linking field observations with remote sensing to ... · 1 © bushfire crc ltd 2007 program a...

1

© BUSHFIRE CRC LTD 2007

PROGRAM A

LINKING FIELD OBSERVATIONS WITH REMOTE SENSING TO DETERMINE GRASSLAND FIRE HAZARD

Stuart AndersonEnsis Forest Biosecurity and Protection, Scion, Christchurch, New Zealand

Elizabeth BothaEnsis Forest Biosecurity and Protection, CSIRO, Canberra

© BUSHFIRE CRC LTD 2007Program A : Linking field observations with remote sensing to determine grassland fire hazard

Outline

1. Grassland curing2. Project overview3. Research methods4. Field data collection5. Remote sensing

2


Grassland curing

• Degree of curing – the proportion of dead material in a grassland fuel complex (%)

(Source: Garvey and Millie 2001)


Why is curing important?• Input into fire behaviour models and fire danger

rating systems in Australia and New Zealand• Grassfires significant in both countries• Prescribed fire also very important• Often poorly assessed

3


Current curing assessment

VisualPoor correlation between visuals and actual curing


Curing assessment: VisualVisual vs destructive - 2005-2007

% curing (destructive measurement)0 20 40 60 80 100

% c

urin

g (v

isua

l est

imat

e)

0

20

40

60

80

100

ACTWANZ

4


Visual• Poor correlation between visuals and actual curing

Destructive• Time-consuming• Difficult

Remote sensing• Cloudiness• Pixel size• Non-uniform fuels• Discontinuous cover/patchiness• Outdated technology

Current curing assessment


Aim:

This project will develop better methods to assess the current and predicted levels of curing in grasslands as an input into fire danger rating systems and fire behaviour models

5


Research approaches

1. Remote sensing2. Curing modelling3. Field sampling


Field data collection

• Field data is ESSENTIAL• Collected across grasslands of Australia

and New Zealand• A major task, but very important• Destructive sampling most reliable –

cost, effort, time• Alternative approach – modified Levy

Rod

6


Field Sites


Data collection: Levy rod method

• Total number of live and dead grass touches against a thin steel rod

• Record at intervals along fixed transects

• Less subjectivity

7


Field sampling: WA (06/07)

Western Australia sites (Levy vs destructive curing estimates)


% c

urin

g (L

evy

estim

ate)

0

20

40

60

80

100

y = 0.58x + 32R2 = 0.59, p < 0.0001n =39, RMSE = 10.22


Field sampling: WA (06/07)Parry Lagoons

(Levy vs destructive curing estimates)


% c

urin

g (L

evy

estim

ate)

0

20

40

60

80

100

R2 = 0.26, p = 0.1132

Silent Grove - Sandstone(Levy vs destructive curing estim ates)

% curing (destructive m easurem ent)0 20 40 60 80 100

% c

urin

g (L

evy

estim

ate)

0

20

40

60

80

100

R2 = 0.49, p = 0.1219

M t H art Sandstone (Levy vs destructive curing estim ates)

% curing (destructive m easurem ent)0 20 40 60 80 100

% c

urin

g (L

evy

estim

ate)

0

20

40

60

80

100

R2 = 0.86, p = 0.0235

Silent Grove - Black Soil (Levy vs destructive curing estimates)


% c

urin

g (L

evy

estim

ate)

0

20

40

60

80

100

R2 = 0.94, p = 0.0066

Lorna Glen(Levy vs destructive curing estimates)


% c

urin

g (L

evy

estim

ate)

0

20

40

60

80

100

R2 = 0.00, p = 0.9073

8


Field sampling: NZ (06/07)

New Zealand sites (Levy vs destructive curing estimates)


% c

urin

g (L

evy

estim

ate)

0

20

40

60

80

100

y = 0.90x + 5R2 = 0.67, p = 0.0021n = 10, RMSE = 7.69


Field Data Collection

1. Levy Rod Assessments

2. Fuel Moisture Content

3. Destructive Sampling (selected sites)

Support from end-user agencies is appreciated

9


Remote Sensing: Background

Wavelength0.5 1.0 1.5 2.0 2.5

Ref

lect

ance

0.0

0.2

0.4

0.6

0.8

1.0

green vegetationdead vegetationbare soil


Satellite data

MODIS MOD09 image producta) 7 spectral bandsb) 500x500m spatial-resolutionc) 8-day temporal resolution

