previously defined. All ancillary digital information was 
converted to ARC/INFO and carefully verified with the 
ERDAS-ARC/INFO Live Link™. The ancillary data were 
then carefully modified as needed. 
One example of the post-classification modeling was the 
reintroduction of linear DLG transportation information into 
the final classification because clustering often delineated the 
transportation network inconsistently. The clumping 
algorithms within supervised and unsupervised classifications 
have a tendency to classify roads bounded by heavy 
vegetation as non-urban. To correct this problem, roads and 
railroads were reintroduced as DLG's and gridded back into 
the final classification. DLG hydrology information was 
also used extensively to emphasize linear streams and to 
produce a lowland mask. Early in the project it became 
apparent that the signatures of lowland forest types and 
upland forest types were being confused in the classification. 
To correct this, a GISMO model was created to mask lowland 
forest and assign the information to the correct class. This 
allowed for the accurate delineation of the lowland rorest 
classes. The orchard class was problematic throughout the 
project study area as well. Because the spectral nature of 
orchard classes is similar to that of forest classes, orchard 
classes were extracted from USGS LUDA data and then 
added to the final classification. 
Overall confidence levels were calculated yielding a lower 
limit of 85.01% and an upper limit of 87.40%. In addition to 
the overall percentage accuracy of 86.25%, a large number of 
test samples were taken to ensure a narrow range of 
uncertainty with a confidence range of only 2.39%. The 
lower limit of 85.01% for the 95% confidence range met the 
85% accuracy criterion for the project, serving as a 
conservative estimate of the overall accuracy for the 
classification. 
In using ancillary data as digital sources to aid the image 
processor in data analysis, it is important to fully understand 
the limitations of the data. Much of the data available today is 
generally outdated, very course in nature, or was created 
under lax digitizing standards. For instance, Digital Line 
Graph data (DLG) must often be rectified to the satellite 
image with which it will be used. Tiger Place boundaries 
provide only course urban boundary information and often 
do not take into account small rural town or urban fringe. 
For these reasons the user must give much thought as to how 
to use ancillary data and never rely on this information as the 
sole source of identification. 
CONCLUSION 
With careful planning and forethought at the beginning of a 
project, it is possible to use ancillary data to extract general 
information from satellite data. These general areas can then 
be analyzed in a more intense fashion, while other extraneous 
data are suppressed, (i.e., breaking out urban data from non- 
urban). The primary objective in the use of ancillary data is 
to increase the speed and accuracy of image analysis. It is 
also essential in a large project to reduce the data to 
managable pieces of information which can be processed and 
assessed for accuracy in a logical, timely progression. 
The case studies discussed in this paper show but two ways 
ancillary data can be used to optimize image processing time 
and analysis. They are examples of the unique ways that 
raster and vector information can be integrated to enhance one 
another. With the integration of raster and vector sources 
available today, the creation of large scale satellite imagery 
classifications can be completed with more accuracy and 
faster than ever before. 
107