Weather based modeling and pre-harvest forecasting of crop yield using statistical and hybrid models

Abstract

Crop yield modeling and forecasting have been an essential step in determining agricultural and economic policy decisions in India. Planning and policy decisions on distribution, price, export-import, storage, and other issues are critically dependent on it. This study aimed to develop a trustworthy crop yield prediction system for rice yield by analyzing the relationship between crop yield and several weather variables. Weather indices, an assimilation of the weekly weather effects on crop yield, were used to study the impact of weather factors on rice yield. Several statistical and neural network models have been developed based on the linearity and non-linearity pattern of the data. The results of statistical models demonstrated that both linear and non-linear patterns were present in the data and the effects of rainfall, minimum temperature, and time variable t were significant. The neural network models outperformed statistical models in terms of accuracy, and the study found hybrid models with three and four hidden nodes were the best-fit models in the districts of Birbhum and Burdwan, respectively. A reliable rice yield estimate can be obtained six to eight weeks before harvest by using the best-fit models for various policy decisions.