21
Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape based on climate, fuels, ignition and topography. Fire detection models can serve as an independent validation source for fire potential models, particularly in under-developed regions. Fire potential and fire detection models both depend on MODIS data.

Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Embed Size (px)

Citation preview

Page 1: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Project Idea• Fire potential models can help stratify and

reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape based on climate, fuels, ignition and topography.

• Fire detection models can serve as an independent validation source for fire potential models, particularly in under-developed regions.

• Fire potential and fire detection models both depend on MODIS data.

Page 2: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Testing VIIRS• Existing eastern fire

potential models • Standardization issues

– AVHRR/MODIS/VIIRS

• Existing fire detection systems – SERVIR– RSAC

FRANKE, Jonas & Gunter MENZRemote Sensing Research Group (RSRG)

Department of Geography, University of Bonn

Bonn, [email protected]

Page 3: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Testing VIIRS

• Existing fire detection systems – SERVIR– RSAC

• Giglio, et al. / Remote Sensing of Environment

87 (2003) 273-282

‘Contextual Fire Detection Algorithm for MODIS’

• Absolute Threshold Test

T4 > 360 K (320 K at night)

• Brightness threshold MODIS 4um (T4) (Bands 21 and 22 [1km]) – No VIIRS exact

replacement

• Brightness threshold MODIS 11um (T11) (Band 31 [1km])– Two VIIRS bands (M15

[742m], I-5 [371m])

Page 4: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

t

0.8*

Page 5: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Monthly Rainfall Totals for July 2003

Page 6: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Weather Station LocationsEvaporation Predictions

St. Bernard

St. CharlesJefferson

MobileBaldwin

PerryPikeGreene

Jackson

Washington

Marion Lamar

StonePearl River

Harrison

Tangipahoa

Forrest

Jones

St. Tammany

Lincoln

George

Clarke

Washington

Wayne

Hancock

WalthallAmite

Lawrence

Orleans

Livingston

CovingtonJefferson Davis

St. John the Baptist

St. Helena

LaFourche Plaquemines

Franklin

Monroe

Ascension ±200 0 200100 Kilometers

Legend

stations

Page 7: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

InlandEvaporation

CoastalEvaporation

0

0

Evaporation Regression ModelsAccepted: Southeastern Geographer

Page 8: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Stoneville, MS Cumulative P-E (average over 40 years) and 2000 estimates

Cumulative summaries - Starting date January 1st each year

-20

-15

-10

-5

0

5

10

15

20

1 13 25 37 49 61 73 85 97 109

121

133

145

157

169

181

193

205

217

229

241

253

265

277

289

301

313

325

337

349

361P-E

(in

)

2000 P_E 40 yrs Avg. P-E

January December

Precipitation – Evaporation (P-Et) Cumulative Inland

Page 9: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Fairhope, AL Cumulative P-E (40 years average) and 1995 estimates

Cumulative summaries - starting date January 1st each year

-5

0

5

10

15

20

25

30

1

17

33

49

65

81

97

11

3

12

9

14

5

16

1

17

7

19

3

20

9

22

5

24

1

25

7

27

3

28

9

30

5

32

1

33

7

35

3

Jan

P-E

(in

)

1995 P-E 40 yrs. Avg. P-E

January December

Precipitation – Evaporation (P-Et) Cumulative Coastal

Page 10: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Road Density/Gravity and Fire Ignition

Very LowLowMedHighVery High

Fire Risk

Page 11: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Road Density and Fire Ignition

Page 12: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Gravity vs. Road DensityGravity and Road

DensityAnnual

CriticalAnnual p-value

Winter

Critical

Winter

p-value

Summer

CriticalSummer p-value

Very Low Gravity and Very Low Road

Density3.51* 0.0085 3.64* 0.0058 3.58* 0.007

Low Gravity and Low Road Density 1.6 0.11 1.56 0.13 2.0* 0.05

Medium Gravity and Medium Road

Density3.09* 0.003 2.82* 0.0064 2.78* 0.007

High Gravity and High Road Density 0.62 0.534 0.67 0.5 0.29 0.77

Very High Gravity and Very High Road

Density0.44 0.664 0.08 0.58 0.42 0.68

Conclusions:Gravity models yield improved estimates of risk at very low levelsRoad density yields improved estimates of risk at medium levels

Page 13: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

18-year Historic AVHRR NDVI 7-day Composites

Departure from average greeness

Page 14: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Physiographic Region Pearson Correlation (NDVI and Average Acre Burned)

0.01 level 0.05 level

Black Prairie -0.842 √

Coastal Zone -0.525

Delta -0.257

Loess Hills -0.817 √

North Central Hills -0.709 √

Pine Belt -0.696 √

South Central Hills -0.581 √

Jackson Prairie -0.534

Tombigbee Hills -0.533

NDVI and fire data averaged by month for each physiographic region

N = 12

Correlation Results – NDVI and Average Acre Burned

Page 15: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Physiographic Region Pearson Correlation (NDVI and Average Acre Burned)

0.01 level

Black Prairie -.390 √

Coastal Zone -.006

Delta -.119

Loess Hills -.293 √

North Central Hills -.383 √

Pine Belt -.212 √

South Central Hills -288 √

Jackson Prairie -.129

Tombigbee Hills -.090

Correlation Results – NDVI and Average Acre Burned

NDVI and fire data averaged by year and month for each physiographic region N = 177

Page 16: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

NDVI Departure from Average

Page 17: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

June 1Terra

June 2Aqua

June 3Terra

June 4Aqua

June 5Aqua

June 6Aqua

June 7Aqua

June 8Terra

Page 18: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

VIIRS Simulation

• ITD and Chuck O’hara

• Florida and Georgia 2007 for tests

• Methods transferrable to Central America?

