Monthly Fire Weather Forecast:
Error Analysis
Verification of the Monthly Fire Weather Forecast system was done
to test the system's accuracy. Climatological data from 1973 to 1990
(a period during which the Climate Analysis Center had prognostic
700mb height pressure anomaly data available) was used in the
verification. Using this data, monthly forecasts were made and
averaged into three-month seasons for 21 geographical regions. The
regions were chosen to be consistent climatologically and
statistically. Within each region, and for each season, the average
afternoon temperature, relative humidity, and Chandler Burning Index
were forecasted.
On the map below, click in a desired region, or select the region,
seasons, and variables from the controls below. A set of charts for
the selected region will be displayed, including the empirical
probability distribution and Forecasted vs. Observed chart for each
selected variable and season.
Because we recently discovered that the error probability
distribution charts erroneously depicted the distribution of
"Observed - Forecasted" rather than "Forecasted - Observed" errors,
as stated on the x-axes, the charts have been regenerated. The
distribution charts now show true "Forecast - Observed" errors.
Additionally, a few other changes were made, such as readjustment of
the axes to start on and have increments that are multiples of
0.05, rather than arbitrary values as before.
Note that there is a vertical yellow stripe in most of the error
distribution charts. This indicates where zero is on the x-axis. The
width of the yellow stripe changes in relation to the range of the
x-axis (the width of the stripe is 0.05 units). In a few of the
distribution charts, namely CBI charts, the stripe is not present.
This occurs when the range of the x-axis is very large.
The scatter plots of observed versus forecasted values are drawn
to appear 'square'--that is, the ranges of the x- and y-axes are the
same for each scatter plot. This enables the viewer to easily see
characteristics of each set of points. Ideally, the points on each
"Observed vs. Forecasted" scatter plot would appear close to a
diagonal line drawn from the lower left corner to the upper right
corner; this would imply that forecasted values closely matched
observed ones.
April 2002
References:
Fujioka, Francis M., McCutchan, Morris H., Meisner, Bernard N.,
and Benoit, John W., 1991. A Process for forecasting
monthly fire weather. Proceedings of the 11th Conference
Fire and Forest Meteorology. Society of American Foresters,
April 16-19, 1991, Missoula, MT. 526-535.
Klein, William H., Charney, Joseph J., McCutchan, Morris H,
Benoit, John W., 1996. Verification of monthly mean
forecasts for fire weather elements in the contiguous United
States. Journal of Climate, volume 9, number 12.
3317-3327.
Lin, K.M.Y, Fujioka, Francis M., 1995. Error analysis of
monthly fire weather forecasts. Proceedings of the 6th
International Meeting of Statistical Climatology. April 19-23,
1995, Galway, Ireland. 1-6.
McCutchan, Morris H., Meisner, Bernard N., Fujioka, Francis M.,
Benoit, John W., and Ly, Benjamin, 1991. Monthly fire
weather forecasts. Fire Management Notes, 1991, Volume
52, Number 3. 41-47.
Back to Monthly Fire Weather Forecast Page
Direct comments or questions about the Monthly Fire Weather Forecast to
ffujioka@fs.fed.us
|
