Changes to the NCEP Meso Eta Analysis and Forecast System: Modified cloud microphysics, assimilation of GOES cloud-top pressure, assimilation of NEXRAD 88D radial wind velocity data
Brad Ferrier*, Ying Lin, David Parrish, Manuel Pondeca*, Eric Rogers, Geoffrey Manikin, Michael Ek+, Mary Hart*, Geoffrey DiMego, Kenneth Mitchell, and Hui-Ya Chuang*
Mesoscale Modeling Branch, *Science Applications International Corporation, and +University Corporation for Atmospheric Research
Environmental Modeling Center, National Centers for Environmental Prediction
A series of changes to the NCEP Mesoscale Eta Analysis and Forecast System are described. These changes are designed to address the following issues:
The changes will consist of the following:
2.0 Extension of 06/18Z Eta runs to 84 h
At the request of NWS field forecasters and NCEP service centers, the 06z/18z eta runs, currently run out to 48-h, will change to be identical to the 00z/12z forecasts, which are run to 60-h in one model execution followed by an extension to 84-h. The so-called "off-time" Eta output beyond 48-h will only be available from the NCEP and NWS ftp servers at this time until Data Review Group (DRG) requests and proper user notification can be accomplished.
3.0 Modifications in Cloud Microphysics and Radiation
This section describes changes in the cloud microphysics and radiation. It first describes the major changes, of interest to most readers, followed by a list of minor changes for those interested in more detailed code documentation.
3.1 Cloud microphysics
In November 2001, the Eta model resolution was increased from 22 km / 50 levels to 12 km / 60 levels (Rogers et al., 2001) and a new grid-scale microphysical parameterization (Ferrier et al., 2002) was implemented. Although timely changes were made in the precipitation assimilation method of Lin et al. (2001) to account for the new microphysics, the model restart files remained unchanged. This meant that model input and output and the posting of forecast fields were not treated in a manner consistent with the new microphysics, but rather with the previous microphysics package (Zhao et al., 1997). The forecast code has now been changed to include I/O for the expanded model restart files, which include additional storage arrays used in the Ferrier microphysics. Old arrays associated with the Zhao microphysics were removed. These modifications allow the new clouds simulated at the end of a 3-h forecast to be read in at the start of the next 3-h forecast during the EDAS, after the assimilation of the observations using 3DVAR. As part of the effort to include full cycling of cloud quantities, the model fields were tested and found to be bit-identical between forecasts initiated from restart files written at different forecast times.
The following minor changes were made to those routines that impact the microphysics:
Longwave radiation is now updated hourly along with shortwave radiation, not every two hours as in the current operational Eta. The following changes were also made in the optical properties of clouds as input to the radiative calculations:
The convective and grid scale cloud fraction changes will increase the occurrence of partly cloudy skies, and a more recent modification of this cloud fraction scheme in the parallel will increase the occurrence of total overcast conditions. (see recent 18-h total cloud cover forecast comparing the operational run [left] and the parallel [EtaX] run [right]). A recent modification of this cloud fraction parameterization in the parallel, which was made to reduce a 2-m cold bias during the winter over northern latitudes (see Sec. 10.4) will also increase total overcast conditions. An example of an 18-h total cloud forecast from the operational run (labeled "OPS", left), an older version of the parallel (labeled "ETAX", middle), and the latest version of the parallel ("ETAX8", right) shows a net increase in cloud coverage in the parallel run (right) compared to the operational run (left). Some of the changes listed above are a temporary fix to reduce the documented high warm season bias in incoming surface solar radiation. A longer-term solution is being sought, which includes incorporating into the Eta model the more sophisticated radiation package recently implemented in the Global Forecast System (GFS), making substantial changes in the convective scheme to allow more low-to-midlevel cloudiness during warm season conditions, and modifying the grid-scale microphysics to slow the rates of sedimentation of ice particles from high-level clouds. Changes are also being made to reduce the adverse impacts of parameterized shallow convection on preconvective forecast soundings (Baldwin et al., 2002).
3.3 Other minor changes
The following is a list of minor changes made in the Eta model that are technical in nature, which are described in more detail on the web, and are not expected to have any discernible impact on the forecast guidance.
