Diagnosis of very short range forecasts errors with the ALADIN limited area model

(Abstract for the workshop on the use of NWP for nowcasting, 24-26 October 2011, Boulder, Colorado)

Gergely Bölöni

A set of ensemble data assimilation (EDA) experiments have been run with the ALADIN
limited area model (LAM) at the Hungarian Meteorological Service (HMS), with the purpose of
diagnosing the contribution of uncertainties (originating from errors in the analysis, lateral boundary
coupling, and the model itself) to the full forecast errors on the very short range. Analysis uncertainty is
assessed via explicit random perturbation of the input observations used in the assimilation scheme
(atmospheric 3DVAR and surface OI) and via implicit perturbations of the background originating
from the previous analysis perturbations. The uncertainty in the lateral boundary conditions is
accounted for by using perturbed members from a global (ECMWF) EDA system for the coupling of
the LAM EDA members. The representation of the model error uncertainty is tackled by using different
physical parametrization packages in the LAM EDA members with main differences in the convection
and micro-physics.

The relative importance of the above error contributions is diagnosed primarily for 6, and partly
for 3 hour forecasts. Given the fact that the diagnostics are computed in spectral space, the dependence
of the above error contributions on the spatial scale has been assessed too. Results show, that very short
range forecast errors (below the range of 6 hours) originate primarily from the analysis uncertainty and
from the model error (physics uncertainty), especially on the small spatial scales, while errors implied
by the lateral boundary conditions are of secondary importance for this forecast range.

Note that the above study was done with the main goal of diagnosing, which uncertainty
representations are the most important while simulating background errors for the computation of the
background error covariance matrix used in the ALADIN 3DVAR assimilation. From the nowcasting
point of view, such diagnostics might precise our picture about how the quality of very short range
forecasts depend on data assimilation-, coupling- and physics developments.