This Office Note describes the post processor for the National Centers for Environmental Prediction Eta model. Preliminary to this discussion is a brief review of the Eta model, emphasizing the model grid and arrangement of variables. A general overview of the post processor design, usage, and capabilities follows. Currently 180 unique fields are available from the post processor. The final section documents these fields and the algorithms used to compute them. Details for using the post processor in conjunction with the model are found in Appendix 1. Appendix 2 lists the various NCEP data sets from which operational Eta model output is available.

The Eta post processor is not a stagnant piece of code. New output fields, improved algorithms, GRIB packing, and code optimization are just a few areas in which development continues. However, it is unlikely that the algorithms discussed in this Office Note will dramatically change.

2. The Eta Model

Since its introduction by Phillips (1957) the terrain following sigma coordinate has become the vertical coordinate of choice in most numerical weather prediction models. A prime reason for this is simplification of the lower boundary condition. Difficulties arise in the sigma coordinate when dealing with steep terrain. In such situations errors in the pressure gradient force computation become significant because two large terms of opposite must be added (Smagorinsky et al., 1967). These errors in turn generate advection and diffusion errors. Numerous methods have been devised to account for this defect of the sigma system. Mesinger (1984) took a different approach in defining the eta coordinate,

In this notation p is pressure and subscripts rf, s, and t respectively refer to reference pressure, the model surface, and the model top (pt = 50 mb). The height z is geometric height. Observe that the sigma coordinate appears as the

case of the eta coordinate. The reference pressure used in the Eta model is

, where

= 1013.25mb, To = 288K,

, g=9.8 m/s2, and R = 287.04J/K-kg. In the eta coordinate terrain assumes a step-like appearance thereby minimizing problems associated with steeply sloping coordinate surfaces. At the same time the coordinate preserves the simplified lower boundary condition of a terrain following vertical coordinate.

The Eta model uses the semi-staggered Arakawa E grid (Fig. 1). Prognostic variables at mass (H) points are surface pressure, temperature, and specific humidity. Zonal and meridional wind components are carried at velocity (V) points. The E grid is mapped to a rotated latitude-longitude grid which is centered at 50N and 111W for the current operational Eta with 22 km resolution. Two rotations are involved. One moves the Greenwich meridian to 111 W. The second shifts the equator to 50 N. Each row of the E grid lies along a line of constant rotated latitude; each column lies along a line of constant rotated longitude. In the operational Eta the shortest distance between like grid points (e.g., d in Fig. 1) is approximately 22 km at the time of this writing. The large box in Fig. 1 delimits the extent of the computational domain. Prognostic variables on the outermost rows and columns are specified by a global model forecast from the previous cycle. The second outermost rows and columns serve to smoothly blend boundary conditions with values in the computational domain. The boundaries are one way interactive.

Model terrain is represented in terms of discrete steps. Each step is centered on a mass point with a velocity point at each vertex. This is suggested by the dashed box in Fig. 1. The algorithm creating the steps tends to maximize their heights (so-called silhouette topography) based on the raw surface elevation data. Topography over the current operational Eta domain is discretized into steps from sea level to 3264 meters over the Colorado Rockies.

The current operational Eta runs with 50 vertical layers. The thickness of the layers varies with greatest vertical resolution near sea level and around 250mb (to better resolve jet dynamics). The top of each step coincides exactly with one of the interfaces between the model's layers. Note that the thickness of the lowest eta layer above the model terrain is not horizontally homogeneous. This presents difficulties when posting terrain following fields, which often exhibit strong horizontal gradients in mountainous regions. Vertical averaging over several eta layers, sometimes coupled with horizontal smoothing, minimizes this effect.

Model variables are staggered vertically as well as horizontally (Fig. 2). Temperature, specific humidity, and wind components are computed at the midpoint of eta layers. Turbulent kinetic energy is defined at the interfaces between layers. A no-slip boundary condition maintains zero wind components along the side of steps. Zero wind points are circled in Fig. 2.

The model uses a technique for preventing grid separation (Mesinger 1973, Janjic 1974) in combination with the split-explict time differencing scheme (Mesinger 1974, Janjic 1979). The fundamental time step for the 22 km operational Eta model is 60 seconds. This is the mass-momentum adjustment time scale. Advection, physical processes, and cumulus convection march at time steps which are integral multiples of the fundamental time step. The horizontal advection algorithm has a built-in strict nonlinear energy cascade control (Janjic, 1984). Vertical advection of moisture is based on the piecewise-linear method (Carpenter et al, 1989).

Fig. 2. Vertical cross section through Eta model with N layers. Mass variables such as temperature and momentum variables such as zonal wind components (T and U respectively) are defined at the midpoint of each eta layer. Ps is the surface pressure. The circled wind components along the side of steps are identically zero as specified by the no-slip boundary used in the model.

The model includes a fairly sophisticated physics package (Janjic, 1990, 1994) consisting of the Mellor and Yamada Level 2.5 scheme (Mellor and Yamada 1974, 1982) in the free atmosphere, the Mellor-Yamada Level 2.0 scheme for the surface layer, and a viscous sublayer over the oceans (Zilitinkevitch, 1970). Surface processes are modeled after those of Miyakoda and Sirutis (1984) and Deardorff (1978). Diffusion uses a second order scheme with the diffusion coefficient depending on turbulent kinetic energy and deformation of the wind field. Large scale and parameterized deep and shallow convection are parameterized through the Betts-Miller-Janjic scheme (1994). The radiation is the NCEP version of the GFDL radiation scheme with interactive random overlap clouds.

The operational Eta runs from an analysis based on the regional 3-d variational analysis (3DVAR), with the first guess for the Eta forecast coming from the Eta-based data assimilation system (EDAS). The fully cycled EDAS has been used to generate better initial conditions that require less time for spinup. Boundary conditions for the model are provided by 6 hour old Aviation model forecasts.

A more complete treatment of the Eta model can be found in Black et al. (1993) and Black (1994). Major changes to the Eta model and the Eta Data Assimilation System since it was implemented in June 1993 are discussed in Rogers et al (1996), Black et al (1997), Rogers et al (1998), Manikin et al (2000), Rogers et al (2000), and Rogers et al. (2001). The presentation above was intended to give the reader a general impression of the Eta model prior to discussing the Eta post processor below.

3. The Eta Post Processor - An Overview

Various changes have been made to the Eta post processor since the codes were first written in 1992. These changes include debugging, adding more posted fields, converting from 1-D indexing to 2-D indexing, paralleling codes to run on multiple CPU, and modifying the post to process output from both eta and sigma modes. The parallelization of the Eta post processor not only reduces the time it takes to process data but also enables the Eta post to handle domains with larger dimensions.

The post processor serves two primary purposes. Foremost, the post processor interpolates forecast fields horizontally and vertically from the model grid to specified pressure levels on specified output grids. These posted fields include standard model output such as geopotential height, temperature, humidity (specific or relative), vertical motion, and u and v wind components. A second function of the post processor is to compute special fields from model variables, such as tropopause level data, FD (flight data) level fields, freezing level information, and boundary layer fields.

With these purposes in mind the Eta post processor was designed to be modular, flexible, and relatively easy to use. A modular approach allows easy introduction of new routines to compute new output fields or test improved algorithms for currently posted fields. The user controls posting of fields by editing a control file. Linking several control files together permits output of data on multiple grids or files. The structure of the control file was based on a similar file used with the NGM.