4.0 Post changes
4.1 Hourly grids
Contingent upon available resources at NCEP Computing Operations (NCO), the following output grids will be provided on an hourly basis out to 36-h forecast times:
4.2 New land-surface field locations
The Eta post now contains (on the native grid) a number of new land-surface-related parameters that are used or generated in the Eta model by the (Noah) land-surface model. Interested users should note that the number of posted land-surface-related fields are expected to increase in future implementations, and these new fields will be contained within a new GRIB Table 130 now being used. Table 130 includes:
Table 130 lists new fields from the land-surface model that are also being posted on the native grid. The number of posted land-surface fields are expected to increase in future implementations, and these new fields will be contained within GRIB Table 130. Additionally, the full complement of land-surface-related parameters are found in the wgrib inventory for Eta (native grid) output. Other output grids may only have a subset of this full inventory, e.g. NLDAS (North American Land-Surface Data Assimilation).
4.3 New atmospheric fields
Variables associated with the new microphysics are in the first five fields in this Table of new Eta post processor fields. Some of these new fields already exist in GRIB parameter Table 2, while others have been added to the new GRIB parameter Table 129. The mixing ratios of rain and snow are 3D fields interpolated to different pressure (p) levels at 25-mb resolution, and are posted from the native (eta) levels in the model. The three storage arrays from the microphysics (FRAIN, FICE, FRIME) are posted from the eta levels primarily for debugging and detailed case-study analysis.
4.4 Cloud base/top pressures
The parallel post algorithm has been changed with regard to the base and top pressures for all types of clouds. When convective and grid-scale cloudiness coexist at a grid point, the current operational post uses only the properties of the convective clouds. The new parallel post uses the lowest of the convective or grid-scale cloud-top pressure (highest cloud top heights), and the highest cloud-base pressure (lowest cloud base heights).
The bases and tops of both grid-scale and convective clouds are posted. As newer, more complicated cloud parameterizations are developed, it's useful to know whether the forecast clouds are grid-scale or convective in nature. Grid-scale clouds form at sufficiently large scales of motion to be resolved well in the model, whereas convective clouds form at smaller scales of motion that are not. Because forecast models have much greater skill in treating the former, it may be useful to know the locations of both types of model clouds to better understand the underlying physical processes of a forecast event. The level of confidence in a forecast tends to be inversely proportional to the relative dominance of parameterized convective processes over more explicitly resolved grid-scale processes.
We now post the bases and tops of both the deep, precipitating convective clouds and the shallow, nonprecipitating convective clouds. The former are associated with thunderstorms and convective showers, the latter with stratocumulus and fair weather cumuli near the top of the boundary layer. The convective scheme used in the Eta model (Janjic, 1994) treats them somewhat differently. Forecasts of shallow convection in the Eta can cover very large areas, particularly over land during the warm season. The National Severe Storms Laboratory (NSSL) and the Storm Prediction Center (SPC) have identified aspects of the Eta's shallow convective parameterization that adversely affect the vertical structure of forecast soundings prior to the onset of deep convection, and the timing and location of the predicted convective storms. Having these fields available to forecasters may help them make better forecast decisions.
4.5 Precipitable water
Calculations of precipitable water in the post now include only the vertical integral of water vapor (specific humidity), not the sum of water vapor and total condensate. This change was made to reflect the definition of precipitable water in the Glossary of Meteorology. Instead, the total depths of cloud water, cloud ice, rain, snow, and total condensate (the sum of all four hydrometeor types) are each calculated over the atmospheric column in the same way as precipitable water. All of these fields are expressed in terms of the equivalent depth of liquid water in mm.
4.6 Extension of freezing level below ground
Based on requests from NCEP/HPC, Alaska and Western Regions, our current procedure of setting no freezing level lower than the model terrain height is changed to produce continuous freezing-level heights. Instead, if either highest or lowest freezing level is below the elevation of the model terrain, the freezing-level height will be estimated by extrapolating below ground the temperature in the first atmospheric layer to 0C assuming a standard lapse rate of 0.0065 C m-1. In this way, freezing level heights will be continuous. The heights will be limited between a lower limit of -1000 m below sea-level and an upper limit of 99999 m, both of which are consistent with the GFS postprocessor.
