Gergely Boloni (boloni.g++at++met.hu) - Andras Horanyi (horanyi.a++at++met.hu)
Computation of background error covariances is required in the context of data assimilation in order to estimate the solidity of first guess information obtained by an earlier model forecast. As in many NWP models, in ALADIN as well the NMC method was proposed to be used for the definition of background errors, that is based on two forecasts with different range (typically 36 and 12 hours) but valid for the same time. The background error statistics are computed then on the differences of the two forecasts accumulated for a few months (typically 3). After the first computations it turned out that the NMC method (originally used in global models) doesn't work efficiently on those smaller scales which are represented in ALADIN, rather on larger scales already treated by ARPEGE analysis. Accordingly, some efforts were made to modify the method in order to increase its efficiency on smaller scales. Finally the so called lagged-NMC method was found convenient to derive mesoscale background error statistics (Siroka at al., 2001). The lagged-NMC method is very similar to the original one (standard-NMC) described above with the modification that the lateral boundary conditions are exactly the same for the two forecasts taking part in the computation of departures. Consequently, providing the forecast differences, the large-scale information are diminished giving a relatively bigger weight to smaller scales. In 2000-2001 both standard and lagged background error statistics were computed in Budapest for the Hungarian domain of ALADIN (ALADIN/HU). Computation of standard statistics gave very similar results comparing to those obtained over other ALADIN domains. Using the lagged method a sensitivity study was also performed, that is exploring the sensitivity of background error statistics to the forecast range and to the difference of forecast ranges used in the NMC method. It means that not only 36h-12h, but all possible combination of forecast differences were created and accumulated as a base of the statistics. As the model is integrated until 48 hours the forecast range can vary between 6 and 48 hours and the difference between them from 6 to 42 hours. The main goal of this study was to choose the optimal statistics for the 3d-var scheme used in ALADIN and also to obtain some information about the predictability properties of the model. Several theoretical conclusions were made concerning the efficiency of the different statistics mainly based on the analysis of variance spectra (Horanyi and Boloni, 2001), however, running single observation experiments, it became clear that the lagged-NMC method is not as powerful in the case of ALADIN/HU as it was shown for ALADIN/LACE. We have the feeling that it is the consequence of the small difference between the resolutions of ALADIN/HU and its driving model ALADIN/LACE, which leads to a strong impact of the coupling information even on smaller scales, which means a too strong error variance reduction in the context of lagged-NMC forecast differences. In order to get a preliminary information about the impact of ALADIN/LACE on ALADIN/HU through coupling, the parallel verification of the two models was carried out that gave very similar scores (Boloni, 2001), thus confirming the results of the single observation experiments.
References
(1) Siroka, M., C. Fischer, V. Cassé and J.-F. Geleyn, 2001: The definition of mesoscale selective forecast error covariances for a limited area variational analysis. Meteor. Atmos. Phys. special issue of the SRNWP Workshop on high resolution modelling, Offenbach, 25-27/10/1999, Submitted.
(2) Horanyi, A. and G. Boloni, 2001: Lagged constant coupling background error statistics: preliminary results for the ALADIN/HU model. Proceedings of the 10th ALADIN workshop "on scientific developments" , 7-8 June 2001, pp. 113-119.
(3) Boloni, G., 2001: Further experiments with the combination of 3DVAR and blending by DFI: tests using incremental digital filter. RC LACE internal report. Available from the author.
|
Home |