Gridded fields of surface variables are an effective and fundamental tool for meteorological analysis and prediction within the NWS operational community. They provide direct measurements of surface conditions, permit inference of conditions aloft, and often give crucial indicators of the potential for severe weather. Surface analyses are particularly valuable at the mesoscale where the frequency, completeness, and density of the surface data are unmatched among in situ observations.
The Rapid Update Cycle (RUC) Surface Assimilation System (RSAS) exploits the resolution of surface data by providing timely and detailed, hourly surface analyses updated twice per hour. Multiple runs per hour allow RSAS to first run earlier in the hour to provide more timely analyses (i.e. 5 minutes past the hour), and then later (21 minutes past the hour), to incorporate late arriving observations. RSAS is the only data assimilation system at NCEP to provide sub-hourly updates to its gridded output.
Other unique aspects of RSAS include speed, minimal disk space requirements, and a close fit to the observations. As a surface analysis-only system, RSAS produces a one-level, analysis-only grid and, therefore, requires very few compute resources. Also, because the system does not initialize a forecast model, the analysis is performed on the actual surface terrain and not along a model topography. Hence no model surface-to-station elevation extrapolations are required, all surface observations may be used, and the fit to the observations is maximized.
This bulletin provides an overview of RSAS, and describes updates to the system awaiting implementation at NCEP.
The domain and resolution configuration of the new version of RSAS represents the single biggest change in the system. Formerly, RSAS provided hourly analyses on a 60-km grid covering the 48 contiguous states and neighboring areas of Canada and Mexico. The new domain, however, incorporates a 15-km grid stretching from Alaska in the north to Central America in the south, and also covers significantly more oceanic areas than did the former domain.
Persistence (the previous hourly analysis) serves as the background for the analysis in areas where surface observations are dense. One-hour persistence provides an accurate forecast and allows the incorporation of previous surface observations into the analysis. It also assures continuity between analyses, especially near stations that report less frequently than hourly. Persistence, however, cannot be used in data-void or data-sparse areas. In these regions, gridded data from NCEP's Eta model are used as a background to ensure that the analysis does not stray far from reality. The Eta grids are linearly combined with 1-h persistence, using weights calculated to produce a smooth transition between data-dense and data-sparse areas. Verification statistics computed for parallel cycles of RSAS, one using a pure-model background, and the other a persistence/model blend, show that the use of persistence significantly improves the ability of the analysis to fit the observations, particularly in the western U.S.
Treatment of the model backgrounds is another area of significant upgrade in the new RSAS. To better fit the observations, for example, the RSAS altimeter analysis now uses an Eta altimeter grid (derived from Eta surface pressure and elevation grids) as its model background, rather than the Eta sea-level pressure grid that was used by the previous version.
Since rough terrain can complicate surface analyses, the RSAS analysis variables were chosen, whenever possible, to minimize the effects of varying terrain. Potential temperature, for instance, is analyzed instead of surface temperature because it varies more smoothly over mountainous terrain when the boundary layer is relatively deep and well mixed, a marked advantage during daytime hours. The major pressure variable is a sea-level pressure computed at each station from altimeter setting observations. Station pressures calculated from the altimeter settings are reduced to sea level using the 700-mb Eta temperature to estimate an effective surface temperature. This reduction generally provides smoother regional, diurnal, and seasonal variation since it avoids the use of actual surface temperatures, which are often unrepresentative of the surrounding conditions. Moreover, more data are available for analysis of the RSAS reduction because more stations report altimeter setting than report sea-level pressure. Similarly, to improve data density, RSAS calculates its own pressure change observations at each station by differencing altimeter reports, which are much more plentiful than pressure change reports.
RSAS also incorporates elevation and potential temperature differences in the optimal interpolation correlation functions used to model the spatial correlation of the surface observations. The resulting functions help to take into account physical blocking by mountainous terrain, and improve the representation of surface gradients. For an example of the effect of the RSAS correlation functions, and the influence of terrain, on a potential temperature analysis, click here. For an example of the ability of the correlation functions to influence the representation of surface gradients, click here.
Most observations contained in its domain are utilized by RSAS. These include standard METARs, Coastal Marine Automated Network (C-MAN) observations, and surface reports from fixed and drifting buoys, and ships, as well as surface observations from the NOAA Profiler Network. RSAS, as it exists now at NCEP, also has the capability to ingest and analyze non-NOAA surface mesonets. As soon as NCEP begins to receive such data, RSAS will be able to use it.
Sophisticated quality control (QC) checks are employed to help screen the surface observations. Observations failing the checks are not ingested or analyzed. The QC checks include validity, internal consistency, temporal consistency, and spatial consistency checks.
RSAS also produces easily accessible hourly, daily, weekly, and monthly QC statistics for various surface networks. The statistics are based on the results of the QC checks, and include frequency of failure for each station, as well as the RMS and mean errors of those failures. Statistical stratifications exist for manual METARs, maritime observations (buoys and ships), profilers, non-ASOS automated METARs, and ASOS METARs. The statistics for ASOS observations are used to generate QC bulletins for the ASOS Program Office, a service that RSAS has been providing since 1993. These statistics have proven useful for identifying ASOS and non-ASOS stations with hardware or software failures. For example, QC statistics in 1997 showed that the profiler surface station at Purcell, OK was reporting bad dewpoint temperature observations 85% of the time, with a persistent bias of approximately 20 degrees. With that information, the Profiler Control Center in Boulder was able to determine that the dewpoint sensor at Purcell had failed. The sensor was then immediately fixed, and the percentage of dewpoint observations failing the QC checks went back down to zero. Time series plots of hourly observational errors as calculated by RSAS QC checks have also proven useful in detecting hardware and software failures. In 1999, for example, time series plots of RSAS sea-level pressure errors indicated a persistent bias in the observations made by a surface station in Galveston, TX. Investigation into the problem revealed incorrect processing software as the cause of the bias.
