88
successful when the DSP and Landsat systems of land categorization are
comparable, and is recommended for all subsequent surveys utilizing
MAGI System data. This is noteworthy because the present capability to
extract the primary and secondary acreages from the MAGI System by
category and to recombine them is cumbersome, reinforcing a user ten
dency to extract unadjusted primary data only.
For this study area, good correspondence between the MAGI System and
Landsat data was achieved for four categories—forest, water, crop/
pasture, and MDR—without _a priori knowledge of the MAGI land use
designations. However, a priori knowledge of and accommodation for
the MAGI designation was necessary to achieve high correspondence for
two of the three remaining cover categories, CII and transitional,
which were not handled uniformly throughout the study area by the DSP
classification scheme. For convenience, large and well-defined areas
(i.e., commercial, institutional and industrial facilities), which
together comprised roughly 15 percent of the total study area, were
blocked out at the photointerpretive stage and given a 100 percent CII
(commercial, institutional or industrial) label regardless of the
actual land cover categories occurring at these locations, a practice
consistent with DSP's traditional emphasis on land use. Although cor
rectly designated by functional land use, these areas also contained
within their property boundaries substantial portions of LDR, the third
category which did not exhibit good correspondence, as well as grass
and forest. For example, 46 percent of the LDR, a spatially dispersed
category by definition, was located on these CII blocks. Roughly half
the ground cover was forest and grassland on individual CII blocks such
as the Baltimore Washington International Airport, Fort Meade, the
National Security Agency, and many other private, state, and federal
institutions. These were solely designated as governmental/institu
tional by MAGI on both the primary and secondary data layers. The
transitional category was also represented only as a primary category
since it was recorded at 60 percent or greater proportions/cell where
it existed, even if representation was actually less. This important
category, though small proportionally, was actually weighted toward
overestimation by DSP, except where it was not counted on the CII
blocks. For these areas where the DSP land use categories were direc
tly compared to the Landsat land cover categories, the correspondence
could not be expected to be large. The result was that LDR as a cover
type (and with much less proportional impact on other categories, such
as forest, as well), was severely underreported with a corresponding
overestimation of CII as a cover type in the MAGI System data base.
For the remaining 85 percent of the study area where both surveys
represented similar land categories, the correspondence of the Landsat
and MAGI results was good to excellent. These results further support
the premise that the observed discrepancies between the MAGI System and
Landsat data on the CII blocks were indeed related to the differences
in the two land interpretation and classification methodologies, a
finding consistent with other research (Fitzpatrick-Lins, 1978).
Overall, the Landsat data provided more reliable "cover type" acreage
estimates than did the acreages from MAGI. Since land cover infor
mation is increasingly important for environmental planning, hydro-
logical models and development strategies, the Landsat data could be
used by DSP to provide information not currently available through MAGI
for some geographical areas. This approach could ultimately yield a
comprehensive means of categorizing all land areas within the state.