Section 7.4 Broad Guidelines

Broad Guidelines

Broad guidelines for interpretation of a series of forecast results

On average, the best practise advice is:

• Make good use of the ensemble, especially (but not only) at longer lead-times.
   
• Give more weight to the more recent forecasts, but don't exclude results from previous runs.
• Larger spread implies larger errors.
• Jumpiness is not a good indicator of likely error.
• Dynamical sensitivity can increase jumpiness at short ranges and should be investigated.
  
• Do not extrapolate trends.
• Make more use of the ensemble mean, rather than the ensemble control member, especially at longer lead times (say ≥ ~ 4 days), to reduce jumpiness.
• At short ranges the ensemble mean (EM) and the ensemble control member tend to “jump together” more often.  The strategy of following the ensemble mean, rather than ensemble control, is less beneficial at short ranges.
• Note that strong gradients are always weakened in the ensemble mean.
• Consider each case according to the synoptic situation and other aspects (eg timing, rapid development).
   

It is important that customers are given consistent advice in the form of:

• the most probable forecast,
• possible alternative developments,
• the probability that a given event will occur,
• the probability that an extreme or hazardous event will occur, 
• general consistency without sudden changes from previous advice. 

This maintains user confidence while retaining forecaster integrity.

Do the opposite to the computer!

General advice could be summarized in a rather unexpected manner; weather forecasters should not try to “compete” with the NWP output on its own terms, but rather do the opposite.  The ensemble control or any one of the ensemble members could be thought of as being deterministic, but it the strength of the ensemble that all members are equally probable and it is the derived probability products which are best able to give a good forecast. The shortcomings of using "deterministic" output  are:

• Deterministic NWP output provides highly detailed synoptic scenarios, irrespective of how predictable they are.  Forecasters are advised not to do the same.  With increasing forecast range, they should not try to add detailed information to the NWP, but rather remove information, providing progressively fewer unreliable details at longer lead-times in their own forecasts.
• Deterministic NWP has to change run by run.  These changes can be quite profound, in particular in the medium-range.  Forecasters are advised to try to dampen any forecast jumpiness in any deterministic forecast guidance that they issue in order to increase the end-user’s confidence.   Forecast jumpiness can also be used constructively, by indicating possible alternative developments.
• Deterministic NWP gives an impression of very high certainty.  Forecasters should also make use of uncertainty.  The public and customers are better served by having an uncertain weather forecast presented as such, rather than with misleading certainty.  This will not only make the weather forecast service much more beneficial to decision-makers, but also make the difficult task of weather forecasting more gratifying for the forecasters themselves.

These three rules apply particularly to extreme weather events, such as cold outbreaks, heat waves, heavy rainfall and strong winds.  Extreme events are more susceptible to unrealistic details and “forecast jumps” and for which uncertainty indices or probabilities are the most appropriate ways to convey forecast information.