The tables give the estimated high frequency skill (skill in predicting the year to year
variations), the mean, standard deviation, and the exceedence threshold values for given
probability levels. (98,95,90,80,70,60,50,40,30,20,10,5, and 2 percentile levels).
These data are derived from the outlooks issued monthly by CPC, together with an estimate
of their skill obtained from the tools used for the prediction.
The forecasts issued on a particular month are grouped together. There are 2 header
records followed by multiple data records (Rows) per monthly forecast. Forecasts are for
three month target seasons, defined by forecast issue time and lead.
The first line on the header record gives the month, day and year (mmdd yyyy) of
forecast issue time, followed by the number of groups and the number of forecast locations
in each group. The final number is a forecast type id flag. 950 is a number assigned to
identify seasonal temperature forecasts. 951 identifies seasonal precipitation forecasts.
The second header label gives column definitions. Each forecast issue time has its own
2 line header record
||Year and month number
that forecast was issued. 1=Jan, 2=Feb. Forecasts are issued around mid-month.
||Approximate lead time of
forecast in months - rounded upward, An actual lead time of about one-half month (the
shortest lead on the outlooks) is rounded to lead=1, for example. The valid target season
can be found by adding lead months to the initial time listed in column 1 and 2, adjusting
for year changes.
||The forecast location to
which data apply. See regdict.txt for definitions.
||The correlation skill
estimate of this forecast. These are approximate high frequency (year-to-year)
correlations, and are not accurate for measuring the trend skill. The actual forecast
(including trend prediction) is more skillful than this value indicates.
(Forecast) to be exceeded by the percent of time listed in the column label.
||The forecast mean value
for the target period.
||The climatological mean
deviation. Temperatures are assumed normal, however precipitation is transformed by a
power transformation (See col. 23). These columns give the values of the transformed
precipitation units and are not in inches!!!! The precipitaion must be transformed before these
values can be used
value. See important note on column 21description above for Precipitation files.
||(Precipitation only) The
power transformation value is stored here. The transformed variable, is (precip)**power
Where **power indicates raised to the power of the value found in column 23. The
transformed variable is approximately Gaussian with mean = median of untransformed
variable (col. 12) and standard deviation in transformed units given in columns 21 and 22.
line in the data set is denoted by a 9999 in the years position - followed by the data set
name of the cd location dictionary file.
Data Set Use
These data are designed to be used in quantitative analysis of CPC forecasts. The
forecasts are only estimates of the distribution implied by the long lead outlooks and are
derived from the tercile probability anomalies (Forecasts for above, near, and below
median). The distribution is most accurate in the vicinity of these tercile boundaries (33
and 66 percent probability of exceedence). Particular caution should be used in the
interpretation probabilities below about 10% or above 90%.