METHOD Apr 
Forest boundaries and areas for the country wide forest mapping and 
monitoring were obtained from visual interpretation of the various standard 
Landsat hard copy products and copied onto transparent material. Base maps of 
with a 10 km. Universal Transverse Mercator (UTM) grid system, gazetted Thi 
forest boundaries, roads, district boundaries and towns were prepared at rat 
1:100,000 and 1:50,000 scales since these were the most convenient for hig 
interpretation purposes. A number of control points for correct positioning had 
and scaling of the aerial photos and satellite imagery relative to the topo- (Ta 
graphic base were also included. These were obtained from the standard 
1:250,000 and 1:50,000 Scales Survey of Kenya topographical maps. Dry Season 
Landsat imagery was preferred due to the clearer distinction between forest and 
and woodland. Kak 
18% 
Using an overhead projector with minimum distortion, imagery in the The 
form of colour transparencies at 1:1,000,000 scale was projected onto a base rep. 
map. Scale adjustments were made in relation to control points and the drg 
forest cover was then drawn on the topographical base from the transparencies. mea: 
Areas designated as forest clear-cuts were then deleaneted on the scenes and ind: 
old aerial photographs taken at different dates going back to 1967. Where witl 
there were no cloud free Landsat scenes available topographic maps at have 
1:50 000 scale provided a convenient base for mapping from a light aircraft tare 
(Doute et al 1981). Boundary shifts as a result of clear-cutting in the for 
critical forest areas were also checked during reconnaissance flights with fore 
light aircraft. Oblique aerial photographs taken during the flights were the 
used to make boundary corrections. Areal measurements were made at least fore 
twice using a planimeter. rema 
Selection of the critical forest areas for monitoring forest cover 
changes was based on the high rates of depletion observed from earlier work area 
(Doute et al 1981) and partly on their location in relation to their surround- high 
ing ecological zone. Representative forests were chosen from both the diff 
marginal and arid areas as well as from agriculturally high potential areas. (197 
f 
In addition to the visual interpretation, digital analysis was done ER 
of a few forest areas using the standard EROS 1600 BPI Computer Compatible | and 
Tapes (CCT's) on the Image Analysis System (CIAS). The system with its the . 
hardware and software has been described by Goodenough (1977). An unsupervised to a 
classification was carried out with seven ''classes' or clusters which were agri 
then correlated to features on the ground (Goldberg et al., 1975). A spatial | it h: 
filtering technique was then implemented; which eliminated border pixels or | with 
smoothed the effect of texture (Goldberg & Goodenough, 1976). Hard copies 
of the classification consisted of colour slides and scaled 1:50,000 or 
1:100,000 alphanumeric printout maps from the line printer. A second resu 
technique used was a contrast stretch which expands the range of the pixel Was 1 
values so that they are displayed over a fuller range of grey tones. Greater at le 
radiometric detail in the forest areas was required and by assigning the a 1% 
display range exclusively to a particular range of forest reflectances it using 
was possible to increase the detial with the forest areas as well as the fores 
boundaries. being 
RESULTS AND DISCUSSION Rhe 
For the country wide forest cover mapping, 270 forests were mapped Suus 
on 72 map sheets (Doute et al., 1981) with a total forest cover, including image: 
mangroves, of 1,370,160 ha. (Table 1). This was divided into indigenous a lig 
forest cover of 1,167,180 ha., plantations 160,000 ha. and mangroves 52,980 ha. throu; 
The indigenous forest cover represents 1.9% of the total land area of Kenya. 
490 
p: -Y—-— EN S Ri 4 pon 
D