The second issue investigated here involves the proper spatial resolu 
tion from Landsat MSS data for use with the MAGI System data. Resam 
pling the Landsat data from the original (geocorrected) 1.54 to 4.6 
acres/cell using a nearest neighbor procedure had a substantial nega 
tive impact on the acreage estimates. Two recent related studies 
(Ramapriyan, et al., 1981; Boyd, Gunther and Lu, 1981) found that a 
similar resampling, or aggregation, scheme using a systematic aligned 
sampling technique over a 5 x 5 pixel region did not significantly 
alter results. However, these results were produced for a simple 2- 
class system which assumed random spatial distribution of classes and 
associated errors. Since we used 4 times as many classes, none of 
which was randomly distributed, these results are not directly appli 
cable here. Further studies are needed to clarify the effect of aggre 
gation on classification error. 
The particular categories expected to be most affected were those which 
were spatially heterogeneous (i.e., crop/pasture, transitional), linear 
in nature (i.e., transitional, commercial strip development, and hydro- 
graphic features), or which had relatively small proportional represen 
tation. Of these, transitional (by a wide margin) and water exhibited 
the greatest error rates in the resampling. CII and LDR fared sur 
prisingly well in this particular location, probably due to the high 
association and clustering of residential and commercial areas in towns 
in the study area. But these accuracies were retained at the expense 
of other categories, such as forest and MDR. 
One explanation for the more satisfactory performance of the full reso 
lution Landsat data over the resampled data, at least for the aggre 
gation type used here is that MAGI System land use data were tabulated 
as opposed to averaged over the 4.6 acre grid cells area or sampled 
from higher resolution subunits. A different aggregation strategy may 
be appropriate for Landsat data at the more common 90 acre grid cells. 
In the future, Landsat data geocorrected to 57m x 57m (as opposed to 
the 79m x 79m correction employed here) are recommended to provide a 
land resolution unit of 1.1 acres, even closer to the functional MAGI 
System resolution. Landsat-4 Thematic Mapper data should provide com 
parable or better resolution data than the high altitude photography 
now used as the basis of land categorization in the MAGI System. 
CONCLUSIONS 
Landsat Data 
(1) A superior land cover survey for all seven cover types examined was 
produced with the Landsat data for the study area. 
(2) With the exception of the CII blocks, comparable land categori 
zation results were obtained from the MAGI System and Landsat data. 
(3) The discrepancies in the results from the MAGI System and Landsat 
data for the CII blocks are related to differences in interpretive 
methodologies, and not to deficiencies in either data source. 
(4) Full spatial resolution data should be used as input to the MAGI 
System. If information for MAGI System-sized grid cells is 
required, the enumeration process should be comparable to the one 
in use by DSP.