Experimental Weekly to Seasonal Fire Danger predictions

1
Experimental Weekly to Seasonal Fire Danger
predictions
J. Roads, P. Tripp, A. Westerling H. Juang,
J. Wang, S. Chen, F. Fujioka ECPC,
NCEP, USFS

• In the mid 90s the USFS requested that we
  produce routine experimental fire danger
  predictions.
• The purpose of this talk is to discuss our
  previous and current research effort to develop
  firedanger predictions from experimental seasonal
  regional predictions.
• FWI Predictions (ca. 1997-2000)
• ECPC predictions
• Initial FDI Efforts (ca. 2001-2004)
• ECPC predictions
• Current FDI Efforts (ca. 2005-2008)
• NCEP ensemble predictions, ECPC analysis

2
ECPC Experimental Predictions

• Atmospheric Forecast Models (ECPC G-RSM)
• GSM (Kanamitsu et al. 2002a,b, NCEP/DOE RII
  model) T62L28, 192x94 transform grid
• G-RSM (Kanamitsu et al. 2005), fully unified and
  parallelized GSM and RSM (28L and 25-50km
  regional resolution)
• Land surface models
• OSU (Pan et al. 1996)
• Noah (Mitchell et al. 2002) modular land surface
  model
• Firedanger Models (USFS)
• Fireweather (Roads et al. 1997)
• Firedanger (Roads et al. 2005)
• ECPC began making experimental, near real-time,
  routine weekly long-range global-regional
  predictions on Sept. 27, 1997 with G-RSM (now
  400 predictions ensemble archive).
• The initial conditions and SST boundary
  conditions (climatology persisted anomaly) for
  these experimental global to regional predictions
  come from the NCEP Global Data Assimilation
  (GDAS) 00UTC operational analysis.

3
FWI depends mostly on RH/WSP (Temp. effect weak)
4
Seasonal FWI Prediction/Validation Correlation
Roads, J.O., S-C. Chen and F. Fujioka, 2001
ECPCs Weekly to Seasonal Global predictions.
Bull. Amer. Meteor. Soc, April 2001. Vol. 82, No.
4, 639-658.
5
(No Transcript)
6
ECPC Firedanger predictions

• The fire danger code depends upon the previous
  history. We must therefore use the best available
  data to drive our validating and initializing
  fire code
• We use our 1 day RSM predictions, which are
  closely related to NCEP analyses, except we can
  more easily access our own predictions in near
  real time.
• Forecast precipitation is a problem. Fortunately,
• CPC precipitation at .25 degrees is now available
  in near-real time and this precipitation is used
  in place of predicted precipitation to update the
  fire danger code everyday.
• We can validate the fire danger seasonal
  forecasts with the validation/initializing fire
  danger values and
• Fire occurrence data (counts, area burned), which
  are available at coarse temporal (monthly) and
  spatial (1-deg.) (cf. Westerling) and this data
  was used to evaluate our fire danger predictions
  for the period 1997-2002.

7
Correlation
Roads, J., F. Fujioka, S. Chen, R. Burgan, 2005
Seasonal Fire Danger predictions for the USA.
International Journal of Wildland Fire, Special
Issue Fire and Forest Meteorology, 14, 1-18.
8
A higher resolution Fire Danger Code
And updated fire statistics (States!)
Model A Annually varying Western grasslands Model
B Mature dense fields of brush Model C Open pine
stands Model D Southeast coastal pine
stands Model F California chaparral Model G Dense
conifer with heavy litter Model H Short needled
conifers Model L Perennial grasses
Model N Florida sawgrass Model O Dense brushlike
fuels of Southeast Model P Closted stands of
long-needled southern pines Model Q Upland
Alsaskan black spruce Model R Deciduous
hardword Model S Alaskan tundra Model T Great
Basin sagebrush grass Model U Closed stands of
western long-needled pines
9
NCEP Global to Regional predictions

• NCEP CFS T62L28 forces NCEP RSM (US 50 km 28
  layers)
• A continuous series of 1-day runs have been made
  from 1982-present, to provide validation data for
  fire danger code
• Five 7-month predictions made monthly (beginning
  2004) starting from 0000 and 1200 UTC of the
  first three days of current month and last two
  days of previous month.
• Experimental prediction effort began Dec. 2004
  and is continuing for next 2 years
• 3 hindcasts (the first two days of current month
  and the last day of the previous month)
  initialized from the NCEP/DOE reanalysis for the
  same month but for each year from 1982-2004, or
  233 hindcasts.
• more hindcast members may be added later if
  model not upgraded.
• In fact, many sensitivity experiments are
  underway
• a new land model
• different bias correction methods

10
Seas. Valid 7 month Fcst
11
US WestTime Series for validation (dark lines)
and 1 month forecasts (red linesNote summer has
largest values
12
Seas. Valid Summer 1983 7 month Fcst
13
Seas. Valid 1994 7 month Fcst
14
US West Anom. Time SeriesNote low frequency
interannual variability reflected in both fire
danger indices, val. and 1 mon. fcst and fire
counts and acres burned
15
Correlations of validations and ln acres burned
are positive but low, we still need to find
better relation between fire measures and fire
danger indices. Given the high correlations
between validation and fcst fire danger indices,
we assume that the correlations for long range
forecasts will be similar.
16
Summary

• The ECPC previously developed an experimental
  global to regional seasonal prediction system
  that provided all the variables needed to drive
  the USFS and other fire danger codes.
• Evaluation of these predictions indicated skill
  in predicting the primary meteorological inputs
  and fire danger indices out to 4 months.
• and modest skill in predicting US West fire
  measures
• We are working with NCEP and USFS to further
  develop US fire danger forecasts
• Daily RSM prediction products and observed
  precipitation from 1982-present are being used to
  develop a fire danger validation set for an
  upgraded fire danger model
• This validation set is used as the initial
  condition for 7-month and historical prediction
  ensembles (523x3).
• Preliminary results are encouraging! Analysis is
  ongoing.
• We also need a unified and global fire danger
  index and global measures of fire activity
• Currently our only available global fire danger
  index is the FWI. More complex indices have been
  developed for individual regions. We need a
  global synthesis, similar to the synthesis that
  is occurring for LSMs.
• Currently our only available fire activity data
  comes from Westerlings manual efforts to gather
  historical info from US govt. and state agencies
  over the US West. Remotely sensed measures of
  fire activity and characteristics are probably
  the ultimate global answer.

17
USFS Fire Danger Indices
Roads, J., F. Fujioka, S. Chen, R. Burgan, 2005
Seasonal Fire Danger predictions for the USA.
International Journal of Wildland Fire, Special
Issue Fire and Forest Meteorology, 14, 1-18.

• SC is an index of the forward rate of spread at
  the head of a fire and is quite sensitive to wind
  speed.
• ER is a number related to the available energy
  per unit area within the flaming front at the
  head of a fire. ER is not affected a by wind
  speed.
• BI is a number related to the contribution of
  fire behavior to the effort of containing a fire.
  BI values represent the near upper limit to be
  expected if a fire occurs in the worst fuel,
  weather and topography conditions for this fuel
  type. SC and IC contribute to the BI.
• IC is a rating of the probability that a
  firebrand will cause a fire requiring suppression
  action. SC is a component of IC.
• KB is a stand-alone index that can be used to
  measure the affects of seasonal drought on fire
  potential.
• FWI was derived by Fosberg (1978) who assumed
  constant fuel (vegetationgrass) characteristics.
  The FWI is most easily applied in practice and
  provides a first look at fire danger globally.