Multi stage wheat yield estimation using multiple linear, neural network and penalised regression models

Abstract

Wheat is second most consumed staple food grain after rice, cultivated nearly 26 Mha areas in the northern part of India. Weather variables like Maximum temperature, Minimum temperature, Relative humidity, Rainfall, Bright sunshine hours, Evaporation etc. have a great impact on crop yield. Weather based pre harvest crop yield estimation is helpful for deciding marketing, pricing, import-export and policy making etc. Wheat yield and weather variable data were collected for last 35 years from Hisar, Ludhiana, Amritsar, Patiala and IARI, New Delhi. Multistage wheat yield estimation was done at tillering, flowering and grain filling stage of the crop by considering weather variables from 46th to 4th, 46th to 8th and 46th to 11th standard meteorological week for model development. Model was developed using stepwise multiple linear regression (SMLR), Principal component analysis in combination with SMLR (PCA-SMLR), Artificial Neural Network (ANN) alone and in combination with principal components analysis (PCA-ANN), Least absolute shrinkage and selection operator (LASSO) and elastic net (ENET) techniques. Analysis was carried out by fixing 70% of the data for calibration and remaining dataset for validation. On examining these multivariate models for stage-wise estimation of wheat yield, percentage deviation of estimated yield by observed yield was ranged between -0.1 to 25.6, 0.9 to 22.8, -0.7 to 22.5% during tillering, flowering, and grain filling stage respectively. On the basis of percentage deviation and model accuracy Elastic net and LASSO model was found better and can be used for district level wheat crop yield estimation at different crop growth stage.