obia for combining lidar and multispectral data to ... · pdf fileivth international...

IVIVthth International Conference, International Conference, & li i& li iLIDAR & SAR applicationLIDAR & SAR application

OBIA for combining LiDAR and multispectral data to characterizemultispectral data to characterize forested areas and land cover

in tropical regionin tropical region

StéphaneStéphane DupuyDupuy11, , GérardGérard LainéLainé1,1, and Thierry Tormosand Thierry Tormos22

(1)(1) CIRAD,CIRAD, UMRUMR TETIS,TETIS, EarthEarthObservationObservation forfor EnvironmentEnvironment andandLandLand Management,Management, Montpellier,Montpellier,

((22)) IRSTEA,IRSTEA, URUR MALY,MALY, PolePole ONEMAONEMA //IRSTEAIRSTEA freshwaterfreshwater hydroecologyhydroecology,,Lyon,Lyon, FF‐‐6933669336,, FranceFranceLandLand Management,Management, Montpellier,Montpellier,

FF‐‐3439834398,, FranceFranceLyon,Lyon, FF 6933669336,, FranceFrance

The French National Agency for Water and Aquatic EnvironmentsThe French National Agency for Water and Aquatic Environments

Context Objective Data Pre‐processing Method Result Conclusion

Study area

375 km²

Mayotte is an island of the ComoroArchipelago located at the entranceof the Mozambique Channel

2GEOBIA 2012 May 7‐9, 2012 ‐ Rio de Janeiro ‐ Brazil

qFrench territory : overseas department


Forest areas in MayotteSince 2002, five forest reserves were established to preserve biodiversity of Mayotte

(5550 ha = 15% of the territory)( y)

Forest complexes are more or less degraded inside reserves because of old and actual human pressures

Agricultural activity in reserves(direct human pressure)( p )

Eroded areas result from slash and Clearing of local


burn cultivation species



(5550 ha = 15% of the territory)

Forest complexes are more or less degraded inside reserves because of old

( y)

and actual human pressures

Presence of Lianas(indirect human pressure)

Invasive specie( p )

C ll d f


Collapsed forest areas



(5550 ha = 15% of the territory)

Forest complexes are more or less degraded inside reserves because of old

( y)

and actual human pressures

Managers need spatial information about :1‐ status (degraded or not) of forest areas inside reserves2‐ presence of human pressures inside and outside reserves

5GEOBIA 2012 May 7‐9, 2012 ‐ Rio de Janeiro ‐ Brazilto prioritize preservation and restoration strategies


Objective: Mapping forest types, their status, and human pressuresp

Classification schemeForest types inside reserve …. and their status

Canopy height :[5‐10] m

Canopy height : ≥10 m

h f d

Not degraded / potentially degraded

Mangrove

Forest plantation (reforestation)

Need toLidar & multispectral

Other forested areas

Human pressures (direct and indirect)Forest plantation (reforestation)

Collapsed forest area induced by lianas

remotely sensed data

Eroded areas (bare soil / herbaceous / fern)

Artificial surfaces ( urban and agricultural)

Other land cover classes


Other land cover classesNatural Low vegetation / Shrub cover / bare soil / bare saline soil / beach dune / water …


Lidar data Lidar cloud points

1 x 1 m, October 2008

DSMDigital Surface Model

DTMDigital Terrain Model

difference

DHMDigital Height ModelDigital Height Model

Median filter(3×3 m)

DCM

DCMDigital Canopy Model

Popescu et al., 2002


DCM


Multispectral data

Spot 5 XS Aerial photographies (orthophotos)Spot 5 XSGreen, Red, Near Infra‐Red and

Medium infra‐Red10 x 10 m June 2005

Aerial photographies (orthophotos)Blue, Green, Red, and

Near Infra‐Red0 5 x 0 5 m November 2008


10 x 10 m, June 2005End of wet season

0.5 x 0.5 m, November 2008Dry season


Thematic dataMMangrove o MangroveMangrove

Eroded areaso Forest plantationso Eroded areas

Forest plantations

o Forest plantationso Collapsed forest areas with lianao Artificial surfaces (Built up area,

R lti f

Main road, Mine dump....)

