770 
ERTS-1/MSS CCT no. 1044-15170 of Sept. 05, 1972 
1315-15233 of June 03, 1973 
1440-15154 of Oct. 06, 1973 
DESCRIPTION OF THE TEST AREA AND GROUND CLASSES 
The Larose Forest test area is located in the Great Lakes-St. 
Lawrence forest region in southeastern Ontario approximately 30 miles east of 
Ottawa. It is characterized by flat terrain with sandy soils once covered by 
the Champlain Sea. Most of the present tree stands are about 35-year-old inte 
plantations consisting of white spruce (Pioea glauca ), red pine (Pinus temp 
resinoso ), white pine (Pinus strobus ) and poplars (Populus ). The test area impc 
which includes the Larose Forest and surrounding agricultural land occupies aeri 
216,000 ERTS pixels, that is about one thirty-fourth of the ERTS image area. of g 
to t 
Though the primary purpose of classification was to delineate c l as 
agricultural land, coniferous forest and decidious forest, four more classes ERTS 
were included in the classification scheme because of their distinctive comb 
spectral signatures: three classes of water (deep water, clear shallow water, yiel 
silted shallow water) and the 1971 landslide at the South Nation River. 
Statistical description of ground classes was based on training sign 
areas which were delineated within each class. Their selection is one of the coni 
most important tasks because they provide the teaching patterns upon which the proc 
computerized classification is based. If they are not representative for 1)• 
respective classes, large classification errors will be introduced regardless feat 
of sophistication of classification procedure used. The problem is especially ERTS 
difficult in forest classification where local variations in species on 
composition, maturity and vigor are frequently caused by microtopography, soil mult 
type, moisture and diseases. Considering the relatively large size of pixels 
(0.6 hectares) it is sometimes impossible to assign them into single classes 
even if medium-scale aerial photographs or ground survey data are available. vect 
Furthermore, the intensity and spectral distribution of visible and 
near-infrared radiation reflected from vegetation are not constant and vary as as 1 
a function of the species' phenological stage, sun elevation, weather 
conditions and sensors' sensitivity at the time of recording. Hence, the 
accuracy estimates of computerized forest classification more often represent 
degree of success in selection of training areas, ERTS scenes and their 
combination than accuracy measures of the classification algorithm. 
The ground truth data for training areas and accuracy estimates were 
provided by a 1960 forest cover map updated through medium-scale (1:10,000) 
1969-72 color aerial photographs and spot field checks. Digital pixels of the 
ERTS image were registered with corresponding image detail through a scaled 
pixel grid. The scale differed along the x and y coordinates as a function of 
the camera attitude. The grid was plotted on a transparency and overlaid on a whe] 
photographic print which had to be corrected for terrain topography if image 
distortion exceeded the size of one-half pixel.