Abstract
The 42.5 microradian angular IFOV of the Thematic Mapper will provide a linear spatial resolution of approximately 30 meters from the nominal altitude of 710 km. This study determined the classification accuracies achieved with MSS data of four different spatial resolutions using two different types of classifiers.
The data were obtained on May 2, 1979 with the NASA NS-001 Thematic Mapper Simulator (TMS) over an area in northeastern South Carolina from a height above ground of 5945 meters. Data sets simulating three different spatial resolutions were computed from the original 15 meter nominal spatial resolution data. The classification accuracies achieved with data of each of the four different spatial resolutions using a "per-point" Gaussian maximum likelihood (GML) classifier were compared. The classification accuracies obtained using simulated 30 meter spatial resolution data with a "per-point" GML classifier were compared to the accuracies achieved with a "per-field" classification approach (i.e., the *SECHO, Supervised Extraction and Classification of Homogeneous Objects, classifier). The "pure field", or "field-center pixel", classification accuracies were determined using training fields and test fields. Accuracy comparisons were conducted with the Newman- Kuels' Range Test on the arcsin transformed proportions. The use of successively higher spatial resolution data resulted in lower overall ("field-center" pixel) classification accuracy. This trend was observed particularly in forest cover types, which are associated with relatively large levels of spectral variability across adjacent pixels. The use of the *SECHO classifier resulted in a higher overall ("field-center" pixel) classification accuracy than was obtained with the per-point GML classifier using the simulated 30 meter spatial resolution data.
Date of this Version
1981