4.7 Other changes
Additional changes made in the post include:
4.8 Links to examples of new atmospheric fields
The EtaX (parallel) forecast web site shows examples of total cloud-top pressure (i.e., the pressure at the tops of both grid-scale and convective clouds), convective cloud- base pressure, convective cloud-top pressure, and deep (precipitating) convective cloud-top pressure; total column-integrated liquid (cloud water and rain) and ice (cloud ice and snow); modified precipitable water field (specific humidity only); and the fraction of precipitation in frozen form.
5.0 Precipitation Product Assessments
5.1 Precipitation-type algorithm comparisons
Several NWS Science Operations Officers (SOOs) raised concerns last winter about the classification of precipitation type in the current NCEP algorithm, particularly over the Great Lakes and Upper Midwest regions. Although this is a topic of ongoing research, we urge forecasters to also look at the fraction of instantaneous precipitation in frozen form (snow, graupel, or sleet), which is calculated directly from the model microphysics. Efforts are underway to include this field within the AWIPS and todistribute it through the NWS Satellite Broadcast Network (SBN). An example is shown here comparing the model precipitation type (right) versus that obtained from the operational Baldwin et al. (1994) algorithm (left), for the 48-h forecast valid at 00Z 29 Jan 2003. The model predicted frozen precipitation (100%of precipitation) and the Baldwin algorithm predicted freezing rain over an area extending from SE Iowa eastward through northern Indiana. The 48-h forecast sounding at Davenport, IA shows the development of a saturated isothermal layer from 900 mb to 950 mb with a temperature very near 0C, in response to cooling and moistening that occurred over a deeper layer between 850 mb and 950 mb during the previous 6-h(see 42-h forecast sounding valid at 18Z 28 Jan 2003). This cooling and moistening is consistent with sublimation and possibly some snow melting, particularly near the beginning of the event, although larger-scale advection and other processes could have also contributed.
5.2 Precipitation-type meteogram locations
Forecast meteograms comparing precipitation type algorithms are available for over 1000 stations. These meteograms, derived from the operational Eta BUFR output, compare precipitation type predicted from the original Baldwin et al. (1994) algorithm, the current operational algorithm, the Ramer (1993) algorithm, and the fraction of frozen precipitation predicted from the model microphysics. The original version of the Baldwin algorithm used a check on the amount of area in the sounding with a wet-bulb temperature (Tw) above 0 C; snow is not allowed if this area exceeds 350 C m-1. The current scheme will not classify the precipitation as snow if the area in the sounding with Tw> -4 C exceeds 3000 C m-1. The Ramer algorithm is based on the ice fraction of precipitation reaching the ground. Rain is diagnosed if the Tw > 2C at the lowest level, and snow is diagnosed if Tw < -6.6C at all other levels. Readers can review the details of how the Ramer algorithm handles all other cases in the conference paper by Cortinas and Baldwin (1999).
5.3 Links to examples of new hydrometeor fields
The EtaX parallel web site shows examples of total column-integrated liquid (cloud water and rain) and ice (cloud ice and snow); modified precipitable water field (specific humidity only); and the fraction of precipitation in frozen form. The fraction of frozen precipitation is calculated from the model compared to the Baldwin classification of precipitation type, as well as instantaneous precipitation rates from the EtaX parallel.
6.0 Assimilation of GOES cloud-top pressure data
The hourly, 10-km cloud-top pressure data, derived from the GOES-8 and 10 sounder radiances, give us valuable information on what the model field should be like (i.e. no cloud) above the cloud top, but reveal little below the cloud top level. In the assimilation of these data, we use them mainly to remove spurious cloud above the observed cloud top level, while making minimal adjustment to the moisture field at the level of observed cloud top (if the model is sub-saturated there). This means that the cloud-pressure assimilation tends to make the model drier.
