Preliminary report on regression equations based on remote sensing data to estimate rice yield

Abstract

Regression equations for estimating rice yield in a field in Hokkaido, Japan using remote sensing data (SPOT5/HRG) are derived. Results of 10-fold cross-validation indicate that multiple linear regression and projection pursuit regression give similar predictive errors (from 55kg/10a to 58kg/10a). These predictive errors may lead to the ability to use the results of such regression equations as alternatives to the visual external examination of rice fields by expert staff dispatched by the National Agricultural Insurance Association. By introducing such techniques, the number of the visual external examination points will be reduced. Further examination shows that the use of a specific cultivar of rice leads to a beneficial regression equation.