Changes to the NCEP Meso Eta Analysis and Forecast System: Increase in resolution, new cloud microphysics, modified precipitation assimilation, modified 3DVAR analysis

Eric Rogers, Thomas Black, Brad Ferrier*, Ying Lin, David Parrish, and Geoffrey DiMego

Mesoscale Modeling Branch and *Science Applications International Corporation

Environmental Modeling Center, National Centers for Environmental Prediction

1. Introduction

A series of changes to the NCEP Mesoscale Eta Analysis and Forecast System are described. These changes are designed to address several issues:

- Provide improved mesoscale guidance (via increases in horizontal and vertical resolution)

- Take full advantage of the parallel architecture and increased computing power of Phase II of the NCEP IBM-SP supercomputer

- Improve Eta model QPF and cloud predictions with a new and more sophisticated cloud microphysics scheme

- Introduce a faster and more scalable version of the Eta 3-dimensional variational analysis (3DVAR) into production, in anticipation of the addition of future high density (spatial and temporal) observations, such as NEXRAD radial wind velocity data.

The planned changes consist of the following:

1) Increase in resolution from 22 km / 50 levels to 12 km / 60 levels

2) A new, more sophisticated cloud microphysics scheme

3) Modifications to the observed precipitation assimilation algorithm in the Eta Data Assimilation System (EDAS) which make it compatible with the new cloud scheme

4) Numerous modifications to the Eta 3DVAR analysis, including the analysis of radiances from the NOAA-16 satellite

2. Increase in Resolution

In September 2000 the Eta model resolution was increased from 32 km / 45 levels to 22 km / 50 levels (Rogers et al., 2000). With this implementation the computational domain of the Eta model was restored to the size which was used in the 48-km Eta from 1995-1998 (see Figure 1 of Rogers et al., 2000). Although the Eta-22 uses 3 times the computing resources of the Eta-32, it still uses a relatively small percentage (< 10%) of the total processors available for production on the NCEP IBM-SP Phase II supercomputer.

Therefore, to improve real-time mesoscale guidance to National Weather Service forecasters and private users, and to take full advantage of the parallel computing resources of the IBM-SP, the Eta model resolution is being increased from 22 km / 50 levels to 12 km / 60 levels. The 12-km computational domain (see Figure 1 of Rogers et al., 2000) is identical to the 22-km domain, but with 3.3 times more grid points. Figure 2.1 shows the vertical distribution of eta levels in the Eta-22 and Eta-12. The greatest increase in vertical resolution is seen near jet stream level and in the boundary layer, with 24 layers below 850 mb in the Eta-12 versus 17 layers in the Eta-22.

Figure 2.2 shows the difference in the Eta grid orography between the Eta-12 and Eta-22 for the Western and Eastern U.S. Note the improved coastline details in the Eta-12 (e.g., Long Island is "longer" in the Eta-12, and in the Chesapeake Bay region).

Preliminary testing of the Eta-12 system was performed at NCEP during the late summer of 2001 without the new cloud microphysics scheme, but with the new 3DVAR analysis (which included the assimilation of NOAA-16 radiances which are not in the 22 km EDAS). An example of the impact of resolution on the prediction of cyclogenesis in the Gulf of Alaska is shown in Fig.2.3. Depicted at 48-h forecasts from the Eta-22 and Eta-12 valid at 1200 UTC 7 September 2001, with the operational Eta-22 sea level pressure analysis valid at that time. The 22 km EDAS (Fig.2.3c) analysed a 998 mb cyclone (identical to the NCEP Hydrometeorological Prediction Center's sea level pressure analysis (not shown). The Eta-12 forecast was much better in predicting the lower central pressure and stronger wind gradient, although the Eta-12 was slower in moving the cyclone towards the coast that was observed. Examination of a real-time Eta-22 parallel testing NOAA-16 radiances only (see Section 4) indicates that some, but not all, of the improvement in the Eta-12 sea level pressure prediction can be attributed to assimilating NCAA-16 radiances. This implies that the increased resolution had an impact on this prediction.

Figure 2.4 illustrates an example of the impact of increased Eta resolution on quantitative precipitation forecasts (QPF). Depicted are 36-h forecasts of 24-h accumulated precipitation from the operational Eta-22 and the Eta-12 (without the new cloud microphysics), along with the NCEP (rain gauge data only) precipitation analysis valid at 12Z 21 September 2001. The greatest difference between the Eta-22 and Eta-12 QPF is seen in the precipitation band extending along the Minnesota-Iowa border eastward towards Chicago, and in the precipitation extending from southern Oklahoma across the Red River into northeast Texas. For the latter the Eta-12 did a much better job of predicted the observed amounts, but did not predict the region of higher precipitation in the extreme northeast corner of Texas. Neither model predicted the widespread region of > 15 mm of precipitation in central Texas. For the northern precipitation band, the Eta-12 did a better job in predicting the band's position along the Wisconsin/Illinois border, but it did not extend the heavier amounts into Chicago, and it overpredicted the precipitation along the Minnesota/Iowa border.

3. New Grid-Scale Cloud and Precipitation Scheme

3.1 Description

A new grid-scale microphysics parameterization was formulated that predicts the variation of six species of water, replacing the simpler Zhao cloud scheme used in current operations (Zhao and Carr, 1997; Zhao et al., 1997). The new parameterization will be referred to as EGCP01, representing the Eta grid-scale cloud and precipitation scheme released in 2001. The following text describes in more detail the differences between the Zhao and EGCP01 schemes that are summarized in Table 3.1.

