Test of forest classification over Bavaria Germany using a SPOTVGT pixel mosaic

1
Test of forest classification over Bavaria
(Germany) using a SPOT-VGT pixel mosaic

• Erwann FILLOL, Pamela KENNEDY, Sten FOLVING

2
Objectives

• To create a cloud-free image of Europe by pixel
  compositing SPOT-VGT S1data
• To evaluate the efficiency of the Maximum NDVI /
  Minimum Red (MaNMiR) pixel compositing method in
  discriminating three types of forest cover
  Evergreen, deciduous, and mixed through classic
  classification methods

3
Databases

• SPOT-VGT S1 45 daily acquisitions for the
  months of July and August, 2000 for all of Europe
• CORINE Land Cover database (CLC) 44 classes
  for 3 hierarchical levels (Artificial surfaces,
  Agricultural areas, Forests and semi-natural
  areas), obtained in part using Landsat TM imagery
  (resolution 100 meters)

4
Pre-processing Compositing MASKING

• Cloud cover (Lissens et al.)
• if ?bluegt 0.36 and ?swirgt 0.16 ? Cloud
• Dilation of cloud cover 5kmx5km window
• Elimination of cloud shadow
• Scan angle limitation
• If ?v gt 45 ? Ground resolution degradation
• SWIR detector defects
• If ?swir gt 0.75 ? SWIR defect
• Hot-Spot and Specular limitation

5
Pre-processing Compositing MASKING

• Cloud cover (Lissens et al.)
• if ?bluegt 0.36 and ?swirgt 0.16 ? Cloud
• Dilation of cloud cover - 5x5 window
• Elimination of cloud shadow
• Scan angle limitation
• If ?v gt 45 ? Ground resolution degradation
• SWIR detector defects
• If ?swir gt 0.75 ? SWIR defect
• Hot-Spot and Specular limitation

6
Pre-processing Compositing MASKING

• Cloud cover (Lissens et al.)
• if ?bluegt 0.36 and ?swirgt 0.16 ? Cloud
• Dilation of cloud cover - 5x5 window
• Elimination of cloud shadow
• Scan angle limitation
• If ?v gt 45 ? Ground resolution degradation
• SWIR detector defects
• If ?swir gt 0.75 ? SWIR defect
• Hot-Spot and Specular limitation

7
Pre-processing Compositing MASKING

• Cloud cover (Lissens et al.)
• if ?bluegt 0.36 and ?swirgt 0.16 ? Cloud
• Dilation of cloud cover - 5x5 window
• Elimination of cloud shadow
• Scan angle limitation
• If ?v gt 45 ? Ground resolution degradation
• SWIR detector defects
• If ?swir gt 0.75 ? SWIR defect
• Hot-Spot and Specular limitation

8
Cloud shadow elimination

• Solar zenith and azimuth angles are known
• Cloud height minimum and maximum are estimated
• The distance dh/tan(90- ?s) and direction of
  the cloud shadow can be estimated

9
Pre-processing Compositing MASKING

• Cloud cover (Lissens et al.)
• if ?bluegt 0.36 and ?swirgt 0.16 ? Cloud
• Dilation of cloud cover - 5x5 window
• Elimination of cloud shadow
• Scan angle limitation
• If ?v gt 45 ? Ground resolution degradation
• SWIR detector defects
• If ?swir gt 0.75 ? SWIR defect
• Hot-Spot and Specular limitation

10
Pre-processing Compositing MASKING

• Cloud cover (Lissens et al.)
• if ?bluegt 0.36 and ?swirgt 0.16 ? Cloud
• Dilation of cloud cover - 5x5 window
• Elimination of cloud shadow
• Scan angle limitation
• If ?v gt 45 ? Ground resolution degradation
• SWIR detector defects
• If ?swir gt 0.75 ? SWIR defect
• Hot-Spot and Specular limitation

11
Pre-processing Compositing MASKING

• Cloud cover (Lissens et al.)
• if ?bluegt 0.36 and ?swirgt 0.16 ? Cloud
• Dilation of cloud cover - 5x5 window
• Elimination of cloud shadow
• Scan angle limitation
• If ?v gt 45 ? Ground resolution degradation
• SWIR detector defects
• If ?swir gt 0.75 ? SWIR defect
• Hot-Spot and Specular limitation

12
Hot-Spot and Specular limitation
To minimise directional effects, the acquisitions
situated near the hot spot and specular zones (
20) are eliminated
13
Resulting mask
Image 26th of August 2000
14
Resulting mask
Image 26th of August 2000
15
Resulting mask
Image 26th of August 2000
16
Resulting mask
Image 26th of August 2000
17
Pre-processing Compositing DOUBLE CRITERIA
COMPOSITING

• Double criteria compositing
• Maximum NDVI (MaN), to eliminate haze and
  unscreened pixels top 15 retained
• Minimum reflectance in the red channel (MiR), to
  limit atmospheric effects and enhance green
  vegetation

DIorio and al., 1991
18
Composite result
19
Composite result
20
Test area Bavaria (Germany)
Corine classification Resolution 100 m
21
Test area Bavaria (Germany)

• Test site selection based on
• little topographic effect
• 3 forest types present coniferous, deciduous,
  mixed
• site is representative of temperate forests

Corine classification Resolution 100 m
22
Test area Bavaria (Germany)
Corine classification Resolution 100 m
23
Classification

• Maximum Like-lihood algorithm
• Training site selected over a homogeneous area
  (according to Corine classification)
• Using channels SWIR, NIR Red
• 3 classes Coniferous, deciduous, mixed forests

24
Spectral separability
SWIR reflectance
SWIR reflectance
NIR reflectance
Red reflectance
NIR reflectance
Red reflectance
25
Classification Results
Corine classification
SPOT-VGT composite
26
Classification Results
Corine classification
SPOT-VGT composite
27
Classification Results
Corine classification
SPOT-VGT composite
Dense forest zones most accurately
classified Over estimation in sparse forest due
to surrounding (pasture)
28
Classification Results
Corine classification
SPOT-VGT composite
29
Classification Results
Corine classification
SPOT-VGT composite
Under estimation in sparse and fragmented
forest. Surrounding Non-irrigated arable land
30
Composition of actual land cover (based on CLC)
classified as Coniferous according to SPOT-VGT
Coniferous from SPOT-VGT
31
Composition of actual land cover (based on CLC)
classified as Broad-Leaved according to SPOT-VGT
Broad-Leaved from SPOT-VGT
32
Composition of actual land cover (based on CLC)
classified as Mixed Forest according to SPOT-VGT
Mixed from SPOT-VGT
33
Composition of actual land cover (based on CLC)
classified as Non-Forest according to SPOT-VGT
Non-Forest from SPOT-VGT
34
Classification sensitivity to sub-pixel forest
density
35
Conclusions and discussion

• High quality composites are possible with
  Spot-VGT
• High potential in discriminating dense
  coniferous wood-land
• Must be careful with area estimation of forest
  cover in Europe, especially in fragmented and
  mixed forest
• Potential of combining medium resolution
  radiometer like IRS-WiFS (200m resolution) and
  low resolution SPOT-VGT