The simplest control file consists of three primary pieces. First is the header block. Here the user specifies the format of the posted fields and the output grid. Data is currently posted in GRIB format. Data may be posted on the staggered E grid, a filled (i.e., regular) version of this grid, or any grid defined using standard NCEP grid specifications. All computations involving model output are done on the staggered model grid. Bilinear interpolation is used to fill the staggered grid. A second interpolation, which is completed in the product generator, is required to post data on a regular grid other than the filled E grid. This interpolation is also bilinear. Those grid points to which it is not possible to bilinearly interpolate a value to receive one of two values. A search is made from the outermost rows and columns of the output grid inward to obtain known values along the edge of the region to which interpolation was possible. Having identified these values, the algorithm reverses direction and moves outward along each row and column. Grid points to which interpolation was not possible are set equal to the known value along their respective row and column. If after this operation corner points on the output grid do not have values, they are assigned the field mean. Depending on the number of output fields requested, the calculation of interpolation weights can take more CPU time than does posting the fields. For this reason, interpolation weights may be pre-computed, saved, and read during post execution. The post retains the ability to compute these weights internally prior to posting any fields. A character flag in the header block controls this feature. A second character flag allows fields on different output grids to be appended to the same output file using the same or different data formats.

The second section of a control file lists available fields. By setting integer switches (0=off, 1=on) the user selects the fields and levels of interest. The current post processor has 180 unique output fields, some on multiple levels. Room exists for posting data on up to 60 vertical levels. In posting fields to an output grid smoothing or filtering of the data may be applied at any of three steps in the posting process. Fields may be smoothed on the staggered E grid, filtered on a filled E grid, or filtered on the output grid. Control of smoothing or filtering is via integer switches. Nonzero integers activate the smoother or filter with the magnitude of the integer representing the number of applications (passes) of the selected smoother or filter. The smoother coded in the post is a fourth order smoother which works directly on the staggered E grid. Once data is on a regular grid, a 25 point Bleck filter is available. A nice property of this filter is its fairly sharp response curve. Repeated applications will remove wavelengths twice the grid spacing while largely preserving field minima and maxima. Additional smoothing of posted fields can be realized in the interpolation process itself. In the current operational post, no smoothing is applied to any fields.

The last section of each control file is the end mark. This is a one line statement which tells the post processor to stop reading the control file and start posting requested fields. By having an explicit end mark the user only needs to specify the fields to be posted rather than all 180 available fields with switches turned off for unwanted fields. The order in which fields are requested is irrelevant to the post processor. However the order in which fields are written to the output file is fixed by the code. Figure 3 charts this ordering. Our discussion of the post processor in Section 4 follows this flowchart.

The Eta post processor is now also able to process the output of the Eta model forecast in sigma mode. There are two options in processing sigma output. The first option interpolates standard model output fields from sigma to pressure coordinates using traditional bilinear interpolation. The second option uses a cubic spline fitting method when performing vertical interpolation. The computation of sea level pressure has been migrated from within the post processor to within the Eta quilt because the sounding post that runs parallel to the main post processor needs sea level pressure as input from the restart files. The sounding post generates the sounding profile. Besides the Shuell and Mesinger sea level pressure reduction, the addition of processing sigma restart files using a spline method results in the third option of deducing sea level pressure in the sigma mode. A logical switch, SPLINE, was put in the namelist fcstdata.parm which is read in by both the Eta quilt and the eta post processor. When SPLINE is set to TRUE, the spline fitting method would be used to perform vertical interpolation, while the bi-linear interpolation method would be used when SPLINE is set to FALSE. The results obtained from the two methods do not appear to be very different for the fields above the ground. However, the underground fields and sea level pressure field are slightly more smooth when using the spline fitting method. Additionally, the locations of cyclone centers differ slightly when using different sea level reduction methods.

4. The Eta Post Processor - Details

The following subsections discuss fields available from the post and the algorithms used to derive these fields. Any model user should understand exactly what is represented by posted model output. Such knowledge allows the user to make more discriminating decisions when using model output. Further, feedback from users can suggest alternative algorithms better suited to their needs.

4.1 Constant eta and pressure fields

One can output data on constant eta or pressure levels. For either option the fields that may be posted are height, temperature (ambient, potential, and dewpoint), humidity (specific and relative), moisture convergence, zonal and meridional wind components, vertical velocity, absolute vorticity, the geostrophic stream function, cloud water/ice, and turbulent kinetic energy. Pressure may also be posted on constant eta layers.

Two options exist for posting eta layer data. Data may be posted from the n-th eta layer. This is simply a horizontal slice through the three dimensional model grid along the n-th eta layer. The slice disregards model topography. A second option is to post fields on the n-th eta layer above the model surface. From the definition of the eta coordinate it is clear that an eta-based terrain following layer is generally not a constant mass layer. Despite differences in layer thickness, examining data in the n-th atmospheric eta layer does have merit. It permits the user to see what is truly happening in the n-th eta layer above the model surface and as such represents an eta-based boundary layer perspective. Additionally, the code can post mass weighted fields in six 30 mb deep layers stacked above the model surface (see Section 4.10).

The height field on the eta interfaces is not one of the output variables from the model forecast and therefore needs to be calculated in the post. The interfaces that overlap with the eta terrain were specified to be the terrain height. The heights above the ground on each eta interface are then integrated using temperature and specific humidity on the eta mid-layers based on the hydrostatic relationship.

The more traditional way of viewing model output is on constant pressure surfaces. The post processor currently interpolates fields to thirty-nine isobaric levels (every 25 mb from 50 to 1000 mb). However, the number of isobaric levels to interpolate fields to can be easily changed by modifying LSM in the parameter statement of the source code and array SPL(LSM) in the namelist fcstdata.parm. Vertical interpolation of height is quadratic in log pressure. For temperature, specific humidity, vertical velocity, horizontal winds, and turbulent kinetic energy, the vertical interpolation is linear in log pressure. Derived fields (e.g., dewpoint temperature, relative humidity, absolute vorticity geostrophic stream function, etc.,) are computed from vertically interpolated base fields.

The following methods are used to obtain the fields on the isobaric levels that lie above the model top (currently 25 mb). Vertical and horizontal wind components above the model top are specified to be the same as those at the uppermost model level. For isobaric levels below the lowest model layer, the first atmospheric eta layer (e.g., the first eta mid-layer above the ground) fields are posted. Turbulent kinetic energy (TKE) is defined at model interfaces rather than the midpoint of each layer. At isobaric layers above the model top, the average TKE over the two uppermost model interfaces is specified. The same is done for pressure surfaces below the lowest model layer using TKE from the first and the second interfaces above ground interfaces.

Temperature, humidity, cloud/ice water, and geopotential heights are treated differently. The temperature averaged over the two uppermost model layers is used as the temperature on all the pressure levels above the model top. The specific humidity at the target level is set so as to maintain the relative humidity averaged over the two uppermost model layers. The cloud/ice water on the isobaric levels above the model top is specified as the cloud/ice water on the model top. Geopotential heights on isobaric surfaces are computed from the temperature and specific humidity using the hydrostatic equation. The treatment is the same for isobaric levels below the lowest model layer except that the averaging is over fields in the second and third model layers above the surface. This is done because including data from the first atmospheric layer imposed a strong surface signature on the extrapolated isobaric level data.

The treatment for the fields that are underground but above the lowest model layer is very similar to the treatment for fields below the lowest model layer. Further detail will be given in section 4.3.