Page 19: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

Comparison of MODIS & VIIRS BandsBand # Band ID Band # Band ID

1 620 - 670 600 - 680 I-1 3.610 Ğ 3.790 M-122 841 - 876 845 - 885 I-2 3.550 Ğ 3.930 I-43 459 - 479 21 3.929 - 3.9894 545 - 565 22 3.940 Ğ 4.0015 1230 - 1250 1230 - 1250 M-8 23 4.020 - 4.080 3.973 Ğ 4.128 M-13

1580 - 1670 M-10 24 4.433 Ğ 4.4981580 - 1610 I-3 25 4.482 Ğ 4.549

7 2105 - 2155 2225 Ğ 2275 M-11 26 1.360 - 1.390 M-98 405 - 420 402-422 M-1 27 6.535 - 6.8959 438 - 448 436-454 M-2 28 7.175 - 7.475

10 483 - 493 478-498 M-3 29 8.400 - 8.700 8.400 Ğ 8.700 M-1411 526 - 536 30 9.580 - 9.88012 546 - 556 545-565 M-4 10.263 Ğ 11.263 M-1513 662 - 672 662-682 M-5 10.050 - 12.400 I-514 673 - 683 32 11.770 - 12.270 11.538 Ğ 12.488 M-1615 743 - 753 739-754 M-6 33 13.185 - 13.48516 862 - 877 846-885 M-7 34 13.485 - 13.78517 890 - 920 35 13.785 - 14.08518 931 - 941 36 14.085 - 14.38519 915 - 965

MODIS Bands 1& 2 are 250 m at nadirMODIS Bands 3-7 are 500 m at nadirMODIS Bands 8-36 are 1,000 m at nadir

VIIRS Bands I-1 & I-2 are 371 m at nadirVIIRS Band I-3 is 371 m at nadir

VIIRS Bands I-4 & I-5 are 371 m at nadir

MODIS VIIRS

6 1628 - 1652

MODIS VIIRS

10.780 - 11.28031

20 3.660 - 3.840

Page 20: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

VIIRS Vis/NIR BandsFire detection, spatial resolution

Band Name

SNR/ NEDT

Ltyp Lmin LmaxSNR/

NEDTLtyp Lmin Lmax

GSD Nadir

(m)

M-1 412 nm 20 nm 352 44.9 30 135 316 155 135 615 742

M-2 445 nm 18 nm 380 40 26 127 409 146 127 687 742

M-3 488 nm 20 nm 416 32 22 107 414 123 107 702 742

M-4 555 nm 20 nm 362 21 12 78 315 90 78 667 742

I-1 640 nm 80 nm 119 22 5 718 371

M-5 672 nm 20 nm 242 10 8.6 59 360 68 59 651 742

M-6 746 nm 15 nm 199 9.6 5.3 41 742

M-7 865 nm 39 nm 215 6.4 3.4 29 340 33.4 29 349 742

I-2 865 nm 39 nm 150 25 10.3 349 371

DNB 700 nm 400 nm 5 3.0E-05 2.0E+02 742

Band Ctr

Band Width

Single or High Gain Low Gain

3

SNR values are as specified for un-aggregated pixel.At nadir SNR will be ~ better after aggregation. (Predicted are better still)

Page 21: Project Idea Fire potential models can help stratify and reduce the number of false positive fire ‘detections’ by assigning probability levels to the landscape

VIIRS S/MW & LW IR BandsFire detection, spatial resolution

SNR values are as specified for un-aggregated pixel.At nadir SNR will be ~ better after aggregation. (Predicted are better still)

Band Name

SNR/ NEDT

Ltyp Lmin Lmax

SNR/ NEDT

Ltyp Lmin Lmax

GSD Nadir (m)

M-8 1.24 0.020 74 5.4 3.5 164.9 742M-9 1.378 0.015 83 6 0.6 77.1 742

M-10 1.61 .06 342 7.3 1.2 71.2 742I-3 1.61 .06 6 7.3 1.2 72.5 371

M-11 2.25 .05 10 .12 0.12 31.8 742M-12 3.70 .18 .396 K 270 K 230 K 353 K 742

I-4 3.74 .38 2.5 K 270 K 210 K 353 K 371M-13 4.05 0.155 .107 K 300 K 230 K 343 K .423 K 380 K 343 K 634 K 742M-14 8.55 0.3 .091 K 270 K 190 K 336 K 742M-15 10.8 1.0 .070 K 300 K 190 K 343 K 742M-16 12.0 1.0 .072 K 300 K 190 K 340 K 742

I-5 11.5 1.9 1.5 K 210 K 190 K 340 K 371

Band CtrBand Width

Single or High Gain Low Gain

3