nbul 2004
nventory:
ds Ltd. on
fillennium
woodland
est with a
aded or
gmented),
lifer, both
x
Highland
w 3.0a.
ıtative for
400 ha,
2) on the
dd be as
st amount
ble 1 and
count for
etation in
> montane
ne form of
) compare
hic ones,
land were
ES
yorland
porland
eat
a
ation
1re
canopy
jus
yniferous
t
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
Arable land
WB Improved pasture
EE Nardus and Molínia
BE Dry heather moorland
I Wet heather mooriand
ES Blanket bog / peat
Montane vegetation
Woodland
rire ei
etters)
J Coniferous plantation; full canopy cover
ss Semi-coniferous forest
! Mainly-confferous forest
KE Broadleaf forest
Figure 1: Location of the selected land cover polygons across
Scotland
3.2 Creation of areas of interest in the NDVI dataset
After the selection. of the relevant land cover data, those
polygons were overlaid with the NDVI data and ‘cut out’ on the
NDVI images. Areas of interest (AOI) were thereby created,
one for each land cover type. A programme using Erdas Macro
Language (EML) was developed, that, using the AOIs, removed
479
cloud and boundary pixels and, from the remaining pixels,
calculated the average NDVI for each area. Finally. the
extracted NDVI data were exported and displayed. The results
were thus the weekly NDVI values for the years 1995-98 for
every land cover polygon selected.
3.3 Further processing steps
The results were displayed using a spreadsheet-programme.
Graphs were made for all polygons. The results were displayed
in such a way as to give an overview of changes between the
polygons of each vegetation type and changes over time for any
polygon.
When looking at individual NDVI curves of a single polygon, it
can be seen that the curves are noisy (figure 2). The main reason
for this may be that pixels with apparent low NDVI values
contain sub-pixel elements of clouds or cloud shadow. Clouds
in a pixel diminish NDVI values. Highly variant atmospheric
effects (water vapour absorption and aerosol scattering) might
also be responsible as no atmospheric correction was applied in
the pre-processing. Another reason might be that the calculation
of the mean NDVI for a polygon would be undertaken, even
when there was only a few cloud-free pixels in the polygon. The
possibility of getting a representative NDVI value for this
polygon on this date is thus low. Pixels with small clouds are
therefore highly weighted. This is even more so where polygons
are made up of a smaller number of pixels.
07
0,6 4
05 4
N
A | X
8 or = j
04 | r \ d
|
NDVI
e
uw
7"
-
———
o
m
x
oa
«c
NDVI
2
d
n
\
A
TT.
A
x —
a
x
wm.
x
un
45 50
o
a
=
o
2
=
"3
th
8
8
m
a
(b)
Figure 2: Mean NDVI of different polygons of wet heather
moorland (a) and arable land (b).
It was thus decided to only include weeks in further
interpretation whose average NDVI had been calculated with as
many pixels as possible. The remaining pixels, however, should
still display a meaningful curve. For some polygons it was
possible to refer to the maximum number of pixels in a polygon,
otherwise a limit was set individually to include all weeks
where the pixels used for the calculation consisted of more than
