United States are
time, on a dwin-
products. They
arch to develop
has led to the
ting natural for-
The result has
U.S. has been
ng of the growth
companies due
monitoring tech-
| be costly and
over large areas.
Isat-type satel-
forest inventory
in a cooperative
the applicability
the study site.
ensive with the
onal feature of
y drained soils.
is resulted in a
tages of growth
hardwood. Ap-
; extensive road
compartment
graphy were also
composition by
1973) Landsat-
zed individually
emporal changes
gery, a contrast
1e 26 February
litions was gen-
black and white
ig. 4.1). There
. y pes of imagery
North Carolina/Williams
cation was accomplished using the standard analysis approach of
btaining spectral signatures and related statistics for these areas, and
a euclidean distance algorithm. Classification of the 4
band winter imagery proved the most useful for obtaining an accurate delineation of hard-
wood and pine forest canopies, while analysis of the 4 bands of summer data resulted in the
delineation of three different levels of pine canopy closure (i.e., closed, partial, and open
canopy). This appraisal of crown closure allows a relative assessment of average tree size in
situations of regular tree spacing due to artificial regeneration. The periodic combination of
this type of information with existing inventory records would provide observations of clo-
sure advancement over time, allowing forest managers to draw conclusions regarding the
growth rate and quality of forest stands.
Computer image classifi
selecting training areas, O
then classifying the study area using
In the final stage of analysis an 8 band multitemporal classification was performed and it re-
sulted in the best overall classification results (Table 4.1 and Fig. 4.2). The winter data con-
tributed to the accurate delineation of hardwood and pine stands and the summer data
simultaneously contributed to the delineation of pine sub-categories based upon the degree
of crown closure. Forest tracts that were clearcut during the six month interval between the
two Landsat overpasses were also easily delineated as they had a rather unique 8 band spec-
tral signature due to the lush vegetation present in August and the lack of vegetation in
February.
The accuracy of the final, 8 band multitemporal classification map was verified by randomly
sampling 232 classified picture elements and comparing them to the corresponding area on a
forest cover-type map generated via independent photointerpretation of 1:36,000 scale color
IR stereo photos. The results of these comparisons show a 9476 agreement for hardwood, a
96% agreement for the four combined pine categories, and a 54% agreement for clearcuts, or
an overall agreement of 90% (Table 4.1). The relatively poor agreement for clearcuts was a
result of using photos which were 18 to 24 months out of phase with the Landsat image data.
For example, clearcut areas that were recently replanted with seedlings were distinguishable on
the high resolution photos and were therefore classified as regenerated by the photo-interpreter.
These results indicate that intensive forest management practices on a large-scale basis can be
monitored with existing Landsat resolution and that forest managers can obtain useful infor-
mation to supplement that which is available via more conventional techniques.
Further Information:
References
Williams, D. L. and G. F. Haver. 1976. Forest land management by satellite:
Landsat-derived information as input to a forest inventory system. NASA/Goddard
Space Flight Center, Greenbelt, Maryland. 36 p.
Williams, D. L. 1976. A canopy-related stratification of a southern pine forest using
Landsat digital data. Proc. of the Fall Convention of the American Society of
Photogrammetry, Falls Church, Virginia. pp. 231-239.
Experimenters
Darrel L. Williams, National Aeronautics and Space Administration, Goddard Space
Flight Center, Earth Resources Branch, Greenbelt, Maryland 20771 U.S.A.