More about ensemble methods
- PERTURBATIONS CAPTURE in part the UNCERTAINTY IN INITIAL CONDITIONS.
- UNCERTAINTY ALSO EXISTS IN MODEL PHYSICS.
- ALSO NEED TO INCLUDE MEMBERS THAT INCLUDE UNCERTAINY IN PARAMETERIZATION SCHEMES.
- ASSUMES MODEL IS UNBIASED WITH EACH SOLUTION HAVING EQUAL SKILL.
- USUALLY THIS IS NOT TRUE….
- During summer the SREF QPF ensembles cluster around the two models used.
- Half usually look like the eta, the other half usually look like the avn
- the quirks in the convective schemes often drive the ensemble expressed uncertainty, not the differences in initial conditions
- the model biases in handling convection show up in each perturbation for that model