecasting of Rice at Kolhapur (Maharashtra)
B S DHEKALE, SAWANT P K AND T P UPADHYE
Agrimet Division, Indian Meteorological Department, Pune 411005, Maharashtra- India
email : bhagyashreedhekale@yahoo.com
ABSTRACT
Crop yield forecasts provide useful information to farmers,
distributors, government agencies and other users and are
needed for policy decisions. In present study pre harvest forecast
models for different weeks after sowing (14th Wk to 20th Wk)
were developed using weather indices based on varied weather
parameters (1985-2009) for estimating rice productivity in
district Kolhapur (Maharashtra). Multiple linear regression
(MLR) technique was used and the cross-validation of the
developed forecast models were tested for their accuracy using
the data of year 2010 and 2011. Out of developed seven
forecasting models 18 th Wk model showed maximum R 2 and
minimum RMSE. The 18th Wk forecast model accounts for 89
% of variation in yield with RMSE 107. On the basis of selected
weather indices in the model, the rice yield was validated for
year 2010 and 11 with -4.9 and 3.6 % variation respectively.
Key words Rice yield, Forecast, MLR techniques and weather
indices
Agriculture is backbone of Indian economy, contributing
about 40 per cent towards the Gross National Product (GNP)
and provide livelihood to about 70 per cent of the population.
Its share in country’s GDP is about 17% against 2-4% in
developed countries. India is the second largest producer
and consumer of rice (nearly 65% of the total population) in
the world. Rice production in India crossed the mark of 100
million MTnes in 2011-12 accounting for 22.81 % of global
production (http://agricoop.nic.in/imagedefault/trade/
Rice%20profile.pdf, last accessed on 20 th Sept 2013).
Forecasting opens menu window on to future. It is a medium
guiding towards plans for the development of a better future.
Crop yield forecasting based on weather, staff scheduling,
business production planning and multistage management
decision analysis are among distinctive examples of
forecasting areas. Reliable pre-harvest forecast of crop yield
is likely to provide valuable information in regard to sale,
storage, export, industries and government for advanced
planning. Understanding the impact of climate variability and
change on crop yields is fundamental to the success of such
research. It is also an essential step towards the development
of key adaptive strategies to scope up with climate change.
Similar work have been done by many scientist viz. Agrawal,
et al., 1980 developed forecasting models for the rice yield in
Raipur district based on weekly data of weather parameters. It
was found that forecasting of rice yield using weather
variables is best possible only two and half months after
sowing for a crop of five month duration. Talukder, et. al.,
2001 developed crop yield forecast model for Amen rice from
42th to 45th week in Bangladesh using backword elimination
procedure. Jain, et al., 1980 found that developed pre harvest
model was reliable to forecast rice yield only after about two
months of sowing. Agrawal, et al., 2001 developed forecasting
model for wheat in Vindhyanchal Plateue zone of Madhya
Pradesh. Agrawal and Mehta, 2007 developed several weather
based forecasting models for crop yield of rice, wheat, sorghum,
maize and sugarcane at selected districts/agro climatic zones/
states of India using regression analysis, discriminant function
analysis and water balance technique. Analysis revealed that
that reliable forecast of crop yield could be provided before
harvest and reliable forecasting of yield could be obtained
when the crops were 12 weeks old i.e. about 2 month before
harvest. Pre harvest forecasting is extremely useful in
formulations of policies regarding stock, distributions and
supply of agricultural produce to various part of the country.
The main objective of the present study was to develop a
simple methodology for forecasting the rice yield before
harvesting using weather parameters.
MATERIALS AND METHODS
The study was carried for Kolhapur district of
Maharashtra. It is located in the Sub Mountain Agroclimatic
zone of Maharashtra. It is stretched between 16.6990° N,
74.2281° E. District has a humid subtropical climate with
variations between summer and winter temperatures. The
study utilized secondary yearly yield data of rice (kg/ha) for
27 years (1985-86 to 2011-12) which were collected from State
Department of Agriculture, Pune and weekly weather data for
respective years was collected from Indian Meteorological
Department (IMD), Pune. Five weather parameters were
included in the study; namely Maximum Temperature (X1),
Minimum Temperature (X2), Total Rainfall (X3), Relative
Humidity at 7.00 hrs (X4), Relative Humidity at 14.00 hrs (X5).
However, weekly weather data related to kharif crop season
starting from a fortnight before sowing up to 15th October
were utilized for the development of statistical models. The
rice yield distribution over the year is shown in Figure 1.
Weather Indices Model:
The pre-harvest forecast model was developed for
different weeks from 14th wk to 20th wk after sowing by using
weather indices as predictors. Weather indices were computed
using varied weather parameters for the year 1985-86 to 2009-
40 Trends in Biosciences 7 (1), 2014