i. Prognostic variables. Both schemes predict changes in water vapor specific humidity. Cloud condensate in the form of cloud water or cloud ice (both cannot coexist) is predicted in the Zhao scheme. Suspended condensate is either in the form of cloud ice at temperatures colder than -15oC or cloud water at temperatures above 0oC. Between 0oC and -15oC, it is assumed to be cloud ice if the cloud top extends above the -15C isotherm, otherwise it is supercooled (i.e., colder than 0oC) cloud water for shallower clouds. The EGCP01 predicts total condensate, which is the sum of cloud water (small droplets that do not fall), rain (larger drops that do fall), and ice (all can coexist at temperatures warmer than -10oC). The ice category in EGCP01 combines attributes of small ice crystals, which are dominant in cirrus and upper tropospheric ice clouds, with larger ice particles in the form of snow (aggregates), graupel (varying degrees of rimed snow), and sleet (ice pellets), which are dominant at lower levels. The fractional contribution of cloud water, rain, and ice to total condensate is determined within the EGCP01 microphysics subroutines using several storage arrays. Future improvements to EGCP01 will include calculating advective tendencies for each hydrometeor category when computer resources are available.

ii. Condensation algorithm. Exchanges between water vapor and cloud condensate (cloud condensation, ice deposition) are calculated in EGCP01 using the Asai (1965) algorithm, which is used in many other high-resolution cloud-resolving and mesoscale models (e.g., Lord et al., 1984; Tao et al., 1989; Reisner et al., 1998; Tremblay and Glazer, 2000). It efficiently adjusts the temperature, moisture, and cloud condensate fields to a threshold relative humidity associated with the onset of cloud condensation, which is assumed to be a function of grid resolution in EGCP01. This threshold relative humidity approaches 100% at high resolutions (1-2 km) when subgrid-scale variability of moisture due to unresolved processes becomes sufficiently small. The Zhao scheme uses the condensation algorithm of Sundqvist et al. (1989). It is a more sophisticated scheme that considers changes in cloud fraction (the fraction of the grid box that is cloudy), grid-averaged cloud mixing ratios, and moistening rates. Both approaches should converge towards each other at increasing resolution, though this is only speculation and has not been investigated.

iii. Precipitation fluxes and storage. Surface precipitation is predicted in Zhao by downward integration of precipitation fluxes from cloud top down to the surface. There is no atmospheric storage of precipitation. Microphysical sources (sinks) of precipitation in the form of rain or snow within a grid box lead to increased (decreased) downward precipitation fluxes from the top to the bottom of the grid box. Whatever remains of the rain (snow) not lost by evaporation (sublimation) below cloud base is accumulated at the surface. The EGCP01 also performs top-down integration of precipitation fluxes because of its computational efficiency; however, it does allow for atmospheric storage of precipitation. Downward fluxes of precipitation are partitioned between local storage in the grid box, which is proportional to the layer thickness, and the fall out of precipitation through the bottom of the grid box. Microphysical sources (sinks) of precipitation within a grid box produce larger (smaller) precipitation fluxes through the bottom of the box and higher (smaller) mixing ratios of precipitation suspended within the grid box. This extension of EGCP01 is most important when representing the gradual sedimentation of ice particles from upper levels, which can be important for properly simulating the advection of ice across mountain ranges and the transport of ice across regions of strong vertical shear often associated with extratropical systems.

iv. Precipitation type and rime physics. Surface precipitation in the form of rain, freezing rain, or snow is predicted in Zhao. These are also predicted in EGCP01, except that the "snow" (a more accurate term is precipitation ice) can be in the form of low-density aggregates ("fluffy" snow), rimed snow, graupel (dense, heavily rimed snow), or sleet (ice pellets). The density of rimed ice varies depending on whether it is formed by freezing of supercooled rain (sleet) or by cloud water riming onto snow. The effective density of precipitation ice is calculated by storing in a local array the contributions made to the growth of ice by riming of supercooled cloud water and freezing of supercooled rain. Implicit in this scheme is the assumption that the collected liquid water is filtrated into the air holes of a porous ice particle, thereby increasing its mass and preserving its volume and cross sectional area.

v. Mixed phase conditions. Although supercooled cloud water is assumed to exist in shallow clouds with cloud-top temperatures warmer than -15oC, the Zhao scheme otherwise does not consider mixed-phase conditions where water and ice particles could coexist. Mixed-phase processes are considered in EGCP01 at temperatures above -10oC (e.g., see Fig. 3.4). At colder temperatures supercooled rain is assumed to freeze to form sleet and supercooled cloud water freezes to form ice. Although earlier versions of EGCP01 allowed mixed-phase processes to exist at colder temperatures, the forecasts were too moist when compared against rawinsonde observations and initial analysis fields during retrospective runs. Future Eta forecasts should be able to provide improved guidance on the potential for aircraft icing, thereby complimenting the more specific aviation-related forecasts provided by the RUC-2.

vi. Freezing and melting processes. The Zhao scheme uses a fairly simple relationship to calculate the melting of snow as a function of temperature only. Small amounts of snow have been found to reach the surface at unrealistically warm temperatures (sometimes approaching 15oC) over isolated regions (Fig. 3.5). No freezing processes are calculated in Zhao, including when clouds change from water to ice in the 0o to -15oC layer. Freezing and melting processes are more sophisticated in EGCP01. Latent heat of fusion associated with the collection of cloud water riming onto ice, freezing of supercooled rain, and melting of ice particles are all considered. Relative humidity effects upon melting rates are also considered, in which condensation onto melting ice will occur in moist conditions and evaporation from wet melting ice will occur in drier conditions (e.g., Musil, 1970; Orville and Kopp, 1977; Lin et al., 1983; Rutledge and Hobbs, 1984; Ferrier, 1994). Vapor condensation onto wet, melting ice in moist conditions will warm the air and increase melting rates, whereas evaporation from wet ice in drier conditions will cool the air and slow melting rates. The EGCP01 also calculates the collection of cloud water by melting ice, and this accumulated water is shed from the melting ice to form rain. Though this process may appear esoteric to some, it can be an important source of rain when grid-scale ascent and cloud condensation occur below the melting level.

