david roy - the global modis burned area product

THE GLOBAL

MODIS BURNED AREA

PRODUCT

David Roy*, Luigi Boschetti#, Chris Justice#

Roy, Boschetti, Justice

http://modis-fire.umd.edu/MCD45A1.asp

* Geographic Information Science Center of Excellence, South Dakota State University, USA

# Department of Geography, University of Maryland, College Park, USA

Darwin, Australia

29 September- 3 October 2008

• Funded as part of NASA MODIS Fire Science Team (Justice et al.) to complement the well established (Collection 1, 3, 5) MODIS 1km active fire product

• Developed for regional-global applications– Green house gas & aerosol emissions estimation – Ecological & LCLUC research (e.g., Fire – Climate – People)

Global MODIS Burned Area Product



– Ecological & LCLUC research (e.g., Fire – Climate – People)– Applied applications (e.g., natural resource management)

• Product prototyping by regional algorithm development during the MODIS Collection 4 era

• Collection 5 systematic global processing at NASA for year 2000 to present

Global Burned Area

Product

Algorithm

• Rolling BRDF based expectation change detection

• Semi-Physically based; less dependent upon imprecise but noise

tolerant classification techniques; very few thresholds

• Automated, without training data or human intervention



• Automated, without training data or human intervention

• Applied independently per pixel to daily gridded MODIS 500m

land surface reflectance time series

=> globally map 500m location and approximate day of burning

Global Burned Area

Product

Conceptual Scheme

ρρρρ

observed

ρ ρ ρ ρ (t|t-1)

>

predictedBRDF Inversion window



time

t-1

ρ ρ ρ ρ (t|t-1)

Global Burned Area

Product

Conceptual Scheme

ρρρρ

observed

ρ ρ ρ ρ (t|t-1)

>




time

t-1

ρ ρ ρ ρ (t|t-1)

ρ ρ ρ ρ (t|t-1)

Global Burned Area

Product

Conceptual Scheme

ρρρρ

observed




time

t

ρ ρ ρ ρ (t+1|t)

>

ρ ρ ρ ρ (t+1|t)

Global Burned Area

Product



Animation: 5 Months of burning, Okavango Delta, Botswana, 2002. Produced using

multitemporal rolling BRDF-based changedetection approach, Roy et al. 2005

Global Burned Area

Product

Median day of burning (2000-2008)



• Burned Area algorithm run globally for first time in MODIS C5 -purposefully running to map burned areas conservatively

• Product QA and Validation underway to incrementally improve product and reduce the impact of known issues

MODIS Burned Area ProductGlobal Evaluation



Inter-Comparison with

MODIS 1km Active Fire Detections

Global

Evaluation

Pro

port

ion A

ctive F

ire

15

104

236

>334

0% <= Tree Cover <= 10%

61,523

40x40km cells

1



Proportion Burned Area

Pro

port

ion A

ctive F

ire

7

Scatter plot of proportion of active fire against proportion burned area detected in 40km x 40km cells. Robust (Theil Sen) regression line is plotted in red; the white blue logarithmic color scale represents the point density distribution of the scatter plot.

