Rice yield forecasting using agro-meteorological variables: A multivariate approach

Abstract

The weather variables impact the crop differently throughout the various stages of development. The weather effect on crop yield thus can be determined not only by the magnitude of weather variables but also on the variability of weather over crop season. Crop yield forecasting methods incorporating weather information provide a better prediction of yield accounting the relative effects of each weather component. Regression analysis is the most frequently used statistical technique for investigating and modelling the relationship between variables. Building a multiple regression model is an iterative process. Usually several analyses are required for checking the data quality as well as for improvement in the model structure. The use and interpretation of multiple linear regression models depends on the estimates of individual regression coefficients. However, in some situations the problem of multicollinearity exists when there are near linear dependencies between/among the independent variables. The Principal Component Analysis (PCA) method has been proposed to address the problem of multicollinearity. Using principal component scores (PC) derived from weather variables as predictor variables helps to obtain better estimate the yield. The discriminant analysis is a multivariate technique involving the classification of separate sets of objects (or sets of observations) and assigning of new objects (or observations) to the groups defined previously. Forecasting of crop yield can also be done using discriminant analysis scores based on the weather variables as regressor.