Several image processing techniques were applied to the data set before Informa 
analysis began. The Seasat Image was corrected for varlations In brightness accurac 
across the scene and a median value filter was applied to reduce the effect of +he SI 
speckle (Blom and Dally, 1982). Variance pictures of both the SIR-A and anal ysi 
Seasat images were also calculated, so that texture Information could be discr in 
included in the analysis. This technique, which is described in Blom and var lanc 
Daily (1982), involves finding the square of the standard deviation of Image accur ac 
tone over a local area, and replacing the value of the center pixel with the an Impc 
computed variance. In this study, variance pictures were computed with window based s 
sizes of 5x5, 9x9, 11x11, 13x13 and 17x17. geol og! 
INTERPRETATION OF THE COMBINED IMAGE DATA SET on, iim 
Several techniques were used to Interpret the combined Image data set. iS ar 
Image enhancement techniques were used to facilitate visual Interpretation of was 339 
the coregistered Images, and multivariate analyses were used to quantitatively 
assess the synergism of the Images. These techniques are described below. C. 
A. Hue, Saturation, Intensity (HSI) Images Tt 
discrin 
Coregistered Seasat and SIR-A images were displayed using the SIR-A Image based c 
to produce color (Hue) and the Seasat image to modulate the intensity of the has nc 
color where the images overlapped. The saturation level was kept constant. I T tholc 
Because of the difference in incidence angles of the two sensors (50 degrees presen! 
for SIR-A and 23 degrees for Seasat), SIR-A images are most sensitive to smal | to Ider 
scale roughness properties of the surface, while Seasat Images are more setting 
sensitive to changes in local slope. Therefore, In the HS! image, variations waveler 
in color were related to roughness variations and variations In Intensity were | ibrar 
related to topography. The combination of surface roughness and topographic interrc 
Information shown in the HS! Image, allowed better separation of the the spe 
lithologic units present. For example, In one case It was possible to been si 
distinguish sedimentary layers that could not be seen In either the SIR-A or but hi 
Seasat image alone. This is a significant result, because It shows how subtle discr ir 
details can be brought out in combined image data sets. It also shows that Inform: 
geologic mapping with radar images will be Improved with the SIR-B sensor, Landsa- 
scheduled for mid-1984, which will obtain data at multiple incidence angles. abil ih 
vol ume: 
B. Linear Discriminant Analysis compone 
Blom et al., (1981) and Blom and Daily (1982) found that the addition of 
the median value-filtered Seasat image Increased classification accuracy in a 
supervised classification scheme by 7% over Landsat Images alone. As M: 
mentioned above, SIR-A Images provide more Information about small scale data : 
surface roughness properties of the units than Seasat because of the larger With + 
incidence angle. In order to determine if the addition of the SIR-A data emphas | 
further improves classification accuracy over the Landsat/Seasat combination, sensor: 
the new Image data set was included in the linear discriminant analysis used various: 
by Blom et al. (1981). The results of the analysis showed that the best  . signatı 
combination for discriminating among units In this portion of the of the swell Backsc: 
were Landsat MSS4, MSS7, and SIR-A. In this analysis, the combination of MSS4 angles 
and MSS7 made it possible to differentiate most of the units to an accuracy of geolog 
96$. However, the outcrop of Kalbab Limestone within the Moenkopi Siltstone 
was classified correctly only 74% of the time, and was misclassified as the 
Moenkopi Siltstone 4$ of the time. The addition of SIR-A data decreased the 
misclassification of limestone for siltstone to less than 1$, and increased TI 
the overall! classification accuracy to 81%. Abrams, 
descr it 
Insti ti 
256