Forecaster intervention with the ENS

It is often taken for granted that forecasters cannot improve on the ENS.  But forecasters can manually intervene by using their experience for a certain location.. They can be guided by verification of previous events to correct tendencies of the ENS to over- or under-forecast probabilities.  These modifications are often appropriate at coastal locations or in mountainous regions.  This is because local effects may be significant and/or the grid point nearest to the location that is used for the meteogram may not be typical nor appropriate. 


Taking account of other state-of-the-art NWP models - Grand and Lagged Ensembles

Ensemble forecasts give the most consistent guidance.  One should not rely on any individual result.

In addition to IFC products, information from other sources (primarily higher-resolution forecasts) can be used where available.  These can be spectral- or grid-point- based, global or limited area, hydrostatic or non-hydrostatic.  The technique of comparing and combining latest and previous solutions applies to all major state-of-the-art NWP models.  The differences in their average forecast qualities are less significant than the daily variability of the values.  Large variations in the results from other models might be important.   These may suggest an indication of extreme or hazardous weather.  The threat should be passed on to users but with a very low probability.

Occasions when there is a clear divergence between the latest and previous developments in the ensemble or ensemble mean occur increasingly rarely -  but if it does the forecasters is in a difficult position.  Nevertheless, there can be some agreement between the spread and jumpiness.

Combining and comparing results from different model runs can be achieved by:  

  • Grand ensembles:  An ensemble of model runs using different models but starting from the same data times.
  • Lagged ensembles:  An ensemble of recent past and latest model runs merged into a grand ensemble.  Lagged ensembles can include results from past ENS runs and/or from different models or both.  (See Fig6.3.5). 


Forecasters may have to assess the probability of an event by balancing probability information from each of several sources.  It is not unusual to have to balance information such as:

  • a 50% probability from the ENS.
  • a 30% probability from a different, but equally reliable, statistical or ensemble system.
  • five of the last six CTRL (or HRES) and/or other state-of-art models have forecast the event.

Forecasters should treat forecasts from different NWP models as part of a “multi-model ensemble”.  This has an advantage because the members differ slightly from IFS in their initial conditions and model characteristics.  Note, on average:    

  • the best NWP model of any kind is not necessarily the best on a particular day.
  • an NWP model that has recently performed significantly better or worse than other models (of about the same average skill) is not likely to continue to do so.

It is difficult to determine the “model of the day” from:

  • several NWP model forecasts.
  • consecutive forecasts from the same model.