Transition year labeling error characterization study

Abstract

Labeling errors made in the large area crop inventory experiment transition year estimates by Earth Observation Division image analysts are identified and quantified. The analysis was made from a subset of blind sites in six U.S. Great Plains states (Oklahoma, Kansas, Montana, Minnesota, North and South Dakota). The image interpretation basically was well done, resulting in a total omission error rate of 24 percent and a commission error rate of 4 percent. The largest amount of error was caused by factors beyond the control of the analysts who were following the interpretation procedures. The odd signatures, the largest error cause group, occurred mostly in areas of moisture abnormality. Multicrop labeling was tabulated showing the distribution of labeling for all crops.