Resulting from :1‐ available topographic data or2‐ previous photo‐interpretation studies based on multispectral images



Two pre‐processing on DCM data1‐For improving segmentation of forest type

Based on DCM Based on Max filter and DCMp g g yp

Max filter(3×3 m)

Object boundaries

(3×3 m)Popescu et al., 2002



Two pre‐processing on DCM data2‐ For enhancing the classification of forest types status (degraded or not)

CooccuNo degraded area

DCM profile on transect

530 5

Haralick texture : GLCM variance all

directionsm) 15

2025

ce23

4

51 m x 51 m.he

ight (m

0510

5 ckvaria

n0

16

Degraded area

Cano

Cano

py

2530

35

Haralic

45

6CaCo

1015

20

> 2 potentially degraded2

311

051

0 50 100 150

< 2 not degraded

01


Image classification techniqueII

OBIAo Multi‐level segmentation approach

II

o Classification based on expert knowledge rules

NN

Softwareo eCognition 8 using multi‐resolution algorithm (region growing technique)


Context Objective Data Pre‐processing Method Result ConclusionOBIA workflow

vector

data

Segmentation (level 1)

Step 1: Thematic classification Step 1

ThematicNo Thematic

max filterTh

ematic Classification

(level 1)

Step 2: Height classification

Mangrove

Eroded areas

Built areas

LiDAR de

DCM

Segmentation (level 2)

l f

Step 2…

erived data

canopy h

e)

Classification (level 2)

Merging same class

≤ 5 m 5 - 10 m ≥ 10 m

≥ 10 m pot. degraded

5 – 10mNo degradedheight co‐occurr

variance

ral data

y + Spot 5 im

age Adjacent objects

Segmentation

Step 3: Land cover classification

Step 3

p g

≥ 10 mno degraded

5 – 10 mPot. degraded

g

rence

Multispe

ctal pho

tography Segmentation

(level 3)

Classification

pWater

Bare soils

bare soils on Erodedareas

…


(aeri

(level 3)

3 successive segmentation‐classification processes…

Context Objective Data Pre‐processing Method Result ConclusionMap accuracy

Control data derived from field measurements

August and October 2009 and January 2010

• Global accuracy : 84 %• Kappa index : 80 %• Kappa index : 80 %

• Highly accurate results for(more than 80%)(more than 80%)‐ forest types‐ shrub and low vegetation• Poorest accuracy for(less than 80 %)‐bare soil on eroded area


‐herbaceous on eroded area

Context Objective Data Pre‐processing Method Result ConclusionInterest for forest managers

Precise localization of eroded areas within reserves

Support managers in prioritization of

reforestation strategies


Context Objective Data Pre‐processing Method Result ConclusionInterest for forest managers

Support managers in prioritization of forest restoration strategies

(cutting fruit tree, promoting local(cutting fruit tree, promoting local species)

Precise localization of forested areas potentially degraded (due to agriculturaldegraded (due to agricultural

activity in this example)



Conclusion

=> OBIA is a suitable framework for exploitingmultisource information in segmentationmultisource information, in segmentationprocess as well as in the classification process

=> LiDAR data was found particularly favorable for characterizingf l d h fforest types in a tropical context, due to the information itprovides on canopy height and its heterogeneity



Outlook=> discriminating forest types according to their composition> discriminating forest types according to their composition (deciduous, evergreen or mixed)

we have attempted in this study with these data but …we have attempted in this study with these data but …

Spot 5 satellite image was not acquiredh i bl d (J h) f

A hi h di t i h t it

at the suitable date (June month) fordiscriminating deciduous to evergreen

A high radiometric heterogeneitybetween the numerous orthophotos

A High presence of shadows onA High presence of shadows onorthophotos

Exploring more radiometric VHSR images acquired at suitable period


p g g q pand acquisition conditions : quickbird or pleiades images for example

Context Objective Data Pre‐processing Method Result ConclusionGEOBIA 2012GEOBIA 2012

Ri d J i B ilRi d J i B ilRio de Janeiro BrazilRio de Janeiro Brazil

OBIA for combining OBIA for combining LiDARLiDAR and multispectral data to and multispectral data to characterize forested areas and land cover in tropical region characterize forested areas and land cover in tropical region

StéphaneStéphane [email protected]@[email protected]@cirad.fr

GérardGérard LainéLainéThierry TormosThierry Tormos

Thanks for your attention!Thanks for your attention!

obia for combining lidar and multispectral data to ... · pdf fileivth international...

Documents