Weather-Based Statistical and Neural Network Tools for Forecasting Rice Yields in Major Growing Districts of Karnataka

Two multivariate models were compared to assess their yield predictability based on long-term (1980–2021) rice and weather datasets over eleven districts of Karnataka. Simple multiple linear regression (SMLR) artificial neural network (ANN) calibrated (1980–2019 data) validated (2019–2020 data), yields forecasted (2021). An intercomparison the revealed better with ANN, as observed deviations smaller (−37.1 21.3%, 4% mean deviation) SMLR (−2.5 35.0%, 16% deviation). Further, district-wise forecasting using ANN indicated an underestimation yield, higher errors in Mysuru (−0.2%), Uttara Kannada (−1.5%), Hassan (−0.1%), Ballari Belagavi (−15.3%) overestimations remaining (0.0 4.2%) 2018. Likewise, 2019 underestimated Kodagu (−0.6%), Shivamogga Davanagere (−0.7%), (−5.1%), (−10.8%) overestimated for other five 4.8%). Such model underestimations are related farmers’ improvement practices carried out under adverse conditions, which not considered by while forecasting. As acceptable range, they prove applicability crop management planning addition its use regional agricultural policy making.