STATISTICAL EVALUATION OF DIFFERENT METHODS FOR PREHARVEST FORECASTING OF GROUNDNUT YIELD IN JUNAGADH DISTRICT OF GUJARAT 2899
Loading...
Date
2019-08
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
JAU, JUNAGADH
Abstract
A timely and reliable forecast of crop yield needs more emphasis for monsoon
dependent country like India where, the economy is mainly based on agricultural
production. Weather is a major factor affecting crop production in advanced
agricultural systems. The large variation in yield from year to year and place to place
is dominated by the weather parameters. In view of fluctuating weather effects, a
timely and reliable forecast of crop productivity could help in deciding the policies. A
proper forecast of production of commercial crops is very important in an economic
system. There is close association between crop productions with prices. An
unexpected decrease in production reduces marketable surplus and income of the
farmers and leads to prices rise. An efficient forecasting is thus a pre-requisite for
food supply information system at district and state level.
The present study has been taken up to, (1) To identify the nature of effect of
weather variables and time period on groundnut yield in Junagadh district of Gujarat,
(2) To explore the possibility of suggesting suitable statistical method for pre-harvest
forecasting of the groundnut yield in Junagadh district of Gujarat and (3) To compare
the efficiency of MLR and ARIMA models.
To estimate the effect of weather variables and time period, yield and weather
data for 31 years (1985 to 2015) were used. The weekly averages of weather variables
viz., maximum temperature (MAX T), minimum temperature (MIN T), morning
ABSTRACTrelative humidity (RH1), afternoon relative humidity (RH2) and weekly total rainfall
(RF) from 24th to 37th standard meteorological week of the respective year were
considered in the study.
In all five approaches were used in the study. Out of these, four approaches
used weather variables which were further categorized as based on generated weather
variables (correlation coefficient as weight and week number as weight) and based on
original weather variables (week wise approach and crop stage wise approach). Three
sets of multiple linear regression equations consisting of 23, 24 and 25 years data
considering the data up to 10, 12 and 14 weeks for each model were fitted. In addition
to this, ARIMA model which used only time series yield data was also used.
Similarly, three sets of ARIMA model consisting 23, 24 and 25 years data for each
model were fitted.
The models based on 10 weeks with 25 years data using correlation coefficient
as weight with generated weather variables and model based on 10 weeks with 23
years data using week-wise approach using original weather variables were
recommended as pre-harvest forecast models for groundnut productivity of Junagadh
district which can predict the groundnut yield 6 weeks before harvest.
In case of ARIMA model, ARIMA (1, 1, 1) with 25 years data can be
considered as the forecasting model for groundnut productivity in Junagadh district of
Gujarat. The comparison between the selected regression models and ARIMA model
on the basis of R2
, R̅2
, RMSE and MAE values showed that regression models were
superior as compared to ARIMA models.
The proposed models are
Model based on correlation coefficient as weight using generated weather
variables
Y= -3536.86 + 1.24Z141 + 5.95Z121 + 15.27Z51 + 1.37Z131 – 41.73Z31 (R̅2 =86.30%)
Model based on week-wise approach using original weather variables
Y = -6605.93 + 105.81 X45 + 456.09 X25 + 2.94 X38 -27.27 X510 - 456.42 X210
(R̅2=85.70%)