Using NCEP's Grid-point Statistical (GSI) Interpolation Analysis System run in the 2DVar mode, the RTMA performs hourly 5km-resolution analysis of surface observations to produce CONUS-National Digital Forecast Database (NDFD) grid matching fields of temperature, specific humidity, dew-point temperature, wind, and surface pressure. In addition, the system maps the NCEP Stage-II precipitation and GOES sounder effective cloud amount to the 5km resolution NDFD grid. The RTMA-2DVar uses the one-hour forecast from the Rapid Update Cycle (RUC) downscaled to the NDFD grid as its first guess. That downscaling comprises an interpolation of the 13km resolution RUC fields to the NDFD grid followed by a reduction to the NDFD topography. The background error covariance model used in the 2DVar analysis is by design anisotropic, with structure functions exhibiting a controlled degree of correlation with the underlying topography. Such an implementation is made possible by the use of recursive filters and the sequential line-filtering "triad" algorithm. For each analyzed weather element, the RTMA also provides an estimate of the corresponding analysis uncertainty. Approximately 80% of all the observations assimilated in the RTMA-2DVar come from mesonet networks. The remaining represent synoptic, METAR, buoy and C-MAN observations, as well as satellite SSM/I wind speeds and QuickSCAT ocean winds.
A detailed description of the RTMA and results of its evaluation will be presented. The challenges associated with the assimilation of the high density, often poor quality-controlled mesonet data will be discussed. A brief reference will also be made to the on-going work aimed at improving the system, including the use of variational quality control for the observations, parameter tuning via cross-validation, and the use of alternate covariance shapes for the background errors.