Skip to Main Content
A study designed on the basis of a fixed-effects, three factor, two-level analysis or variance (ANOVA) was conducted to quantify the effect of Thematic Mapper (TM) sensor improvements on classification accuracy using TM data acquired over the Washinton, DC, area on November 2, 1982. The TM data were systematically degraded spatially, spectrally, and radiometrically to simulate the effect of changing each individual sensor parameter separately, and in conjunction with other sensor characteristics, to ultimately simulate Multispectral Scanner (MSS) data characteristics. The greatest level of variance was accounted for by the spectral waveband variable, providing an average increase in classification accuracy of 5.85 percent. This increase constituted a 21-percent relative improvement from TM data with respect to Landsat MSS data [i. e., percent relative improvement = (high accuracy value-low accuracy value)/low accuracy value X 100]. The second greatest amount of variance was accounted for by the Â¿radiometricÂ¿ variable (i. e., bit quantization level). This provided a 5.25-percent increase in percent correctly classified pixels, which constitutes a 19-percent relative improvement of TM over MSS data due to quantization level. Spatial resolution accounted for the lowest source of variability in the observed classification accuracies, with an overall average decrease of 0.7 percent. This constituted a 2-percent relative degradation from TM data with respect to MSS data. Only the differences found for the spectral waveband combinations and the quantization level were statistically significant at Â¿ levels of 0.01 to 0.001.