6.1 The approach
Prior to running each 3-h EDAS forecast, a preprocessing program reads in 3-h of GOES cloud-top data from the PREPBUFR file and distributes the observations into the appropriate assimilation hour and horizontal grid box within the Eta domain.
At each physics time step during the EDAS forecast, condensate (water or ice) is removed from the model above the GOES cloud top level, or removed from the entire column if the satellite data indicate that this point is cloud-free. The water vapor mixing ratio is also set to no more than grid-scale saturation for those grid points where the satellite observations indicate that no cloud is present. To prevent grid-scale saturation at these locations, water vapor is also removed by setting the specific humidity to ice saturation at -10C or water saturation at warmer temperatures. At the model level closest to the observed cloud top, subsaturated air is moistened at a rate that just brings it to saturation in one hour.
If there is a need to create a cloud layer in the precipitation assimilation procedure where no forecast cloud is present, it is created by moistening the air below the satellite-observed cloud top.
6.2 Impact of assimilating cloud-top pressure
The assimilation of cloud-top pressure data was tested by itself in the 32km Eta and in combination with other components in this implementation of the 12km Eta parallel (see Table of EtaX tests described in Sec. 9.0). An example of the impact on precipitation forecasts is the 24h forecast for the 12Z 6 Jul 2002 cycle, where small improvements in the forecast precipitation field can be seen in Minnesota, Nebraska, and Mississippi. 24h and 24+36+48+72h precipitation scores show generally improved precipitation forecasts (as measured by equitable threat scores), and a slightly lower precipitation bias, which is not surprising given that we are removing more water than we add back in.
Upper-air forecasts of wind, temperature, RH and height also appear to be improved by the assimilation of the cloud-top pressure information.
7.0 Assimilation of the Stage II/IV hourly precipitation
Since its implementation in July 2001, the EDAS precipitation assimilation uses the NCEP Stage II hourly precipitation analysis as the source of observed precipitation. In the new package, the NCEP Stage IV precipitation analysis will be the primary source of the precipitation input, supplemented by Stage II. Both analyses are produced at NCEP, and they are based on hourly radar and gauge observations.
7.1 Merging the Stage II/IV hourly analyses for precipitation assimilation
The NCEP Stage IV analysis merges the regional multi-sensor precipitation analyses produced by the twelve River Forecast Centers (RFC) with the CONtiguous US (CONUS). It benefits from some manual quality control performed at the RFCs, and is generally considered to be of higher quality than the Stage II product. It is, however, not as timely as the Stage II, which is produced at NCEP at approximately 40 minutes after the top of the hour. The timeliness of the Stage IV depends on the transmission of the regional analyses, which is generally delayed several hours. The latest available Stage IV analyses often have only partial coverage since analyses from RFCs arrive at different times of the day.
Prior to the model integration for each 3-h EDAS segment, the available Stage II/IV analyses for that time period are collected to produce a merged analysis by using Stage IV values when available (unless the point is within NWRFC domain - see below). Data gaps are filled in with available Stage II data. If there is no Stage IV data available at all for that hour, then Stage II alone is used. If both analyses are not available, which often occurs during the last hour of the operational EDAS due to time constraints, then no precipitation data is assimilated. Stage II data alone is used in the NWRFC domain (using the RFC domain mask), because their Stage IV analyses often contain spurious precipitation "bulls eyes" and appear to be less reliable than the Stage II analyses.
Here is an example of the Stage II/IV merged analyses for the 12Z 20030112 cycle. The pre-forecast data assimilation period covers 00-12Z 12 January 2003. During the first nine hours, the Stage IV data coverage is relatively complete. Stage IV coverage decreased for the hour ending 10Z and was filled in by Stage II data. No Stage IV data was available during the last two hours, ending at 11Z and 12Z, so Stage II analyses were used instead.
7.2 Impact of assimilating Stage II/IV data
The merging of the Stage II/Stage IV analyses for EDAS has been tested alone on the 32km Eta parallel and in combination with the other components of this implementation on the 12km Eta parallel (see Table of EtaX tests described in Sec. 9.0). During the first five weeks of testing on the 32km Eta, the contribution from NWRFC was not masked out in the Stage IV analyses, so the precipitation bull's eyes in the NWRFC domain made it into the merged analyses. Most of them were filtered out by the gross error check before the merged product was used in the assimilation.