4.2 Sea level pressure

4.2.1 Eta mode

Sea level pressure is one of the most frequently used fields posted from any operational model. Because large portions of the Eta Model domain are below the model terrain, some assumptions must be made to obtain sea level pressure at an underground grid point. Although there is no one best way to specify the underground sea level pressure as well as other underground fields, it is desirable to specify these fields so that they are representable and somewhat smooth. Although, as mentioned previously, the computation of sea level pressure is now carried out in the Eta Quilt instead of the Eta Post, the computation of sea level pressure will still be discussed here for completeness. The question here is which of a myriad of reduction algorithms to use. Different reduction algorithms can produce significantly different sea level pressure fields, given similar input data. The traditional approach is to generate representative underground temperatures in vertical columns and then integrate the hydrostatic equation downward. Saucier (1957) devotes several pages detailing the then current U.S. Weather Bureau reduction scheme. Cram and Pielke (1989) compare and contrast two reduction procedures using surface winds and pressure. References for other schemes may be found in their paper.

Sea level pressure is available from the Eta model using either of two reduction algorithms. One is based on a scheme devised by Mesinger (1990). The other is the standard NCEP reduction algorithm. The methods differ in the technique used to create fictitious underground temperatures.

The standard reduction algorithm uses the column approach of vertically extrapolating underground temperatures from a representative above ground temperature. The algorithm starts with the hydrostatic equation in the form

Mean sea level pressure, pmsl, is computed at mass points using the formula

. The function

, where

is the average of

at the model surface and mean sea level. The remaining question is how to determine these

's.

In the NGM

(sfc) and

(msl) are first set using a 6.5 K/km lapse rate from the first sigma layer. A similar approach was not successful in the Eta model due to the discontinuous nature of the step topography. Virtual temperatures are averaged over eta layers in the first 60 mb above the model surface. The resulting layer mean virtual temperature field is in turn horizontally smoothed before extrapolating surface and sea level temperatures.

In both the NGM and Eta,

(sfc) and

(msl) are subject to the Shuell correction. Whether this correction is applied depends on the relation of the extrapolated

's to a critical value

Once mean sea level pressure is computed, a consistent 1000 mb height field is obtained using the relation

. This simple relationship itself can be used to obtain sea level pressure given 1000 mb geopotential heights and an assumed mean density. In the post, the mean density, ?*, is computed from

* and p* (the average in log pressure of p(sfc) and p(msl)).

In contrast to the traditional column approach, the Mesinger scheme uses horizontal interpolation to obtain underground virtual temperatures. He made an assumption that sea level pressure should be obtained to maintain the shape of the isobars on surfaces of constant elevation. Therefore, it is physically more reasonable to create underground temperatures using atmospheric temperatures surrounding the mountain rather than extrapolating downward from a single temperature on the mountain. The step-mountain topography of the Eta model simplifies coding of this approach. The algorithm starts from the tallest resolved mountain and steps down through the topography. Virtual temperatures

on each step inside the mountain (i.e., underground) are obtained by solving a Laplace equation:

Atmospheric virtual temperatures on the same step surrounding the mountain provide consistent, realistic boundary conditions. The relaxation method is used to smooth the virtual temperature on all the grid points. However, only the underground virtual temperature is replaced by the smoothed virtual temperature. In the Eta Post, (4.2) is solved by applying an eight-point averaging to the virtual temperature fields on each eta mid-layer:

where A and B are constants, and ihw and ihe are increments in i directions for the grid points that are to the west and east of the grid point (i,j). Currently, the eight-point averaging is applied to the virtual temperature fields 500 times. Once all underground temperatures have been generated, the hydrostatic equation (e.g., (4.1)) is integrated downward to obtain sea level pressure. Note that the thickness dz used to calculate the sea level pressure is not based on the actual geopotential heights, but the heights of the standard interfaces which are computed using standard ground level atmospheric temperature 288 K and standard lapse rate 6.5 K/km based on the hydrostatic relationship.

For selected sites the Eta model posts vertical profile (sounding) data plus several surface fields. The posting of profile information is not part of the post processor. Sea level pressures included in the profile data are available only from the Mesinger scheme in the Eta mode. The standard and Mesinger schemes can produce markedly different sea level pressure fields given the same input data. This is especially true in mountainous terrain. The Mesinger scheme generally produces a smoother analysis, much as one might produce by hand.

4.2.2 Sigma mode

As mentioned previously, there are two options when processing the Eta model output in sigma mode, SPLINE and NON-SPLINE, which then produces three different sets of sea level pressure. Similar to Eta sea level pressure reduction, the first option generates sea level pressure using both Shuell and Mesinger reduction algorithms. Recall that the Mesinger sea level reduction involves computation of eight-point averaging on constant eta levels. Therefore, because the sigma interfaces are often steep over the mountains, the temperature fields are first interpolated/extrapolated from sigma to pressure vertical coordinates using the bi-linear interpolation/extrapolation before the smoothing of the temperature fields is performed. The underground temperature is then obtained by solving the Laplace equation of temperature on the constant pressure levels. All the other procedures used to obtain Mesinger sea level pressure are similar to those in the eta mode.

The procedures for generating sea level pressure using the second option, the spline fitting method, are described as follows. First, the spline fitting method is used to interpolate height fields from sigma to pressure levels. Note that the spline fitting can only perform interpolation but not extrapolation. Therefore, when the lowest pressure level falls under the lowest sigma level over a specific grid point, the bi-linear extrapolation is done to obtain the height at the lowest pressure level. The sea level pressure at each grid point is then obtained by finding the pressure level at which height is equal to zero using the spline fitting method.

4.3 Subterranean fields

The treatment for the underground fields is very similar to the treatment for fields that are above the surface but below the lowest model layer as described in section 4.1. The horizontal wind components on the first atmospheric eta layer above ground are posted for all the pressure levels below the ground. The underground turbulent kinetic energy is specified as the average of the first and second layers above ground. The fictitious underground temperatures on the constant eta levels generated during deduction of Mesinger sea level pressure were not used as output temperature for the Eta post. Instead, the underground temperature calculated during Shuell reduction is currently posted as an underground temperature on the constant eta levels. Note that there are no underground fields on the constant sigma levels. The underground temperature on isobaric levels is then interpolated/extrapolated bilinearly with respect to pressure using underground temperature on the constant eta levels or specified as the average of the second and third layers above ground on the constant sigma levels. Underground specific humidity is adjusted to maintain the average of the second and third lowest atmospheric eta layer relative humidity.

4.4 Tropopause level data.

The post processor can generate the following tropopause level fields: pressure, temperature (ambient and potential), horizontal winds, and vertical wind shear. The greatest difficulty was coding an algorithm to locate the tropopause above each mass point. The procedure used in the Eta post processor is based on that in the NGM. Above each mass point a surface-up search is made for the first occurrence of three adjacent layers over which the lapse rate is less than a critical lapse rate. In both the NGM and Eta model the critical lapse rate is 2 K/km. The midpoint (in log pressure) of these two layers is identified as the tropopause. A lower bound of 500 mb is enforced on the tropopause pressure. If no two layer lapse rate satisfies the above criteria the model top is designated the tropopause. Very strong horizontal pressure gradients result from this algorithm. Horizontal averaging over neighboring grid points prior to or during the tropopause search might minimize this effect. To date, this alternative has not been coded. It might be more accurate to describe the current algorithm as one locating the lowest tropopause fold above 500 mb.

Linear interpolation in log pressure from the model layers above and below the tropopause provides the temperature. Recall that velocity points are staggered with respect to mass points. Winds at the four velocity points surrounding each mass point are averaged to provide a mass point wind. These mass point winds are used in the vertical interpolation to tropopause level. Vertical differencing between horizontal wind field above and below the tropopause provides an estimate of vertical wind shear at the tropopause.

