Land Surface Characterization Monitoring land cover and land cover change

(1)

Canadian Centre for Remote Sensing

Land Surface Characterization

Monitoring land cover and land cover change

Workshop of ABCC Program Ottawa, Canada

23 - 24 September 2009

Canadian Centre for Remote Sensing

Rasim Latifovic, Darren Pouliot and Ian Olthof

(2)

Regional scale land cover characterization

EO based methodology for producing highly consistent annually updated land cover time series at regional scale to support environmental status and trends assessments and ecosystem process modeling.

Improved EO techniques for monitoring land cover change provide consistent and accurate multi year land cover information.

Better understanding of cause – effect relation /land cover change – global change

ABCC Project Title ABCC Project Title

Expected Results and Objectives Expected Results and Objectives

Outcome(s) and Potential Impact Outcome(s) and Potential Impact

- data collection and processing;

- methodology development;

- land cover change mapping;

- analysis and synthesis.

Canada coverage

1999-2010 data record from MODIS, Landsat-TM, and other higher resolution images.

Timeframe for Activities & Stage(s) Completion Date(s)

Study area Study area

Data Collection Plan & Timetable Data Collection Plan & Timetable

(3)

Data collection and processing Data collection and processing

• Long Term Satellite Data Record

• Medium resolution coverage

• Cal\Val ground truth data

(4)

AVHRR NOAA 1km Record: 1982-2009 Transition: MetOp 2006-2020

-

---- -

MODIS NASA 0.25 km Period: 2000-2009 Transition: NPOESS,VIIRS

Long Term Satellite Data Record Long Term Satellite Data Record

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

400 900 1400 1900 2400

Wave length (nm)

Leafe reflectance AVHRR VIS/NIR/SWIR

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

400 900 1400 1900 2400

Wave length (nm)

Leafe reflectance MODIS Land bands

(5)

SPOT/VGT 1 km Period: 1998-2009

Transition: PROBA, SENTINEL 3 2012

MERIS ESA 1,0.3km Period: 2008-2009

Transition: SENTINEL 3 2012

Long Term Satellite Data Record Long Term Satellite Data Record Long Term Satellite Data Record Long Term Satellite Data Record

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

400 600 800 1000 1200 1400 1600 1800 2000 Wave length (nm)

Leafe reflectance MERIS bands

(6)

17000 17500 18000 18500 19000 19500

1980 1985 1990 1995 2000 2005 2010 2015 Year

(NDVI+1)*10000

AVHRR 1km MODIS 250m SPOT VGT 1 km

Cross calibration for multi sensors applications Cross calibration for multi sensors applications Cross calibration for multi sensors applications Cross calibration for multi sensors applications

AVHRR 1km MODIS 250m SPOT VGT 1 km

17000 17500 18000 18500 19000 19500

1980 1985 1990 1995 2000 2005 2010 2015

Year

(NDVI+1)*10000

AVHRR Calibrated to VGT 1km AVHRR uncalibrated 1 km

AVHRR/2 NOAA-06 - 14

AVHRR/3 NOAA16-18

(7)

E PoS length Int T

Medium resolution coverage Medium resolution coverage Medium resolution coverage Medium resolution coverage

S PoS

0501 0511 0521 0601 0611 0621 0701 0711 0721 0801

E PoS

0721 0801 0811 0821 0901 0911 0921 1001 1011 1021 length

30 40 50 60 70 80 90 100 110 120 130 140 150 160 170

• 2005-2010 SPOT 4-5 ortho data from CTI GeoBase

• Determine annual length of Peak of Season (PoS) from AVHRR for 2005-2009

– Lowess smoothed 10-day NDVI – 95% threshold of max NDVI

• Filter SPOT scenes to include only PoS

• Normalize to long-term median composite from 1 km VGT

(8)

1383 normalized SPOT 4-5 scenes, each acquired within peak-of-season window for its corresponding location in Canada, are included in the current database.

This sample represents approximately 42% of Canada’s landmass (excluding water).