The most significant impact from introducing the Stage IV analysis into precipitation assimilation is the improvement of the model precipitation field during the EDAS, as shown in the EDAS precipitation scores (blue: cloud top assimilation run; red: control run). Since the model's soil moisture field is driven by the amount of model precipitation during EDAS, this means the soil moisture field would also be improved. When daily accumulations (12Z-12Z) of the Stage II/IV analyses are compared to the CPC 1/8 degree gauge-based analysis, widely considered to be the most reliable measure of daily precipitation, sometimes the Stage IV is clearly better than Stage II, while at other times the Stage II is better. Although both analyses are comparable, our overall assessment is that the Stage IV analyses are better.
There is a small positive impact on the precipitation forecast. Impact on upper air fields is minimal.
8.0 Assimilation of NEXRAD 88D radial wind data
Radial wind data from the network of 88D radars across the US are assimilated into the EDAS using the 3DVAR analysis. Observations for the first 4 radar tilts (0.5, 1.5, 2.5, and 3.5 degrees) are obtained from the NWS Multicast of the Level III products gathered from all WSR-88D doppler radar sites. These data are averaged into an hourly super-observation (composite) data set at a horizontal resolution of 5 km in the radial direction by 6 degrees in the azimuthal direction (azimuth). Quality control of the radial winds is now fully functional and is linked to Bill Collins' QC of the WSR-88D VAD winds, which includes, among other things, checks for migrating bird contamination.
Dave Parrish has used a new "minimal information" technique to incorporate radar wind data in the 3D-Var code. The vertical component of the radar beam width is assumed to increase with radar range at a rate of 20 m km-1, which is roughly 20% larger than the actual beam width in order to account for some uncertainty in beam propagation. Winds at all model levels (in coordinates) intercepted by the radar beam are adjusted so that the observation is as close as possible to an acceptable range of wind speeds derived from straight-line fits of the radial wind observations. All winds out to the maximum range of the radar are used. All the quality marks of wind observations derived from VAD (velocity-azimuth display) analyses are colocated in the vertical in 500-m bins with the corresponding radial winds. The radial winds are not used if there is no VAD observation or if it fails a quality-control (QC) check. This approach combines the QC algorithm for removing radial winds contaminated by birds with other checks used on the VAD winds. The radial winds are not used if (1) the beam envelope extends below the Eta model terrain height, (2) the super observation error is larger than 6 m s-1, or (3) it fails the same gross checks used for conventional winds. These radial wind runs show little positive or negative impact in the verification statistics, so it is certainly safe to include these winds treated this way in the 3DVAR.
8.1 Modification of radiance processing
The assimilation of radiances in the 3DVAR analysis is done with codes adapted from the global model 3DVAR. The radiance processing code has been updated to reflect changes previously made to the global system. These changes include a more sophisticated quality control of microwave channel data, which allows more data to be used over both land and water, and the inclusion of NOAA-16 infrared channels (currently turned off).
Finally, this package includes changes to processing procedures for polar-orbiting satellite radiances. First, the 3DVAR ingest for satellite data was switched from IEEE format to BUFR format. This will allow the addition of radiances from NOAA17, which are only available in BUFR. Second, there will no longer be explicit thinning of the radiances. Instead, all radiances will be used and an additional quality control procedure has been added to sense how high in the atmosphere the influence function extends for each channel. If the influence function extends too far beyond the model top, then that channel's data are not used. This is important for the Eta Model because the model top is much lower than in the GFS (37 mb for Eta versus 0.2 mb for GFS). With the addition of NOAA17 and no thinning, there is roughly a factor of twenty increase in the amount of radiance data used. The data counts went from around 50000 per 3hr cycle before the change to roughly 1000000 per 3hr cycle after the change. This added substantial processing to the analysis, but since the radiance processing part of the 3DVAR code is highly scalable, the run time was still kept in the required window by applying additional processors to the job. These changes were run in parallel for 3 weeks during the transition of the 12km EtaX parallel run to the new Central Computing System (CCS). The addition of all the new radiance data had a very small positive impact in upper air fits of the 3hr guess to the data, but otherwise there were no significant changes.
