Use of weighted regression for estimating rice yield by satellite data

Abstract

Use of weighted regression is attempted for deriving regression equations which aim at estimating rice yield by satellite data give by ASTER sensor. It is assumed that satellite data and rice yield data are obtained in 2001 and 2002, and in 2004, all satellite data and a part of rice yield data are given. Then, the unobserved rice yield data in 2004 is to be estimated. For this purpose, multiple regression equations are obtained and predictive errors are estimated. The results indicates that use of weighted regression allows us to make the most of three-year data for beneficial estimation. The procedure suggested here is an attempt to utilize weighted regression for realizing regression which reflects the degree of validity of respective data, as well as a method for coping with inhomogeneity of variance of data. Moreover, use of additive model instead of multiple regression equation confirms the effect of weighted regression. The results given by additive model are shown to be more desirable than those by multiple regression equation.