insights on the statistical variability of experimental ...wildfire/2010/pdfs/doug mcrae.pdf ·...

Insights on the Insights on the statistical variability statistical variability of experimental fire of experimental fire behavior data using behavior data using airborneairborne--infraredinfrared

Douglas J. McRae, Susan G. Douglas J. McRae, Susan G. ConardConard, , JiJi--zhongzhong Jin, Anatoly I. Jin, Anatoly I. SukhininSukhinin, , Tom W. Blake, and Galina A. Tom W. Blake, and Galina A. IvanovaIvanova

Natural Resources Canada

Canadian Forest Service

Wildland Fire Canada 2010, 6 October 2010, Kitchener, ON.

Adapted from: Stocks, B.J. 1987. Fire behavior in immature jack pine.

Can. J. For. Res. 17: 80–86.

The problemThe problem

TABLE 4. Fire impact on forest fuels and fire behavior characteristics of the SharpsandCreek experimental fires in immature jack pine.

2910.663.011.470.000.730.7418

2599027.304.643.361.111.310.9414

1713620.164.633.001.041.300.6612

4090349.443.842.921.400.970.5511b

5992.101.910.950.000.420.537

1078514.643.742.601.270.940.395

471710.742.011.550.890.440.222

Total fuel

Crown fuels

Ground fuels

Total surface

Frontal fire

intensity (kW/m)

Rateof

spread (m/min)

Depth ofburn (cm)

Fuel consumed (kg/m2)

Fire No.


Assumption:

Values for fire behavior databases are given as though they are a ‘magical’constant for a particular fuel type and burning condition rather than an average with statistical ranges associated with it.

1. How accurate are these averages?

2. What is the standard deviation/standard error level of each average?

3. What is the absolute range in these values?

Obvious questions:


Reason for questions:

1. Suppression activities – Concerns regarding rate of spread/intensity values: is it safe to send fire crews to work on a firefront?

Rate of spread average of 6.5 m/min

5.0 - 8.0 versus 5.0 – 15.0 m/min

• The lack of adequate sampling due to field expenses and expense of

monitoring equipment.• Equipment failures.• Reliance on visual observations based on fixed-point measurements.• Research personnel safety concerns.• Mother nature never cooperates (e.g., wind lulls and gusts, wind direction

changes, etc.).

Fire behavior is difficult to quantify in the past because of:


FIRE BEAR ProjectFIRE BEAR Project(Fire Effects in the Boreal Eurasia Region)(Fire Effects in the Boreal Eurasia Region)

• To better understand fire in central Siberia, the FIRE BEAR Project was created as a forest fire research study to provide answers to basic questions on fire management.

• Replicated 200 x 200-m experimental burn plots on Scots pine (Pinus sylvestris) / lichen (Cladonia sp.) / feather moss (Pleurozeum schreberi) forest sites.

• Fuel and fire behavior on these fires was quantified.

FIRE BEAR ProjectFIRE BEAR Project(Fire Effects in the Boreal Eurasia Region)(Fire Effects in the Boreal Eurasia Region)

Rate of spread: 2.5 m/minFireline intensity: 620 kW/m Rate of spread: 4.9 m/min

Fireline intensity: 2259 kW/m

Rate of spread: 5.6 m/minFireline intensity: 5220 kW/m

Boguchany, Russian, June 18, 2002

Plot 1 3:29:11 PM

1000

900

800

700

600

500

400

300

200

100

4 Geo-referencing fires


Plot 1 3:31:43 PM

1000

900

800

700

600

500

400

300

200

100

Ignition line

Wind


Plot 1 3:34:42 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:37:15 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:39:28 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:39:32 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:39:35 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:41:56 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:44:07 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:46:39 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:49:14 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:51:27 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:53:45 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:56:12 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 3:58:34 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 4:00:52 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 4:03:27 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 4:05:59 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 4:06:05 PM

1000

900

800

700

600

500

400

300

200

100


Plot 1 4:08:39 PM

1000

900

800

700

600

500

400

300

200

100

Isolines showing firefront locations

(seconds)

Infrared data analysis

10,000 sample points/ha

(1-m resolution)

Rate of spread(m/min)

Infrared data analysis

All data

Rate of spreadRate of spread

0

500

1000

1500

2000

2500

3000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31

Rate-of-spread classes (m/min)

Nu

mb

er

of

ob

serv

ati

on

s

≥31

All data (n = 21434)

All data


0

500

1000

1500

2000

2500

3000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31


Nu

mb

er

of

ob

serv

ati

on

s

8.9 m/min(SE=0.08)

All data (n = 21434)

Sample size = 507

Maximum value = 912 m/min

≥31


Variability in rates of spread can be caused by:

• Differences in fuel structure.• Differences in soil (ground fuel) moisture.• Gusts and lulls in wind speed.• Changes in wind direction.• Channeling and acceleration effect on wind.• Junction zone effects.• Edge effect of experimental plot.• Impact of tree density on solar radiation and fuel dryness.• Spotting.• Analysis problems.

