Implementation of a 3D variational data assimilation system at the Canadian meteorological centre. Part II: The regional analysis

Abstract

This paper describes the implementation of the 3D variational (3D‐var) analysis in the Regional Data Assimilation System (RDAS) of the Canadian Meteorological Centre. The RDAS, a 12‐h data assimilation cycle, is run twice daily to provide analyses to the variable resolution Global Environmental Multi‐scale (GEM) model. The same incremental 3D‐var algorithm is used for both the regional and global data assimilation systems. In this algorithm, the innovations are calculated with respect to the full‐resolution model while the analysis increments are calculated at a lower resolution on a Gaussian grid. Background error correlations for the global and regional data assimilation systems are examined. It is shown that the resolution of the analysis increments is determined to a great extent by the horizontal correlation lengths. Although the horizontal resolution of the regional model is three times higher than the global model, the correlation lengths for both models are similar. Consequently, the same horizontal resolution for the analysis increments is used in both data assimilation systems. A pre‐implementation evaluation showed that the RDAS maximizes the coherence between the analysis and the forecast model. This results in a higher‐resolution and more consistent analysis with respect to the regional model. The forecasts issued from the RDAS are generally improved, especially the temperature and the geopotential above 300 hPa. The RDAS also reduces the precipitation spin‐up observed during the first 12 hours when initiated with an analysis from the global data assimilation system. Finally, the impact of the digital filter on the analysis from the RDAS is small, indicating that the regional analysis is already well balanced.