Considering ensemble based forecasting of let's say stream flow, which one is better; a forecast with high ensemble spread or low ensemble spread. Its a bit ambiguous: low ensemble spread could mean that all the ensemble members are agreeing and the forecast could be more close to reality but it could also miss the real value by big amount. On the other hand, high ensemble spread could mean that there is high uncertainty but the real value could be forecast within a the ensemble range. So what is a better forecast, high spread or low spread?
This is dependent on accuracy of the ensemble mean. It will be best that the higher the ensemble mean forecast skill is, the larger the ensemble spread is. Otherwise, the larger spread will be absolutely noise.
Hi Michael,
in addition to the answers already given:
It is often said that the spread of the ensemble members around the ensemble mean should equal the root mean square error of the ensemble mean forecast. However, this is only true if the forecast system and model are all perfectly realistic.
Another simple requirement for a good forecast system is to ensure that the total variance in the ensemble predictions (ensemble mean variance plus ensemble spread about the mean) matches the observed historical variability.
Best
Adam
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Atmosphere