RSAS also maintains two text files, a reject list and accept list, which allow the subjective override of the automated QC checks. Observations included in the reject list are not analyzed by RSAS, while observations included in the accept list are always analyzed by RSAS, regardless of the outcome of the automated QC checks. The QC statistics are not affected by the subjective intervention lists, with the sole exception that observations on the reject list will be labeled as "suspect" and not used to check the spatial consistency of neighboring observations. This allows personnel to continue to monitor the performance of the stations contained in the lists.
Statistics on the RSAS analysis fit to observations are given below for eight different geographic regions and ten variables (RSAS sea-level, NWS sea-level, altimeter, 3h pressure change, u-component of the surface wind, v-component of the surface wind, potential temperature, temperature, dewpoint temperature, and dewpoint depression) for the period from December 2000 to January 2002. For comparison purposes, statistics are also given for the 60km RSAS, and for Eta (grid 221 - the RSAS background grid) and RUC2 (grid 236) surface analyses. It should be noted that both the Eta and RUC2 are full three-dimensional assimilation systems with forecast models, and have the advantage of vertical integration and integration with the assimilating model. The Eta utilizes a 32km, North American-scale grid, the RUC2 a 40km, CONUS-scale grid. RUC2 surface analyses are available hourly, Eta surface analyses every 6 hours.
The statistical analysis indicates that even at 60 km, RSAS fits the surface observations, in nearly every case, more closely than either the 40 km RUC2, or the 32km Eta. There are several reasons for this. First, the RSAS correlation functions take into account elevation and surface potential temperature differences and, therefore, preserve surface gradients better. Also, because RSAS does not initialize a forecast model, the analysis is performed on the actual terrain and not along a model topography. Hence, no model surface-to-station elevation extrapolations are required, and all surface observations may be used. The new, 15km RSAS improves upon the 60km fit to the observations, in many cases significantly, and also provides surface analyses in Alaska, Canada, Mexico, and Central America.
In June 2001, the Reno, NV WFO was sent instructions on how to configure their AWIPS system to ftp and and display 15 km RSAS grids. Although the grids were available twice per hour, Reno chose to download them only once per hour (at approximately 25 minutes past the hour), after the second analysis update. Complete AWIPS capability (real-time loops, observation overlays, etc.) was available to the evaluating forecasters upon completion of the installation instructions. The Reno office was chosen for its surrounding complex terrain.
On July 3, 2001, the Reno Science and Operations Officer (SOO), Mary Cairns, filed a report which stated that after looking at the RSAS grids for several weeks, the Reno forecasters were in agreement that the new 15 km grids were much better than the 60 km operational grids, in that they provided detail not observable at 60 km. RSAS sea-level pressure was particularly popular. Mary concluded her report by asking for continued access to the grids.
NCEP's Aviation Weather Center was sent instructions on how to configure their experimental AWIPS system to ftp and and display 15 km RSAS grids in October of 2001. Their evaluation mainly focused on surface temperature fields using METAR observations and satellite IR sampled temperatures as the ground truth. Fred Mosher, the AWC SOO, reported that the new RSAS was able to correctly identify mountains and the corresponding cooler temperatures. For example, Fred reported that: in West Virginia, there was an observation of snow with an air temperature of 37 degrees F, while the old 60-km grids had temperatures in the low 40 range, the 15 km RSAS correctly had a temperature of 37 at the surface observing site with slightly cooler temperatures in the higher elevations. In the clear air in western North Carolina, the 15 km RSAS had the colder temperatures observed on the satellite IR temperatures in the mountains around Asheville, while the 60 km RSAS did not have any indication of colder temperatures there. Looking at the near offshore waters and using ship reports and satellite sea surface temperatures as ground truth, again the 15 km RSAS seemed to correctly capture the surface temperature conditions off shore much better than the 60 km grids.
Overall, Fred reported that the analysis grids provided by the new RSAS version are a "huge improvement" over the currently operational version, and that the "derived fields are much smoother and appear to have a much better definition of the underlying meteorology".
The 2001 DACFO report includes a recommendation that the 15 km RSAS grids being produced at NCEP be placed on the SBN, and delivered to the field through AWIPS.
Miller, P.A. and S.G. Benjamin, 1988: A Scheme for Analyzing Surface Observations over Heterogeneous Terrain. Preprints, Eighth Conference on Numerical Weather Prediction, Baltimore, Maryland, February 22-26, 1988, 178-184.
Benjamin, S.B. and P.A. Miller, 1990: An Alternative Sea-Level Pressure Reduction and a Statistical Comparison of Geostrophic Winds Estimated with Observed Surface Winds. AMS Monthly Weather Review, 118, 2099-2116.
Miller, P.A. and S.G. Benjamin, 1992: A System for the Hourly Assimilation of Surface Observations in Mountainous and Flat Terrain. AMS Monthly Weather Review, 120, 2342-2359.
Miller P.A., L.L. Morone, 1993: Real-time quality control of hourly reports from the Automated Surface Observing System. Preprints, Eighth Symposium on Meteorological Observations and Instrumentation, Anaheim, California, January 17-22.
Miller, P.A. and R.L. Fozzard, 1994: Real-time Quality Control Monitoring of hourly surface observations at NOAA's Forecast Systems Laboratory. Preprints, Tenth Conference on Numerical Weather Prediction, Portland, Oregon, July 17-22, 1994.