Using GEOBIA to assess crown diameter classes of Acacia tortilis in Bou-Hedma, Tunisia
1. Using GEOBIA to assess crown diameter
classes of Acacia tortilis in Bou-Hedma, Tunisia
Kevin DELAPLACE a, Frieke VAN COILLIE a, Robert DE WULF a,
Donald GABRIELS b, Koen DE SMET c, Mohammed OUESSAR d,
Azaiez OULED BELGACEM d, Houcine TAAMALLAH d
Ghent, September 24, 2010
EARSeL Joint SIG Workshop: Urban - 3D - Radar - Thermal Remote Sensing and Developing Countries
2. Partners
a GhentUniversity, Belgium
Laboratory of Forest Management and Spatial Information Techniques
(kevin.delaplace, frieke.vancoillie, robert.dewulf)@ugent.be
b GhentUniversity, Belgium
Department of Soil Management and Soil Care
donald.gabriels@ugent.be
c Flemish
Government, Belgium
Environment, Nature and Energy Department
koen.desmet@lne.vlaanderen.be
d Institut des Régions Arides de Médenine, Tunisia
med.ouessar@ira.agrinet.tn, (Azaiez.ouledbelgacem, taamallah.houcine)@ira.rnrt.tn
3. Context
Acacia raddiana forest steppe in Bou-Hedma national park, Tunisia
A. raddiana persists on edge of desert keystone species pre-Saharan Tunisia zone
Past: Bou-Hedma no attention concerning biodiversity, protection and use
Result: desertification caused by excessive livestock grazing, partial land
clearance, ploughing soil erosion
1955: only few old trees left
Sixties: first actions to combat desertification/erosion
700 ha fenced
Tree nursery Gradual restoration of vegetation
Two Integral Protection Zones
1977: part of the network of Biosphere Reserves of UNESCO
1985: regeneration actions
Context – Aim – Study area – Imagery- Method – Results - Conclusions
4.
5. Context
Important activity in Bou-Hedma: restoration of original woodland
combating desertification
afforestation and reforestation
Cooperation between Flemish Government & Direction Générale des
Forêts of Tunisia
reforestation 50,000 ha with A. raddiana in historical geographic range
Part of larger project Kyoto Clean Development Mechanism (CDM)
scientific follow up of plantations, ecological and socio-economical consequences
Context – Aim – Study area – Imagery- Method – Results - Conclusions
6. Context
Restoration of A. raddiana induce local climate change
Consequences likely to occur
Change of soil temperature
Formation of humus by partial leaf shedding in summer
Influence on the cloud formation process, inducing precipitation
Interception of water by tree leaves and trunks, influencing erosion processes
However, scientific data remain scarce
Some phenological and ecophysiological studies
Little knowledge of composition, structure and by extension dynamics useful to
test efficiency of different management scenarios
Context – Aim – Study area – Imagery- Method – Results - Conclusions
7. Aim
In order to evaluate future trends assess current situation
Aim
RS-based monotemporal assessment of amount of A. raddiana
Estimation of diameter and height class distributions of A. raddiana
Research questions
What is the most appropriate (semi-)automated image processing procedure?
Are we able to establish models to estimate individual tree attributes?
Are we able to provide an estimate of the structure of the Acacia raddiana forest
steppe at Bou-Hedma National Park?
Context – Aim – Study area – Imagery- Method – Results - Conclusions
8. Study area
Location
34°24’ N to 34°32’ N
09°23’ E to 09°41’ E
Context – Aim – Study area – Imagery- Method – Results - Conclusions
9. Study area
Location
34°24’ N to 34°32’ N
09°23’ E to 09°41’ E
Altitude
90 m - 814 m above sea level
Context – Aim – Study area – Imagery- Method – Results - Conclusions
10. Study area
Location
34°24’ N to 34°32’ N
09°23’ E to 09°41’ E
Altitude
90 m - 814 m above sea level
Climate: Mediterranean arid with
temperate winters
average annual rainfall 180 mm
average temp 17.2°C
mean max temp 38°C
mean min temp 3.9°C
Context – Aim – Study area – Imagery- Method – Results - Conclusions
11. Field survey
Field conditions A. raddiana
Two spatial configurations
Plantations
‘Natural’
Individuals & tree groups
Presence of Eucalyptus sp.
> 400 A. raddiana trees randomly sampled
Tree positions (UTM WGS84)
Tree bole diameter (basal and breast height)
Total tree height
Crown diameter (two perpendicular directions)
Context – Aim – Study area – Imagery- Method – Results - Conclusions
12. Imagery
GeoEye-1 scene
Spatial resolution: 0.5 m PAN, 2 m MS
Radiometric resolution: 11 bits per pixel
Spectral resolution: PAN, Blue, Green, Red, NIR
Acquisition date:
Optimum: discrimination trees, soil and vegetation largest
Dry season: no lush vegetation, low cloud cover, maximum leaf development
1st of August 2009 at 10:18 GMT
Context – Aim – Study area – Imagery- Method – Results - Conclusions
13. Imagery
GeoEye-1 scene
Spatial resolution: 0.5 m PAN, 2 m MS
Radiometric resolution: 11 bits per pixel
Spectral resolution: PAN, Blue, Green, Red, NIR
Acquisition date: 1st of August 2009 at 10:18 GMT
Cloud cover: 0%
Preprocessing:
Geometric correction by GeoEye Inc.
