|
| |
|
HOME >
Monitoring and Data >
Oceanic & Atmospheric Data >
Forecast Verfications
|
| |
|
| |
|
Climate Prediction Center: Forecast Verifications
|
| Verification of official forecast and tools |
| |
|
Monthly forecasts
|
|
Seasonal forecasts
|
|
|
| |
Conventions
- Seasonal and monthly forecasts can be either: Below=1,
Near-normal/neutral=2, Above=3 and Climatological=4. The
climatological (4) is a forecast of equal probabilities
(1/3, 1/3, 1/3) of the below/near-normal/above clases.
- The 3 categories (1, 2, and 3) all have 1/3 climatological
probabiliity change
of occurring except for "dry" stations. For dry stations,
the change of no precipitation is greater than 1/3. For
these stations, the precipitation is scored on a 2 class
system. The lower class has a probability of
max(2/3, climatological probability of no precipitation).
The upper class has the remaining probability.
- The monthly and seasonal forecasts are verified three
ways. (1) Cl is ignored. (2) Cl is verified as normal
and (3) Cl is verified as forecast of (1/3,1/3,1/3)
probabilities. The "Cl verified as normal" is only
shown in the contingency tables where it shows the
verification of the the CL class. Otherwise CL is
either not verified or verified as (1/3,1/3,1/3)
forecast depending on the title of the plot.
-
The contingency tables have column labels of:
below-obs-count, near-normal-obs-count, above-obs-count, total-obs.
The row labels are below-fcst-count, near-normal-fcst-count,
above-fcst-count, total-fcst-count. The upper row of tables
show the stations counts. The bottom row shows the percentages.
The left column shows the table where CL is not verified and
the right column shows the table where CL is verified as near-normal.
|
|
Note: The "Optimal Climate Normals" (OCN), "Canonical Correlation Analysis"
(CCA), "Screening Multiple Regression" (SMT) and "Seasonal-Forecast model"
(SFM) forecast tools produce standardized-anomaly forecasts as well as
"inflated" forecasts which are plotted for the forecasters. The scores for the
"standardized anomaly" forecasts will score differently than
"inflated forecasts" which are shown here.
|
|
Forecasts, especially those generated by statistical methods, have a
tendency to have too many normal forecasts. (When the explained variance
is small, the minimum RMS forecast is often the normal class.)
However to maximize the Heidke score (and hit rate), you can do better
by "inflating" the scores. For example, suppose a forecast is
for the 55% percentile (normal category) and the expected squared
forecast error is the same as the climatological variance.
(I.e., the forecast is the climatological distribution with
a slight warm/wet shift). In this case, the "above normal"
class is the best forecast (most probable). One often accounts
for this factor by "inflating" the forecasts. In theory, each
tool would have its own inflation factor. In practice all the
tools were inflated equally.
| | |
| Scores for
standardized anomaly forecasts
| | |
|
comments: Wesley.Ebisuzaki@noaa.gov
|
|
|
|
|
www.nws.noaa.gov 
