5. Photographic overlap
The need for stereo coverage depends on the ability to identify vegeta-
tive forms, variability on ground cover, and stress characteristics on
the photography. Frequently, recognition characteristics of cropland
were adequate to recognize monoscopically (20 percent for coverage).
Range shrubs, plants and forest types usually require stereoscopic
interpretation for accurate delineation.
RESULTS
Quality of Classification
Examination of the Goose Lake data revealed two sources of errors in the
original classification which could be attributed to errors in ‘'ground-
truth" data. These errors included pixels which were properly grouped
(clustered), but not adequately or correctly described; and pixels which
were incorrectly clustered.
Statistical accuracy of this original classification varied from about 50
percent to 90 percent. Reclassification of the selected plots revealed
the two sources of errors. The revised classification was not applied to
the complete study area, so a comparison of accuracy levels was not pos-
sible. This will be closely followed on the Five-County Study.
The improved classification was possible because individual pixels were
located on the photos and on-the-ground. The location of individual
pixels or pixel groups is made possible by transfer procedures described
next.
Data Transfer System
Comparison of classification efforts on the Goose Lake project indicate
that classifications based upon ground-truth data enhanced by low flight
CIR photography resulted in correction of class descriptions and proper
clustering of many pixels which were in error on the original classifica-
tion.