Diagnosis of Stochastic Physics with Ensemble Data Assimilation

Abstract:

Stochastic physics is generally essential to obtain sufficient error-growth rates in global atmospheric models. It is generally tuned to maintain overall statistical 'reliability' in ensemble forecasts. However, for ensemble forecast systems to be worthwhile, they must be reliable in a flow-dependent sense too. Non-linear interactions make the root causes of deficiencies in such flow-dependent reliability more difficult to trace at forecast ranges >2 days, and so we discuss a new diagnostic that can be applied at the short ranges associated with data assimilation. Results demonstrate that the current stochastic physics scheme in the ECMWF model may be over-active in sub-tropical anticyclone regions, where mid-tropospheric meteorology is largely a balance between time-mean decent and radiative cooling. On the other hand, the stochastic physics may be under-active in convective regions. This later result has implications for the prediction of heightened forecast uncertainty over Europe following mesoscale convection over North America. The success in localising deficienices in stochastic physics suggests that this diagnostic it will be useful for the future assessment of more stochastically-formulated physical parametrizations in future.