4.5 FD level fields.

Flight level temperatures and winds are posted at six levels, namely 914, 1524, 1829, 2134, 2743, and 3658 meters above the model surface. At each mass point a surface-up search is made to locate the model layers bounding the target FD level height. Linear in log pressure interpolation gives the temperature at the target height. Again, wind components at the four velocity points surrounding each mass point are averaged to provide a mass point wind. The wind averaging is coded so as to not include zero winds in the average. This can happen in mountainous terrain where the no slip boundary condition of the model maintains zero winds along the side of steps. Experimentation demonstrated that the averaging of winds to mass points minimize point maxima or minima in posted FD level wind fields. The process is repeated for all six flight level heights.

4.6 Freezing level data.

The post processor computes two types of freezing level heights and relative humidities at these heights. The calculation is made at each mass point. To obtain the first type of freezing level height, a search is made for the two model layers over which the temperature first falls below 273.16 K when moving up from the model surface. The second type of freezing level, the highest freezing level, is the last level whose temperature falls below 273.16 K and the temperature above it is above 273.16 K when moving from surface up. Vertical interpolation gives the mean sea level height, temperature, pressure, and specific humidity at this level. From these fields the freezing level relative humidity is computed. These fields are used to generate the FOUS 40-43 NWS bulletins containing six hourly forecasts of freezing level heights and relative humidities for forecast hours twelve through forty-eight. The surface-up search algorithm is designed so that posted freezing level heights can never be below the model terrain. This differs from the LFM algorithm in which underground heights were possible.

4.7 Sounding fields.

Several lifted indices are available from the Eta model. All are defined as being the temperature difference between the temperature of a lifted parcel and the ambient temperature at 500 mb. The distinction between the indices hinges on which parcel is lifted. The surface to 500 mb lifted index lifts a parcel from the first atmospheric eta layer. This lifted index is posted as the traditional LFM surface to 500 mb lifted index. The thinness of the first atmospheric eta layer in certain parts of the model domain imparts a strong surface signal on temperatures and humidities in this layer. In particular strong surface fluxes can create an unstable first atmospheric layer not representative of the layers above. The surface to 500 mb lifted index generally indicates larger areas of instability than other Eta lifted indices.

A second set of lifted indices are those computed from constant mass or boundary layer fields. The post can compute mass weighted mean fields in six 30 hPa deep layers stacked above the model surface. Lifted indices may be computed by lifting a layer mean parcel from any of these layers. Of six possible lifted indices, the operational Eta posts that obtained by lifting a parcel from the first (closest to surface) 30 mb deep layer.

The last lifted index available from the post processor is similar to the NGM best lifted index. In the NGM, the best lifted index is the most negative (unstable) lifted index resulting from lifting parcels in the four lowest sigma layers. The Eta best lifted index is the most negative lifted index resulting from lifting parcels in the six constant mass layers.

Two integral, sounding based fields are available from the Eta post processor: convective available potential energy (CAPE) and convective inhibition (CINS). As coded in the post processor, CAPE is the column integrated quantity (Cotton and Anthes 1989)

The parcel to lift is selected as outlined in Zhang and McFarlane (1991). The algorithm locates the parcel with the warmest equivalent potential temperature (Bolton, 1980) in the lowest 70 mb above the model surface. This parcel is lifted from its lifting condensation level to the equilibrium level, which is defined as the highest positively buoyant layer in the Eta post. During the lifting process positive area in each layer is summed as CAPE, negative area as CINS. Typical is Atkinson's (1981) definition of CAPE

with z* being also the equilibrium level. Apart from the difference in integration limits this definition of CAPE and the one coded in the post processor produce qualitatively similar results. This is easily seen from the power series expansion of

, which shows the integrands to be related.

Posted CAPE values can indicate a greater potential for convection than may be realized. The search to determine which parcel to lift starts from the first eta layer above the surface. As mentioned above, the thinness of this layer over certain parts of the domain imparts a strong surface signal on temperatures and humidities in this layer. Instabilities in the first atmospheric eta layer may not be representative of the layers above. This should be kept in mind when using CAPE values posted from the operational Eta.

Random overlap clouds are included in the Eta model radiation package. This code is based on that in the NCEP global spectral model (Campana and Caplan (1989), Campana et al. (1990)). Both stratiform and convective clouds are parameterized. Key variables in the parameterization are relative humidity and convective precipitation rate. Clouds fall into three categories: low (approximately 640 to 990mb), middle (350 to 640 mb), and high (above 350 mb). Fractional cloud coverage for stratiform clouds is computed using a quadratic relation in relative humidity (Slingo, 1980). The operational Eta posts time-averaged stratiform and convective cloud fractions.

In addition to cloud fractions the post processor can compute lifting condensation level (LCL) pressure and height above each mass point. These calculations appear quite sensitive to the definition of the parcel to lift. Experiments are ongoing to find an optimal definition of this parcel. Under certain situations the convective condensation level or level of free convection may be more indicative of cloud base heights. The modular design of the post processor simplifies the development of such routines. Currently neither LCL pressure nor heights are posted from the operational Eta post.

The components of storm motion and storm-relative helicity are important sounding parameters available from the Eta model. The actual helicity computation is taken from Davies-Jones et al. (1990) and is computed over the lowest 3 km, giving a value in m2/s2. The computation of the storm motion vector (which is used to compute the helicity value), however, can be done several different ways. Prior to March 2000, the Eta used the technique described in Davies and Johns (1993) in which storm motion was defined as 30 degrees to the right of the 0-6 km mean wind at 75% of the speed for mean speeds less than 15 m/s and 20 degrees to the right for speeds greater than 15 m/s. This method was found to perform very well in `classic' severe weather events in which the vertical wind shear profile is primarily in the upper-right quadrant of the hodograph.

This method, however, doesn't produce as desirable results for events with `atypical' vertical wind shear profiles such as those with weak winds, shifted towards the lower-left quadrant of the hodograph, or having unusual orientations (such as northwest-flow events). The Internal Dynamics method (hereafter ID) performs much better in the unusual shear cases and at least as well as the older method in the classic cases and is now used in the Eta model.

The ID method assumes that supercell motion has two components - advection of the storm by a mean wind and propagation to the right or left of the vertical wind shear due to internal storm dynamics. The code first determines the 0-6 km mean wind in the profile. The code next finds the 0 - 0.5 km and 5.5 - 6 km mean winds to represent the tail and head of the shear vector. The storm motion vector is then defined as 7.5 m/s to the right of the 0-6 km mean wind constrained along a line which is both perpendicular to the 0-6 km vertical wind shear vector and passes through the 0-6 km mean wind. A left-moving storm would be 7.5 m/s to the left of the 0-6 km mean wind, but since most Northern Hemispheric supercell storms are right-movers, the left-moving case is ignored.

4.8 Surface-based fields.

The post processor can output surface pressure, temperature (ambient, dewpoint, and potential), and humidity (specific and relative). Surface temperatures and humidities are strictly surface based and should not be interpreted as being indicative of shelter level measurements. The model carries running sums of total, grid-scale, and convective precipitation. The accumulation period for these precipitation amounts is set in the fcstdata.parm file prior to the model run. Interpolation of accumulated precipitation amounts from the model grid to other output grids utilizes an area conserving interpolation scheme. Other surface-based fields that can be posted include instantaneous and time-averaged incoming and outgoing radiation, roughness length, friction velocity, and coefficients proportional to surface momentum and heat fluxes.