9.0 Verification Procedures
This table summarizes the tests made in this implementation package. Each test involved compares forecasts from the 32-km real-time parallel runs and/or the 32-km retrospective runs using the modified Eta/EDAS modeling system against a 32-km control run using the current operational Eta/EDAS system. Each change that led to improved forecasts at 32 km was added to the 12-km real-time parallel run.
At NCEP/EMC, operational models and various parallel model runs are routinely verified against surface (not including precipitation, which is done separately), upper-air, and precipitation observations.
Model forecast fields are verified against a myriad of surface and upper-air observational data that include:
upper air winds from pibals, profilers, satellite derivations and doppler radar VAD product.
Model fields are interpolated to the location of the observation for comparison. Further information about the surface and upper-air verification is documented in the Model Verification System at NCEP and EMC Verification Database.
Daily (12Z-12Z) precipitation verification is performed using 0.125 degree precipitation analyses over the contiguous United States (based on 7,000-8,000 daily gauge reports) that are quality-controlled with radar and climatological data. The verification is done on 80-km and 40-km grids for the NCEP operational models and various international models, and on a 12-km grid for the NCEP Nonhydrostatic Mesoscale Model (NMM) runs. Precipitation fields (forecast and observed) are mapped to the verification grids, from which forecast/observed/hit statistics are collected over different verification domains (CONUS and 13 subregions). A total of 26 different scores can be calculated, including equitable threat, bias, probability of detection, false alarm rate, and odds ratio. Further information about precipitation verification can be found on the NCEP/EMC Precipitation Verification page.
The extensive verification database allows evaluation of model performance using a variety of means, for example, the RMS errors of 12, 24, 48 and 60h vector wind forecast over CONUS and equitable threat and bias scores for the 48h precipitation forecast over Eastern U.S., both from the 12km parallel experiment. Links to the verification pages for various parallel and retrospective runs are provided in the table summarizing the various tests conducted for this implementation.
A real-time parallel test of the full change package at 12 km resolution started at 1200 UTC 14 August 2002 (see the summary of the changes in the parallel). The EtaX parallel was restarted (from the operational 12-km EDAS) at 1200 UTC 19 September 2002 in order to fix an error in the interpolation of some of the surface fields to the 12-km Eta grid. Statistics for the EtaX are used only after this date. The cycled EDAS was broken in the middle of November by CFL violations, and the parallel was restarted at 0000 UTC on 14 November 2002. Roughly a month later hardware errors on the IBM-SP caused the parallel to fail numerous times, and it was restarted at 0000 UTC on 16 December 2002. Finally, the cloud fraction parameterization within the radiation code was modified, starting at 1200 UTC on 20 February 2003, in order to fix the 2-m cold bias that formed in the parallel over northern latitudes during the winter (see Sec. 10.4).
10.1 Precipitation skill scores
The equitable threat (ETS) and bias scores of 24-h accumulated precipitation between the 12-km Eta and (parallel) EtaX (for all forecast ranges from 22 September 2002 - 5 January 2003 over the CONUS) show very little difference between the Eta and EtaX, with a very slight decrease in ETS at lower thresholds and a very slight increase at higher thresholds. The low bias in the EtaX is slightly worse. Casual inspection of recent model runs indicates a tendency for the EtaX to forecast a larger area of very light precipitation, such as that shown in the 00Z 5 February 2003 runs for the 0-24h and 24-48h accumulated precipitation. The EtaX produced slightly better scores over the Eastern US and slightly worse scores over the Western US. The precipitation scores prior to December 2002 showed slightly better performance by the EtaX prior to the onset of the surface cold bias.
