Modeling fire occurrence as a function of landscape Loboda T (loboda@umd.edu), Carroll M, DiMiceli C - Geography Department, University of Maryland

Introduction and Background

Risk of Ignition (ROI) Potential Burning (PB)

Future Climate–induced Changes in Fire Occurrence

References and Acknowledgements: This work has been supported by the NIH-NIEHS Climate Change Grant #1RC1ES018612-01. [1] Giglio L, Descloitres J, Justice CO & Kaufman YJ, 2003. An enhanced contextual fire detection algorithm for MODIS. Remote Sensing of Environment, 87 (2-3), 273-282. [2] Loboda T, Csiszar I, 2007. Reconstruction of Fire Spread within Wildland Fire Events in Northern

Eurasia from the MODIS Active Fire Product. Global and Planetary Change, 56 (3-4): 258-273. [3] Friedl MA, McIver DK, Hodges JCF, et al. (2002). Global land cover mapping from MODIS: algorithms and early results. Remote Sensing of Environment, 83, 287-302. [4] Hansen MC, DeFries RS, Townshend JRG et al., 2003. Global percent tree cover at a spatial resolution of 500 m: first results of the MODIS

vegetation continuous fields algorithm. Earth Interactions, 7 (10), 1–15. [5] Turner JA, Lawson BD, 1978. Weather in the Canadian Forest Fire Danger Rating System. A user guide to national standards and practices. Environment Canada, Pacific Forest Research Centre, Victoria, BC. BC-X-177; [6] Van Wagner CE; Pickett TL, 1985. Equations and FORTRAN program for the Canadian Forest Fire Weather Index

System. Canadian Forest Service, Ottawa, ON. Forestry Technical Report 33. [7] Pal, J. S., et al. (2009), The ICTP RegCM3 and RegCNET: Regional Climate Modeling for the Developing World, Bull. Amer. Meteo. Soc., 88, 1395.

NH33B

-1563

Wildland fire is a prominent component of ecosystem functioning worldwide. Nearly all ecosystems experience the impact of

naturally occurring or anthropogenically driven fire. Here, we present a spatially explicit and regionally parameterized Fire

Occurrence Model (FOM) aimed at developing fire occurrence estimates at landscape and regional scales. The model provides

spatially explicit scenarios of fire occurrence based on the available records from fire management agencies, satellite

observations, and auxiliary geospatial data sets. Fire occurrence is modeled as a function of the risk of ignition, potential fire

behavior, and fire weather using internal regression tree-driven algorithms and empirically established, regionally derived

relationships between fire occurrence, fire behavior, and fire weather. The FOM presents a flexible modeling structure with a set

of internal, globally available default geospatial independent and dependent variables. However, the flexible modeling

environment adapts to ingest a variable number, resolution, and content of inputs provided by the user to supplement or replace

the default parameters to improve the model’s predictive capability. A Southern California FOM instance (SC FOM) was

developed using satellite assessments of fire activity from a suite of Landsat and Moderate Resolution Imaging

Spectroradiometer (MODIS) satellite data, Monitoring Trends in Burn Severity fire perimeters, and auxiliary geospatial information

including land use and ownership, utilities, transportation routes, and the Remote Automated Weather Station data records.

Above: A pyrocumulus

cloud over Downtown Los

Angeles (photo - Michael

Castillo)

Below: Station Fire in La

Crescenta, California (AP

Photo/Jae C. Hong)

Southern California Instance of the Fire Occurrence model (SC FOM)

Fire Weather

Julian Date

Fire detections Projections of fire detections on the respective axes

235

225

215

205

195

185

175

165

155

145

135 128.0 128.5 129.0 129.5

62.75

63.0

63.25

63.5

MODIS active fire detections

clustered by the FSR algorithm

Landsat ETM+ image path 122 row

16 from 08/19/02

130.0

MODIS)active fire detections [1] were processed through Fire Spread

Reconstruction Algorithm [2], illustrated below. This algorithm clusters

individual fire ignitions in space-time to identify contiguous fire events and

proxy ignition points. In this study we used MODIS fire detections

acquired between 2001 and 2009 to train the FOM.

Monthly fire ignition points were used to assign monthly ignition weights to

annual probability of burning, developed within the PB module, by 0.01

intervals:

𝑅𝑂𝐼𝑤𝑒𝑖𝑔𝑕𝑡 =𝑚𝑜𝑛𝑡𝑕𝑙𝑦 𝑖𝑔𝑛𝑖𝑡𝑖𝑜𝑛𝑠 𝑤𝑖𝑡𝑕𝑖𝑛 𝑃𝐵 𝑧𝑜𝑛𝑒

𝑡𝑜𝑡𝑎𝑙 𝑖𝑔𝑛𝑖𝑡𝑖𝑜𝑛𝑠

The combined MTBS and MODIS burned area record 2001-2009 was

enhanced through additional processing using the MODIS active fire

detection to model daily fire progression. MODIS active fire detections

were clustered using the Fire Spread Reconstruction algorithm and used

to build daily fire progression surfaces, illustrated below.