0

Slope = 0.173

Intercept = 0.002

r = 0.71

0 1 Global

2001-2002

Global

Evaluation

5

15

24

>29

10% < Tree Cover <= 30%

417

40x40km cellsP

roport

ion A

ctive F

ire

1



3

0

Slope = 0.359

Intercept = 0.006

r = 0.87

Pro

port

ion A

ctive F

ire

Proportion Burned Area0 1 Global

2001-2002

Global

Evaluation

13

78

168

>233

30% < Tree Cover <= 60%

2,197

40x40km cellsP

roport

ion A

ctive F

ire

1



6

0

Slope = 0.562

Intercept = 0.006

r = 0.65

Pro

port

ion A

ctive F

ire


2001-2002

Global

Evaluation

28

296

811

>1242

60% < Tree Cover <= 100%

1,403

40x40km cellsP

roport

ion A

ctive F

ire

1



10

0

Slope = 2.276

Intercept = 0.002

r = 0.67

Pro

port

ion A

ctive F

ire


2001-2002

Global

Evaluation

17

120

282

>403

0 < LAI <= 1

49,798

40x40km cellsP

roport

ion A

ctive F

ire

1



7

0

Slope = 0.225

Intercept = 0.003

r = 0.67

Pro

port

ion A

ctive F

ire


2001-2002

Global

Evaluation

10

54

110

>148

1 < LAI <= 2.5

15,762

40x40km cellsP

roport

ion A

ctive F

ire

1



5

0

Slope = 0.351

Intercept = 0.006

r = 0.80

Pro

port

ion A

ctive F

ire


2001-2002

Global

Evaluation

8

34

64

>83

2.5 < LAI <= 5

1,224

40x40km cells P

roport

ion A

ctive F

ire

1



4

0

Slope = 0.371

Intercept = 0.005

r = 0.79

Pro

port

ion A

ctive F

ire


2001-2002

Global

Evaluation

38

481

1438

>2283

LAI > 5

2,855

40x40km cellsP

roport

ion A

ctive F

ire



13

0

Slope = 1.260

Intercept = 0.003

r = 0.38

Pro

port

ion A

ctive F

ire

Proportion Burned Area Global

2001-2002

Global

Evaluation

Australia500m burned areas1 month 2002

This pattern clearly evident …



Global

Evaluation

Australia1km active fires1 month 2002

This pattern clearly evident …



Global

Evaluation

Brazil,



Brazil, Southern Para,500m burned areas1 month 2002

Global

Evaluation

Brazil,



Brazil,Southern Para,1km active fires1 month 2002

Global

Evaluation

Pattern confirmed by Global Land Cover AnalysisVegetation Class

(MODIS land cover

product)

Burned Area(MCD45)

[km2]

Active fires(MOD14)

[km2]

Area

[km2]

Avg. Unmapped

Area MCD45

[%]

Avg. Unmapped

AreaMOD14

[%]

Evergreen needle leaf

forest

1.18E+04 2.54E+04 5.67E+06 73.81% 65.72%

Evergreen broadleaf

forest

4.84E+04 1.61E+05 1.46E+07 70.00% 46.04%

Deciduous needle leaf

forest

2.50E+03 8.82E+03 9.59E+05 76.79% 30.22%

Deciduous broadleaf

forest

8.87E+04 7.86E+04 2.33E+06 34.41% 57.68%

Mixed forests 4.87E+04 4.87E+04 6.82E+06 63.13% 24.26%



Global

Evaluation


Closed shrublands 3.67E+04 2.14E+04 8.00E+05 26.43% 38.46%

Open shrublands 4.19E+05 3.27E+05 2.66E+07 39.97% 17.31%

Woody savannas 9.76E+05 7.07E+05 1.10E+07 46.49% 35.30%

Savannas 1.37E+06 8.10E+05 1.02E+07 32.30% 30.24%

Grasslands 3.46E+05 2.03E+05 1.35E+7 45.41% 20.10%

Croplands 2.76E+05 3.43E+05 1.61E+7 44.20% 26.02%

Barren or sparsely

vegetated

3.71E+04 2.58E+04 2.21E+7 15.51% 23.86%

Other 7.60E+03 2.68E+04 3.63E+06 67.46% 56.81

TOTAL 3.66E+06 2.78E+06 1.31E+8 42.96% 29.82%

Pattern confirmed by Global Land Cover AnalysisVegetation Class

(MODIS land cover

product)

Burned Area(MCD45)

[km2]

Active fires(MOD14)

[km2]

Area

[km2]

Avg. Unmapped

Area MCD45

[%]

Avg. Unmapped

AreaMOD14

[%]

Evergreen needle leaf

forest

1.18E+04 2.54E+04 5.67E+06 73.81% 65.72%

Evergreen broadleaf

forest

4.84E+04 1.61E+05 1.46E+07 70.00% 46.04%

Deciduous needle leaf

forest

2.50E+03 8.82E+03 9.59E+05 76.79% 30.22%

Deciduous broadleaf

forest

8.87E+04 7.86E+04 2.33E+06 34.41% 57.68%




Global

Evaluation


Closed shrub lands 3.67E+04 2.14E+04 8.00E+05 26.43% 38.46%

Open shrub lands 4.19E+05 3.27E+05 2.66E+07 39.97% 17.31%

Woody savannas 9.76E+05 7.07E+05 1.10E+07 46.49% 35.30%

Savannas 1.37E+06 8.10E+05 1.02E+07 32.30% 30.24%

Grasslands 3.46E+05 2.03E+05 1.35E+7 45.41% 20.10%

Croplands 2.76E+05 3.43E+05 1.61E+7 44.20% 26.02%

Barren or sparsely

vegetated

3.71E+04 2.58E+04 2.21E+7 15.51% 23.86%

Other 7.60E+03 2.68E+04 3.63E+06 67.46% 56.81

TOTAL 3.66E+06 2.78E+06 1.31E+8 42.96% 29.82%

MODIS Burned Area ProductValidation Protocol

How Accurate ?



How Accurate ?

Validation

Protocol

Product Validation being undertakenwith GOFC GOLD –fire regional networks

• To Stage 2 Validation “the product accuracy has been assessed over a widely distributed set of locations and time periods via several ground truth and validation efforts”

– emphasis on sampling a range of continentally representative conditions – including where the burned area product has apparent limitations, i.e., in

regions with high forest cover, high LAI, and in croplands.

