Surface Data Assimilation with the Canadian Fast-Cycling High-Resolution Ensemble Kalman Filter

An hourly-cycling high-resolution Ensemble Kalman Filter (HREnKF) has been implemented for southern Ontario area to assimilate Meteorological Terminal Air Report (METAR) data, in addition to the conventional observations assimilated operationally at Canadian Meteorological Center (CMC). There are more than 200 METAR stations widely distributed in the 900 km x 900 km domain, reporting surface weather conditions, such as wind at 10 meters height and temperature at 2.5 meters, hourly or even more frequently (i.e. SPECI reports). In order to incorporate such a large amount of information into numerical prediction, a 96-member hourly-cycling HREnKF was developed to run in research mode at CMC to assimilate those data into a 2.5-km resolution Global Environmental Multiscale - Limited Area Model (GEM-LAM).

In this study, background error statistics close to the surface are examined above all. Firstly, due to the restriction of surface layer parameterizations, the background ensemble spread is usually small near the surface, which gives much more weights on background fields than surface observations, thus preventing data from being assimilated. In order to increase surface ensemble spread, 24-member ensemble analyses at 2.5km resolution produced by the Canadian Land Data Assimilation System (CaLDAS) are used as a subset of surface boundary conditions. Additionally, geographic parameters, such as roughness length, are also perturbed. Those efforts are important for maintaining sufficient surface ensemble spread, as well as yielding reasonable forecast error correlations. Secondly, the background error correlations, especially in the vertical direction are studied, since it is crucial for the HREnKF to use surface observations to correct higher model layers. Surface data assimilation benefits the entire model in two ways: driving higher levels during model integration (forecast step of HREnKF); and correcting higher level errors through spatial error correlations during the analysis step. The later requires significant vertical error correlations to transfer information from surface to upper levels.

Having thus adjusted HREnKF's near surface background error statistics, an ensemble of analyses are generated by HREnKF, and corresponding ensemble forecasts are produced. The impact of surface data assimilation on analyses and forecasts are verified against precipitation and METAR observations. Results from this study highlights the fact that background surface error statistics are different on water and on land due to distinct surface properties. Ensemble CaLDAS analyses and geographic parameter perturbation we found beneficial for realistically increasing ensemble spread near the surface. Although the vertical error correlations are different for different model variables, they are significant enough to support the correction of higher model levels by assimilating surface data. Further verification of analyses and forecasts will be presented at the conference.