bul 2004
€ OCCUI-
r natural
Ocessing
e terrain
ing pine
| as class
the areas
laced by
e. Over
me step
ate, fol-
for 1989
(dry sea-
Ocessing
of MSS-
» COarser
as their
id infra-
on MSS-
1s to be
:ps-1075
egarding
studies,
ar as the
TM-data
aries
ty in the
concen-
ithin the
1e 1930s
Maragoli
s shown
y are not
F the 11
d by the
all time
‘est con-
'condary
ver time
- ha of
compare
)0 ha in
in 2001.
loss in
past 30
at "Near
y forest"
bserved:
ompared
nted via
> typical
f forest
for each
he forest
ced op-
ding the
^ar natu-
forest^,
and
are ob-
e see the
11 Forest
the arca
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
covered by “Near natural + old secondary forest“ and *Secon-
dary forest” is generally decreasing while the area of “Bushland
/ shrubs^ is continously increasing since 1975. South Nandi
Forest is marked by a tremendous loss of forested area: 15,000
ha of class 1 and 2 in 1972 as compared to 7,200 ha in 2001.
The rather sudden increase in “Bushland / shrubs® area between
1979/80 and 1984 can be disputed regarding its height. The
steep step in the numerical values is more likely a result of
change in sensor systems used. Small island areas of bushland
found as gaps within forest formations due to selective logging
are not detected by MSS because of its coarser resolution as
compared to ETM+/TM and therefore ,disappear“ in the forest
classes.
4,3 Change-Detection-Analysis
In some cases the extent of forest is reaching across the official
forest boundaries. In order to consider these forested areas also,
the forest cover as marked on the topo maps in 1:50,000 scale
(based on aerial photography from 1967) are digitized. The
digitized forest cover is overlayed by the official forest bounda-
ries from the 1930s (GIS-functionality Union) and the result
buffered by 1 km. The derived area extent is in total used for a
complex change-detection-analysis. Thus on the one hand it is
ensured, that also the widest extent of forest formations within
the complete time series is considered in analysis. On the other
hand, the mass of the cultivated land surrounding the forest
areas with its difficulties of in particular distinguishing between
agricultural land and grassland is excluded. This area is of no
importance for the current research objectives of the BIOTA
project.
While when looking at the areas enclosed by official forest
boundaries area sizes covered by the land cover classes were
derived for every single forest area and time step individually,
for analysis of the buffered forest areas the total area delineated
is assessed by directly comparing each two time steps. Via the
p
land cover classes of high total area portions
Land Cover Change 1972/73 to 2001
30.000
25.000 -
20.000
|
15.000
total area in ha
|
10.000
5.000 —
i
0 —
6 10
local GIS-function Combine a matrix is generated that assigns
an unique value for every possible combination of the 12 land
cover classes. Not only the classifications of the neighbouring
time steps (1972/73 compared to 1975, 1975 to 1979/80, etc.)
are evaluated but also the complete time series in three time
steps (1972/73 to 1989 and 1989 to 2001) as well as its start
directly with its end (1972/3 to 2001, sce Figure 4). Besides
change matrices again diagrams have been prepared. For each
land cover class two bars are placed in one another. The broader
bar in the back represents the area covered by that land cover
class for the latter of the two compared time steps. The narrow
bar in the front reveals the portions of the land cover classes by
which that area was covered at the earlier time step of the two
under consideration. The unchanged portion is always displayed
as the lowest part of the bar. Above that base the areas that have
changed the land cover class are found. For deriving the total
area covered by a certain land cover class at the earlier time
step, the portions (which are still absolute values) of this class
within the different narrow bars are to be added.
Via these diagrams statements regarding both quantity and
direction of changes in land cover are possible. Some peculiari-
ties which can be attributed to classification difficulties attract
attention. For the two classes 6 (“Grassland‘“) und 10 (“Agricul-
tural land“) a high proportion of the other land cover class is
shown for the earlier time step (see inner bar). The diagram thus
reveals areas which change between either being covered by
agricultural land or by grassland. This confirms the conclusion
made earlier, that areas classified as grassland (in particular
along the edges of the forest areas) are to some extent more
likely areas used for agriculture. For the class „Others“ the
unchanged proportion is rather low but the portions of ,,Grass-
land" and Agricultural land“ for the earlier time step are in
contrast quite large. This is to be interpreted that the areas cov-
ered by that class changes over time which is most unlikely
because e.g. roads and settlements contributing to this class
"Others" are not expected to change positions to a large extent.
TTY
height of bar represents area covered
So by a certain land cover class in 2001
|^
E IN. portions of land cover classes found in
Fd ~ 1972/73 on this area covered by a single
land cover class in 2001 (unchanged portion
is displayed as the lowest part of the bar)
land cover classes of low total area portions
6.000
5.000
|
4.000
|
3.000
2.000 E
1.000
8 gk
Figure 4: Land cover changes 1972/73 to 2001 for total area of Kakamega Forest and associated
forest arcas buffered by km (for colours see Figure 2)
469