Static surface fields may also be posted. These are the geodetic latitude and longitude of output grid points, the land-sea mask, the sea ice mask, and arrays from which three dimensional mass and velocity point masks may be reconstructed. The land-sea mask defines the land-sea interface in the model. Three dimensional mass and velocity point masks vertically define model topography. For operational models the practice is to post model output atop background maps. This assumes that the model geography matches that of the background map. A one to one correspondence between the two is obviously not possible. The same remark holds true in the vertical. These comments should be kept in mind when interpreting output from any model.

4.9 10-meter winds and 2-meter temperatures.

The post processor computes anemometer level (10 meter) winds and shelter level (2 meter) temperatures. Gradients of wind speed and temperature can vary by several orders of magnitude in the surface-layer. Direct application of the Mellor-Yamada Level 2.0 equations in the surface-layer would require additional model layers to adequately resolve these gradients. A computationally less expensive approach is to use a bulk layer parametrization of the surface-layer consistent with the Mellor-Yamada Level 2.0 model. Lobocki (1993) outlined an approach to derive surface-layer bulk relationships from higher closure models. Assuming a horizontally homogenous surface layer at rest the Monin-Obukhov theory maintains that dimensionless gradients of wind speed and potential temperature at height z (in the surface-layer) may be represented as a function of a single variable

. The length scale L is the Monin-Obukhov scale. A second important surface-layer parameter is the flux Richardson number Rf. which quantifies the relative importance of two production terms in the turbulent kinetic energy equation. Using the Mellor-Yamada Level 2.0 model, Lobocki derived a fundamental equation relating internal or surface-layer parameters

and Rf. with external or bulk characteristics of the surface-layer. Equations consistent with this fundamental equation relating the wind speed, U, or potential temperature,

, between two levels, z1 and z2, in the surface layer are

The functions

and

are integrated forms of similarity functions for dimensionless differences of the quantity U or

across the layer z1 to z2

Specifically, for S = U or

where

is a constant,

, and

. The function

is given by equation (48) in Lobocki's paper for S = U and (49) for S=

When applying these equations to compute anemometer level winds or shelter level temperatures the height z2 refers to values in the first eta layer above ground. The height z1 refers to the target level in the surface layer (either 10 or 2 meters). The dependence of

on the Monin-Obukhov height

introduces a physically reasonable stability-based variability in computed anemometer level winds and shelter temperatures. In the absence of strong synoptic forcing both anemometer level winds and shelter temperatures exhibit a typical diurnal cycle.

4.10 Boundary layer fields.

The Eta model does not explicitly forecast fields in a boundary layer. Additionally, the thickness of the n-th eta layer above the model terrain varies horizontally. The post processor computes mass-weighted mean fields in six 30 mb deep layers above the model surface. Note that since the thickness of the n-th eta layer above the surface varies horizontally the number of layers used in computing mass weighted means is not horizontally homogenous. Variables that can be posted from any or all of the six layers are pressure, temperature (ambient, potential, and dewpoint), humidity (specific and relative), moisture convergence, horizontal wind components, vertical velocity, and precipitable water. The precipitable water is that amount obtained by integration over the constant mass layer. The operational Eta posts all possible boundary layer fields in the first (lowest) 30 mb layer above the surface. Additionally, temperature, relative humidity, and winds are posted from the third and sixth constant mass layers.

Considerable time was spent developing an algorithm to mimic the behavior of LFM boundary layer winds. Boundary layer winds from the LFM did not exhibit a diurnal cycle typical of those from the NGM and Eta model. Rather, LFM boundary layer winds appeared geostrophic with a superimposed cross isobaric turning towards lower pressure. To reproduce this effect using the Eta model, the code starts with geostrophic winds computed from heavily smoothed sea level pressure or 1000 mb heights. The resulting geostrophic wind components are turned using the classic Ekman spiral equations (Section 8.5.2 of Haltiner and Williams, 1980). A rotation parameter controls the amount of the cross contour flow. After much experimentation a suitable rotation parameter along with appropriate smoothing was found to produce a wind field very comparable to the LFM boundary layer winds.

4.11 LFM and NGM look-alike fields.

In addition to posting standard data on pressure surfaces or deriving other fields from model output, the post processor generates fields specific to the LFM and NGM using Eta model output These fields are written to the output file using LFM or NGM labels. The primary reason for including these look-alike fields was to ensure compatibility of posted Eta model output with existing graphics and bulletin generating codes.

The post computes equivalents to fields in the NGM first (S1=0.98230), third (S3=0.89671), and fifth (S5=0.78483) sigma layers data as well as layer mean relative humidities and a layer mean moisture convergence field. Recall the definition of the sigma coordinate,

Given the pressure at the top of the model and the forecast surface pressure ps, target sigma levels are converted to pressure equivalents. Vertical interpolation from the eta layer bounding each target pressure provides an eta-based approximation to the field on the target sigma level. This calculation is repeated at each horizontal grid point to obtain eta-based sigma level S1, S3, S5 temperatures, S1 relative humidity, and S1 u and v wind components. Since surface pressure is carried at mass points a four point average of the winds surrounding each mass point is used in computing the S1 u and v wind components. A check is made to ensure that zero winds are not included in this average. S3 and S5 relative humidities are layer means over the eta layers mapping into sigma layers 0.47 to 0.96 and 0.18 to 0.47, respectively.

The FOUS 60-78 NWS bulletins are generated from the NGM look-alike fields and other posted fields. These bulletins contain initial condition and six hourly forecasts out to forecast hour 48 for thirteen parameters at sites over the U.S., Canada, and coastal waters. Table 1 identifies which Eta fields are used in generating these bulletins.

Table 1: Posted Eta model fields used to generate FOUS 60-78 NWS bulletins.
Fous 60-78 entry	Posted Eta field used
PTT (accumulated precipitation)	total accumulated precipitation
Rl (sigma layer 1 relative humidity)	NGM look-alike S1 relative humidity
R2 (0.47 to 0.96 layer mean relative humidity)	NGM look-alike S3 relative humidity
R3 (0.18 to 0.47 layer mean relative humidity)	NGM look-alike S5 relative humidity
VVV (700 mb vertical velocity)	700 mb vertical velocity
LI (best (NGM four layer) lifted index)	Eta best (six layer) lifted index
PS (sea level pressure)	standard reduction sea level pressure
DDFF (sigma layer 1 wind speed and direction)	NGM look-alike S1 u and v winds
HH (1000-500 mb layer thickness)	1000 and 500 mb geopotential heights
Tl (sigma layer 1 temperature)	NGM look-alike S1 temperature
T3 (sigma layer 3 temperature)	NGM look-alike S3 temperature
T5 (sigma layer 5 temperature)	NGM look-alike S5 temperature

LFM look-alike fields include three layer mean relative humidities and a partial column precipitable water. An approach similar to that used for the NGM is not directly applicable. The distinction arises due to the vertical structure of the LFM. The approach taken here was to assume a sigma based vertical coordinate in the LFM and identify appropriate sigma levels bounding LFM layer mean fields. The sigma levels used for layer mean relative humidities are 0.33 to l.00, 0.66 to l.00,and 0.33 to 0.66. For precipitable water the range in sigma is 0.33 to 1.00. Given these sigma bounds the same sigma to eta mapping used for the NGM fields is applied here.