Medium resolution coverage Medium resolution coverage Medium resolution coverage Medium resolution coverage

1km SPOT VGT 9-year median R, NIR, SWIR

Selected SPOT 20m scene Normalized SPOT 20m scene

(9)

First Acquisition – Aug 19-22, 2010 Planned Acquisition –Province of Quebec

ScanSAR Wide A Coverage in Dual Polarization ScanSAR Wide A Coverage in Dual Polarization ScanSAR Wide A Coverage in Dual Polarization ScanSAR Wide A Coverage in Dual Polarization

PJ Farris-Manning

(10)

Ground truth data Ground truth data

(11)

Land cover characterization:

methodology development methodology development

• Land cover time series

• Abrupt land cover change (annual change)

• Gradual land cover change (multi year change)

(12)

Land cover time1

A imagem não pode ser exibida. Talv ez o computador não tenha memória suficiente para abrir a imagem ou talv ez ela esteja corrompida. Reinicie o computador e abra o arquiv o nov amente. Se ainda assim aparecer o x v ermelho, poderá ser necessário excluir a imagem e inseri-la nov amente.

Change detection abrupt and long term

Map update Land cover time 2 Low spatial and high temporal resolution (yearly update)

Integrated earth observation data sources Land Cover Monitoring SystemLand Cover Monitoring SystemLand Cover Monitoring SystemLand Cover Monitoring System

Land cover time1

Change detection abrupt and long term

Map update Land cover time 2 High spatial and low temporal resolution (5 to 10 year update)

Calibration\Validation\Sampling

Integrated earth observation data sources

(13)

Land cover time series trough land cover update Land cover time series trough land cover update Land cover time series trough land cover update Land cover time series trough land cover update

198519851985 19901990 19951995 20002000 1985 19901990 19951995 20002000

Land CoverImage Data

Change Area

Land cover

map to be updated AVHRR Composite data (Red, NIR, NDVI) for year to be updated

Inter-map consistency = 93 %

Land Cover

(14)

Land Cover Change Land Cover Change Land Cover Change Land Cover Change

Abrupt (1-5 years) Progressive\Gradual (> 5 years)

• Fire

• Biomass harvesting

(agriculture, forest, other)

• Vegetation composition change

• Re-vegetation of disturbed lands

• Pollution related forest decline

• Storm damage (e.g. severe winds\ice)

• Permanent flooding

• Land slides

• Drought

• Urban development

• Pollution related forest decline

• Continuous low to moderate insect defoliation

• Consistent long term hydrologic changes

• Permafrost changes –active layer increase\decrease, subsidence

• Coastal\shoreline erosion

(15)

Change Detection Methodology Change Detection Methodology Change Detection Methodology Change Detection Methodology

2) Change features:

Feature₁,Feature_2,,Feature₃, ……Feature₁₆₇

3) Bootstrap feature selection 1) Radiometric normalization

4) Decision tree change classification – annual change results

5) Change persistence evaluation

Pouliot, D., R. Latifovic, R. Fernandes, and I. Olthof. (2009). Evaluation of annual forest disturbance monitoring using a static decision tree approach and 250 m MODIS data. RSE, 113:1749-1759.

Disturbances

(16)

Fraction of forest change within an 8 km pixel from 2001-2006

Abrupt land cover change (annual change) Abrupt land cover change (annual change) Abrupt land cover change (annual change) Abrupt land cover change (annual change)

Percent forest change within Landsat footprint 2000-2006

(17)

MODIS 2001 MODIS 2006 Change

Decision Tree Change Detection:

Decision Tree Change Detection: Mountain pine Mountain pine beetle damage

beetle damage

Decision Tree Change Detection:

Decision Tree Change Detection: Mountain pine Mountain pine beetle damage

beetle damage

Photo by Lorraine Maclauchlan, Ministry of Forests, Southern Interior Forest Region, Canada

2006 2005 2004 2003 2002

Disturbance year

(18)

• A system based on moderate to coarse resolution high temporal frequency EO data for monitoring seasonality events and features of snow, ice, and vegetation completing existing in situ networks.