Data minus outer 10-m perimeter edge effect and erroneous high

rates of spread (± 2 standard deviations of mean)

>31`


0

500

1000

1500

2000

2500

3000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31


Nu

mb

er

of

ob

serv

ati

on

s

Data minus outer 10-m perimeter edge effect

9.0 m/min(SE=0.11)

≥31

Data minus outer 10-m perimeter edge effect and erroneous high

rates of spread (± 2 standard deviations of mean)

>31`


0

500

1000

1500

2000

2500

3000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31


Nu

mb

er

of

ob

serv

ati

on

s

Data minus outer 10-m perimeter edge effect and rates of

spread greater than ± 2 standard deviations

7.9 m/min(SE=0.04)

≥31

0

500

1000

1500

2000

2500

3000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31


Nu

mb

er

of

ob

serv

ati

on

s


63% 27%

7.9 m/min(SE=0.04)

ISI

0 2 4 6 8 10

Rate

of

sp

rea

d (

m/m

in)

0

1

2

3

4

5

6

7

8

9

10

Application to ModelsApplication to Models

RoS = 1.008 + (0.741 * ISI)R2=0.87; SE=0.90

95%

Error bars indicate one standard error from the mean

ISI

0 2 4 6 8 10

Ra

te o

f s

pre

ad

(m

/min

)

0

5

10

15

20

25

Application to ModelsApplication to Models

Data range for each fire

X

X

X

X

X

X

X

X

X

X

X

XX XX

X

XXX

0

10

20

30

40

50

60

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Depth-of-burn classes (cm)

Nu

mb

er

of

ob

serv

ati

on

s

0 21 3 1554 9876 13 14121110

Depth of burnDepth of burn

DMC= 19DC = 189

3.3 cm(SE=0.09)

0

10

20

30

40

50

60

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16


Nu

mb

er

of

ob

serv

ati

on

s

0 21 3 1554 9876 13 14121110

6.3 cm(SE=0.15)


DMC= 19DC = 189

DMC= 35DC = 399

3.3 cm(SE=0.09)

0

5

10

15

20

25

30

1 2 3 4 5 6 7 8 9 10 11 12 13

Fuel consumption class (kg/m 2)

Nu

mb

er

of

ob

se

rva

tio

ns

1.0 kg/m2

(SE=0.04)

Total fuel consumptionTotal fuel consumption

2.0 kg/m2

(SE=0.22)

0 21 3 54 9876 121110

0

5

10

15

20

25

30

1 2 3 4 5 6 7 8 9 10 11 12 13

Fuel consumption class (kg/m 2)

Nu

mb

er

of

ob

se

rva

tio

ns

3.2 kg/m2

(SE=0.23)

DMC= 51DC = 393

Total fuel consumptionTotal fuel consumption

1.0 kg/m2

(SE=0.04)

2.0 kg/m2

(SE=0.22)

0 21 3 54 9876 121110

ConclusionsConclusions

Fire behavior

• Fire behavior is a highly variable phenomena at the microsite level (e.g., 1- m pixel).

• Due to the lack of statistical data in the past, most current models do not indicate the actual ranges of fire behavior.

• Remote sensing using infrared cameras allows for multiple sampling to take place, which allows for adequate sample numbers to allow for statistical analysis.

• For fire crew safety, realize that there is a range of possible values around any average!

• Other applications for infrared monitoring (e.g., fuel consumption, carbon emission).

Thank you!

Wildland Fire Canada 2010, 6 October 2010, Kitchener, ON.

