You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Remote sensing for resources development and environmental management
Damen, M. C. J.

is an added advantage.
The main objective now is to develop further and
use the method for accurate and precise inventory
ing, mapping and monitoring of land use and land
cover for regional as well as national resource
management. Specifically, therefore, the method is
being used, among other things, to estimate:
1. crops (and other class types) hectarages;
2. crops (and other class types) distribution and
intensity zones;
3. crop mix zones; and
4. spread of cultivation into the rangeland.
Outlined below is a summary of the procedures and
preliminary results for a land use and land cover
survey for Meru District in 1985. Similar surveys,
whose results are briefly compared, have been
accomplished for 9 other districts and more are in
the process.
The method entailed multistage sampling techniques
using low altitude aerial vertical photographs.
The district covering approximately 9850 km^ was
divided into 5x5km square sampling units using the
UTM grid system and east-west sampling flight lines
were placed 5km apart. Two sample photographs were
taken systematically at every 2.5km UTM grid inter
sections in the high potential region and at 5km in
the marginal rangelands. Flight height was
approximately 488m above ground level. Navigation
used the Global Navigation System (GNS) fitted in
the twin-engine Partenavia aircraft.
The photographic system used was a 35mm Nikon F3
camera with a 20mm lens and 35mm ektachrome 200
slide films. This gave positive transparencies
covering approximately 560m x 850m on the ground,
with a scale of 1:24,380. The positive trans
parencies were each overlayed and fixed with 91
(7x13) systematically placed dots on a transparency
and observed under a hand held slide viewer or
sometimes projected onto a screen. Crop and other
cover classes coincident with each dot were identi
fied by photointerpretation. The area covered by
each class type was then estimated by the pro
portion of dots for that class type to the total
number interpreted in the 5x5km sample unit and
expressed as a percentage. This was the basis for
the construction of overlays for each class type in
terms of distribution, density, and crop mix zones.
A framework of procedures for regional and field
familiarization before and during interpretation
were established and followed to increase accuracy
and consistency in interpretation. Before actual
district-wide photointerpretation began preparatory
procedures involved literature review; increasing
relevant local knowledge through discussions;
choosing one or two training photosamples and
making trial runs on them in interpretation and
then ground checking them for every major ecologi
cal region in the district; and this last exercise
enhanced the interpreters' familiarization of the
land use and land cover classification system which
was being used. Procedures were also laid down on
how to classify dots falling on non-pure class
types like mixed fields, field boundaries and
single trees to avoid personal bias.
The identification of land use and land cover
types for every dot on the photograph was done for
at least the first three levels of classification
coarseness as described by Anderson et al (1976)
and modified for local adoption. Accuracy and
consistency of identification of class types were
assessed using 'confusion tables' as outlined by
Kalensky (1975,1978) and linear regression analysis
between photointerpretation and field sample checks
at every level of identification using aggregate
data from every photo-sample field checked. The
correlation coefficient was considered a measure of
consistency in interpretation. Near perfect inter
pretation would have a regression line of Y = X and
a correlation coefficient (r) > 0.90.
Mapping was partly done using Map Analysis
Package (MAP) a Geographical Information System
(GIS) software developed at Harvard and Yale
A total of 443 photographs were analyzed for 263
sample units. Three areas of the district, Mount
Kenya National Park and Forest; Meru National Park
area; and the Northern Conservation Area, were not
surveyed mainly because their uses and cover types
are well documented and partly because of terrain
and air turbulence.
For accuracy and consistency tests twenty-seven
(27) photo samples were interpreted and ground
checked by each of the three photointerpreters.
The dots overlay method was used in assessing their
interpretation accuracy and consistency. Each dot
was fixed and examined by each interpreter and
therefore making it possible to compare their
performance. A total of 2457 dots were checked in
the 27 photos. Correlations were carried out
between photointerpretation and ground truth for
those class types with sufficient data and occur
rence for the three interpreters separately and for
the three combined to give an average. Interpreta
tion for the major cover types was fairly similar
between the three interpreters.
The interpretation accuracy test showed that land
under cultivation as a class was interpreted with
an average accuracy of 91% with r=0.93; improved
and unimproved pasture 77% with r=0.89; bushlands
71% with r=0.90;; forests 84%; and the rest of the
third level classes had few (<50) dots and photo
sample occurrence for analysis.
For specific crops and cover classes, maize was
interpreted with an average accuracy of 83% with
r=0.94; coffee 86%; tea 85%; bananas 62%; cotton
49%; ploughed land 72%. The rest had insufficient
data for analysis.
Table 1 below gives the percent cover for the
major class types condensed from over 35 specific
class types identified in the district. For
example, of the cultivated land, 38% was under
maize, 14% was under coffee, 12% was ploughed, 9%
was fallow and the remainder 27% was distributed
among 16 different crops. Distribution and
intensities for each of the extensive class type
was mapped and contouring approximated using MAP.
The systematic dot method was therefore found to be
quick and reliable for estimating areas of each
class type for the whole district; a difficult dot
could be discussed among the interpreters to
reach a consensus. Thus it is contended that
district-wide interpretation accuracy and con
sistency should be higher than that given above by
the accuracy tests for most class types.
Previous analyses for the other districts
invariably involved the projection of the slide
photos onto a screen which had fifty random dots
on which class types were identified. For each
district an attempt was made to sample survey the
whole district surface area but large tracts of
water bodies and extensive gazetted forest
reserves were often omitted to reduce costs and
expedite the process to cover much more critical
areas. Table 1 is also a summary of the percent
area under the major land use and land cover types
found in the sampled area for the nine districts.
(The indicated areas in brackets are the sampled
areas only, which may not necessarily be the area
of the district).