• Southern Africa – Completed - SAFNET

• Australia – in progress (planning this meeting, Friday)



• Australia – in progress (planning this meeting, Friday)

• Europe – partnership with JRC-EC

• South America – Redlatif, INPE & Imazon collaboration

• Boreal – NERIN planning

• China & India – planning

Validation

Protocol

Image 1:

Landsat ETM+

Sept. 4th



Validation

Protocol

Image 2:

Landsat ETM+

Oct 6th



Yellow vectors = ETM+ interpreted burned areas occurring between the two ETM+ acquisitions

Validation

Protocol

MODIS 500mBurned Areas

Sept. 4to Oct. 6



White vectors = ETM+ interpreted burned areas occurring between the two ETM+ acquisitions

Validation

Protocol

Burned Area ProductInter-Comparison and Validation

Considering 3 Global Burned Area Products:



Considering 3 Global Burned Area Products:

L3JRC – SPOT VEGETATION 1km

GLOBCARBON – SPOT VEGETATION and ATSR 1km

MODIS – MODIS 500m

Inter-comparison

and Validation

Landsat ETM+ validation scenes distributed from dry savanna to wet miombo woodland to quantify product accuracy over range of representative biomass burning conditions

ETM+ scene

~185 * 185 km

Each ETM+ scene has a local SAFNet collaborator

11 scenes = ~3% of southern



11 scenes = ~3% of southern African surface

Inter-comparison

and Validation

L3JRC – A/S/O 2001



Inter-comparison

and Validation

GlobCarbon –A/S/O 2001



Inter-comparison

and Validation

MODIS – A/S/O 2001



Inter-comparison

and Validation

All L3JRC Data (294148 km2)

>143

L3JR

C P

roport

ion B

urn

ed

10

53

106

0.6

0.8

1.0

Africa 20015km x 5km

grid cells



Landsat ETM+ proportion burned

L3JR

C P

roport

ion B

urn

ed

5

00.2 0.4 0.6 0.8 1.0

0.2

0.4

0.0

Slope = 0.136

Intercept = 0.136

r2 = 0.128

Inter-comparison

and Validation

All Globcarbon Data (294148 km2)

Glo

bcarb

on P

roport

ion B

urn

ed

9

38

72

>95

0.6

0.8

1.0


grid cells




Glo

bcarb

on P

roport

ion B

urn

ed

4

0

0.2

0.4

0.0 0.2 0.4 0.6 0.8 1.0

Slope = 0.509

Intercept = 0.013

r2 = 0.509

Inter-comparison

and Validation

All MODIS C5 Data (294148 km2)

MO

DIS

Pro

port

ion B

urn

ed

5

16

27

>33

0.6

0.8

1.0


grid cells




MO

DIS

Pro

port

ion B

urn

ed

3

0

0.2

0.4

0.0 0.2 0.4 0.6 0.8 1.0

Slope = 0.75

Intercept = -0.005

r2 = 0.746

Inter-comparison

and Validation

Summary – Global MODIS Burned Area Product

• This is a new product – processed systematically for 2000 to present

• User manual & information on how to access, are posted on the MODIS Fire website

• Users need to understand accuracy and utility of burned area products. • Given the proliferation of products available we are advocating that a validation

protocol be endorsed as an international standard, with open sharing of validation data

• Southern African validation indicates that– MODIS burned area product captures 75% of actual area burned



– MODIS burned area product captures 75% of actual area burned– MODIS burned area product outperforms other (L3JRC and GLOBCARBON) global

burned area products– All products considered have lower errors of commission than omission

• We intend to validate the MODIS burned area product globally– emphasis on sampling a range of continentally representative conditions – including where the burned area product has apparent limitations, i.e., in regions with

high forest cover, high LAI, and in croplands– This Friday - OZ validation planning

• Changes to the Collection 5 algorithm are planned– Change descriptions will be posted on the MODIS Fire website

Final Slide

Thanks



C5 monthly

burned

area

(MCD45)

product



Greece

August

2007

Example

refinement

C5 monthly

burned

area

(MCD45)



(MCD45)

product

Greece

August

2007

Scheme of the detection

Day

BRDF inversion window Search duration (daily Z- score calculations)

S search

S test n n extra

NpassNused



Algorithm

Background

Day

first burn candidate

Contextual iterative procedure:

Burn candidates are selected if they provide persistent evidence of fire occurrence.

The measured persistence varies depending on gaps in the reflectance

time series and the timing of the fire relative to non-missing data, an iterative rather than simple thresholding approach is used

Validation Plan

• Currently CEOS stage 1 validation– “the product accuracy has been estimated using a small number of independent

measurements obtained from selected locations and time periods and ground-truth/field program efforts”

• Stage 2 by the end of Collection 5



– “the product accuracy has been assessed over a widely distributed set of locations and time periods via several groundtruth and validation efforts”

• Stage 3 in Collection 6– “product accuracy has been assessed and the uncertainties in the product well

established via independent measurements in a systematic and statistically robust way”

Validation

david roy - the global modis burned area product

Documents