10.2 Rawinsonde skill scores
Overall, the EtaX fits to rawinsonde observations have improved slightly for temperatures, relative humidities, heights, and winds. Root-mean-square (RMS) temperature errors from the parallel EtaX and the operational Eta are compared against rawinsonde observations for 12-h, 24-h, 48-h, and 60-h forecasts, which show a slight improvement in the EtaX temperature forecasts at all levels and over all forecast periods. The RMS errors in relative humidity shows very slight improvement in the EtaX, with reasonably modest moisture biases in both versions of the model (within ±3 %) over the entire period, with the parallel being slightly moister over all levels. Both versions tend to become dry (moist) relative to the observations below (above) the 700-mb level. Errors in the heights and vector winds are reduced slightly at nearly all levels for all forecast periods. Overall, the change package yielded slight improvements in the upper-air statistics over the CONUS. The parallel has developed a cold bias over Alaska in the lowest levels, which gets progressively worse with forecast time. This cold bias is believed to be in response to the development of a cold bias in the surface skin temperatures and 2-m temperatures, which will be discussed in more detail in the next two sections.
10.3 Surface verification
The performance of the parallel change package based on fits to the surface observations is mixed. Early results for the period 22 Sep 2002 to 30 Nov 2002 showed noticeable improvements in the diurnal variation of 2-m temperature over the Western US (00Z and 12Z cycles) and Eastern US (00Z and 12Z cycles), slight improvements in 2-m relative humidities over the West (00Z and 12Z cycles) and the East (00Z and 12Z cycles), and little change (a very slight degradation) in 10-m winds over the West (00Z and 12Z cycles) and the East (00Z and 12Z cycles). Since the surface verification package cannot currently treat missing data, statistics could not be calculated after 12Z 9 Nov 2002 because of the CFL violations in the parallel. More recent statistics are again available after 18 December 2002 (because of the other interruptions in the parallel) and unfortunately show the development of a cold bias in the 2-m temperatures over the Western US (see 00Z and 12Z cycles) and Eastern US (see 00Z and 12Z cycles).
Time series of RMS and bias errors from 36-h forecasts of 2-m temperatures (00Z cycles only) over the Northern Rocky Mountain (NMT) region is shown for the first two months of the parallel from 19 Sep through 15 Nov 2002. It indicates that the parallel (EtaX) is slightly colder than the operational Eta, with both having similar performance (in terms of RMS errors) before November. Thereafter, however, the cold bias in the parallel increased, as can be seen in a similar plot for the subsequent two-month period from 16 Nov 2002 through 15 Jan 2003, where the EtaX is averaging 0.6C colder than the operational Eta, which already has an average cold bias exceeding -1C. Similar time series over other regions suggest that this cold bias is primarily associated with northern areas. For example, the time series of RMS and bias errors of 36-h forecast 2-m temperatures for the period 19 Sep through 15 Nov 2002 over the Southern Plains (SPL) region shows improved performance by the parallel, with a reduction in the 2-m warm bias by over 0.6C and improved (lower) RMS errors. Differences in 2-m temperatures are generally small over this region for the later period of 16 Nov 2002 through 15 Jan 2003.
10.4 Sensitivity experiments addressing 2-m wintertime cold bias
A series of single runs were made, using initial conditions from 12Z 24 January 2003, to determine the cause of this 2-m cold bias in the parallel over northern regions during the cold season. For reference, the differences in 2-m temperatures between the EtaX and the operational Eta can be viewed at 6-h intervals over the Eastern US (you may need to move the cursor over the box or click the animate button at the top) and over the Western US. The 2-m temperature differences between the EtaX run and the operational Eta are shown in color, and the contour lines show the forecast 2-m temperatures from the operational Eta. These fields can also be compared against the surface observations available at the same 6-h time intervals. Similar figures are shown for other runs discussed below. The sequence of surface temperature differences reveal a large area of colder surface temperatures in the parallel (EtaX) over Central and Eastern Canada, and over the Western Plains to the lee of the Rockies. The following experimental runs were made (see links for results):
11.0 Summary remarks
Reducing the cloud optical depths and increasing the cloud amounts (cloud fractions) in the parallel were found to reduce the surface cold bias over northern latitudes during wintertime conditions. A change was recently made (1200 UTC 20 Feb 2003) in the parallel to improve the consistency in parameterized cloud fractions between the radiation and microphysics. This modification is expected to increase fractional cloud coverage, leading to improved (warmer) surface temperatures over northern latitudes during winter and improved (cooler) surface temperatures over the CONUS during the warm-season. Our future plans are also to make further improvements in the next implementation using a new radiation package.
