Weather based wheat yield prediction using machine learning

Abstract

Wheat crop is highly affected by the influence of weather parameter, adverse weather drastically reduce wheat yield. Thus, there is need to develop and validate weather-based models using machine learning for its reliable prediction at multiple stages. Wheat yield and weather data during crop growing period were collected from IARI, New Delhi, Hisar, Amritsar, Ludhiana and Patiala. The yield prediction model was developed using stepwise multi linear regression (SMLR), support vector regression (SVR), least absolute shrinkage and selection operator (LASSO) and hybrid model LASSO-SVR, and SMLR-SVR in R software. Analysis was done by fixing 70% of the data for calibration and remaining data for validation. Wheat yield prediction was done at tillering, flowering and grain filling stage for wheat crop by considering 46^th to 4^th, 46^th to 8^th and 46^th to 11^th standard meteorological week for model development. On examining these models for yield prediction at tillering, flowering and grain filling stage, LASSO performed excellent having nRMSE value ranged between 0.02 % at grain filling stage for IARI, New Delhi to 8.36 % for Hisar at flowering stage. The model performance of SVR is increased if hybrid model in combination with LASSO and SMLR is applied. Hybrid model LASSO-SVR has shown more improvement in SVR model compared with SMLR-SVR.