Mapping of arid regions in N' Africa, middle East and Southeast Asia using VGT S10

1
Mapping of arid regions in N. Africa, middle East
and Southeast Asia using VGT S10
Michael Cherlet
2
Mapping of arid regions in N. Africa, middle East
and Southeast Asia using VGT S10
3
Mapping of arid regions in N. Africa, middle East
and Southeast Asia using VGT S10
4
Specific Problematic for Mapping Land Cover
in Arid Areas
Low cover vegetation gtgt 3 - 40 (LCCS
sparse to open) mixed with background soil
S10 NDVI products gtgt high variability of
NDVI not explained only by vegetation
5
IGBP
6
Specific Problematic for Mapping Land Cover
in Arid Areas
timing of seasonal variability related to
vegetation is difficult to determine - erratic
character of rainfall in space and
time - influence of two climatic zones N
gt Mediterranean influence S gt tropical
ITCZ influence not possible to choose
best period for vegetation development throughout
year gtgt difficult to use S1
7
Specific Problematic for Mapping Land Cover
in Arid Areas
Using SPOT VGT S10 or longer composites based on
MVC atmospheric, aerosol or clouds
contamination is limited in S10 over arid areas
(no persistence) BRDF effect which is
probably enhanced in relation to
topography Spectral behaviour related to
lithology and geology (colour) confusi
on between low cover vegetation and sandy
soils/sand-stones
8
Contamination on S10
Threshold on ratio MIR/BO improves classification
of unsure class
9
Specific Problematic for Mapping Land Cover
in Arid Areas
Using SPOT VGT S10 or longer composites based on
MVC atmospheric, aerosol or clouds
contamination is limited over arid areas
(no persistence) BRDF effect which
is probably enhanced in relation to
topography Spectral behaviour related to
lithology and geology (colour) confusi
on between low cover vegetation and sandy
soils/sand-stones
10
NDVI
In general, but locally of importance
increases confusion of e.g. sandstone outcrops
and vegetation
11
Specific Problematic for Mapping Land Cover
in Arid Areas
Using SPOT VGT S10 or longer composites based on
MVC atmospheric, aerosol or clouds
contamination is limited over arid areas
(no persistence) BRDF effect which
is probably enhanced in relation to
topography Spectral behaviour related to
lithology and geology (colour) confusi
on between low cover vegetation and sandy
soils/sand-stones
12
Final Approach still open Three methods tried
1.
1.producing yearly composites - NDVI image Max,
Min, amplitude statistics (st.
dev.) (cloudmask) - NDWI image Max, Mean,
Min, amplitude statistics (methods
tested) - Minimum B0, B2, B3, Mir
differentiation of different zones/masks using
Max NDVI thresholds ( cover)
Sensor sensitivity 0.01
13
- non-supervised classification (isoclass) within
masks using yearly derived products - grouping
of non-vegetation vs vegetation classes
and re-iterate isoclass and regrouping (min
3) based on subjective interpretation of all
available data and field knowledge - final
grouping of all non-vegetation and
vegetation masks - differentiation of a.
physical features using isoclass on bands and
regrouping within non-vegetation b.
different life forms within vegetation part
using NDVI time series statistics and
ancillary data
14
Orange 3 - 6 cover (GP length?) gt LCCS
sparse herbaceous
Aquam 6 - 10 cover (GP length?) gt LCCS
herbaceous
green1 10 - 20 cover (GP length?) gt LCCS
green2 20 - 40 cover (GP length?) gt LCCS
15
2.
2. - producing yearly composites - NDVI image
Max, Min, amplitude statistics
(st.dev.) - NDWI image Max, Mean, Min,
amplitude statistics - Min B0, B2, B3,
Mir - stratification of land-units based on
classification of bands (isoclass and
re-grouping) - non-supervised classification
(isoclass) within landunits using yearly derived
products - grouping of non-vegetation vs
vegetation classes and re-iterate isoclass
and regrouping (min 3) based on subjective
interpretation of all available data and field
knowledge - final grouping of all
non-vegetation and vegetation masks -
differentiation of a. physical features using
isoclass on bands and regrouping within
non-vegetation, optimizing first
stratification b. different life forms
within vegetation part using NDVI time
series statistics and ancillary data
Used to attach further info to vegetation
classes
16
3.
3. Determination of vegetation character of
individual pixels based on detection of
significant NDVI change during year 2000 by
separation of background noise from signal
using long term time series to establish
noise level per pixel ()
() in cooperation with Univ. UCL, Belgium
17
Result of the iterative process
Pixel FLAGGED NDVI gt Mean nSTD
Reflects a status of CHANGE in probable
vegetation cover related to its dry season
status (whatever that is .... Soil or
vegetation....)
() in cooperation with Univ. UCL, Belgium
18
Avg 2STdev
19
Temporal mask and spatial mask
Needs refining to be used as base probable
vegetation - non vegetation
20
Conclusions methods 1 2 - straightforward
techniques - need for ground knowledge -
subjective - not very repeatable method 3 -
still to be validated technique - fine tuning
required - objective - repeatable -
ground knowledge only required in final stage