Application of Artificial Neural Network in Predicting Farmers’ Response to Water Management Decisions on Wheat Yield

Abstract

Water management usually involves decision-making with respect to allocation, scheduling and application of
available water to different crops over an irrigation season so as to get maximum economic returns. A study was
carried out in the Kaithal irrigation circle for prediction of farmers’ decisions regarding total depth of irrigation
water, fraction of groundwater and delay in sowing on yield of wheat crop under varying conditions of
groundwater and soil salinity using Artificial Neural Networks (ANN). Three ANN algorithms i.e. gradientdescent
back propagation (BP), Levenberg-Marquardt (LM) and radial basis functions (RBF) with various
architectures were used. It was found that radial basis function with a spread constant of 0.1 performed better
in predicting wheat yield. Also, it was observed that ANN algorithm predicted wheat and rice yields better
correlated to observed yields (r2=0.63 and 0.74) in comparison to regression model (r2=0.37 and 0.52)