 |
GEMPAK / N-AWIPS
Displaying Ensemble Grids
|
Why use Ensembles?
The relative skill of various forecast models decreases with time as errors
in the model accumulate. The initial state of a model is dependent on
measurements to provide a sample of the actual conditions, and this state
is just one possible solution which depends on the individual observation
location, time, and instrument uncertainty. By perturbing the initial state of the model
slightly, the model can simulate the effect of initial uncertainty on the
forecast. A set of forecasts based on a number of perturbations are referred to as
ensemble members. Generally, the skill of the model can be improved to further forecast
periods by combining the ensemble members to provide a forecast, while at the same time
providing a quantitative measure of the spread in possible outcomes. We assume that
no one outcome will be perfect because the initial state of the model will not be perfect;
however, by combining the results, sensitivity due to the input measurements themselves can
be moderated.
What types of Ensemble data are there?
- control (initial analysis and/or forecast)
- member (perturbations of the control run)
- average (computed from members)
- spread (computed from members)
- probabilistic (computed from members)
Special ensemble functions
GEMPAK provides a special set of functions, all of which are named beginning with ENS_,
to do specific calculations over multiple members of an ensemble. The constitution of
the ensemble is specified as a GDFILE entry by listing file names and aliases, separated by
commas and enclosed in curly brackets ({}). For specific functions available see the
GPARM online documentation.
Example
The GFS model provides deterministic output at the 72 hour forecast time for 6 hour
accumulated precipitation (P06M) and boundary layer CAPE as shown in GDPLOT2 using:
GDFILE = gfs004
GDATTIM = f072
GLEVEL = 0 ! 180:0
GVCORD = none ! pdly
SCALE = 0
GDPFUN = p06m ! cape
TYPE = f ! c
CONTUR = 3/3
CINT = 300
LINE = 2/1/2
FINT = .25;2.5;6.35;12.7;19.05;25.4;31.75;38.1;44.45;50.8;63.5;76.2;101.6;127;152.4;177.8
FLINE = 0;21-30;14-20;5
The figure above shows several large areas of precipitation with low CAPE values. We
also see several areas with large cape values and little precipitation.
By utilizing the ensemble members, we can quantify the probability of precipitation
exceeding .25mm (red contour lines), and CAPE values exceeding 500 J Kg^-2 (yellow
shading) using 20 members of the global ensemble forecast system (gefs) in GDPLOT2 using:
GDFILE = {gefs}
GDATTIM = f072
GLEVEL = 0 ! 180:0
GVCORD = none ! pdly
SCALE = 0
GDPFUN = ens_prob(gt(p06m,.25)) ! ens_prob(gt(cape,500))
TYPE = c ! f
CONTUR = 3/3
CINT = 0.2
LINE = 2/1/2
FINT = ! .5;1.2
FLINE = ! 0;5/7
Observing the plot above, we can visually detect several regions where precipitation
probability and CAPE values might suggest likely areas of thunderstorm activity
where the two contour regions intersect. We can quantify the combined probability
by using the logical operator AND() to compute the combined probability of
both conditions as shown below:
GDFILE = {gefs}
GDATTIM = f072
GLEVEL = 0
GVCORD = none
SCALE = 0
GDPFUN = ens_prob(and(gt(p06m,.25),gt(cape@180:0%pdly,500)))
TYPE = f
CONTUR = 3/3
CINT =
LINE =
FINT = .1/.1
FLINE = 0;23-13/7
