Climate-based forecasting of maize yield in punjab using machine learning

The present study was designed to predict Kharif maize yield based on climatic conditions over 28 years (1990–2017) period in Punjab. The two regions of Punjab selected for the current study were North Punjab i.e. Gurdaspur, Amritsar and Hoshiarpur districts and Centre Punjab i.e., Ludhiana, Jalandhar and Kapurthala districts. As per the study objectives, the required meteorological data were acquired from meteorological observatories and automatic weather stations installed at respective districts. The statistical decomposition of time series maize yield data discovered the presence of technological trends in both North Punjab and Centre Punjab regions. The detrended yield data was used as dependent variable and meteorological factors as independent variables to develop multiple linear regression models following machine learning approach. The coefficient of determination (R²) exhibited that the models can explain 82% and 76% of the yield variability with selected weather factors in North and Centre Punjab, respectively. The model evaluation using test data showed that predicted yields were highly correlated with observed data with less error for both the regions.