Abstract
To estimate rice yield using remote sensing data, predictive error should be reduced and the variation in estimates caused by fluctuating data should be abated. Furthermore, the extraction of singular data from dataset is beneficial for data analysis. Hence, the bootstrap method is applied to the estimation of comprehensive predictive error given by a regression equation, predictive error for each datum, and the degree of dependence of estimates on fluctuating data. "Unbiased bootstrap predictive error" and "0.632 bootstrap predictive error" are recommended tools for estimating comprehensive predictive error yielded by a regression equation. Moreover, predictive error may be reduced by weighting data using each predictive error. A graphical method of determining each predictive error and the responses of estimates to perturbing data is also shown. The applications of the above techniques to the derivation of regression equations using real remote sensing data and rice yield data give expected results.