0

10

20

30

40

50

60

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16


Nu

mb

er

of

ob

serv

ati

on

s

0 21 3 1554 9876 13 14121110

6.3 cm

(SE=0.15)


DMC= 19DC = 189

DMC= 51

DC = 393

0

10

20

30

40

50

60

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16


Nu

mb

er

of

ob

serv

ati

on

s

0 21 3 1554 9876 13 14121110


Low Moderate Severe

3% 76% 21%

DMC= 51

DC = 393

6.3 cm

(SE=0.15)

71 22.5 90 52 52

50 23.5 70 44 44

52 24.5 55 35 35

44 25.5 44 27 27

35 26.5 50 35 35

27 27.5 41 30 30

35 28.5 42 32 32

30 29.5 45 32 32

32 30.5 507 377 46

32 Total 21434 14015 13684

46

0

2

4

6

8

10

12

14

16

18

1 2 3 4 5 6 7 8 9 10 11 12 13

Series1

Series2

Series3

2.0 kg/m2

(SE=0.22)

Fuel consumptionFuel consumption

0

500

1000

1500

2000

2500

3000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31

Series1

Series2

Series3

1.0 kg/m2

(SE=0.04)


0

5

10

15

20

25

30

1 2 3 4 5 6 7 8 9 10 11 12

Fuel consumption class (kg/m2)

Nu

mb

er

of

ob

se

rva

tio

ns

3.2 kg/m2

(SE=0.23)


ResultsResults

Values in parentheses show the range in consumption values

Table 2. Fuel consumption values (plot averages with standard error) observed for each experimental Siberian Scots pine fire

1.50±0.15

(0.53-7.79)

1.16±0.05

(0.12-3.65)

0.11±0.01

(0.01-0.27)

0.30±0.15

(0.01-7.33)

0.02±0.02

(0.00-0.13)

20

3.03±0.23

(1.17-9.79)

2.45±0.12

(0.09-10.36)

0.25±0.25

(0.07-0.51)

0.44±0.18

(0.00-7.55)

0.07±0.01

(0.00-0.13)

14

2.03±0.22

(0.26-10.47)

1.50±0.08

(0.02-5.53)

0.30±0.06

(0.02-1.50)

0.41±0.17

(0.00-7.43)

0.07±0.01

(0.00-0.16)

13

0.93±0.04

(0.49-12.43)

0.74±0.04

(0.00-2.00)

0.18±0.04

(0.07-0.37)

0.04±0.01

(0.00-0.30)

0.00±0.04

(0.00-0.12)

3

1.68±0.12

(0.32-4.32)

1.36±0.06

(00.0-4.90)

0.11±0.02

(0.00-0.28)

0.18±0.06

(0.00-2.28)

0.03±0.01

(0-0.12)

2

1.80±0.16

(0.34-5.23)

1.45±0.08

(0.03-4.63)

0.16±0.03

(0.03-0.55)

0.28±0.10

(0.02-3.28)

0.03±0.01

(0-0.09)

1

Total

Forest

FloorLitterDead & DownVegetation

Consumption (dry weight) by category (kg/m2)

Fire

No.

ResultsResults

Equilibrium (steady-state) fire behavior characteristics (plot averages with standard error) observed for each Siberian experimental Scots pine fire

25 757 ± 3145

(6629-157376)

2790 ± 341

(718-17049)

6.5±0.05

(2.4-16.5)

4.2±0.10

(1.7-8.3)20

55 938 ± 4646

(19344-196851)

5220 ± 434

(805-18372)

5.6 ± 0.07

(3.6-14.8)

6.3 ± 0.15

(1.2-15.0)14

36 450 ± 4474

(1564-209443)

1214 ± 149

(52-6981)

2.0 ± 0.34

(0.8-5.7)

4.6 ± 0.15

(0.9-10.5)13

14 878 ± 828

(6230-24920)

620 ± 34

(260-1038)

2.5 ± 0.25

(1.1-9.4)

3.3 ± 0.09

(1.1-5.9)3

27 662 ± 2316

(25235-83261)

2259 ± 189

(207-6800)

4.9 ± 0.35

(1.2-9.9)

4.4 ± 0.13

(1.1-9.8)2

27 662 ± 2316

(25235-83261)

2259 ± 189

(207-6800)7.9 ± 0.04

(1.3-17.9)

5.6 ± 0.20

(0.5-9.4)1

Total fire intensity

(kJ/m2)

Firelineintensity

(kW/m)

Rate of spread

(m/min)

Depth of burn

(cm)

Fire

No.

Values in parentheses show the range in values

insights on the statistical variability of experimental ...wildfire/2010/pdfs/doug mcrae.pdf ·...

Documents