Coordinate system: UTM (WGS 84)
Context – Aim – Study area – Imagery- Method – Results - Conclusions
15. Methodology GeoEye-1
MS bands
Object-based approach
Image Objects
Multi-resolution Object feature
GeoEye-1
/ contrast-split calculation / NN
PAN band 0
segmentation classification
Measured tree attributes Input and reference data per reference object
per reference object Acacia objects
Object Features Crown diameter Tree height
Crown diameter Segment 1 f1, f2, ………f200 5.2 2.5
Tree height Segment 2 f1, f2, ………f200 6.3 3.2
… …..
Segment i f1, f2, ………f200 4.1 2.1
Estimated tree attributes per Acacia object
Small CD/TH/…
Regression
Accuracy
models
Large CD/TH/…
Context – Aim – Study area – Imagery- Method – Results - Conclusions
16. Methodology
Image Objects
Multi-resolution
GeoEye-1
/ contrast-split
PAN band 0
segmentation
Multi-resolution segmentation (eCognition Developer 8):
Parameter Value
Scale 40
Shape 0.2
Compct 0.5
Context – Aim – Study area – Imagery- Method – Results - Conclusions
17. Methodology
Image Objects
Multi-resolution
GeoEye-1
/ contrast-split
PAN band 0
segmentation
Contrast-split segmentation (eCognition Developer 8):
Parameter Value
Contrast mode Edge difference
Min rel area dark 0.1
Min rel area bright 0.1
Min contrast 0
Min object size 10
Context – Aim – Study area – Imagery- Method – Results - Conclusions
18. Methodology GeoEye-1
PAN & MS
bands
Image Objects Acacia objects
Object feature
calculation / NN
0 classification
Type Feature (200)
Customized FDI (NIR-(R-B)), SAVI, NDVI
Layer values Mean, Min, Max, Border contrast, Contrast
to neighbour pixels, Edge contrast of
neighbour pixels, StdDev to neighbour
pixels, Circular mean
Geometry Area, Assymetry, Density, Compactness
Texture after Homogeneity, Mean, Correlation, Ang 2nd
Haralick moment, Entropy
Context – Aim – Study area – Imagery- Method – Results - Conclusions
19. Methodology GeoEye-1
PAN & MS
bands
Image Objects Acacia objects
Object feature
calculation / NN
0 classification
KIA=0.96
A. raddiana sp. Eucalyptus sp. Soil
Context – Aim – Study area – Imagery- Method – Results - Conclusions
20. Methodology GeoEye-1
object features
Measured tree attributes Input and reference data per reference object
per reference object
Object Features Crown diameter Tree height
Crown diameter Segment 1 f1, f2, ………f200 5.2 2.5
Tree height Segment 2 f1, f2, ………f200 6.3 3.2
… …..
Segment i f1, f2, ………f200 4.1 2.1
Measured Acacia’s: train & test sets
Regression Correlation analysis
Accuracy
models Tree attribute Object feature
Crown diameter Area
Tree height Area
Bole diameter Area
Object feature Object feature
Area GLCM Entropy Layer 4 (90°)
Context – Aim – Study area – Imagery- Method – Results - Conclusions
24. Methodology
Regression
Accuracy
models
Crown diameter estimation
60 Measured crown diameter (field data)
50 Crown diameter derived from area (number of pixels)
40
Frequency
Crown diameter derived from GLCM Entropy Layer 4
(90°)
30
20
10
0
[0,2] ]2,4] ]4,6] ]6,8] ]8,10] ]10,12] ]12,14] ]14,16]
Crown Diameter Classes (m)
Context – Aim – Study area – Imagery- Method – Results - Conclusions
25. Methodology
Acacia objects Estimated tree attributes per Acacia object
Small CD/TH/…
Regression
models
Large CD/TH/…
characterising forest structure (arrangement of diameters), height and density
Context – Aim – Study area – Imagery- Method – Results - Conclusions
26. Results
Acacia objects Estimated tree attributes per Acacia object
Small CD/TH/…
Regression
models
Large CD/TH/…
Structure (Crown diameter)
Context – Aim – Study area – Imagery- Method – Results - Conclusions
27. Results
Acacia objects Estimated tree attributes per Acacia object
Small CD/TH/…
Regression
models
Large CD/TH/…
Density
Characteristics original forest steppe, 1900
1925 (Zaafouri et al, 1996)
Kernel of 1 ha (200×200 pixels, 0.5 m resolution) Tree attribute Value
Total of 2596 non-overlapping kernels Total tree height 9 - 10 m (max 10 m)
Excluding border effect Trunk height 3 – 4 m at most
mean density = 8.4 trees/ha Mean basal diameter 15 – 20 cm
2.3 – 3 m
max density = 95 trees/ha (plantations) BD of biggest trees
Density 4 – 25 trees/ha
Context – Aim – Study area – Imagery- Method – Results - Conclusions
28. Conclusions
Answer to RQ
Object-based segmentation/classification approach suitably addressed
estimation of amount of A. raddiana
Established models adequately estimate of individual tree attributes
Estimation of structure, height and density Acacia raddiana forest steppe at
Bou-Hedma National Park
Future activities
Fine-tuning of segmentation / classification steps
Thorough validation
Multitemporal assessment inclusion of phenological information optimal
image acquisition time to estimate amount of Acacia’s and their attributes
Context – Aim – Study area – Imagery- Method – Results - Conclusions
29. THANK YOU FOR YOUR ATTENTION
(kevin.delaplace, frieke.vancoillie, robert.dewulf)@ugent.be
Laboratory of Forest Management and Spatial Information
Techniques, Ghent University, Belgium
http://dfwm.ugent.be/forman
Ghent, September 24, 2010
EARSeL Joint SIG Workshop: Urban - 3D - Radar - Thermal Remote Sensing and Developing Countries