For the other microphysical processes, both schemes predict cloud condensation and evaporation, ice deposition and sublimation, rain evaporation, autoconversion (self-collection) of cloud water to rain, and cloud water collection by falling rain. The Zhao scheme also includes the autoconversion of cloud ice to snow and collection of cloud ice crystals by snow, but these last two processes are treated in an overly simplified manner. Observational and numerical modeling studies support a clear distinction in the liquid phase between small, non-precipitating cloud droplets and precipitating raindrops at sizes between 80 and 100 microns. Studies of ice in clouds, however, indicate that no such distinction is justified, where instead there is a continuum in the mass and fall-speed characteristics of ice particles as a function of size that is further complicated by a myriad of different habits. This is clearly a challenge for all bulk microphysical schemes, but particularly for operational NWP models where the sophistication of the physical parameterizations are limited by computing speed and production schedules. The following added features of EGCP01 described in Table 3.2 address these challenges.

1: For purposes of speed and efficiency, the EGCP01 assumes that the sizes of the ice particles vary as a function of temperature based on observations of precipitating layer clouds (Fig.3.1). Exponential distributions are assumed for all precipitation particles (rain and precipitation ice). Based on ice observations of precipitating clouds from the Pacific Northwest (Houze et al., 1979), the mean diameter for the ice is assumed to decrease from 1 mm at 0oC to 50 microns at -55oC. This mean diameter is only a first guess. A final mean diameter is calculated so that the number concentrations of ice are between 0.1 and 20 L-1. Although the observations were made only at temperatures warmer than -40oC, it was decided for convenience to extend these relationships to colder temperatures. Cloud ice is composed of small, exponentially distributed ice crystals with a fixed mean diameter of 50 microns. The number concentrations of small ice particles are assumed to be a factor of four higher than that of large ice particles in ice saturated conditions. In drier conditions below ice saturation, only precipitation particles are assumed to exist. Preliminary findings indicated that ice sublimation rates were too large when a fixed fraction of small ice particles was assumed in subsaturated air, leading to moist biases in the upper-level humidity fields compared to rawinsonde observations. This rudimentary aspect of the scheme is intended only to crudely mimic observations of cirrus and other cold clouds. Improvements will be made in this area once computing resources are available to better resolve these cloud processes. Results from GCSS (GEWEX Cloud System Studies), ARM (Atmospheric Radiation Measurement) and other programs will also lead to improved understanding and representation of ice clouds in NWP models.

2: The mass and fall speed characteristics of ice particles vary with temperature and moisture associated with the predominance of different habits over different layers of the atmosphere. As a simpler introductory step in incorporating this effect into a NWP model, a composite set of mass-diameter (m-D) and fall speed-diameter (V-D) relationships were derived for unrimed ice crystals and aggregates. The m-D and V-D relationships for unrimed ice are an average of the relationships for bullets, columns, and plates from Starr and Cox (1985) following the work of Heymsfield (1972). Composite relationships for aggregates are an average of the three relationships for unrimed aggregates in Locatelli and Hobbs (1974). Since the mass and fall-speed relationships for unrimed ice crystals and aggregates overlapped near 1.5 mm diameter, the final mass and fall speeds of ice use the average values of unrimed ice crystals smaller than 1.5 mm diameter and that for aggregates at larger sizes. The average ice-particle masses and fall speeds are stored at 1-micron intervals for sizes ranging from 20 microns to 2 cm, and these values are used to calculate various particle moments. For example, the mass for a spherical particle of constant density is associated with the third moment, or D3 with D representing the ice-particle diameter. Different moments of the ice-particle spectrum are associated with various physical processes, such as ventilation (for vapor deposition and sublimation), collection (cloud water riming, collection by rain), mass content, and precipitation rate. What almost all bulk-microphysical models require is expressing these relationships as functions of the mean diameter of the exponential distribution. These quantities are obtained in EGCP01 by integrating the various particle moments over different mean diameters and storing the solutions in lookup tables at 1 micron resolution. The EGCP01 also predicts changes in the bulk density of rimed ice, taking into account the increase in fall speed as a function of particle density (see item iv) following the general formulation of Böhm (1989). The calculations are fairly complex and the solutions are also stored in lookup tables. A lot of detailed physics are included in these tables. An analogous set of lookup tables are also used for calculating various microphysical processes involving rain.

3: Those microphysical processes that lead to rapid removal of condensate are expressed in implicit exponential functions, which are more stable with time since the Eta uses large time steps (8-10 min) in most atmospheric physics calculations. The errors were found to exceed 1 percent when the total decrease in hydrometeor mixing ratio (rate times time step) exceeded roughly 5% of its initial value (e.g., the loss of cloud water due to collection by falling rain). In this regard, the modified top-down integration of precipitation fluxes described in item iii and in Table 3.1 is an inexpensive alternative to other bulk microphysical schemes, which explicitly calculate the flux convergence of precipitation using the flux form of the second-order vertical advection equation. The latter requires the use of increasingly smaller time steps with finer vertical resolution, particularly for hydrometeors with appreciable fall speeds like sleet or rain. The EGCP01 can therefore be run over a wide range of vertical and horizontal resolutions and different time steps without concern for numerical stability.

As a consequence of the first two items in Table 3.2, the "seeding" or downward transport of ice from upper to lower levels is better represented in EGCP01. The performance of the current operational microphysics, like many other bulk schemes, is somewhat dependent on properly selecting tunable coefficients that represent the conversion from cloud ice to snow. This is expected to be much less of a problem in the new grid-scale scheme.

The benefits of the new grid-scale scheme are demonstrated in Figs. 3.2 through 3.14 and summarized in Table 3.3. The next two sections will discuss the formation of unrealistic, locally intense precipitation maxima in the current microphysics and the excitation of spurious gravity waves in the model.

