hardcopy (three copies) of each quad, and conversion 
services of raster files into ARC/INFO vector format. 
se St Two - Lake Michigan Ozone Stu MOS 
To quantify the ozone source-receptor relationships in the 
area surrounding Lake Michigan, a non-profit corporation of 
the four states surrounding Lake Michigan jointly funded an 
emissions study for the Lake Michigan area. The study 
required development of a current land use/land cover data 
base of Illinois, Wisconsin, Indiana and Michigan, and was 
governed by the Lake Michigan Air Director's Consortium. 
This data provided source information to geographically 
locate ozone emissions and to estimate biogenic emissions 
from a variety of vegetative sources. 
Due to the legal ramifications of this project, the Consortium 
required that the land use/cover classification be completed 
within a 10 month period. This time line allowed a maximum 
of four weeks, per scene, for all land use/cover processing 
and accuracy assessment. For this reason it was necessary to 
explore the use of ancillary data as a resource for breaking the 
data into logical components that could be analyzed 
independently. 
It is commonly known that certain urbanized and non- 
urbanized classes reflect visible and infrared light in a very 
similar spectral signature. Previous experience had 
determined that separating these areas would minimize any 
possible confusion of these classes. For example, quarries, 
bare asphalt and beach usually reflect light similarly, therefore 
they were separated before automated classification 
techniques were implemented. 
The ancillary data used for this project were grouped into 
three categories: aerial imagery, analog maps, and digital data 
sources. These digital data sources included four different 
types of digital data: 
USGS 1:100,000 scale Digital Line Graph (DLG) 
*1:250,000 Land Use and Land Cover Data (LUDA) 
*1:250,000 scale Digital Elevation Model (DEM) 
* United States Bureau of Census 1990 Post-Census 
TIGER data 
Aerial photography included the use of two types of aerial 
photography and one type of slide film. Aerial photography 
included photography from the National High Altitude 
Photography (NHAP-2) program and National Aerial 
Photography Program (NAPP). This photography was 
acquired to provide as much coverage as possible while 
adhering to budgetary constraints. The photography served a 
dual purpose: first, to provide a data source for initial 
classification accuracy, and second, to serve as a source of 
ground truth when computing the final land use and land 
cover classification accuracy assessment. The aerial coverage 
was divided into two data sets. The first set was used to 
extract signatures and perform initial accuracy assessments. 
The unused photography was set aside to perform the final 
accuracy assessments. 
As an enhancement to the national photography acquired, 
United States Department of Agriculture - ASCS 35mm color 
compliance photography was ordered as well. This film 
proved valuable as a source of information to support the 
aerial photography. In addition, ASCS-578 and ASCS- 
156EZ crop report information was also available for this 
area. Individual county ASCS offices provided lithographed 
"photomaps" delineating each field and specific crop type for 
any desired year. Because of the dynamic nature of 
agricultural practices (i.e., crop rotation), this information 
provided an accurate representation of crop types for various 
areas. This information greatly enhanced the ability of the 
106 
analyst to identify crop type and increase classification 
accuracy. 
The LMOS project required that a significant amount of TM 
data be processed quickly and accurately. Therefore it was of 
great importance that the methodology for the classification be 
established and adhered to at the onset of the project. The 
Consortium requested that an initial test classification be 
completed which set the methodology for the complete 10 
month study. This classification effort was unique in that a 
very complex set of pre-classification procedures were 
established to allow the image analyst to immediately begin 
processing. The data were processed by project teams with 
assigned tasks for completion of various phases of the 
project. 
Pre-processing included tasks as simple as loading data onto 
the computer system and analyzing each band for data 
anomalies and cloud cover. Each data set was verified for 
acceptability prior to pre-processing. Once the data were 
verified, they were then clipped to data boundary limits and 
separated into urban and non-urban data sets using Census 
TIGER political boundaries or Place Boundaries. Although 
TIGER Place Boundary information is very general, it allows 
for the accurate separation of a majority of the data into urban 
versus non-urban data sets. The non-urban areas were also 
carefully checked for small urban areas which would not have 
been represented in the TIGER files. These small urban areas 
were separated with on-screen "heads up" digitizing using the 
ERDAS-ARC/INFO Live Link?M, 
Once the data were separated into urban and non-urban sets, 
the data were processed into principal components using a 
standard Principal Components Analysis (PCA) algorithm. 
The PCA was used primarily as a data reduction technique to 
eliminate redundant spectral information, thereby reducing the 
amount of data and speeding computer processing time. 
When the principal components analysis was completed, a 
varimax rotation was applied to the data. The varimax rotation 
manipulated the transformation coefficients to correlate more 
closely with specific bands in the original TM data. This 
allowed the analysts to more easily interpret what a specific 
component band represented. 
After the data sets were separated and the principal 
components analysis completed, the data were processed 
through the ERDAS software program ISODATA (Iterative 
Self Organizing Data Analysis Technique). ISODATA 
generates a set of mean vectors and covariance matrices for 
each distinct spectral cluster. This unsupervised classification 
algorithm was used to derive a set of clusters which 
represented general features within the data set. The clusters 
were then used by the analyst to determine and refine manual 
signatures used in the final supervised classification. 
This unique pre-processing classification methodology 
allowed for a great deal of repetitive process work to be 
completed prior to analysis by a professional. While other 
scenes were in the classification phase of the project, the pre- 
processing was completed. The image analysts were therefore 
able to move from one scene to the next with little distraction, 
establishing a tight progress cycle. 
The final project classification utilized numerous levels of 
ancillary information. The addition of this information 
provided an excellent opportunity to utilize the latest 
technologies in GIS modeling. The GIS modeling package, 
GISMO'M, was used to assure that various procedures were 
performed under consistent conditions each time the process 
was run. This allowed the project teams to more quickly 
complete multiple GIS processes in an efficient manner. In 
addition, use of this model assured that scenes matched in 
overlap areas. 
Post-classification techniques were primarily GIS modeling 
functions using much of the ancillary digital information 
previoi 
conver 
ERDA 
then ca 
One e 
reintro( 
the fin: 
transp: 
algoritl 
have € 
vegetal 
railroac 
the fin 
also us 
produc 
appare 
upland 
To cor 
forest : 
allowe: 
classes 
project 
orcharc 
classes 
added t 
Overall 
limit of 
the ovei 
test sa 
uncerta 
lower I 
8590 a 
consen 
classifk 
In usin 
process 
the limi 
general 
under 1 
Graph. 
image 
provide 
do not 
For the. 
to use a 
sole sot 
With cz 
project, 
inform: 
be anal; 
data are 
urban). 
to incre 
also es 
managa 
assesse: 
The cas 
ancillar 
and an: 
raster ai 
another 
availab 
classifi 
faster tl