5. Summary

In this Office Note we have reviewed the output capabilities of the Eta post processor. Prior to the post processor description was a brief review of the Eta model with an emphasis on the horizontal and vertical layout of model variables. Given this background, we previewed the Eta post processor in general terms. Key points included the modular design of the post processor and ease of use. The user controls the post via a control file, specifying not only which fields to post but also on which grid to post the data and the format to use. Following this was a field by field description of the algorithms used to derive posted fields. Users of output from any model should understand how the output is generated, as this information allows the user to better use posted model output.

Development continues on the Eta model and work will continue on the post processor as well. As users become more familiar with the Eta model it is envisioned that their feedback will suggest the addition or deletion of routines. Such communication can play an important but often overlooked role in development.

6. References

Atkinson, B.W., 1981: Meso-scale Atmospheric Circulations, Academic Press, New York, 495 pp.

Betts, A.K., 1986: A new convective adjustment scheme. Part I: observational and theoretical basis. Quart. J. Roy. Meteor Soc., 112, 677-691.

__, and M.J. Miller, 1986: A new convective adjustment scheme. Part II: single column tests using GATE wave, BOMFX, and artic air-mass data sets. Quart. J. Roy. Meteor Soc., 112, 693-709.

Black, T., D. Deaven, and G. DiMego, 1993: The step-mountain eta coordinate model: 80-km Early version and objective verifications. NWS Technical Procedures Bulletin No. 412, NOAA/NWS, 31 pp.

__, 1994: NMC NOTES, The New NMC Mesoscale Eta Model: Description and Forecast Examples. Wea. Forecasting, 9, 265-278

__, M. Baldwin, K. Brill, F. Chen, G. DiMego, Z. Janjic, G. Manikin, F. Mesinger, K. Mitchell, E. Rogers, and Q. Zhao, 1997: Changes to the Eta forecast systems. NWS Technical Procedures Bulletin No. 441, NOAA/NWS, 10 pp.

Bolton, D., 1980: The computation of equivalent potential temperature. Mon. Wea. Rev., 108, 1046-1053.

Campana, K.A., and P.M. Caplan, 1989: Diagnosed cloud specifications, Research Highlights of the NMC Development Division: 1987-1988, 101-111.

__, P.M. Caplan, G.H. White, S.K. Yang, and H.M. Juang, 1990: Impact of changes to cloud parameterization on the forecast error of NMC's global model, Preprints Seventh Conference on Atmospheric Radiation, San Francisco, CA, Amer. Meteor. Soc., J 152-J 158.

Carpenter, R.L., Jr., K.K. Droegemeier, P.W. Woodward, and C.E. Hane, 1989: Application of the piecewise parabolic method (PPM) to meteorological modeling. Mon. Wea. Rev., 118, 586-612.

Cotton, W.R., and R.A. Anthes, 1989: Storm and Cloud Dynamics, Academic Press, New York, 880 pp.

Cram, J.M., and R.A. Pielke, 1989: A further comparison of two synoptic surface wind and pressure analysis methods. Mon. Wea. Rev., 117, 696-706.

Deardorff, J., 1978: Efficient prediction of ground temperature and moisture with inclusion of a layer of vegetation. J. Geophys. Res., 83, 1989-1903.

Haltiner, G.J., and R.T. Williams, 1980: Numerical Prediction and Dynamic Meteorology. John Wilely & Sons, New York, 477 pp.

Janjic, Z. I., 1974: A stable centered difference scheme free of two-grid-interval noise. Mon. Wea. Rev., 102, 319-323.

__, 1979: Forward-backward scheme modified to prevent two-grid-interval noise and its application in sigma coordinate models. Contrib. Atmos. Phys., 52, 69-84.

__, 1984: Nonlinear advection schemes and energy cascade on semistaggered grids. Mon. Wea. Rev., 112, 1234-1245.

__, 1990: The step-mountain coordinate: physical package. Mon. Wea. Rev., 118, 1429-1443. Keyser, D.A., 1990: NMC Development Division Rotating Random Access Disk Archive - User Documentation. [Available from NMC, 5200 Auth Road, Washington, D.C., 20233].

Lobocki, L., 1993: A procedure for the derivation of surface-layer bulk relationships from simplified second-order closure models. J. Appl. Meteor, 32, 126-138.

Manikin G., M. Baldwin, W. Collins, J. Gerrity, D. Keyser, Y. Lin, K. Mitchell, and E. Rogers, 2000: Changes to the NCEP Meso Eta Runs: Extended range, added input, added output, convective changes. NWS Technical Procedures Bulletin No. 465, NOAA/NWS, 85 pp.

Mellor, G.L., and T. Yamada, 1974: A hierarchy of turbulence closure models for planetary boundary layers. J. Atmos. Sci., 31,1791-1806.

__, and__, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys. Space Phys., 20, 851-875.

Mesinger, F., 1973: A method for construction of second-order accuracy difference schemes permitting no false two-grid-interval wave in the height field. Tellus, 25, 444-458.

__, 1974: An economical explicit scheme which inherently prevents the false two-grid-interval wave in forecast fields. Proc. Symp. on Difference and Spectral Methods for Atmosphere and Ocean Dynamics Problems, Novosibirsk, Acad. Sci., Novosibirsk, Part II, 18-34.

__, 1984: A blocking technique for representation of mountains in atmospheric models. Riv. Meteor Aeronautica, 44,195-202.

__, 1990: Horizontal pressure reduction to sea level. Proc. 21st Conf. for Alpine Meteor, Zurich, Switzerland, 31-35.

Miyakoda, K., and J. Sirutis, 1983: Impact of sub-grid scale parameterizations on monthly forecasts. Proc. ECMWF Workshop on Convection in Large-Scale Models, ECMWF, Reading, England, 231-277.

Phillips, N.A., 1957: A coordinate system having some special advantages for numerical forecasting. J. Meteor., 14, 297-300.

Rogers, E., T. Black, D. Deaven, G. DiMego, Q. Zhao, M.Baldwin, and Y. Lin, 1996: Changes to the operational Early Eta analysis/forecast system and the National Centers for Environmental Prediction, Wea. Forecasting, 11, 391-413.

__, M. Baldwin, T. Black, K. Brill, F. Chen, G. DiMego, J. Gerrity, G. Manikin, F. Mesinger, K. Mitchell, D. Parrish, Q. Zhao, 1997: Changes to the NCEP Operational "Early" Eta Analysis/ Forecast System. NWS Technical Procedures Bulletin No. 447, NOAA/NWS, 85 pp.

__, T. Black, W. Collins, G. Manikin, F. Mesinger, D. Parrish and G. DiMego, 2000: Changes to the NCEP Meso Eta Analysis and Forecast System: Assimilation of satellite radiances and increase in resolution. NWS Technical Procedures Bulletin No. 473, NOAA/NWS, 85 pp.

__, M. Ek, Y. Lin, K. Mitchell, D. Parrish, and G. DiMego, 2001: Changes to the NCEP Meso Eta analysis and Forecast system: Assimilation of observed precipitation, upgrades to land-surface physics, modified 3DVAR analysis. NWS Technical Procedures Bulletin No. 473, NOAA/NWS, 14 pp.

Saucier W.J., 1957: Principles of Meteorological Analysis, Dover Publications, New York, 438 pp.

Slingo, J.M., 1980: A cloud parameterization scheme derived from GATE data for use with a numerical model. Quart J. Roy. Met. Soc., 106, 747-770.

Smagorinsky, J.J., L. Holloway, Jr., and G.D. Hembree, 1967: Prediction experiments with a general circulation model, Proc. Inter. Symp. Dynamics Large Scale Atmospheric Processes. Nauka, Moscow, U.S.S.R., 70-134.

