Development of ARIMA and weather yield models for pre-harvest forecasting of mustard crop in Haryana
Loading...
Date
2016
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
CCSHAU
Abstract
An efficient crop forecasting infrastructure is pre-requisite for information system about food supply,
especially export–import policies, procurement and price-fixation. The statistical modelling approaches viz.,
ARIMA, multiple linear regression and principal component analyses were used to achieve the district-level
mustard yield estimation in Haryana. The study was categorized into two parts i.e. the development of ARIMA and
zonal weather-yield models for pre-harvest mustard yield forecasting in Hisar, Bhiwani, Sirsa, Fatehabad,
Mahendergarh, Rewari, Jhajjar and Gurgaon districts of Haryana. The ARIMA models have been fitted using the
time-series mustard yield data for the period 1966-67 to 2008-09 of Hisar, Mahendergarh and Gurgaon districts
and 1972-73 to 2008-09 of Bhiwani district, 1975-76 to 2008-09 of Sirsa district and 1997-98 to 2008-09 of
Fatehabad and Jhajjar districts. However, the fortnightly weather data have been utilized from 1980-81 to 2008-09
for building the zonal weather-yield models. Models have been validated using the data on subsequent years i.e.
2009-10 to 2013-14, not included in the development of the models. ARIMA(0,1,1) for Hisar, Bhiwani and Sirsa
districts and ARIMA(1,1,0) for Mahendergarh, Jhajjar, Fatehabad and Gurgaon districts were fitted for estimating
district-level mustard yield(s) in Haryana. The zonal weather-yield models were fitted by taking DOA yield as the
dependent variable and fortnightly weather parameters (or PC scores) along with trend yield/CCT/dummy
variables as regressors.
The predictive performance(s) of the contending models were observed in terms of the percent
deviations of mustard yield forecasts in relation to the observed yield(s) and root mean square error(s) as well. The
level of accuracy achieved by zonal yield model(s) using crop condition term as categorical covariate along with
weather variables was considered adequate for estimating the district-level mustard yield(s) well ahead of the
harvest time. The model incorporating maximum and minimum temperatures, rainfall, relative humidity, sunshine
hours and crop condition term (categorical variable) as predictors had the highest predictive accuracy. Though, the
zonal yield models using trend yield and PC scores as regressors showed the superiority over the multiple linear
regression analysis based on trend yield and weather variables but sometimes the percent relative deviations were
too high to be considered worth for yield forecasting purpose. Zonal yield models incorporating CCT and weather
variables consistently showed the satisfactory results in capturing the percent relative deviations pertaining to
mustard yield forecasts and performed well with lower error metrics as compared to the remaining models in all
time regimes for the districts under consideration.
Description
Keywords
Autocorrelation function, Partial Autocorrelation function, Maximum and Minimum Temperature, Rainfall, Sun shine hours and Relative humidity, Principal component scores, crop condition term, categorical covariate, ARIMA and zonal weather- yield models.