Additional tools that can be used in prognostic assessment are offered below. These tools assist when the methods described above are providing little or inconsistent signals.
The Model Bias Web page at http://www.hpc.ncep.noaa.gov/mdlbias offers an interactive interface to display cumulative bias (high/low warm/cold) for a given model parameter over the past 5 or 10 consecutive 00Z or 12Z model runs.
For example, you can view where the GFS
has been systematically too high with heights over North America at fhr
72 (below).
Below is an example of the PW bias in the
Eta at fhr 72. The blue indicates the Eta has consistently under
forecast PWs by at least a quarter inch and the red indicates where the
Eta has over forecast the PWs by about .25" over the last 10 cycles.
You can also loop these bias images through
forecast hours and watch a bias "grow" with forecast time. Below
is a loop of the 500 mb height bias from fhr 24 through fhr 168 in 24h
increments. Note the growth in bias with increase lead time.
Since these images are averages of the most recent 5 or 10 forecasts for a given fhr, bias signals (shading) often tend to be muted when the pattern has been progressive over the past 5 or 10 forecast cycles.
A strong bias in a stable pattern implies a greater likelihood the bias is typical for the pattern. This means you can more confidently subtract the bias from the current forecast to use as a staring point. However, a strong bias signal in a pattern that has been progressive over the past 5 or 10 cycles implies the bias is event skewed. This means that one or two of the forecasts over the past 5 or 10 cycles busted so bad its contaminating the statistical output... and therefore you should not try to apply the bias as a correction to the current model output.
Explore this web site and view the on line help to get a feel for which bias statistics are available for use. This is a good objective tool to utilize when no other tool is offering you any "advice" on which way to lean for a preference.
Verification analysis is a tool which is
utilized more in the medium forecast range can also be of assistance in
the short time frame. Verification analysis shows how a model converged
on a solution over subsequent model runs. It is different than a
trend loop.
In this example, blue contours represents the verification and yellow represents forecasts. Note that we loop these from the a specific fhr back to fhr 00. This loop is a good way to estimate at what fhr the models start to go awry for a specific feature. GENERALLY the models do a decent job within 72 hours for synoptic patterns in the warm season but can start to loose their mind beyond 48-60 hours in a progressive pattern during the cool season.
Another tool that will assist in identifying
pattern regimes are standard deviation plots. Not only do these plots
provide a way to quantify the pattern in terms of departure for normal
(in standard deviations), but they can help identify unlikely model solutions.
Normalized Standard Deviation values more than +/-4 indicate an EXTREMELY
unlikely event.. or more realistically the model is out of whack.
This image is not relevant to our case,
however, it does show how you can view a 500 mb pattern in terms of deviation
from normal. The blue/red lines
show where the 500 mb heights are deviating
from normal in a -/+ sense. A great web page explaining more on how
to apply these operationally can be found at http://www.hpc.ncep.noaa.gov/training
Tie in this information when you conduct verification to increase the chance
you can anticipate how a model will behave in given pattern.
QPF BOMBS
During your model assessment you will
likely encounter a feature referred to as a "QPF bomb". QPF bombs
result from grid scale feedback in a model and occur primarily in
GFS output during the warm season. It is critical
for you to recognize this feature and understand why it occurs as a QPF
bomb can contaminate a model solution. You can find a very
short tutorial on these features at http://www.hpc.ncep.noaa.gov/qpfbombs
Final note on prognostic assessment. Occasionally, even after using the techniques above you are still left undecided on a preference for a solution. At this point a model choice can be made based on the intended model strengths and weakness. For example, it is known that the Eta is not designed to handle tropical systems. It may be better to go with the higher resolution model - especially if the forecast of mass fields from both the Eta and GFS are similar.
You must therefore keep up to date with model changes. How you ask ? The best way to do this is to