DVD – Duration of vegetation dormancy

RLG – Rate of leaf growth

DPG – Duration of peak growing season

DGS – Duration of the growing season 0.7

0.9 _DPG

RLG PGS RSLRLS

Seasonal change Seasonal change Seasonal change Seasonal change

DGS – Duration of the growing season RLS – Rate of leaf senescence

LGS – Leaf growth start LGE – Leaf growth end

PGS – Peak of growing season LSS - Leaf senescence start LSE - Leaf senescence end SMS – Snow\ice melt start SME – Snow\ice melt end

SAS – Snow\ice accumulation start SC – Snow covered

-0.5 -0.3 -0.1 0.1 0.3 0.5

0 50 100 150 200 250 300 350 400

Day of year

AVHRR NDVI ^RSL^RLG

DVD

LGE PGS

LGS

LSS

LSE

SMS

SME SAS SC

DVD

DGS

Courtesy of http://pheno-eye.org RLS

(19)

1. Time series is divided into seasons based on

1. Season extraction 2. Initial parameter estimation

Scaled NDVI Scaled NDVI

Julian Day Julian Day

1. Time series is divided into seasons based on the least squares polynomial and harmonic basis fit of Jönsson and Eklundh (2004) Computers and Geosciences, 30:833-845.

2. Initial Lowess filter is applied to estimate logistic function parameters.

3. Double logistic function is fit:

4. Weights are adjusted through an iterative process to adjust fit to upper envelope.

(

¹ ⁽ ¹ ⁽ ¹⁾

) (

¹ ² ⁽⁾ ²⁾

)

( _mn _s _t _i^mx _s^mn_t _i

e e

NDVI NDVI NDVI

t

NDVI ₋ _× ₋ _× ₋

+ + +

+ −

=

3&4 logistic fit & adjustment

Scaled NDVI

Julian Day

(20)

12000 12500 13000 13500 14000 14500 15000 15500 16000 16500 17000

1980 1985 1990 1995 2000 2005 2010 2015

Year

(NDVI+1)*10000

Gradual land cover change (multi year change) Gradual land cover change (multi year change) Gradual land cover change (multi year change) Gradual land cover change (multi year change)

Sturm, M., C. Racine, and K. Tape (2001). Increasing shrub abundance in the Arctic. Nature, 411, 546-547.6 AVHRR Calibrated to VGT 1km MODIS 250m SPOT VGT 1 km

Pouliot, D., R. Latifovic, and I. Olthof. (2008).Trends in vegetation NDVI from 1 km AVHRR data over Canada for the period 1985-2006. International Journal of Remote Sensing, in press.

(21)

0%

5%

10%

15%

20%

25%

30%

35%

Ice cover [%]

Trend analysis sea ice extent in Hudson and James Bay Trend analysis sea ice extent in Hudson and James Bay Trend analysis sea ice extent in Hudson and James Bay Trend analysis sea ice extent in Hudson and James Bay

1985 1993

1994

1986 1987 1988 1989 1990

1995 1996 1997 1998 1999 2000

2003 2004 2005 2006 2007 2008

2001 2002

1991 1992

2009 0%

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Time [year]

Latifovic, R. et .al. (2005) Generating historical AVHRR 1 km baseline satellite data records over Canada suitable for climate change studies. Canadian Journal of Remote Sensing, 31:324- 621 348.

(22)

– Maintain and improve Long Term Satellite Data Record – Completion of the medium resolution (20m) coverage of

Canada

– Reference database generation for training and validation – Further development and improvements on land cover

change detection and updating methodologies

– Testing methods for land cover change monitoring (various

Future Development Future Development Future Development Future Development

– Testing methods for land cover change monitoring (various options of combining coarse and fine resolution data)

– Development or related Canada-scale products at 0.25 km – Enhance collaborative efforts within ABCC involving

exchange of data and methods between partners.