May well discard this section.

Using Deterministic and Probabilistic Forecasts

IFS models produce a wide range of output products available online through the website in chart form or or by dissemination or extraction in a GRIB format.  Presentation through ecCharts allows output data to be combined and displayed in a user-friendly way tailored to the needs and requirements of the user.  An individual ensemble member forecast or the HRES forecast is a deterministic forecast. 

Relation between deterministic and probabilistic forecasts 

The ECMWF forecast products can be used at different levels of complexity, from categorical, single-valued forecasts to probabilistic, multi-valued forecasts.  They can be used as guidance to forecasters but also to provide direct input to elaborate decision-making systems.  The choice largely depends on user demands but is also influenced by the traditions, and constraints, of the particular meteorological service.  However, the main aims of forecasters in interpretation of available data is to:

  • identify the predictable scale,
  • dampen forecast jumpiness,
  • estimate the overall confidence
  • draw attention to possible alternative developments, in particular those which involve extreme or hazardous weather events.

Some guidance is given on how best to use the forecast products using ensembles.

Issuing reliable categorical weather forecasts is of crucial importance for any meteorological service during normal weather conditions.  It builds trust with the public.  If they have confidence in the ability of a weather service to successfully forecast conditions in normal weather conditions, they will be more likely to trust its forecasts, even probabilistic ones, in cases of extreme weather.  The provision of categorical and probabilistic forecasts to the public and end-users therefore support and complement each other.

Categorical forecasts imply a confidence that may not be justified and ECMWF suggests a more probabilistic approach should be used.  Nothing undermines public confidence more than “jumpy” forecasts where forecasts change, sometimes radically, and in particular in connection with anomalous or extreme weather events.   A bad five-day forecast will be identified as such only after five days; a “jumpy” forecast will be identified immediately to the exasperation of the user.   Although ensemble forecasts must, by necessity, be "jumpy" to some extent, there is no reason to convey this “jumpiness” to the public by basing a forecast solely on the very latest NWP output.  This can best be avoided by making active use of uncertainty information derived from recent ensemble forecasts.

Probabilistic frameworks can and should be adapted whenever specific user requirements have to be taken into account; some examples are given in the appendix.  Any member of the ensemble could be considered deterministic, but in general the skill of a singe member is less than the skill of the ensemble mean and a single member gives no information on the confidence that may be placed on its results.

Differences between short range and medium range operational use of NWP

 There are some fundamental differences between how forecasters work with NWP model output in the short range and in the medium range.

  • Using the models:
    • In the short range, forecasters use real-time observations which have not been used by the NWP models (e.g. due to their late arrival) to determine which NWP guidance is verifying with observations or satellite data (the “Model of the Day” approach).  Forecasters then use their meteorological knowledge and experience to determine to what degree the NWP needs to be modified.
    • In the medium range, upstream influences mean this “Model of the Day” approach cannot be applied.  Instead, forecasters have to choose between, or combine information from, ENS and other NWP sources.
  • Considering errors:
    • In the short range, forecasters rarely have to question the existence of predicted synoptic systems; in the medium range such systems might not come into existence at all.
    • In the medium range forecast errors are usually dominated by non-systematic errors related to the positions and intensities of atmospheric systems, rather than systematic errors within the NWP model construction.
  • Measurement of skill:
    • In the short range, the skill of forecasts is measured against persistence.
    • In the medium range, the skill of forecasts is measured against climate.
  • Technique selection:
    • In the short range, meteorological knowhow is more important. 
    • In the medium range, statistical knowhow is more important.