View Source

Data file consists of score values and corresponding metadata in an ASCII format.

Every score value is described by the full set of key attributes, like its parameter, station id, month, step etc. Parameters describing one score value at one station are organised into a record. Each record corresponds to one score value. A record is a collection of pairs key=value separated by commas. A record spans one line. Value of the key which is not given in the current record is inherited from the previous record, except for the value parameter v which has to be present in each record.

A record has the following format:

centre=centre, model=model_id, d=yyyymm, t=time, s=forecast_step, st=station_id, lat=latitude, lon=longitude, lam=model_grid_latitude, lom=model_grid_longitude, se=station_elevation, me=model_orography_elevation, par=parameter, sc=score, th=event_thresholds, n=sample_size, v=score_value

If the value of any key is unknown it is encoded as "na". However, every record has to have a valid score_value (the "v" key); if the value is not known such record should not be included in the data file.

Example

For information on how these reports are assembled at ECMWF please refer to ECMWF implementation notes.

centre=ecmf,model=hr_0001,d=201602,t=0,s=0,st=97146,lat=-4.1,lon=122.43,lam=-4.147,lom=122.484,se=50,me=163,par=tcc,sc=ct,th=2/6,v=0/0/0/0/0/7/0/0/21
t=3,s=3,v=0/0/0/0/0/4/0/0/24
t=6,s=6,v=0/0/0/0/0/2/0/0/26
t=12,s=12,v=0/0/0/0/0/6/0/0/22
t=15,s=15,v=0/0/0/0/0/3/0/0/25
t=18,s=18,v=0/0/0/0/0/4/0/0/24
t=0,s=24,v=0/0/0/0/0/5/0/0/23
s=0,sc=mae,th=na,v=60.92
t=3,s=3,v=59.92
t=6,s=6,v=62.01
t=12,s=12,v=60.59
t=15,s=15,v=59.81
t=18,s=18,v=62.08
t=0,s=24,v=65.01
s=0,sc=me,v=-60.92
t=3,s=3,v=-59.92
t=6,s=6,v=-62.01
t=9,s=9,n=26,v=-66.37
...

Values of keys

centre (4-characters string) is the WMO identifier of the originating centre (ammc, cwao, ecmf, edzw, egrr, kwbc, lfpw, rjtd, rksl, rums etc);
model_id (a string not containing a comma or vertical bar) is free model identifier assigned by the originating centre (to distinguish between potentially different models provided by the centre);
yyyymm is the month of the mean, where yyyy is the year and mm is the month (01-12);
time is the validity time (in hours UTC) of the forecasts verified;
forecast_step is the length of the forecast (in hours);
station_id (a number) is the WMO ID of the observation station verifying the forecasts;
latitude is the latitude of the observation station verifying the forecasts;
longitude is the longitude of the observation station verifying the forecasts;
model_grid_latitude is the latitude of the model grid point used to extract the forecast at the observation location (*);
model_grid_longitude is the longitude of the model grid point used to extract the forecast at the observation location (*);
station_elevation is the elevation of the observation station above the mean sea level in meters;
model_orography_elevation is the elevation of the model orography at the observation location (*);
parameter is the verified model output parameter:

score is the name of the verification score or statistic:

score

description

me

mean error (bias)

mae

mean absolute error

rmse

root mean square error

ct

contingency table values,
scanned from bottom up, then left to right

The rank of a contingency table is defined by number of values in the key event_thresholds
(rank_of_ct = len(event_thresholds)+1)

NB. Please note how those contingency tables are constructed.
For an event like "10m wind speed >15m/s" a corresponding contingency
table would traditionally be

	obs	non-obs
fc	H	F
non-fc	M	R

but following the definition of 2x2 contingency table on the left the table should be

	ob<15	15≤ob
fc<15	R	M
15≤fc	F	H

2x2	ob<th1	th1≤ob
fc<th1	A	B
th1≤fc	C	D

v=C/A/D/B

3x3	ob<th1	th1≤ob<th2	th2≤ob
ob<th1	A	B	C
th1≤ob<th2	D	E	F
th2≤ob	G	H	I

v=G/D/A/H/E/B/I/F/C

4x4	ob<th1	th1≤ob<th2	th2≤ob<th3	th3≤ob
ob<th1	A	B	C	D
th1≤ob<th2	E	F	G	H
th2≤ob<th3	I	J	K	L
th3≤ob	M	N	O	P

v=M/I/E/A/N/J/F/B/O/K/G/C/P/L/H/D

event_thresholds is a value or values of threshold(s) defining events for contingency tables; number of values in event_thresholds defines the rank of the contingency table; multiple values are separated by a forward slash /

event_threshold	description
15	threshold for a 2x2 contingency table, e.g. if par=ff10m this is an event of 10m wind speed
5/10/15	thresholds for a 4x4 contingency table for 10m wind speed

sample_size is number of observations used to compute the monthly mean at the given station;
score_value is the value or values of the score mean computed based on the forecast initiating at time time UTC, verifying in the month yyyymm, for the forecast length forecast_step hours; in case of n_xn contingency table these are the n² values delimited by forward slashes (see the score table above); if possible, the value should be printed using the printf string format specifier "g" (or equivalent).

Remarks

(*) If the model grid or orography changed during the reported month (due to a model upgrade etc) the values of lam, lom and me should be those of the latest model run.