Presentation about the research project “Detecting forestdegradation with multi-spectral, seasonal and

multi-temporal classification trees in North CentralVietnam”

Belinda Freiheit

Student Research ColloquiumUniversity of Eberswalde

Research project conducted at the Geomatics Lab of the Geographical Institute of the HumboldtUniversity in Berlin, Supervisor: Dr. Dirk Pflugmacher

25. April 2013

1 / 11

Outline

1 I-REDD+ Project

2 Project area

3 Research topic

4 Methodology

5 Results

6 Discussion

7 Outlook

2 / 11

I-REDD+ Project

Impacts of Reducing Emissions fromDeforestation and Forest Degradation andEnhancing Carbon Stocks (I-REDD+)

I-REDD+ addresses issues at national tolocal level in four countries: Laos, Vietnam,China (Yunnan) and Indonesia

An interdisciplinary team with researchersand professionals from 9 countries worktogether

WP3 - Remote sensing based monitoring ofchange in land use and biomass

Figure: Paddy rice fields (a typical land use in the uplands ofVietnam; Source: HU Berlin 2013

3 / 11

Project area

Con Cuong is a district ofthe Nghe An Provincelocated in the NorthCentral Vietnam

Characteristics of the area:Small-scale agricultural andagro-forestry land usepractice, mountainous

Figure: Map of the Con Cuong district with the GPS reference plots

4 / 11

Research Topic

Detect forest cover changesin the project area withmeans of temporal Landsatscenes (1989 - 2012)

Data sources: 133multi-temporal Landsat images(TM 4/5 and ETM+), GPSpoints from field trip in2011/2012Challenge: frequent cloud cover,inaccurate reference data, spatialresolution, complex land cover!

104,478836

104,478836

104,523802

104,523802

104,568768

104,568768

104,613734

104,613734

104,658700

104,658700

104,703667

104,703667

104,748633

104,748633

104,793599

104,793599

104,838565

104,838565

104,883531

104,883531

104,928497

104,928497

104,973463

104,973463

105,018429

105,018429

105,063395

105,063395 18,75

4661

18,79

9627

18,79

9627

18,84

4593

18,84

4593

18,88

9559

18,88

9559

18,93

4525

18,93

4525

18,97

9491

18,97

9491

19,02

4457

19,02

4457

19,06

9423

19,06

9423

19,11

4389

19,11

4389

19,15

9355

19,15

9355

19,20

4322

19,20

4322

19,24

9288

19,24

9288

19,29

4254

19,29

4254

19,33

9220

19,33

9220 Nghe An1:350.000

0 6 123Kilometer

ConCuong NgheAnStudy sites

127047 2009-02-14 Landsat TMRGB

Red: ETM_Meta_Band_5Green: ETM_Meta_Band_4Blue: ETM_Meta_Band_3

Diem

Moi

Figure: Landsat scene of the Con Cuong district in 2009 (dry season); Source: HU Berlin 2013

5 / 11

Methodology

6 / 11

Results I

Visual interpretation of the 9 x 9 pixel image extracts of the multi-temporalTC Angle Landsat images

Explore which reflectance values represent cleared land (comparing threshold)

Model-based cluster analysis to discriminate between non-forested andforested areas (threshold: 0.278 degree)

Forest resources account for 419 plots out of 748 plots with vegetation cover

Figure: Extract of four image chips out of 133 in total for plot no. 8 for the years 1989, 1996, 2008 and 2009 (from left to right)

7 / 11

Results II

The four classes comprising natural vegetation (according to the LULCclassification):Closed canopy forests TEBC ( more than 60% canopy cover),Open forests TEBO (less or equal 60% canopy cover),Bamboo trees OWB (naturally regenerated),Bushes SEBO (mixed with small to medium trees)

Figure: Box and whisker plot of the TC Angle value of the four vegetation classes

8 / 11

Results III

RF 1: with multi-spectral data from the dry season of 2012RF 2: with multi-spectral data from the rainy season of 2011RF 3: with multi-spectral data of both seasonsRF 4: with multi-temporal, multi-spectral data for the rainy seasonRF 5: with multi-temporal, multi-spectral data with temporal information (slope,intercept) for the rainy season

Figure: Overview of the results from the RF-CART with the vegetation classes as dependent variable

9 / 11

Discussion

1 Limitations when working with optical sensors in tropical climate regions

2 Differing illumination and view angles (topographic effects)

3 Difficulty of distinguishing marginal variances of reflectance values within therespective vegetation classes

4 Field data exhibit inadequate land cover classes

10 / 11

Outlook

Enlarge the study area to the Nghe An Province (eventually includingbordering areas in Laos)

Literature review about accomplished research and projects in the region

Focus on a thematic question: land cover changes and its impact on naturalforests in the context of REDD+

Study on shifting cultivation cycles to assess proper classification

Review appropriate methods in RS: vector change analysis, post-classificationetc.

11 / 11