Comparison Statistical Rice Yield Prediction with Multiple Weather Parameters

Abstract

Advance knowledge of harvestable products, especially essential food crops such as rice, wheat, maize, and pulses, would allow policymakers and traders to plan procurement, processing, pricing, marketing, and related infrastructure and procedures. There are many statistical models are being used for the yield prediction with different weather parameter combinations. The performance of these models are dependent on the location’s weather input and its accuracy. In this context, a study was conducted at Agro Climate Research Centre, Tamil Nadu Agricultural University, Coimbatore during Kharif (2020) season to compare the performance of four multivariate weather-based models viz., SMLR, LASSO, ENET and Bayesian models for the rice yield prediction at Tanjore district of Tamil Nadu State with Tmax, Tmin, Mean RH, WS, SSH, EVP and RF. The results indicated that the R2, RMSE, and nRMSE values of the above models were ranged between 0.54 to 0.79 per cent, 149 to 398 kg/ha, 4.0 to 10.6 per cent, respectively. The study concluded that the Bayesian model was found to be more reliable followed by LASSO and ENET. In addition, it was found that the Bayesian model could perform better even with limited weather parameters and detention of wind speed, sunshine hours and evaporation data would not affect the model performance. It is concluded that Bayesian model may be a better option for rice yield forecasting in Thanjavur districts of Tamil Nadu.