3.2 Local precipitation maxima associated with grid-scale processes.

During the past several months the Eta model has on occasions produced highly localized areas of intense precipitation ("bulls eyes"). These precipitation bulls eyes most often occur when convective and grid-scale cloud schemes are both active, in which the rate of drying and stabilization by the convection is not enough to prevent the column from being at or near saturation over a deep layer at lower and middle levels. Once the column becomes saturated over a deep layer, precipitation evaporation ceases and grid-scale processes become highly efficient. The situation is exacerbated in the current operational microphysics due to the instantaneous fall out of precipitation and the lack of atmospheric storage of slower falling ice particles. Figure 3.6 shows an example of precipitation bulls eyes produced from the Zhao grid-scale microphysics from the 10-km central nest. These intense precipitation maxima occurred in the 10-km central nest between 36 h and 39 h into the forecast. These very intense, localized features are not produced in forecasts using the EGCP01, primarily because precipitation ice descends more gradually with time. Localized production of ice at upper levels, such as in association with the front in central Illinois and Indiana, will be dispersed by the winds and not necessarily result in the immediate production of surface rainfall as in the operational model. These precipitation bulls eyes in the operational models will also produce fast-moving gravity wave modes, such as those shown for this case in Fig. 3.7. An example of precipitation bulls eyes produced from the current grid-scale microphysics in both the 10-km and the 22-km Eta model runs are shown in Figs. 3.8 and 3.9, respectively. Spurious gravity waves associated with this case will be discussed in more detail in the next subsection.

3.3 Temperature tendencies associated with cloud processes.

The current operational Eta model periodically updates the temperature field due to convective and grid-scale processes at time intervals associated with physics calculations every 540 s. These episodic updates in temperature will suddenly warm the atmospheric column through the release of latent heat associated with precipitation. The column-integrated warming is particularly dramatic in areas where grid-scale precipitation bulls eyes form. The result of this sudden warming is the excitation of fast-moving gravity waves. This can be seen in the forecasts from the operational control runs as oscillations in the vertical motion field (Figs. 3.7, 3.10, and 3.11), noise in the geopotential heights (Figs. 3.12 and 3.13), and noise in the surface pressure fields (Figs. 3.8 and 3.9). These waves are effectively eliminated in the forecasts using the new grid-scale microphysics by calculating the temperature tendencies from convective and grid-scale cloud processes and then incrementally updating the temperature fields every time step. Comparing the forecasts of vertical motion (Figs. 3.10 and 3.11) and geopotential heights (Figs. 3.12 and 3.13), the amplitude of the spurious gravity waves associated with the current operational Eta model are stronger in the higher resolution runs because intense precipitation features are more explicitly resolved by the grid-scale microphysics.

3.4 Vertical eddy mixing of condensate

Turbulent mixing in the vertical now modifies total condensate (now the sum of all cloud and precipitation fields) in the same way as temperature and moisture. The main impact of this change is to reduce the extensive areas of light precipitation (mist or drizzle) over the oceans, such as off the Baja California coast (Fig. 3.14) and on occasions over the Gulf of Mexico, by enhancing the mixing of cloudy air with drier above cloud top. It should be noted that vertical eddy mixing of condensate did not contribute much to reducing the area of light precipitation given by the bias scores in Fig. 3.2.

Table 3.1 Comparing grid-scale microphysical schemes

Feature Current scheme (Zhao) New scheme (EGCP01)
Prognostic variables Water vapor, cloud condensate
(cloud water or cloud ice)
Water vapor, total condensate
(cloud water + cloud ice + rain + snow/graupel/sleet)
Condensation algorithm Sundqvist et al. (1989) Asai (1965) algorithm used in Tao et al. (1989), Reisner et al. (1998), and others
Precipitation fluxes and storage Top-down integration of precipitation fluxes. No precipitation storage, instantaneous fall out of rain and snow. Top-down integration, precipitation fluxes partitioned between local storage in grid box and fall out through bottom of grid box. Storage of all forms of precipitation, especially important for snow.
Precipitation type and rime physics Rain, freezing rain, or snow. Cloud-water riming onto snow is not considered. Rain, freezing rain, snow, graupel, or sleet. Rime physics calculates variable ice density from "fluffy" snow to sleet.
Mixed-phase conditions No mixed-phase conditions. Condensate is either ice or water. Mixed-phase conditions at temperatures >-10oC.
Freezing and melting processes Freezing processes are not considered.
Uses a simple melting equation.
Includes cloud riming onto ice, freezing of supercooled rain. More sophisticated melting physics affected by relative humidity, includes heat and moisture exchanges.

Table 3.2 Added features of new grid-scale microphysics (EGCP01)

1. Assumes size of precipitation ice is primarily a function of temperature (Ryan, 1996).
2. Computationally efficient lookup tables store complex calculations for rain and precipitation ice. Fall speed of rimed ice increases as a function of calculated rime density, results stored in lookup tables.
3. Numerically stable calculations of rapid rates for large time steps.

Table 3.3 Summary of improvements of new grid-scale microphysics (EGCP01)

  • Less overprediction of light precipitation (Figs. 3.2 and 3.14).
  • Less overprediction of upper-level cloudiness (Fig. 3.3).
  • Representation of mixed-phase processes (Fig. 3.4).
  • Improved forecasts of precipitation type (Fig. 3.5).
  • Fewer, less intense grid-scale precipitation bulls eyes in 10-km and 22-km simulations (Figs. 3.6, 3.8, and 3.9).
  • Much less spurious gravity-wave excitation (Figs. 3.7, 3.10, 3.11, 3.12, and 3.13).
  • Designed for higher resolution mesoscale models, physics more compatible with RUC-2.

4. Precipitation assimilation

4.1. Description of changes

The precipitation assimilation has been revamped to work with the new cloud microphysics scheme. The precipitation assimilation in the current operational Eta/EDAS was designed to work in tandem with the Eta's convective parameterization and grid-scale cloud microphysics, which has, until now, been the Zhao scheme (Zhao, 1997). The change to the new grid-scale cloud package necessitates extensive changes to the precipitation assimilation scheme.

The essence of precipitation assimilation in the Eta Data Assimilation System (EDAS) is to use observed precipitation to adjust the model's latent heating, moisture and cloud water fields during the 12-hour pre-forecast assimilation period. To that end, at each time step during EDAS, for each grid point where precipitation observations are available, we compare the model precipitation (Pmod ) against the observations (Pobs ), make adjustment to the model precipitation field and corresponding adjustments to the models temperature, water vapor and cloud fields.

