Abstract :
A number of developments, such as those reported by Badhwar et al. [4], have resulted in a signature extendable technology. This technology, which we refer to as temporal profile technology, is simple and efficient and automatically recognizes crops by utilizing the Kauth-Thomas transform of Landsat, multidate data, and parameters derived from a model of each crop´s greenness-time trajectory. These parameters have overcome the lack of crop separability and stability encountered with technology utilizing spectral values of individual Landsat bands. The temporal profile technology was able, without manual intervention or retraining, to classify Landsat MSS data and estimate within 3 percent the area of corn and soybeans within 5 x 6 nm segments over a large geographic region within the U. S. Corn Belt and Mississippi Delta for three crop years [ 5]. In this current paper we test this algorithm over a set of data in the Argentina corn and soybeans region, and show that it will, with at most minor modifications, apply directly to Argentina. Small-grains evaluations by Badhwar [ 5] and Lennington et al. [181] strongly indicate that the temporal profile technology will work well for that important crop group. We also develop a theoretical framework for signature extendability. We show that the feature space generated by the temporal profile parameters will satisfy two conditions, separability and identifiability, required for a signature-extendable technology and will apply to any group of crops in any region that differ in their seasonal cycles.
