One problem in the estimation of crop acreages from classification results of remotely sensed data points (pixels) is that misclassification is likely to arise in recognizing the crops represented by these pixels. When the observed crop proportion in a sample data set is obtained by computing the frequency of classifications into a crop type, a biased estimate of the true crop proportion in the agricultural area is, in general, obtained. To reduce bias, an adjustment is needed with respect to the amount of classification error present in the results. In our paper we discuss the problem of estimating the classification errors and define crop proportion estimates which are adjusted for the classification errors. Considering the particular case of a two-crop problem, we obtain asymptotically unbiased estimate of a crop proportion, derive an upper bound for its mean square error, and determine sample sizes that minimize the sampling cost but provide estimate with a specified precision.

Date of this Version