• All forecasts are affected by errors. The errors originate from observations (lack of observations, errors affecting observations), model (non resolved scales, imperfections in the model theory and numerics, etc) and the chaotic nature of the atmosphere. 

• Forecast errors increase throughout the forecast. The rate of increase is situation dependant

• As the errors increase with forecast ranges the predictable scales become increasingly large.
• Errors affect many aspects of the forecast: Occurrence of events, Spatial evolution of features, Timing evolution, Magnitude/intensity. Monitoring of the forecast evolution provides clues on the predictability of each aspect.
• Probabilistic forecasts are the best tool to assess uncertainty and level of confidence in the forecasts. The use of a single forecast is truth hiding especially for long forecast ranges.

Guidelines:

• Improve Knowledge of the model being used (Forecast User Guide, IFS known issues, Severe event catalogue, Model upgrades, Forecast verification)
• Monitor the forecast evolution and continuously identify predictable scales

• Dampen forecast jumpiness

• Continuously estimate the overall confidence

• Pay attention to alternatives in particular those related to extreme events

• HRES is best used when viewed as part of the ensembles

• Avoid over interpreting non predictable scales (forecast range dependent) even in case of good consistency

• Medium range forecasts should be probabilistic. Detailed deterministic forecasts should get more weight at shorter forecasts ranges only if the ensemble uncertainty is reduced.