When Pmod > Pobs, we try to take back the "extra" precipitation and the latent heating associated with it. When Pmod < Pobs, we try to nudge the model into producing more precipitation by first checking if more convection is possible; after that, if we still need the model to produce more precipitation, we try to do so in the grid-scale mechanism.

A major difference from the old precipitation assimilation is that, in the new scheme, the adjustment in the grid-scale goes through an iterative process, as shown in Table 4.1. This gives the model a chance to adapt to the adjustments more smoothly.

A schematic overview/flowchart of the precipitation physics/precipitation assimilation procedure is listed in Table 4.1. Detailed information about the precipitation assimilation procedure is provided in Table 4.2.

4.2. Impact of changes

The model precipitation during EDAS improved somewhat in mid-ranges (0.1"- 0.75"/day, which contributes the bulk volume of precipitation reaching the ground. See http://www.emc.ncep.noaa.gov/mmb/ylin/imp200111/pscore/EDASPPT.wretro.precip.jpg). The large precipitation bias at the high end of daily precipitation thresholds (3"/day) has been corrected because we now use a lower limit when reading in the precipitation analyses to filter out very large values due to bad gauge data.

Comparisons of precipitation over periods of a month (cold and warm season retrospectives vs. control runs vs. observations) and the resulting impact on the soil moisture are shown in http://www.emc.ncep.noaa.gov/mmb/ylin/imp200111/#PCPEDS.

Statistics showing impact on precipitation forecast, upper-air and near-surface fields can be found in http://www.emc.ncep.noaa.gov/mmb/ylin/imp200111.

Table 4.1. Schematic overview of the old vs. new cloud physics/precipitation assimilation procedure

Steps

Old scheme

New scheme

1.

Grid scale condensation/evaporation of cloud

 

2.

Convective precipitation

3.

Grid-scale precipitation

Grid-scale cloud physics driver: for each horizontal grid point, call the columnar grid-scale microphysics module (GSMCOLUMN), which contains both cloud and precipitation calculations

4.

Precipitation assimilation:for each horizontal grid point, compare Pmod against Pobs, adjust the following model fields:

        Pmod (i.e. precipitation accumulation on the ground)

        Temperature

        Water vapor

        Cloud (cloud water + ice)

Precipitation assimilation: for each horizontal grid point,

1)      compare Pmod against Pobs, adjust the following model fields:

        Pmod (i.e. precipitation accumulation on the ground)

        Temperature

        Water vapor

        Precipitation and cloud water mixing ratios

2)      Call GSMCOLUMN again.Add any precipitation generated in this step back to the accumulated precipitation

3)      Compare Pmod (from the GSMCOLUMN call above) against Pobs, then adjust the following fields again:

        Temperature

        Water vapor

        Precipitation and cloud water mixing ratios

 

 


Table 4.2.Detailed comparison of old vs. new precipitation assimilation procedure

 

Table 4.2a.IfPobs = 0 butPmod > 0:

 

Steps

Old scheme

New scheme

1.

Zero out all precipitation accumulated through this physics time step (Pmod).

2.

Take back the latent heat generated from the past physics time step from the modelís temperature profile. Adjust water vapor mixing ratio (qv) so the relative humidity (RH) remains unchanged.

3.

Reduce cloud water mixing ratio to no more than the minimum amount required to produce preciptation.

Zero out all precipitation in this column. Zero out all cloud ice.

Bring cloud water to below the auto-conversion threshold.

4.

 

Call GSMCOLUMN again.Add any precipitation generated in this step back to the accumulated precipitation.

5.

 

Repeat steps 2-4.



Table 4.2b. IfPmod >Pobs > 0:

Steps

Old scheme

New scheme

1.

Reduce modelís precipitation accumulation by taking back the Ďexcessive precipitationí (Pobs - Pmod).

2.

Adjust the modelís temperature profile by modifying the latent heat release in the last physics time step proportionally (multiplying Tlat by the factor Pobs/Pmod).

3.

RH adjustment

Modify water vapor mixing ratio so that outside of the cloud, RH stays unchanged, and within the cloud, RH is the new cloud schemeís grid-scale threshold RH for clouds.

4.

Reduce cloud water mixing ratio (which includes both cloud water and cloud ice) to no more than the minimum amount required to produce preciptation.

Multiply the modelís cloud ice and precipitation mixing ratios by the factor Pobs/Pmod.

 

5.

 

Call GSMCOLUMN again.Add any precipitation generated in this step (mod) back to the accumulated precipitation.

6.

 

Check model precipitation produced in step 5 (mod) against Pobs:

        If mod >Pobs,repeat steps 2-5.

        If mod <Pobs and Pobs/mod < 10, go to Table 4.2c, step 6.

        If Pobs/mod > 10, go to Table 4.2c, step 5.



Table 4.2c. If Pmod <Pobs:

 

 

Steps

Old scheme

New scheme

Convective adjustment

1.

Go to the Betts-Miller-Janjic convection scheme to see if deep convection is possible.If so, accelerate the convective process with the goal of producing an amount of convective precipitation (Pcnv) equal to Pobs, up to the limit allowed by the convective scheme.

 

Grid-scale adj. decisions

2.

If Pcnv = Pobs, adjustment for this time step is complete. If Pcnv < Pobs, then we will go to the grid scale precipitation process to generate the rest of ďmodel precip deficitĒ (Ptarget = Pobs - Pcnv), as below:

3.

Check the modelís grid scale precipitation produced in this time step prior to precipitation adjustment (Pgrd) against Ptarget.

4.

If Pgrd > Ptarget, follow the steps listed in Table 4.2b.

Specifying a parabolic grid-scale heating profile

5.

