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 begin
  to produce routine experimental weekly to
  seasonal fire danger predictions. This happened
  around the time that the WWW became a new
  outreach medium, when we started collaborating
  with NCEP modelers (Kanamitsu and Juang), and
  when desktop scientific computing became
  available.
• 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 (Kalnay et al. 1996) T62L28, 192x94 Gaussian
  grid
• RSM97 (Juang et al. 1997) regional model, US,
  60k 18 levels
• Firedanger Models (USFS)
• Fosberg Fireweather Index (Roads et al. 1997)
• ECPC began making experimental, near real-time,
  routine weekly long-range global-regional
  predictions on Sept. 27, 1997 with GSM (150
  seasonal prediction ensemble archive used for
  initial paper).
• 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.
• For validation, we used 1 day GSM predictions,
  which are closely related to NCEP analyses,
  except we could more easily access our needed
  output in near real time.
• 1-day forecast precipitation was a problem, we
  used instead the Xie-Arkin pentads interpolated
  to weekly and seasonal

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
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.

1. SC is an index of the forward rate of spread at
   the head of a fire and is quite sensitive to wind
   speed.
2. 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.
3. 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.
4. IC is a rating of the probability that a
   firebrand will cause a fire requiring suppression
   action. SC is a component of IC.
5. KB is a stand-alone index that can be used to
   measure the affects of seasonal drought on fire
   potential.
6. 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. It
   is a grassland approximation to BI.

7
ECPC Experimental Predictions

• Atmospheric Forecast Models (ECPC G-RSM)
• GSM (Kalnay et al. 1996) T62L28, 192x94 Gaussian
  grid
• RSM97 (Juang et al. 1997) regional model, US,
  60k 18 levels
• Firedanger Models (USFS)
• Fosberg Fireweather Index (Roads et al. 1997)
• USFS Firedanger Indices (Roads et al. 2005)
• We used our expanding historical archive of
  experimental, near real-time, routine weekly
  long-range global-regional predictions began on
  Sept. 27, 1997 with G-RSM (then 300 prediction
  ensemble archive) for the new evaluations.
• 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.

8
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 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,
• Daily 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
  every day.
• We validate the fire danger seasonal forecasts
  with the validating/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.

9
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.
10
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 (along with Higgins
  precipitation) the initialization/validation data
  for fire danger code
• Ten 7-month predictions made monthly (beginning
  2004) starting from 0000 and 1200 UTC of the
  first 5 days of current month and last 5 days of
  previous month.
• Experimental prediction effort began Oct. 2004
  and will continue for at least next 2 years
• 3 hindcasts (the first two days of month and the
  last day of the previous month) initialized from
  the NCEP/DOE reanalysis for the same month but
  each year 1982-2004, or 233 mon. hindcasts.
• more hindcast members may be added later if
  model not upgraded.
• In fact, many sensitivity experiments are
  underway
• a new land model, etc.
• Please see poster 2.12 Juang and Wang for details

11
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
12
MJJAS Valid Mar. 1 Fcst
13
Forecast biases are fairly similar for all of the
indices, we are still trying to track down which
of the input variables is causing this behavior,
we suspect humidity and pcp biases
14
US WestTime Series5 month running means.
Validation (dark lines) and 2-month lead
forecasts, 25 (red lines)Note summer has
largest values
15
US West Anom. Time Series5 mon. running
meanNote low frequency interannual variability
reflected in fire danger indices, val. (black)
and 2 mon. lead fcst (red lines)
16
Seas. Valid MJJAS 1983 Mar. 1 Fcst
17
Seas. Valid MJJAS 1994 Mar. 1 Fcst
18
Correlations of MMJJA validations with Mar. 1
forecastDifferent indices have slightly
different preferred regions
19
Correlations of validations and ln acres burned
are positive but low, we still need to find
better relations between fire measures and fire
danger indices.
20
Correlations of validations and fire counts seem
higher and we will be investigating this measure
as well as the acres burned.
21
The correlations for long range forecasts are
similar but lower. Interestingly, all indices
have skill in similar regions, may be related to
skill of certain input variables.
22
Summary

• The ECPC previously developed an experimental
  global/regional fireweather/firedanger prediction
  system
• There was skill in predicting the primary
  meteorological inputs and fire danger indices out
  to 4 months for many places.
• and modest skill in predicting US West fire acres
  burned
• We are now working with NCEP and USFS to further
  develop US fire danger forecasts
• Daily RSM products and observed precipitation
  from 1982-present now provide a much longer term
  fire danger initialization/validation set for an
  upgraded fire danger model and upgraded fire
  statistics
• This validation set is used as the initial
  condition for 7-month and historical prediction
  fire danger ensembles (1023x3).
• Preliminary results for Mar. 1 forecast of MJJAS
  encouraging! Analysis is ongoing. Need to now
  include ensemble forecasts/hindcasts
• We also need a 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 firedanger model synthesis.
• 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. Global historical measures of
  fire activity and characteristics are needed.