Baldwin, M., R. Treadon, and S. Contorno, 1994: Precipitation type prediction using a decision tree approach with NMCs mesoscale eta model. Preprints, 10th Conf. On Numerical Weather Prediction, Portland, OR, Amer. Meteor. Soc., 30--31.
Baldwin, M. E., J. S. Kain, and M. P. Kay, 2002: Properties of the convection scheme in NCEP's Eta model that affect forecast sounding interpretation. Weather and Forecasting, 17, 1063-1079.
Bigg, E. K., 1953: The supercooling of water. Proc. Phys. Soc. London, B66, 688-694.
Cortinas, J. V, Jr., and M. E. Baldwin, 1999: A preliminary evaluation of six precipitation-type algorithms for use in operational forecasting. 6th Workshop on Operational Meteorology, Halifax, Nova Scotia, 207-211.
Ferrier, B. S., Y. Jin, Y. Lin, T. Black, E. Rogers, and G. DiMego, 2002: Implementation of a new grid-scale cloud and precipitation scheme in the NCEP Eta model. Preprints, 15th Conf. On Numerical Weather Prediction, San Antonio, TX, Amer. Meteor. Soc., 280-283.
Harshvardhan, D. A. Randall, T. G. Corsetti, and D> A. Dazlich, 1989: Earth radiation budget and cloudiness simulations with a general circulation model. J. Atmos. Sci., 46, 1922-1942.
Hong, S.-Y., H.-M. Juang, and Q. Zhao, 1998: Implementation of prognostic cloud scheme for a regional spectral model. Mon. Wea. Rev., 126, 2621-2639.
Janjic, Z., 1994: The step-mountain Eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev., 122, 927-945.
Lin, Y., M. E. Baldwin, K. E. Mitchell, E. Rogers, and G. J. DiMego, 2001: Spring 2001 changes to NCEP Eta analysis and forecast system: Assimilation of observed precipitation data. Preprints, 14th Conf. On Numerical Weather Prediction, Fort Lauderdale, FL, Amer. Meteor. Soc., J92-J95. Lin, Y. L., R. D. Farley, and H. D. Orville, 1983: Bulk parameterization of the snow field in a cloud model. J. Climate Appl. Meteor., 22, 1065-1092.
Ramer, J., 1993: An empirical technique for diagnosing precipitation type from model output. Preprints, 5th International Conf. On Aviation Weather Systems, Vienna, VA, AMS, 227--230.
Randall, D. A., 1995: Parameterizing fractional cloudiness by cumulus entrainment. Preprints, Workshop on Cloud Microphysics Parameterizations in Global Atmospheric Circulation Models, Kananaskis, AB, Canada, WMO, 1-16.
Rogers, E., T. Black, B. Ferrier, Y. Lin, D. Parrish, and G. DiMego, 2001: Changes to the NCEP Meso Eta Analysis and Forecast System: Increase in resolution, new cloud microphysics, modified precipitation assimilation, modified 3DVAR analysis. NWS Technical Procedures Bulletin. [Available at http://wwwt.emc.ncep.noaa.gov/mmb/mmbpll/eta12tpb/ or from the National Weather Service, Office of Meteorology, 1325 East-West Highway, Silver Spring, MD 20910].
Slingo, J. M., 1987: The development of a cloud prediction model for the ECMWF model. Quart. J. Roy. Meteor. Soc., 113, 899-927.
Xu, K.-M., and D. A. Randall, 1996: A semiempirical cloudiness parameterization for use in climate models. J. Atmos. Sci., 53, 3084-3102.
Zhao, Q., T. L. Black, and M. E. Baldwin, 1997: Implementation of the cloud prediction scheme in the Eta model at NCEP. Wea. Forecasting, 12, 697-712.
Zhao, Q., and F. H. Carr, 1997: A prognostic cloud scheme for operational NWP models. Mon. Wea. Rev., 125, 1931-1953.