s d'une limitations of different approaches.
és sous
seront Perhaps the leading remote sensing research organizations, many of whom are
represented at the present symposium, will participate in this experiment. The results of
the experiment might be published in a compendium with contributions from all
cooperators. Perhaps discussion of the results could also be the topic of a special
n working group meeting.
igitaler
wierig, Procedures
Es wird
on über A test area and training sets or calibration areas will be defined and each
undlage cooperator will receive a "ground truth” map or statistics for the latter. Each
cooperator would have to acquire, by purchase or loan, the computer compatible tapes
s selected.
| eines
konnte, A number of possibilities for choice of the test area exist, including the
alibrie- selection from the general area of this Freiburg Symposium. The criteria for selection
von der should include:
stellung ae :
gkeiten l. Availability of reliable ground truth.
rd dem 2. Availability of good Landsat data.
id Kon- 3. The presence, within a portion of a Landsat scene, of several important
ng wird land use classes.
Standardization of objectives is to be achieved by selecting classes for
ationen interpretation according to the well-known system of Anderson et al. (1976). This
uch das hierarchical classification system recognizes nine classes at Level I:
Beitrag 1. Urban or built-up land
2. Agricultural land
3. Range land
y, Forest land
3 3. Water
, that it 6. Wetland
date on 7. Barren land
e higher 8. Tundra
' of the 9. Perennial snow or ice
rithms.
rmation For Level II, classification detail is increased. For example the forest land is
re often subdivided into:
he main :
/ of the Deciduous forest land
loes the Evergreen forest land
-riment, Mixed forest land
i by the Ru À
t data is If a Level II classification is successful (accuracy of at least 85% — Anderson 1976) a
ut after further breakdown should be attempted.
Common standards will have to be set on "minimum size of area" to be
recognized for each class, and for the evaluation of accuracy. Total classification
accuracy as well as accuracy per level and per class will be computed. The computation
he most will follow the proposal by Kalensky and Scherk 1975, which includes a mapping accuracy
put data index and the confusion matrix.
Institutions or individuals having access to image interpretation systems are