If Pgrd = 0 or if Ptarget /Pgrd > 10, we create a layer of precipitating cloud:

 

5a

See if the model has a layer of cloud that is at least 200mb in thickness.If not, create such a layer by specifying that cloud base is 150mb above the surface, and cloud top/cloud thickness is dependent on the amount of grid scale precipitation (Ptarget) we need to generate:

        If the hourly precipitation rate corresponding to Ptarget is more than the equivalent of 2.81mm/hr, cloud top is at 200mb

        If Ptarget is equivalent to between 0.375 and 2.81mm/hr, cloud depth is 450mb (or less, if cloud top has reached 200mb)

        If Ptarget is less than the equivalent of 0.375mm/hr, cloud depth is 300mb (or less, if cloud top has reached 200mb)

5b

Specify a parabolic latent heating profile within the aforementioned cloud layer.The shape of the parabola is determined by the amount of latent heat that would be generated if Ptarget has been condensed in this time step.

5c

If Ptarget is more than the equivalent of 1mm/hr, keep RH at 80% within the cloud.

Keep RH at the grid-scale threshold RH within the cloud.

5d

Within the cloud, keep the cloud mixing ratio above the threshold for generating precipitation.Add the Ptarget to the precipitation accumulation (in the Zhao scheme, the precipitation falls out instantly).

No adjustment to cloud water or cloud ice.Distribute grd vertically (assuming a straight rain water mixing ratio profile) between the freezing level and the surface (if the entire column is freezing, distribute Ptarget in the two levels at the surface and immediately above.

 

 

5e

Adjustment is complete

Go to step 7.

 

6.

Here we have Pgrd > 0 and Ptarget /Pgrd 10:

 

 

6a

Within the cloud layer, change the temperature profile by multiplying the latent heating with the ratio Ptarget /Pgrd.

 

 

6b

Keep RH constant.

Modify water vapor mixing ratio so that outside of the cloud, RH stays unchanged, and within the cloud, RH is the new cloud schemeís grid-scale threshold RH for clouds.

 

 

6c

 

Multiply the modelís cloud ice and precipitation mixing ratios by the factor grd/Pgrd.

 

7.

 

Call GSMCOLUMN again.Add any precipitation generated in this step to the accumulated precipitation.

 

8.

 

Check model precipitation produced in step 7 (grd) against the target grid-scale precipitation we want to generate (grd):

        If grd >grd, perform steps 2-4 in Table 4.2b.

        If grd <Pí grd and grd/grd 10, repeat step 6 in this table.

        If Pobs/mod > 10, repeat step 5 in this table.

5. Changes to the Eta 3-dimensional variational analysis

5.1. Updating of radiance processing

In order to use NCAA-16 radiance data and AMSU-B moisture channels (on both NCAA-15 and NCAA-16), it was necessary to bring the Eta 3DVAR radiance processing up to date with the current operational global 3DVAR. The most significant change made in the global system was to simplify the bias correction scheme, moving from 6 to 4 predictors per channel. There were also some very minor adjustments to quality control, mostly to eliminate obviously cloudy IR data early in the processing to increase computational efficiency and to make the thinning of the data more likely to choose a clear spot. These changes were run in parallel to the operational 22km system during August- September 2001 to make sure that the new radiance processing was not significantly different from the currently operational Eta 3DVAR. In terms of quantitative skill scores, the overall impact of the NCAA-16 radiances was small (details can be found at http://www.emc.ncep.noaa.gov/mmb/mmbpll/etaxstats/) . An example of the impact of these data on a case of oceanic cyclogenesis is shown in Figure 5.1. Depicted is the 48-h forecast of sea level pressure from the operational Eta-22 and the parallel Eta-22 using NOAA-16 radiances valid at 1200Z 7 September 2001, along with the operational Eta analysis valid at that time. Assimilating NOAA-16 radiances led to the prediction of a 4 mb deeper cyclone in the parallel run when compared to the operational Eta-22.

5.2. Addition of WSR-88D radar radial wind processing

The capability to assimilate radial winds from WSR-88D radars was added and tested with a 10km nested Eta run. The data are "super" observations generated from NEXRAD. The superobs convert the 1-km x 1-deg azimuth radial wind values to averages over 5km x 12deg azimuth x 1 hr. No quality control was done in the initial test. Other than obvious deterioration due to bird migration, there was small positive impact. The radar data will not be turned on in this implementation, but we plan to do further tests with simple extension of VAD quality control algorithm (Manikin et al., 2000) to the radar wind superobs, with a possible implementation early next year.

5.3. More efficient recursive filter code for background error

A new version of the recursive filter code is being implemented, which is more scaleable, and has anisotropic capability. For this implementation, isotropic settings are used which give bit-reproducibility with the recursive filter code currently running in the operational 3DVAR code. Preliminary tests are under way which will allow the moisture correlations to be anisotropic. This is being done using an extension of the so-called "NMC method" for generating background error statistics from a large number of 24 - 48hr global forecasts verifying at the same time. This was the procedure used to generate the current operational Eta background error. The number of tunable parameters resulting from the anisotropic extension is unchanged at 5, so tuning should be straightforward. After tests with anisotropic moisture are complete, similar definitions of anisotrophy will be extended to the other analysis variables--stream function, potential function, unbalanced temperature and unbalanced surface pressure.

6. Results

A real-time parallel test of the full change package was started at 0000 UTC 13 October 2001. A detailed set of quantitative skill scores for this parallel test can be found at http://www.emc.ncep.noaa.gov/mmb/mmbpll/eta12stats.newcloud/ . Overall, Eta-12 fits to rawinsondes (in terms of root-mean-square errors) are the same as or slightly worse for temperatures and heights, and generally worse for winds (not shown). However, the greatest impact of the Eta-12 on skill scores is seen in precipitation skill scores (positive) and in root-mean-square errors for relative humidity (negative). Some details are discussed below.