Stackpole, J., 1990: NMC Handbook [Available from NMC, 5200 Auth Road, Washington, D.C., 20233].

Zhang, G.J., and N.A. McFarlane, 1991: Convective Stabilization in Midlatitudes. Mon. Wea Rev., 119, 1915-1928.

Ziltinkevitch, S.S., 1970: Dynamics of the planetary boundary layer. Gidrometeorologicheskoe Izdatelystvo, Leningrad, 292 pp. (in Russian).

7. Appendix 1: Using the Eta Post Processor

7.1 Introduction

In this appendix we discuss in greater detail how to use the Eta post processor. We assume the reader knows how to run the Eta model. The peculiarities of any single user application necessarily limits how specific our treatment can be. It is hoped that enough information is given to get the reader started using the Eta post processor.

7.2 Model and post processor source

The source code for the most current operational eta post processor can be found in /nwprod/sorc/eta_etapost.fd on the IBM. The source code for the forecast model can be found in /nwprod/sorc/eta_etafcst.fd. The makefile in each of the two directories is used to compile the post and forecast codes respectively to generate an executable. Both the post and forecast source codes have been parallelized to run on multiple processors. The use of multiple processors has made it possible to run the forecast and post the model output for domains with larger dimensions which could not have been accomplished by the serial code due to lack of memory space. Furthermore, the clock time needed to forecast and post Eta model is greatly reduced.

7.3 Namelist FCSTDATA

Prior to running the model the user sets runtime variables in namelist FCSTDATA. These settings affect both the model and the post processor. A sample FCSTDATA is shown in Fig.7.1.

The model integration starts at hour TSTART and runs through hour TEND. If the forecast is part of the data assimilation cycle, then TCP is set to be equal to TEND and is the hour at which the restart file restrt03 is generated. When RESTART is set to true, then the model uses the full restart file as the input initial condition. Otherwise, the model is initialized with the nfc file (the so-called "cold-start" restart file). The times (measured in hours) at which to output forecast files are set in array TSHDE. NMAP is the number of posting times specified in array TSHDE. The maximum number of posting times is currently ninety-nine. The only restriction on the output times is that they be between TSTART and TEND. Array SPL specifies isobaric levels (in Pascals) to which the post processor can vertically interpolate certain fields. The number of elements in array SPL is set by parameter LSM in include file parmeta. The ordering of pressure levels directly maps to level switches in the post processor control file. This will be covered later when we discuss the control file in the next section. The variables NPHS and NCNVC represent how often model surface processes and cumulus convection are executed in terms of number of model time steps. The frequencies in hours at which short and long wave radiative processes are set by NRADSH and NRADLH. The variable NTDDMP specifies how often (in hours) the divergence damping process is called. Through variables TPREC, THEAT, TCLOD, TRDSW, TRDLW, and TSRFC, the user specifies the accumulation period (in hours) for accumulation arrays. Note that the accumulation periods operate independently of the posting times set in TSHDE. They define the frequency at which accumulated quantities are reset to zero:

7.4 The Control File

The user interacts with the post processor through a control file which has a setup similar to the one used with the NGM. By editing the control file the user selects which fields to post, to which grid to post the fields, and in what format to output the fields. If fields are to be posted to a grid other than the model grid, interpolation weights may be computed beforehand and read in. Depending on the number of output fields, calculation of interpolation weights can require more CPU time than the time it takes to post the fields. Obviously, operational Eta runs utilize pre-computed interpolation weights. However, this is not necessary. The post retains the ability to compute interpolation weights itself prior to posting fields. By stringing together several control files the user may request that the same or different fields be posted on different output grids. In turn, these different grids may be in the same or different output files.

The simplest way to describe the control file is by means of an example. Figure 8.2 shows a portion of the operational Eta control file. A control file consists of three basic components: the header, body, and end mark. In the header the user sets the output grid type, the data set name, the data format, a new file flag, output grid specifications, and two additional input/output flags. Following the header, the user specifies which fields and levels to post. The post processor has a fourth order smoother and a 25 point Bleck filter through which data may be passed. The user controls these features by setting integer switches in the body. The order in which the post processes requested output fields is fixed by the code, but the order in which the user requests the fields is irrelevant. The body of a control file only needs to list those fields the user wants. To allow for this flexibility, every control file must end with an end mark. The end mark line tells the post processor to stop reading the control file and start posting requested fields.

The key to understanding the header is remembering what the variable name at the start of each line means. KGTYPE is a non-negative integer representing the type of output grid. The convention here is to use Office Note 84 grid types with two exceptions. The first exception is grid type 99999 which is the staggered E grid, regardless of the horizontal resolution. The second exception is grid type 00000. This grid type instructs the code to post the requested field(s) on a filled E grid. In the upper portion of Fig. 7.2, grid type 94 is the 22km domain as defined in GRIB documentation (Office Note 388). The string "START OF THIS OUTPUT GRID" is simply added for readability. The post processor ignores everything in the header outside of the parentheses. Each line of the header contains the data format the post expects to read. Proper spacing is crucial.

DATSET is the root around which the post creates output filenames. To this root the post appends the format of the output file and the forecast hour xx as specified in namelist FCSTDATA. Through the character string OUTYPE the user specifies the data packing to be used when writing the output file. Two formats are available: unpacked sequential binary and GRIB 1. Setting OUTYPE to NO tells the post to write data using unformatted FORTRAIN writes. If DATSET equals NOHEAD when OUTYPE equals NOPACK, no headers are written to the binary output file. Otherwise, a grid header starts each file and each output field is preceded by a record denoting the type and level of the field. Setting OUTYPE to GRIBIT produces a GRIB 1 packed dataset.

When GRIB output is requested, the two digit forecast time is appended to DATSET to form the the first part of output filename. For example, if the environmental variable tmmark in the running script for the Eta post processor (e.g. Fig. 7.3) is set to tmyy, the first output file generated by the Eta post using the control file in Fig. 7.2 is conventionally named EGDAWPxx.tmyy, where xx is the forecast time and yy is equal 00 in the forecast and is negative in the EDAS. For sequential binary output, .SbinFxx is appended to DATSET. Variable NUFILE allows the user to specify whether fields requested in the body are to be appended to a currently open output file or if a new output file is to be opened. It is a simple YES/NO switch.

The indented variables in the header deal with the output grid and pre-computed interpolation weights. PROJ, NORTH, IMOUT, JMOUT, POLEI, POLEJ, ALATVT, ALONVT, and XMESHL are the basic set of parameters by which standard NCEP software defines different types of geographical grids. PROJ is a character*6 string denoting the type of output grid projection. Three projections are currently supported: POLA for polar stereographic, LOLA for latitude-longitude, and LMBC for Lambert conformal conic. If the user wants grid relative winds on the native model grid, PROJ must equal LOLA. NORTH is a logical flag for the northern (. TRUE.) or southern (. FALSE.) hemisphere. (IMOUT, JMOUT) are the number of west-east and south-north grid points (directions relative to the rotation specified by ALONVT). Grid point (1,1) is in the southwest corner of the grid; (IMOUT, JMOUT) in the northeast corner. POLEI and XMESHL define the north-south and west-east grid increment on transformed grid respectively. POLEJ and ALONVT are geodetic longitude and latitude of the center of the E-grid. ALATVT is only required for Lambert conformal grids. It is the latitude at which the projection is tangent to the surface of the earth.

