An ensemble Kalman filter (EnKF) has been implemented for atmospheric data assimilation. It assimilates observations from a fairly complete observational network with a forecast model that includes a standard operational set of physical parameterizations. To obtain reasonable results with a limited number of ensemble members, severe horizontal and vertical covariance localizations have been used. It is observed that the error growth in the data assimilation cycle is mainly due to model error. An isotropic parameterization, similar to the forecast-error parameterization in variational algorithms, is used to represent model error. After some adjustment, it is possible to obtain innovation statistics that agree with the ensemble-based estimate of the innovation amplitudes for winds and temperature. Currently, no model error is added for the humidity variable, and, consequently, the ensemble spread for humidity is too small. After about 5 days of cycling, fairly stable global filter statistics are obtained with no sign of filter divergence. The quality of the ensemble mean background field, as verified using radiosonde observations, is similar to that obtained using a 3D variational procedure. In part, this is likely due to the form chosen for the parameterized model error. Nevertheless, the degree of similarity is surprising given that the background-error statistics used by the two procedures are rather different, with generally larger background errors being used by the variational scheme. A set of 5-day integrations has been started from the ensemble of initial conditions provided by the EnKF. For the middle and lower troposphere, the growth rates of the perturbations are somewhat smaller than the growth rate of the actual ensemble mean error. For the upper levels, the perturbation patterns decay for about 3 days as a consequence of diffusive model dynamics. These decaying perturbations tend to severely underestimate the actual error that grows rapidly near the model top.

SUMMARYIn this study several approaches for obtaining more accurate background-error covariances for atmospheric data assimilation are evaluated. Experiments are conducted by replacing the covariances in the operational threedimensional variational analysis system at the Canadian Meteorological Centre. In the current system, these covariances are computed using the so-called NMC method that is known to suffer from several deficiencies. The approaches evaluated in this study attempt to more realistically sample the probability distribution of background error by simulating (using a Monte Carlo approach) the error generated at each stage of the forecast-analysis process. The ensemble Kalman filter and a simpler approach applied to an existing forecast-analysis system are both used to generate these error samples. In addition, error samples are generated directly from the covariances of the operational system to allow the effects of sampling error to be quantified. Several strategies for estimating the full covariance matrix from a relatively small number of error samples are then employed. Approaches include the use of a spatially localized ensemble representation of the correlations that allows the usual assumptions of homogeneity and isotropy to be relaxed. In addition, the use of a weighted average between such a covariance matrix and a covariance matrix with homogeneous and isotropic correlations is evaluated. Several diagnostic results from the estimated background-error covariances are presented in addition to verification statistics computed from two-week forecast-analysis experiments. Modest forecast improvements are obtained by using the new background-error covariance estimates, mostly in the southern hemisphere. However, additional results suggest that further improvements may be gained by increasing the number of error samples and a preliminary quantitative estimate of the expected gain is computed.

An intercomparison of the Environment Canada variational and ensemble Kalman filter (EnKF) data assimilation systems is presented in the context of global deterministic NWP. In an EnKF experiment having the same spatial resolution as the inner loop in the four-dimensional variational data assimilation system (4D-Var), the mean of each analysis ensemble is used to initialize the higher-resolution deterministic forecasts. Five different variational data assimilation experiments are also conducted. These include both 4D-Var and 3D-Var (with first guess at appropriate time) experiments using either (i) prescribed background-error covariances similar to those used operationally, which are static in time and include horizontally homogeneous and isotropic correlations; or (ii) flow-dependent covariances computed from the EnKF background ensembles with spatial covariance localization applied. The fifth variational data assimilation experiment is a new approach called the Ensemble-4D-Var (En-4D-Var). This approach uses 4D flow-dependent background-error covariances estimated from EnKF ensembles to produce a 4D analysis without the need for tangent-linear or adjoint versions of the forecast model. In this first part of a two-part paper, results from a series of idealized assimilation experiments are presented. In these experiments, only a single observation or vertical profile of observations is assimilated to explore the impact of various fundamental differences among the EnKF and the various variational data assimilation approaches considered. In particular, differences in the application of covariance localization in the EnKF and variational approaches are shown to have a significant impact on the assimilation of satellite radiance observations. The results also demonstrate that 4D-Var and the EnKF can both produce similar 4D background-error covariances within a 6-h assimilation window. In the second part, results from medium-range deterministic forecasts for the study period of February 2007 are presented for the EnKF and the five variational data assimilation approaches considered.

An intercomparison of the Environment Canada variational and ensemble Kalman filter (EnKF) data assimilation systems is presented in the context of producing global deterministic numerical weather forecasts. Five different variational data assimilation approaches are considered including four-dimensional variational data assimilation (4D-Var) and three-dimensional variational data assimilation (3D-Var) with first guess at the appropriate time (3D-FGAT). Also included among these is a new approach, called Ensemble-4D-Var (En-4D-Var), that uses 4D ensemble background-error covariances from the EnKF. A description of the experimental configurations and results from single-observation experiments are presented in the first part of this two-part paper. The present paper focuses on results from medium-range deterministic forecasts initialized with analyses from the EnKF and the five variational data assimilation approaches for the period of February 2007. All experiments assimilate exactly the same full set of meteorological observations and use the same configuration of the forecast model to produce global deterministic medium-range forecasts. The quality of forecasts in the short (medium) range obtained by using the EnKF ensemble mean analysis is slightly degraded (improved) in the extratropics relative to using the 4D-Var analysis with background-error covariances similar to those used operationally. The use of the EnKF flow-dependent error covariances in the variational system (4D-Var or 3D-FGAT) leads to large (modest) forecast improvements in the southern extratropics (tropics) as compared with using covariances similar to the operational system (a gain of up to 9 h at day 5). The En-4D-Var approach leads to (i) either improved or similar forecast quality when compared with the 4D-Var experiment similar to the currently operational system, (ii) slightly worse forecast quality when compared with the 4D-Var experiment with EnKF error covariances, and (iii) generally similar forecast quality when compared with the EnKF experiment.

A major set of changes was made to the Environment Canada global deterministic prediction system during the fall of 2014, including the replacement of four-dimensional variational data assimilation (4DVar) by four-dimensional ensemble–variational data assimilation (4DEnVar). The new system provides improved forecast accuracy relative to the previous system, based on results from two sets of two-month data assimilation and forecast experiments. The improvements are largest at shorter lead times, but significant improvements are maintained in the 120-h forecasts for most regions and vertical levels. The improvements result from the combined impact of numerous changes, in addition to the use of 4DEnVar. These include an improved treatment of radiosonde and aircraft observations, an improved radiance bias correction procedure, the assimilation of ground-based GPS data, a doubling of the number of assimilated channels from hyperspectral infrared sounders, and an improved approach for initializing model forecasts. Because of the replacement of 4DVar with 4DEnVar, the new system is also more computationally efficient and easier to parallelize, facilitating a doubling of the analysis increment horizontal resolution. Replacement of a full-field digital filter with the 4D incremental analysis update approach, and the recycling of several key variables that are not directly analyzed significantly reduced the model spinup during both the data assimilation cycle and in medium-range forecasts.