Machine processing techniques which utilize remotely sensed data to estimate crop acreage have been extensively evaluated in several large scale experiments, beginning in 1972 with the Crop Identification Technology Assessment for Remote Sensing (CITARS) and more recently, the Large Area Crop Inventory Experiment (LACIE). Landsat multispectral scanner data acquired over several major global agricultural regions, has been processed using state-of-the-art technology. The performance has been quantified in two important situations (1) using in situ acquired ground observations to obtain classifier training samples; (2) using image analysis of Landsat data to obtain classifier training samples without the aid of ground observations. This collection of experiments has produced a considerable body of knowledge regarding the performance of machine processing technology under these conditions and have defined and prioritized areas for further needed research.
This paper will review and summarize the results of these experiments, discuss the nature of the problems surfaced by them, and address potential directions for current and future research.
Date of this Version