6.1. Precipitation skill scores

Figure 6.1 shows the equitable threat (ETS) and bias scores for 24-h accumulated precipitation for all Eta-22 and Eta-12 forecasts (all ranges) from 15-30 October 2001 over the contiguous United States (CONUS). The ETS is higher at all thresholds in the Eta-12, accompanied by lower bias. When these scores are calculated for the Eastern and Western U.S. (Figure 6.2) we see that the greatest positive impact for this period is seen in the Western U.S. Examination of individual forecasts during this period (not shown) imply that the improved QPF for this period from the Eta-12 in the Western U.S is due to better forecasts of orographically induced precipitation, which can be attributed. to the more accurate depiction of surface terrain in the Eta-12.

An example of improved QPF in the Eta-12 which is attributable to the new cloud microphsyics is shown in Figure 6.3. Depicted is the 24-h accumulated precipitation valid at 1200 UTC from the operational Eta-22 and parallel Eta-12 with the full change package. The observed precipitation analysis indicates that the only significant rainfall in the CONUS is occurring in southeast Florida, northern Idaho, and western Washington. No precipitation fell in the mid-Atlantic states. The Eta-22, however, predicted a swath of precipitation > 10 mm to fall from eastern North Carolina northeastward into southern Delaware, with measurable amounts as fall north as southern New Jersey. The Eta-12 also predicted a swath of precipitation in this region, but the total amount forecast was < 5 mm and the swath was placed further south than in the Eta-22. The 500 mb flow (not shown) was west-southwest over the mid-Atlantic states, with a short-wave trough located to the north over Michigan and Lake Huron. There were no appreciable differences (not shown) between the Eta-22 and Eta-12.

Figure 6.4 shows the 12-h forecast of convective precipitation valid at 1200 UTC 20 October from the operational Eta-22 and the Eta-12. In the Eta-12, all of the spurious precipitation over North Carolina and the southern Delmarva Peninsula is convective, which in the Eta-22 a significant portion of the precipitation is grid-scale. The implication is that the improved cloud scheme in the Eta-12 reduced the magnitude of the Eta QPF error.

6.2. Relative humidity skill scores

Figure 6.5 shows the relative humidity root-mean-square (RMS) errors versus rawinsondes over the CONUS for all 12, 24, 48, and 60-h forecasts from the operational Eta-22 and Eta-12 (with the full change package) for 15-30 October 2001. Generally, the relative humidity RMS errors in the Eta-12 are 5-10% higher than the Eta-22 throughout the troposphere. This result can be attributed to the differences between the new cloud microphysics scheme in the Eta-12 and the Zhao et al. (1997) scheme in the operational Eta. In the Zhao scheme the threshold for condensation is 75% RH over land and 80% RH over water. In the new scheme, this threshold is much higher, and it is scaled to the grid resolution. At 12-km resolution the threshold in the new scheme is 97.7%. With this higher threshold, the Eta forecast will predict a more realistic cloud pattern (generally less cloud) but with the higher threshold for condensation there will be more ambient water vapor, i.e., higher relative humidity.

6.3. 22-km retrospective rests with the full change package

Two sets of one month retrospective tests have been performed at 22-km resolution. Both a warm season (12 August - 12 September 2000) and a cold season (February 2001) period was run. The control run for these two period includes all the changes to the Eta system (Rogers et al., 2001 : new land-surface physics, assimilation of analyzed precipitation) implemented in July 2001. Quantitative verification scores for these retrospectives can be found at http://www.emc.ncep.noaa.gov/mmb/ylin/imp200111/ (August - September 2000) and http://www.emc.ncep.noaa.gov/mmb/mmbpll/retronmic.cold/ (February 2001).

6.4. Impact on 10-m wind : 2100 UTC 20 October 2001

Figure 6.6 shows an example of differences between the operational Eta-22 and the Eta-12 for a low-level wind forecast. This case was brought to our attention by Steve Zubrick of the WSFO in Sterling, VA. The top row shows the 45-h forecast of sea-level pressure from the Eta-22 and the Eta-12 valid at 2100 UTC 20 October 2001, along with the verifying EDAS analysis. The bottom row shows the accompanying 10-m wind forecasts, and the verifying surface wind observations. Although both forecasts predicted a weak pressure gradient in the Baltimore/Washington area, the Eta-12 had a much better forecast of the overall pressure gradient, predicting westerly-southwesterly winds. The Eta-22 erroneously predicted the center of the weak anticyclone to be over central Maryland and northern Virginia, resulting in a forecast of northerly winds in this region. The observed surface winds indicate that the Eta-12 produced a better forecast.

7. Conclusions and closing comments

Due to insufficient computing resources, EMC has been unable to run any warm or cold season retrospectives with the full change package at 12-km resolution, specifically the August-September 2000 and the February 2001 periods discussed in Section 6.3. As this Technical Procedures Bulletin is being written a 12 km rerun of February 2001 will be started. Results from this test will be made available on-line at http://www.emc.ncep.noaa.gov/mmb/mmbpll/eta12retro.cold/

Pending CAFTI approval, the implementation of the Eta-12 is scheduled for 27 November 2001. All products from the Eta model which are currently available via the NWS Satellite Broadcast Network (SBN), the operational NWS Gateway, and the NCEP ftp server will remain unchanged. To give users access to gridded output close to 12-km resolution, output on the new 12-km AWIPS grid #218 (same areal coverage as the 40 km grid #212 and the 20 km grid #215) will be generated which will be a clone of the output on grid #215. Initially this new grid will be available only via the NCEP ftp server, with plans to place it on the NWS SBN in the near future.

8. References

Asai, T., 1965: A numerical study of air-mass transformation over the Japan sea in winter. J. Meteor. Soc. Japan, 43, 1-15.

Bennetts, D. A., and P. Ryder, 1984: A study of mesoscale convective bands behind a cold front. Part II: Cloud and microphysical structure. Quart. J. Roy. Meteor. Soc., 110, 467-487.

Böhm, H. P., 1989: A general equation for the terminal fall speed of solid hydrometeors. J. Atmos. Sci., 46, 2419-2427.