The user may sidestep this method to defining an output grid by setting READLL to YES. This tells the post to read an input file containing the geodetic latitude (glat) and longitude (glon) of output grid points. The post can read multiple (glat,glon) files starting from unit number 30. The structure of the (glat,glon) file expected by the post is ( ( (glat (i,j ),glon(i,j ) ), i=l, imout), j=l, jmout) using FORMAT 5 (gl2.6, lx). This option of the post has not been exhaustively tested since most users desire data on standard NCEP grids.

Whenever the user is not posting data on the model grid, it is recommended that interpolation weights be pre-computed. The user tells the post to read an external weight file by setting READCO to YES. If READCO equals NO the post processor will compute all necessary interpolation weights prior to posting any fields. Source code to pre-compute interpolation weights resides in ~wd22tb/etafcst/post/e2gd. See the Read me files in this directory for details. The user must ensure that the order in which interpolation weights are assigned in the template is the order in which the grids are listed in the control file (see Section 1.5 for elaboration).

The bulk of the control file is the body. This is where the user specifies which fields to post and optionally the degree of smoothing or filtering to apply to the posted fields. There are over 150 unique fields that may be posted from the Eta model. This, of course, is subject to change in response to model development and user needs. As mentioned above only those fields which are desired need to be listed in the control file. Each field specification consists of two lines. The first line, the identifier line, starts with a brief description of the field. The post processor ignores this. Following this are blocks Q= (xxx) and S= (xxx), which are now ignored by the code.

The SMTH block on the identifier line controls the smoothing or filtering. In most applications the model to output grid process involves two steps. First the staggered E grid is filled to make a regular grid. This is then interpolated to the output grid. Multiple pass smoothing or filtering of the data may be activated at any of three places in this process. The first element of the SMTH block activates a fourth order smoother that works on the staggered E grid. A positive, nonzero integer tells the post to apply this smoother to the field the indicated number of times. A more heavy handed multiple pass smoother was found necessary to produce pleasing vorticity fields. Thus, when smoothing a vorticity field, it is this smoother, not the fourth order smoother, that is applied. Once data are on a regular grid a 25 point Bleck filter may be applied in one of two possible places. The second integer segment in the SMTH block controls the filtering of data on a filled E grid. The last integer block of SMTH activates the Bleck filter on the output grid. The Bleck filter is designed to remove scales smaller than twice the grid separation. It has a fairly sharp response curve and will largely preserve field maxima and minima even with several applications.

Following the identifier line is the level line (L =) where the user requests data on particular levels. There is room for output on as many as sixty levels. Some fields (e.g., total precipitation, shelter temperature, tropopause pressure) are single level fields. For single level fields the integer 1 in the place immediately following the left parenthesis activates output of the field. In general the integer 1 activates output at a given level; 0 turns off output at that level. However, there are exceptions which are noted below.

For isobaric fields (fields for which S= 8) the pressure levels to which data may be posted are controlled by namelist FCSTDATA read in at the start of an Eta model integration. The order in which pressure levels are specified in FCSTDATA maps directly to the left to right ordering of integers on the level line. For example, using the FCSTDATA shown in Fig. 8.1, moving left to right across the level line are pressure levels 50, 75, 100, 125, .... ,975, and 1000 mb. Fields may be posted to different pressure levels by editing namelist FCSTDATA.

from Fig. 8.2. The field is geopotential height on isobaric surfaces. The Q and S integers are set for Office Note 84 packing. For each requested level two passes of the Bleck filter will be applied to data on the filled E grid and the output grid. Heights at all 39 isobaric levels as listed in Fig. 8.1 will be posted.

Two options are available for data on constant eta layers. Setting the n-th integer on the level line instructs the post to extract data on eta layer n. By "eta layer n" we mean the n-th eta layer using the top down vertical indexing of the eta model. At times it may be of interest to see what a selected field looks like at the n-th atmospheric (i.e., above ground) eta layer. This is a terrain following perspective. To activate this option, set the n-th integer (left to right) on the level line to any integer between 2 and 9, inclusive. For example, if a user wanted pressure data on the first, second, and fourth atmospheric eta layers the settings could be

In addition to eta layer and isobaric level data, multiple levels may be requested for FD (Flight level) fields and boundary layer fields. There are six FD levels. The ordering on the level line is from the lowest (914 m MSL) to the highest (3658 m MSL) FD level. Boundary layer fields are available from six 30 mb deep layers. The ordering on the level line is from the lowest (nearest the surface) to the highest constant mass layer. Two types of convective available potential energy (CAPE) and convective inhibition (CIN) are available. The first type (type one) starts the vertical integration from the lowest above ground model layer. The second type (type two) searches the six 30mb constant mass layers for the layer with the highest equivalent potential temperature. The CAPE and CIN calculation then starts from this level using this layer mean parcel. Type 1 CAPE and CIN are activated by setting the leftmost integer on the level line to 1. The second leftmost integer controls posting of type two CAPE and CIN. Both types may be requested. All other fields are single level, meaning that the leftmost integer activates (1) or deactivates (0) posting of that field.

The last section of a control file is the end mark. This single line tells the post processor to stop reading the control file and start posting requested fields. The key word on the line is DONE. The post scans each line read from the control file for this string. It is the only way to specify the end of a control file.

As shown in Fig. 8.2, individual header-body-end control files may be chained together to output data in numerous ways. The post reads the control files sequentially. If pre-computed interpolation weights are to be read in, the user must ensure that their assigned unit numbers correspond to the order in which the grids appear in the combined control file. One last detail involves the end mark at the end of the last control file. The post knows it has processed everything when it reads an end of file mark (EOF) from the control file. This EOF must immediately follow the last DONE statement. If not, the post will unsuccessfully try to process what it thinks is the next set of control cards.

7.5 The Template

The post processor was designed to run as a stand-alone executable. Figure 8.3 shows a script that can be used to run the Eta post processor on the IBM. The file itag is used to specify the posting times for the Eta post processor. The file contains three 2-digit numbers xx. The first two-digit number indicates the first forecast time the user wishes to post, the second one is the interval between the posting times, and the third one specifies how many forecast times the user wishes to post. As shown in the template the post reads as input (1) namelist FCSTDATA, (2) the constants nfile (nhb), (3) a restart file, and (4) the control file.

7.6 Summary

We have described how to use the Eta post processor in conjunction with the model. The post processor was designed to run as a stand-alone executable. The user controls the post by editing a control file. In this file the user specifies which fields to post, smoothing options, data format, and output grids. When running the product generator to output the grids on the non-native model grid it is recommended to pre-compute interpolation weights.

While the post processor can generate numerous output fields, it will never post every possible field. Every user will eventually find need for some field not available from the post. When the inevitable happens, several options exist.

First, any user can edit copies of the model and post processor to generate the desired output fields. The arrays needed to calculate the field must be added to the restart file generated by the model. Subroutine CHKOUT writes the model restart file. Post processor routine INITPOST which reads this file must be correspondingly edited. The new field must also be added to the control file. Lastly, code to generate the desired field must be added to the post processor. Where this code is added is not particularly important. However, post processor subroutine MISCLN has traditionally served this "catch-all" purpose. A list of available posted fields can be found in the routine POSTDATA.f.

For those who do not wish to tinker with the post processor code an alternative solution is to compute the desired field(s) directly in the model. If this is deemed too expensive, the model could simply write the information required to generated the new field(s) to an output file. An external piece of code written by the user would compute the new field(.s) from information in this output file. If the user wanted to pack the new field(s) in GRIB 1, appropriately edited versions post processor subroutines GRIBIT (and their supporting routines) could be added to the user's post processor code.

Table of Contents

1. Introduction