2007 fire scars Modeled date of burning

0 20 km

Oct 21

Oct 22

Oct 23

Oct 24

Oct 25

Oct 26

Oct 27

Oct 28

Oct 29

The Remote Automated Weather Station (RAWS) data 2001-2010 were

interpolated across the study area following the protocol developed for the

Canadian Fire Weather Index (CFWI) calculations [5, 6] to develop a 10-

year record of CWFI at the daily time step. Weather weights were

calculated following:

𝐹𝑊𝑙𝑜𝑎𝑑𝑖𝑛𝑔 = 𝑓𝑟𝑎𝑐𝑡𝑖𝑜𝑛 𝑎𝑟𝑒𝑎 𝑏𝑢𝑟𝑛𝑒𝑑 𝑤𝑖𝑡𝑕𝑖𝑛 𝑎 𝐶𝐹𝑊𝐼 𝑧𝑜𝑛𝑒 𝑓𝑟𝑜𝑚 𝑎𝑙𝑙 𝑏𝑢𝑟𝑛𝑒𝑑

𝑓𝑟𝑎𝑐𝑡𝑖𝑜𝑛 𝑜𝑓 𝐶𝐹𝑊𝐼 𝑧𝑜𝑛𝑒 𝑓𝑟𝑜𝑚 𝑡𝑕𝑒 𝑡𝑜𝑡𝑎𝑙 𝑎𝑟𝑒𝑎

Subsequently, 2 regression equations were fitted:

CFWI < 80: weight = 0.041*exp(0.0557*CFWI) (R2 = 0.94)

CFWI ≥ 80: weight = 0.000008*exp(0.1616*CFWI) (R2 = 0.98)

Monitoring Trends in Burn Severity (MTBS) burn perimeters from 1984 – 2000 and combined MTBS and

MODIS burned area products from 2001 - 2009 were used to model Potential Burning (PB) as a function of

land cover (MODIS land cover product [3]), percent tree cover (MODIS Vegetation Continuous Fields [4]), and

topography including slope, aspect and elevation obtained from the Shuttle Radar Topography Missions Digital

Elevation Model and aggregated to the MODIS nominal 500 m resolution (below left).

• Regression tree classifications were developed from annual distribution of burned areas between 1984

and 2009. These classifications partitioned the spectral space by the likelihood of burning driven by

the burned (all burned pixels from a given year) and unburned (randomly selected within study area 3

times the number of burned pixels) training samples.

• Probabilities of burning were calculated per each class generated by the decision tree for each year.

• PB = mean annual probabilities 1984-2009 (above right).

Percent tree cover

Slope (percent)

Land cover type

Elevation (m)

Aspect (degrees)

Fire Occurrence Index

Fire Occurrence Index (FOI) is calculated daily :

𝐹𝑂𝐼 = 𝑃𝐹𝐵 𝑎𝑛𝑛𝑢𝑎𝑙 ∗ 𝑅𝑂𝐼𝑤𝑒𝑖𝑔𝑕𝑡 𝑚𝑜𝑛𝑡𝑕𝑙𝑦 ∗ 𝐹𝑊𝐼𝑤𝑒𝑖𝑔𝑕𝑡 𝑑𝑎𝑖𝑙𝑦

Burns 2000-2009

Burns 1990-1999

Burns 1984-1989

Fire Occurrence Record

in southern California

FOM validation for 2010 fire season

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

200

1

200

2

200

3

200

4

200

5

200

6

200

7

200

8

200

9

201

0

201

1

201

2

201

3

201

4

201

5

201

6

201

7

201

8

201

9

202

0

202

1

202

2

202

3

202

4

202

5

202

6

202

7

202

8

202

9

203

0

203

1

203

2

203

3

203

4

203

5

203

6

203

7

203

8

203

9

204

0

'>50'

'41-50'

'31-40'

'21-30'

'11-20'

'0-10'

Future weather conditions, modeled at 25 km resolution using the Regional Climate Model (RegCM Version

4.1) [7], do not project a noticeable change in frequency of high fire weather conditions (expressed through

CFWI), compared to the 2001-2010 period (below). It is likely that the San Diego county will experience

approximately 2 extreme fire seasons each decade by 2040 (see graph below).

CFWI

MODIS fire detections July 30, 2010

Fire Occurrence Index (FOI) on July 30, 2010

The largest fire event of

2010 season = 92% of

fire detections on July

30, 2010

0.00000

0.00005

0.00010

0.00015

0.00020

0.00025

0.00030

0.0

1

0.0

6

0.1

1

0.1

6

0.2

1

0.2

6

0.3

1

0.3

6

0.4

1

0.4

6

0.5

1

0.5

6

0.6

1

0.6

6

0.7

1

0.7

6

0.8

1

0.8

6

0.9

1

0.9

6

1.0

1

1.0

6

1.1

1

1.1

6

1.2

1

1.2

6

1.3

1

1.3

6

1.4

1

1.4

6

>1.5

fire

pix

els

/to

tal

FO

I zo

ne

pix

els

Fire Occurrence Index

2010 fire occurrence as a

function of

Fire Occurrence Index

Weather variables from the RAWS dataset were ingested into the FOM to

produce FOI projections for 2010 fire season. Results (above) show that

higher FOI are directly linked to higher probability of fire occurrence.