Image source: http://www.ga.gov.au

Image source: http://terra.nasa.gov/Brochure/Sect_4-7.html

10


Remote Sensing: What the satellite sees

Wavelength0.5 1.0 1.5 2.0 2.5

% R

efle

ctan

ce

0.0

0.5

1.0

Wavelength0.5 1.0 1.5 2.0 2.5

% R

efle

ctan

ce

0.0

0.5

1.0

Wavelength0.5 1.0 1.5 2.0 2.5

% R

efle

ctan

ce

0.0

0.5

1.0

Wavelength0.5 1.0 1.5 2.0 2.5

% R

efle

ctan

ce

0.0

0.5

1.0


Remote Sensing

Wavelength0.5 1.0 1.5 2.0 2.5

Ref

lect

ance

0.0

0.2

0.4

0.6

0.8

1.0

green vegetationdead vegetationbare soil

MO

DIS

Ban

d 1

MO

DIS

Ban

d 2

MO

DIS

Ban

d 6

MO

DIS

Ban

d 7

1212

BBBBNDVI

+−

=

67

BBSWIRRatio =

11


Spectral unmixing

NDVI0.0 0.2 0.4 0.6 0.8 1.0

B7/

B6

0.0

0.2

0.4

0.6

0.8

1.0 Bare soil

Dead vegetation Green vegetation


NO DATA (CLOUDS)100%PV

100%NPV

100%BS

12


Parry Lagoons, WA(May 2006 - May 2007)

DateMay Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May

fraction non-photosynthetic vegetationfraction photosynthetic vegetationfraction bare soil

WA Spectral Unmixing

CLOUDS


Results: WA (Kimberley)Silent Grive (black soil), WA

estimated fraction cured vegetation-1.0 -0.5 0.0 0.5 1.0 1.5

mea

sure

d fr

actio

n cu

red

vege

tatio

n

-1.0

-0.5

0.0

0.5

1.0

1.5Silent Grove (black soil), WA

(April - November 2006)

DateApr Jul Oct

est. fraction non-photosynthetic vegetationfraction cured grassest. fraction photosynthetic vegetationfraction green grass

y = 0.5 + 0.76xR2 = 0.97, p = 0.0004RMSE = 0.0415, n = 6

Silent Grove (black soil), WA

13


Results: WA (Goldfields)

Lorna Glen, WA(June 2006 - June 2007)

DateJul Oct Jan Apr Jul

Lorna Glen


mea

sure

d fr

actio

n cu

red

vege

tatio

n

-1.0

-0.5

0.0

0.5

1.0

1.5

y = 0.62 + 0.15xR2 = 0.81, p = 0.0004RMSE = 0.0337, n = 10


Results: QldRyan's Farm


mea

sure

d fr

actio

n cu

red

vege

tatio

n

-1.0

-0.5

0.0

0.5

1.0

1.5

Ryan's Farm, Qld(April - November 2006)

DateApr Jul Oct

est. fraction non-photosynthetic vegetationfraction cured grassest. fraction photosynthetic vegetationfraction green grass

y = 0.61 + 0.5xR2 = 0.79, p = 0.0004RMSE = 0.1159, n = 10

14


Results: ACT/NSWBraidwood, NSW

(October 2006 - June 2007)

DateOct Jan Apr

Braidwood, NSW


mea

sure

d fr

actio

n cu

red

vege

tatio

n

-1.0

-0.5

0.0

0.5

1.0

1.5

y = 0.53 + 0.24xR2 = 0.33, p = 0.0396RMSE = 0.1872, n = 13


Results: ACT/NSWMajura, ACT

(January 2006 - February 2007)

DateJan Apr Jul Oct Jan

Majura, ACT


mea

sure

d fr

actio

n cu

red

vege

tatio

n

-1.0

-0.5

0.0

0.5

1.0

1.5

y = 0.64 + 0.31xR2 = 0.68, p = 0.0066RMSE = 0.0659, n = 8

15


Preliminary results

1. Spectral patterns of dead/live vegetation correlate with ground observations

2. No 1:1 relationship between unmixing results and ground data – model will have to be calibrated separately for each region

3. Best results in tropical savanna area where model was developed

4. Each region may require different end-members to optimize the spectral unmixing algorithm.

5. We need more ground observations to validate this method –especially of the grasslands in the southern part of Australia.


Acknowledgements

Wendy Anderson (UNSW@ADFA)Jennifer Hollis (DEC, WA)Ruth Gibbs (Ensis/CSIRO)Kelsy Gibos (formerly Ensis/Scion)Juan Pablo Guerschman (CSIRO Land and Water)Leigh Douglas (formerly Ensis/CSIRO)Peter Ellis (Ensis/CSIRO)Francis Hines and Adam Fountain (formerly UNSW@ADFA)End-user agencies:

Australia: Dept of Environment & Conservation (WA), Qld Fire & Rescue Service, ACT Government, private landownersNew Zealand: Dept of Conservation, National Rural Fire Authority, NZ Forest Owners Assn, Local Government NZ, NZ Defence Force, private landowners

linking field observations with remote sensing to ... · 1 © bushfire crc ltd 2007 program a...

Documents