CITARS was an experiment designed to quantitatively evaluate crop identification performance for corn and soybeans in various environments using a well-defined set of automatic data techniques. These techniques differed mainly by the procedure used to obtain signatures from training data (e.g., clustering) and by the method of classification employed (e.g., linear or quadratic decision boundaries and equal or unequal class weights).

Each technique was applied to LANDSAT-1 data acquired over six Indiana and Illinois test sites throughout the growing season in an attempt to recognize and estimate proportions of corn and soybeans using both local and non-local (i.e. extended) training statistics.

As a result of these analyses the significance of factors which contribute to classification performance was determined. In this paper the results of (1) the differences indifferent ADP procedures; (2) the linear vs. quadratic classifier; (3) the use in classification Of prior probability information derived from historic data: (4) differences in local and non-local recognition and the associated use of preprocessing: (5) the use of multitemporal data: (6) the effects Of classification bias and mixed pixels in proportion estimation: (7) the effects of site characteristics including crop, soil, and atmospheric effects: and (8) the effects of crop maturity are presented and discussed.

Date of this Version