85 
Landsat Data Analysis 
Data analysis of a May 1978 Landsat tape of central Maryland (Path 16, 
Row 33; ID: 30081-15073) was conducted on IDIMS. An "unsupervised" 
process which employed a Gaussian maximum likelihood clustering 
algorithm was used to generate most of the 68 unique spectral class 
signatures identified in the analysis from an area of mixed land cover/ 
land use comprising about 30 percent of the study area. These spectral 
classes were matched with eight Level I and II DSP general land cover/ 
land use categories derived by reorganization of DSP's 76-category 
Level IV land use classification scheme (which is based on 21 primary 
categories and 69 subcategories; see Appendix). These eight categories 
are forest, crop and pasture, water, transitional (disturbed land, con 
struction), low-density residential (LDR), medium-density residential 
(MDR), commercial/industrial/institutional (CII), and wetlands. 
The study area boundary was digitized from USGS 7% minute quadrangle 
maps using the GES. The boundary was also overlayed in IDIMS on the 
MAGI data base and on Landsat images to enable digital extraction of 
the same areas in these data sets. Clouds and their shadows within 
the study area boundary were located on the original Landsat image and 
affected areas digitally removed from both the Landsat and MAGI images. 
For the remaining cloud and shadow-free areas, the Landsat land cover 
acreages were determined for each cover type within the study area 
(74,364 acres) at both the full resolution of 1.54 acres/pixel for geo 
metrically corrected data and the MAGI special data base cell size of 
4.6 acres/cell. For the latter, spatial degradation was accomplished 
by a nearest neighbor resampling algorithm (Turner, Applegate, and 
Merembeck, 1978) which produced a revised grid with the number of 
Landsat-derived pixels reduced by a scalor factor of 0.579, the Landsat 
to MAGI linear dimension conversion. Using a systematic sampling 
approach, each new pixel was assigned the cover category belonging to 
the closest central original pixel. 
Landsat/MAGI System Data Comparisons 
The DSP and Landsat data sets were compared in several different ways. 
First, the acreage totals for each of the seven cover types within the 
study area (wetlands were not significantly represented) from the full 
resolution IDIMS classification and the comparable MAGI data base were 
determined and compared. Next, due to inconsistencies in the DSP land 
use designations, acreage totals were recalculated after removal of the 
Baltimore Washington International Airport, industrial parks and cer 
tain institutional properties. These areas were designated in the MAGI 
data base by particular commercial/industrial/institutional (CII) land 
use categories which bore no relationship to the land cover actually 
occurring there and are referred to as CII blocks elsewhere. In the 
accuracy assessment performed on the IDIMS classification for each 
cover type in the study area, the Landsat-derived data were assumed to 
be the observed values and the 1978 DSP land use data the expected 
values for the computation of percent error, defined as I[(observed - 
expected)/expected] x 100]. The impact of resampling the Landsat data 
on the acreage estimates for the seven land cover/land use categories 
was also examined by comparison of the original Landsat acreages with 
the resampled data and with the MAGI data. The required minimum level 
of correspondence for all comparisons was 90 percent to meet DSP's 
accuracy standards. 
The statistics generated by these analyses are summarized in Tables I-V.