Regression models of dryland wheat yields from water supplies in ustifluvent in Punjab, India

Abstract

Abstract Seven years' data on yield and water supply parameters of adequately fertilized wheat grown on the foothill soils (Kandi region) of Punjab were used for statistical modelling of wheat yields from water supplies. Multiple regression analysis was carried out to relate grain yield to total water supply or its components. Seasonal water use (the available water depletion from the soil profile plus the growing season rainfall) accounted for 67 and 62% of the variation in yield on loamy sand and sandy loams soils, respectively. Interestingly, total water supply (the available water stored in the soil profile at seeding time plus the growing season rainfall) was a better predictor of yield (77% on loamy sand and 85% on sandy loam) than seasonal water use by the crop. Splitting of the total water supply into its two components did not improve yield predictability of the regression model. The available water stored in the soil profile at seeding time alone accounted for 36 and 52% of the variation in yield on the two soils. The growing season rainfall was divided into six 25-day periods and the yield was related to the distribution of the water supply during the periods of time following seeding; this improved the predictability of the regression to 87 and 91% on loamy sand and sandy loam soils, respectively. The larger improvement in the predictability of the regression model for loamy sand soil compared with sandy loam soil by dividing the growing season rainfall suggests that crops grown on the former are more sensitive to timings of water supply. Validation of this regression model on independent data seemed quite satisfactory for the two soils.