Transcript Slide 1

From Ignition to Spread
Wildland Fire Forecasting and Color Maps
Haiganoush K. Preisler
Pacific Southwest Research Station
USDA, FS
Managing fire on populated forest landscapes
October 20 - 25, 2013
Banff International Research Station
For Mathematical Innovation and Discovery
Overview
• Uncertainties in Fire danger maps
1. 1-7 day forecasts
2. Seasonal (1-month ahead) forecasts
3. 1-year ahead
4. 1-2 hours ahead (will not be covered in this talk requires a statistical model of fire growth. Future
research.)
• Maps of risk that incorporate loss to societal or
ecological values
One-day Forecast
One-day Forecast
One-day Forecast
It is hard to perform goodness-of-fit analyses of these maps
Need probability models to perform validation
(8/23/2013)
EROS = Earth Resources Observation System
Pr[Fire size > C | ignition ] = f x,y,t( FPI )
FPI = Fire Potential Index is a
moisture-based vegetation flammability indicator.
= f(living vegetation greenness, 10-h dead fuel moisture)
(8/23/2013)
Alternative color legend
By using the alternative color legend we are able to note the amount of uncertainty
in the maps AND at the same time demonstrate the goodness-of-fit of the forecasts
7-day forecast
• Fire occurrence data from MTBS: Monitoring Trends in Burn Severity
Satellite imagery of burned area for fires > 500 acres (in East) and > 1000
acres (in west) starting from 1984 – present
• Explanatory variables: 1) location 2) day-in-year 3) Forecasted FPI values for
upcoming 7-days evaluated daily on a 9km2 grid cell surrounding ignition pt.
• Model: spatially and temporally explicit logistic regression at 1kmx1kmxday
grid cells.
A legend that includes some uncertainty. Goodness-of-fit analysis still to be done.
Seasonal Forecast (one-month ahead)
Large Fire Forecast Probabilities for the month of August, 2013
based on explanatory variable values up to July 31, 2013
Explanatories
used:
• Moisture
Deficit
• ENSO, TEMP
• Elevation
• Lightning
Scenario
Anthony Westerling
UC Merced
Observed Fraction of large fires
Goodness-of-fit for the one-month-ahead forecasts based on large fire
occurrences (>200ha) in California and Nevada between 1985-2008
Predicted Probability of a large fire
Observed Fraction of large fires
Same as previous slide but with the Predicted values grouped
Predicted Probability of a large fire (Grouped)
Alternative legend
demonstrating expected
amount of uncertainty and
degree of goodness-of-fit
of the forecasts to historic
data
Forecasting one-year-ahead fire risk
1) Use season specific historic averages based on historic large fire occurrences:
Historic large fire occurrence from MTBS data
Forecasting one-year-ahead fire risk
2) Use a model that includes a trend
over the years
Data – Corsica & Sardinia
(Alan Ager and Michele Salis)
Risk to social, economic and ecological values
Alan Ager (WWETAC)
Western Wildland Environmental Threat Assessment Center
• Color maps to help managers with their fuel treatment
decisions
• Maps based on fire risk AND on #people/homes/type of
habitat at risk
• Produce maps by simulating the process from ignition to
spread
The process to be simulated
Spatial-temporal Marked Point Process {x,y,t,u}
Likelihood for discretized process (km×km×day)
Once an ignition location
and fire size is simulated
then fire perimeters/scars
may be simulated using a
fire growth model
Simulated (red)
Observed (orange)
fire perimeters
(Farsite, FSPro)
Mark Finney
Distribution of Fire Sizes
Observed vs Simulated Quantiles
Although simulated fire sizes seem to be a good approximation of observed fire sizes,
goodness-of-fit of fire growth models still needs to be done.
Simulated fire perimeters/scars are then overlapped with other
polygons with high value (e.g., owl habitat; old growth trees;
houses)
The number of houses, owl habitat or people being affected by
each simulated fire are then used, together with the simulated total
area burned in a given region to produce risk maps based on a
measure of loss of interest.
Number of people exposed (power of 10)
Number of people exposed vs total area burned
by simulated fires ignited on FS land
95th %
Grouped total area burned (power of 10)
Criteria based on
expected burn
area and #people
affected
There is a large amount
of variation in this color
map too. Both spatial
(between districts) and
temporal (between
years) variation as seen
in the boxplots of the
next slide.
Total area burned per district per year (power of 10)
5
4
3
2
Boxplot colors
match the colors in
the previous map
References
• D.R. Brillinger, H.K. Preisler, and J.W.Benoit. (2003). Risk assessment: a forest fire example. In
Science and Statistics: A Festschrift for Terry Speed. D.R. Goldstein [Ed.]. pp: 177- 196.
• Preisler, H.K., D.R. Brillinger, R.E. Burgan, and J.W. Benoit. (2004) Probability based models for
estimation of wildfire risk. Journal of Wildland Fire, 13, 133-142
• Brillinger, D. R., Preisler, H. K., and Benoit, J. (2006) "Probabilistic risk assessment for wildfires.
Environmetrics, 17 623-633.
• Preisler,H.K., Westerling, A.L. (2007). "Statistical model for forecasting monthly large wildfire
events in western United States". Journal of Applied Meteorology and Climatology 46, 1020-1030.
• Preisler, H.K.,Chen, S.C. Fujioka, F., Benoit, J.W. and Westerling, A.L. (2008). "Wildland fire
probabilities estimated from weather model-deduced monthly mean fire danger indices".
International Journal of Wildland Fire17: 305-316.
• Preisler, H.K., Burgan, R.E., Eidenshink, J.C, Klaver, J.M., Klaver, R.W. (2009) ‘Forecasting
distributions of large federal-lands fires utilizing satellite and gridded weather information’
International Journal of Wildland Fire 18, 517-526.
• Preisler, H.K., Westerling, A.L. Gebert, K. and Munoz-Arriola, F. and Holmes, T. (2011) ‘Spatially
explicit forecasts of large wildland fire probability and suppression costs for California.’
International Journal of Wildland Fire. 20:508-517
• Preisler, H.K. and A.A.Ager. (2012) ‘Forest fire models’ in A. H. El-Shaarawi and W. Piegorsch (eds.)
Encyclopedia of Environmetrics Second Edition, John Wiley and Sons Ltd: Chichester, UK.