Assessing the Accuracy of Fire Spread Model Outputs

1
Assessing the Accuracy of Fire Spread Model
Outputs
Scott Sweet Winrock International Program
Assistant/GIS Analyst
1621 N. Kent St. Arlington VA, 22201 USA Phone
703.201.9172 ssweet_at_winrock.org
www.ussartf.org/images/ Wildland_Fire_3.jpg
2
Overview

• Give a brief history of fire management policy.
• Talk about the importance of management ignited
  fires.
• Review the Cerro Grande fire in North Central New
  Mexico.
• Explain FARSITE (Fire Area Simulator Model).
• Describe model inputs and data processing flow.
• Discuss validation of the model outputs.
• Show results of the analysis.

3
Following a catastrophic fire in 1910 and the
loss of many lives, the Department of the Army
was put in charge of combating wildland fires.
4
Early US Forest guidelines adopted the, 1000AM
Policy, meaning all forest fires should be put
out by 1000AM the day following detection.
An early lookout
Graves Peak Lookout - 1929
Old Powell Fire Warehouse
Sighting in on a wildfire in the 1950s
Photos courtesy of USDA Forest Service 
Clearwater National Forest, Lochsa Ranger
District, Powell Ranger Station, MT
5
The Firefighters
http//www.wildlandfire.com/index.html
6
Air tankers dropping fire retardant
Support Crews and Equipment
http//www.wildlandfire.com/index.html
http//www.wildlandfire.com/index.html
7
(No Transcript)
8
(No Transcript)
9
The militaristic methodology of attacking
wildland fire was prevalent up until the 1960s
when fire ecologists started examining the
benefits of natural fire.
figures courtesy of John Rogan
10
A dense build up of fuels cause more intense
fires.
11
Understory burns without ladder fuels are less
intense.
12
Management ignited fires are ignited in order to
meet a land management objective such as debris
removal or wildlife habitat improvement.
Prescribed natural fires are those that are
allowed to burn under an approved plan to
preserve the natural role of fire in the
ecosystem. (Pyne et al 1996, p 48)
13
A 1000 acre burn to reduce hazardous fuels was
planned for May 4th, 2000 in the Bandelier
National Monument.
http//www.newmexico.org/maps/
14
Fire Danger Maps
15
Fire Danger Maps
16
Fire Danger Maps
17
Fire Danger Maps
18
Fire Danger Maps
19
Fire Danger Maps
20
Fire Danger Maps
21
Fuel Moistures
22
Fuel Moistures
23
Fuel Moistures
24
Fuel Moistures
25
Fuel Moistures
26
Fuel Moistures
27
Fuel Moistures
28
http//www.nps.gov/cerrogrande/
29
Los Alamos
30
http//www.nps.gov/cerrogrande/
31
The Cerro Grande fire destroyed 235 homes and
damaged many other structures including parts of
the Los Alamos National Laboratory.
Some of the areas that burned were known or
suspected to be contaminated with radionuclides
and chemicals.
http//www.racteam.com/Experience/Projects/CerroGr
ande
32
Research Objectives

• Relay needed information to all parties involved
  for meeting the needs of prescribed fire
  operations through the use of fire prediction
  modeling.
• Validate how accurately these models can predict
  fire spread.
• Compare model output accuracy to an area of equal
  size radiating out in a circle.
• Compare model outputs to a null model, or no
  burn prediction.

33
(No Transcript)
34
More validation is certainly necessary because
the many potential sources of error can confound
the comparisons. These tests have highlighted the
need to consider error associated with (1) the
model input data (fuels and weather), (2) the
spatial and temporal resolution of the inputs,
and (3) the observed fire progression maps used
for comparison.