Table 1. Rice yield forecasting model equations using RESULTS AND DISCUSSION
weather indices model
On the basis of weather data available for pre-harvest
Pre R2 week i.e. from 14th to 20th of week, seven forecasting model
Model Model
Harvest (Adj R2) F cal RMSE equations for estimating the yearly rice yield were developed
No. Equations
Week No.
using stepwise regression analysis. These equations were
Y= 5818.92 + 0.85
I. 14
0.998* Z141 (0.84)
17.66 11 1.67 developed using generated variables (based on correlation
Y= 3825.18 +
coefficient as weights) as independent variables. The model
0.85 equations for forecasting the yield of rice are presented in
II. 15 1.001* Z141 + 14.56 11 2.26
(0.84)
0.73* Z241 Table 1.
Y= 2232.13 +
0.995* Z141 + 0.78 The stepwise regression analysis results obtained in
III. 16 20.61 11 4.32
0.63 *Z241 + (0.77) Table 1 showed that all the generated variables entered in
0.86* Z50 different model equations affected the rice yield significantly
Y= 2868.65 + through F-test. The value of Adj. R2 for estimating the rice
1.251* Z141 + 0.79
IV. 17
0.76*Z241 + (0.78)
18.68 11 4.62 yield in different models (1 to 7) varied from 81 to 89 per cent
0.84 *Z50 with maximum 89 per cent for Model V (5) indicating 89 per
Y= 4486.99 + cent variation in the rice yield due to generated variables Z141,
0.966*Z141 + 0.89 Z241 and Z50 together. The constituent of each of these
V. 18 20.35 10 6.85
3.01 Z11 + (0.89)
0.96*Z50 generated variables is as follows:
Y = 5346.23.13 0.88 18
VI. 19 + 0.943 * Z141 23.48 11 0.02
+ 2.87 Z11
(0.87) Z141   r1w Tmax w RHI w ,
Y = 7568.66 + w1
0.88
VII. 20 0.968*Z141 + 22.27 10 9.70
(0.87) 18 18
1.06 121
Z 241   r24 w Tmin w RHI w , Z 50   r50 w RHII w
w 1 w1
10. The cross-validation of the developed forecast models
were tested their accuracy using the year 2010-11 and 2011- where,
12. The weather indices were computed as un-weighted indices r1w= coefficient of correlation between yield (Y) and
and weighted indices, where weight being correlation product of 1st and 4th weather variable able (viz. Average Weekly
coefficient between yearly crop yields (de-trended) and Maximum Temperature (X1) and Relative Humidity at 7.00 hrs
weather parameters with respective weeks. The forms of (X4))
weather indices were given as below:
r24w = coefficient of correlation between yield (Y) and
Stepwise regression analysis was used for weather product of 2nd and 4th weather variable (viz. Average Weekly
indices, which were significantly affected by detrended yield Minimum Temperature (X2) and Average Weekly Relative
for different weeks (Draper and Smith, 1981; Gomez and Gomez, Humidity at 7.00 hrs (X4))
1966). Developed forecasted models were compared for
selection of best week for forecasting rice yield. r50w = coefficient of correlation between yield (Y) and 5th
weather variable (viz. Average Weekly Relative Humidity at
Comparison of the developed statistical forecasting 14.00 hrs (X5))
models:
For the comparison of developed statistical forecasting
models Root Mean Square Error (RMSE) were calculated on
the basis of test data set as follows:

In addition to RMSE, forecast error percentage was

calculated for validation years (2010-11 and 2011-12) for
developed models.