• Mark Finney, creator of the FARSITE
• Fire Area Simulator-Model

35
The general flow of a FARSITE simulation.
36
Raster landscape input layers required from the
GIS for FARSITE simulation.
37
Building the landscape file in FARSITE required
conversion to ASCII file formats
38
(No Transcript)
39
(No Transcript)
40
(No Transcript)
41
(No Transcript)
42
Table used to create raster images and model
inputs
Grd-crn Height the distance from the ground to
the base of the live tree crown.
43
(No Transcript)
44
(No Transcript)
45
(No Transcript)
46
(No Transcript)
47
Data from the Los Alamos National Laboratory were
used to create weather files from May 5th through
May 19th 2000.
48
Data from the Los Alamos National Laboratory were
also used to create wind files from May 5th
through May 19th 2000 at 15 minute intervals.
49
Fuel Moisture is used in FARSITE to calculate
conditions for predicting fire behavior.
50
(No Transcript)
51
(No Transcript)
52
Ignition
53
Validation run comparing the first model output
(burn 1) to the actual burn perimeter (truth).
75 correct
54
Ignition
55
Validation run comparing a circle with the same
area as the first model output (circle 1) vs. the
actual burn perimeter.73 correct
56
Adjustment files allow the user to fine tune the
model.
57
(No Transcript)
58
Ignition
59
Validation run comparing the second model output
(burn 2) to the truth. 80 correct
60
(No Transcript)
61
Validation run comparing a circle with the same
area as the second model output (circle 2) vs.
the actual burn perimeter.
62
(No Transcript)
63
Validation run comparing a no burn scenario
versus the truth.
64
This figure shows the validation comparison
between burn 1 vs. truth to circle 1 vs. truth.
Red pixels would need to be moved 22 to 44 km in
order to put the burn in the proper location.
43.8 km
65
This figure shows the validation comparison
between burn 2 vs. truth to circle 2 vs. truth.
66
Multiple Resolution Analysis comparing burn 1 to
the, no burn scenario.
80.12
75.3
21.7 km
67
Conclusions

• The creator of FARSITE said spatial validation
  was an important next step and I have presented a
  method for going about doing that.
• This type of analysis is the only way to specify
  how accurately models predict in space.
• Model predictions had a higher percentage of
  correct pixels than circular growth with equal
  acreage.
• Model predictions had a lower percentage of
  correct pixels than predicting no fire.
• The FARSITE model has many parameters, thus
  demands large amounts of information that is not
  always available. Therefore, strict validation
  is nearly impossible.
• This study shows the importance of validating
  models in comparison to different scenarios
  rather than simply stating how accurate outputs
  are without considering error of location at
  multiple resolutions.

68
Continued Research

• Raster outputs including time of arrival, fire
  line intensity, flame length, rate of spread,
  heat/area, reaction intensity, crown fire
  activity, and spread direction need to be
  validated.
• Model output using detailed fuel moisture data,
  and multiple wind and weather streams need to be
  incorporated.
• High resolution land cover maps with custom fuel
  models should be considered.

69
Acknowledgments

• Advisors John Rogan, Gil Pontius, Eugenio
  Marcano and Ron Eastman.
• Randy Balice and Steve Koch of the Los Alamos
  National Laboratory.
• James Toledano and the IDRISI crew.
• Jay Miller, Steve Yool, and Steven Pyne
  University of Arizona, Arizona State University.
• Mark Finney creator of the FARSITE model.
• Kevin Denman creator of the Assess10 program
  used for analysis.

70
Scott Sweet Winrock International Program
Assistant/GIS Analyst 1621 N. Kent St. Arlington
VA, 22201 USA Phone [phone] ssweet_at_winrock.
org

• Additional Readings
• 1. R G Pontius Jr. 2002. Statistical methods to
  partition effects of quantity and location during
  comparison of categorical maps at multiple
  resolutions. Photogrammetric Engineering Remote
  Sensing 68(10) p. 1041-1049.2.
• R G Pontius Jr and J Malanson. 2004. Comparison
  of the structure and accuracy of two land change
  models. International Journal of Geographical
  Information Science. in press.
• 3.  R G Pontius Jr, D Huffaker and K Denman. in
  review. Useful techniques of validation for
  spatially-explicit land-change models. Ecological
  Modeling.