Fig. 1. Rice yield (kg/ha) during 1985-86 to 2011-12

 DHEKALE et al., Weather Based Pre-harvest Forecasting of Rice at Kolhapur (Maharashtra) 41

Table 2. Forecast error percentage for developed weather indices models

Model No. Pre Harvest Week No. Year Actual yield Forecasted yield Forecast Error (%)
2010 2906.8 3060.7 -5.3
I. 14 Wk
2011 2663.0 2688.3 -1.0
2010 2906.8 3022.0 -4.0
II. 15 Wk
2011 2663.0 2844.5 -6.8
2010 2906.8 3102.0 -6.7
III. 16 Wk
2011 2663.0 2846.1 -6.9
2010 2906.8 3195.0 -9.9
IV. 17 Wk
2011 2663.0 2672.4 -0.4
2010 2906.8 3050.1 -4.9
V. 18 Wk
2011 2663.0 2567.8 3.6
2010 2906.8 2880.7 0.9
VI. 19 Wk
2011 2663.0 2412.7 9.4
2010 2906.8 3004.3 -3.4
VII. 20 Wk
2011 2663.0 2678.3 -0.6

harvesting rice yield).

Fig. 2. Actual and forecasted yield of 18 Wk model
These results indicated that variable generated by
correlation coefficients as weights can prove to be a meaningful
analytical approach in developing rice yield forecasting
models. Also RMSE for model V was found minimum with
value 106.85kg/ha. On the basis of Adj. R2 values and RMSE
obtained for different models given in Table 1 the best fit
forecasting model for estimating the pre-harvest rice yield
was model V (18th week of crop season or two months before
Forecast error percentage was calculated (Table 2.) for
developed models for validation year ( 2010-11 and 2011-
12).The forecast error for model V was observed as -4.9 and
3.6 percent for 2010 and 11 respectively.
This study reveals that stepwise multiple linear
regression techniques can be successfully used for pre-harvest
crop yield forecasting. Reliable and accurate pre harvest
forecast of rice can be obtained using weather indices model
after 18wks of sowing. This methodology can be applicable in
many crops viz. cereals, oil seeds, sugarcane etc. to develop
district-wise pre-harvest forecasting models and these
forecasts have significant value in agricultural planning.
LITERATURE CITED
Agrawal, R. and Mehta, S.C. 2007. Weather based forecasting of crop
yields, pest and diseases – IASRI Models. J. Ind. Soc. of Agril. Stat.
61 (2):255-263.
Agrawal, R., Jain, R.C. and Mehta, S.C. 2001. Yield forecast based on
weather variables and agricultural inputs on agro-climatic zone basis.
Indian J. Agric. Sci. 71(7):487-490.
Agrawal, R., Jain, R.C. Jha, M.P. and Singh, D. 1980. Forecasting of
rice yield using climatic variables. Indian J. Agric. Sci. 50(9):680-
68 4.
Draper, N.R. and Smith, H. 1981. Applied regression analysis, second
edition, John Wiley and sons, New York.
Gomez, K.A. and Gomez, A.A. 1984. Statistical Procedures in
Agricultural Research. 2nd edition, New York, Chichester, etc.:
Wiley paperback, pp. 680
Jain, R.C., Agrawal, R. and Jha, M.P. 1980. Effect of climatic variables
on rice yield and its forecast. Mausam, 31(4):591-596.
Talukader, R.K., Maniruzzaman, A.F., Sarker., K.S. 2001, Production
Forecast of Aman Rice Using Crop Yield-Weather Regression Model.
Ministry of Food, Dhaka Early Warning and Food Information
System Project, report 12.

Fig. 3. Deviation of actual yield from forecasted yield of 18

Recieved on 19-10-2013 Accepted on 27-11-2013
Wk model