Ferrier, B. S., 1994: A double-moment multiple-phase four-class bulk ice scheme. Part I: Description. J. Atmos. Sci., 51, 249-280.

Gordon, G. L., and J. D. Marwitz, 1986: Hydrometeor evolution in rainbands over the California valley. J. Atmos. Sci., 43, 1087-1100.

Heymsfield, A. J., and C. M. R. Platt, 1984: A parameterization of the particle size spectrum of ice clouds in terms of ambient temperature and ice water content. J. Atmos. Sci., 41, 846-855.

Heymsfield, A. J., 1972: Ice crystal terminal velocities. J. Atmos. Sci., 29, 1348-1357.

Houze, R. A., P. V. Hobbs, and P. H. Herzegh, 1979: Size distribution of precipitation particles in frontal clouds. J. Atmos. Sci., 36, 156-162.

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.

Locatelli, J. D., and P. V. Hobbs, 1974: Fall speeds and masses of solid precipitation particles. J. Geophys. Res., 79, 2185-2197.

Lord, S. J., H. E. Willoughby, and J. M. Piotrowicz, 1984: Role of a parameterized ice-phase microphysics in an axisymmetric non-hydrostatic tropical cyclone model. J. Atmos. Sci., 41, 2836-2848.

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, convective changes. NWS Technical Procedures Bulletin No.465. [ Available at http://205.156.54.206/om/tpb/465.htm or from the National Weather Service, Office of Meteorology, 1325 East-West Highway, Silver Spring, MD 20910].

Marecal, V., D. Hauser, and C. Duroure, 1993: Airborne microphysical measurements and radar reflectivity observations near a cold frontal rainband observed during the FRONTS 87 experiment. Atmos. Res., 29, 170-207.

Musil, D. J., 1970: Computer modeling of hailstone growth in feeder clouds. J. Atmos. Sci., 27, 474-482.

Orville, H. D., and F. J. Kopp, 1977: Numerical simulation of the life history of a hailstorm. J. Atmos. Sci., 34, 1596-1618.

Reisner, J., R. M. Rasmussen, and R. T. Bruintjes, 1998: Explicit forecasting of supercooled liquid water in winter storms using the MM5 mesoscale model. Quart. J. Roy. Meteor. Soc., 124, 1071-1107.

Rogers, E., 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. [ Available at http://wwwt.emc.ncep.noaa.gov/mmb/mmbpll/eta22tpb/ or from the National Weather Service, Office of Meteorology, 1325 East-West Highway, Silver Spring, MD 20910].

Rogers, E., 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.479. [ Available at http://www.emc.ncep.noaa.gov/mmb/mmbpll/spring2001/tpb/ or from the National Weather Service, Office of Meteorology, 1325 East-West Highway, Silver Spring, MD 20910].

Rutledge, S. A., and P. V. Hobbs, 1984: The mesoscale and microscale structure and organization of clouds and precipitation in midlatitude cyclones. Part XII: A diagnostic modeling study of precipitation development in narrow cold-frontal rainbands. J. Atmos. Sci., 41, 2949-2972.

Ryan, B. F., 1996: On the global variation of precipitating layer clouds. Bull. Amer. Meteor. Soc., 77, 53-70.

Starr, D. O'C, and S. Cox, 1985: Cirrus clouds. Part I: A cirrus cloud model. J. Atmos. Sci., 42, 2663-2681.

Stewart, R. E., J. D. Marwitz, J. C. Pace, and R. E. Carbone, 1984: Characteristics through the melting layer of stratiform clouds. J. Atmos. Sci., 41, 3227-3237.

Sundqvist, H., E. Berge, and J. E. Kristjánsson, 1989: Condensation and cloud parameterization studies with a mesoscale numerical weather prediction model. Mon. Wea. Rev., 117, 1641-1657.

Tao, W.-K., J. Simpson, and M. McCumber, 1989: An ice-water saturation adjustment. Mon. Wea. Rev., 117, 231-235.

Tremblay, A., and A. Glazer, 2000: An improved modeling scheme for freezing precipitation forecasts. Mon. Wea. Rev., 128, 1289-1308.

You, L., and Y. Liu, 1989: The synoptic structures and microphysical processes of precipitation systems in winter in Wulumuqi area, Xianjang. The Fifth WMO Science Conf. on Weather Modification and Applied Cloud Physics, Beijing, China, Weather Meteorological Organization, 65-68.

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.

Appendix A: Subjective Evaluation of the Eta-12 (S. Zubrick, WSFO Sterling, VA)

Caveats...

First, these comments only reflect my subjective viewing of the Eta 12km model runs as depicted on the web site. I never did get time to get the 12km BUFR data nor the grib files.

Second, these comments reflect only my viewing and not that of the other staff here at Sterling. Although I sent an e-mail to the staff on Oct 19, 2001 announcing the NCEP web site containing the parallel Eta 12km runs, my observation is that few forecasters took time to look at the new model.

Third, I've tried to follow the Eta 12km model from the Oct 19 through Nov 6. I feel I viewed about 75% of the parallel runs and where possible, compared the forecasts to observations (both surface and remote-sat, radar).

Fourth, the Mid-Atlantic region (which I'm obviously MOST interested in! :)) is in the middle of drought. Precipitation opportunities have been few and far between. Most folks are interested in the model's ability to aid in precipitation forecasting. As such, I feel that the few short weeks I've looked at the parallel runs, I've not yet really tested the ability of the new model to adequately forecast precipitation. However, since precipitation seems to be a rarity here in the Mid-Atlantic, I managed to look at other areas of the country (mainly the Ohio Valley and southeastern U.S.) to see how the new 12km model handled precipitation events in those areas. My comments will reflect this more "global" view.

Fifth, I did not perform any precipitation type observations for this analysis.

Finally, it would have been nice to get the grib data so that I could look at some of my personal favorite diagnostics or ability to overlay remotely sensed data (e.g., satellite images). That was not done. While the web site was adequate, it did not include everything I might look at. Then again, as the models go downscale in grid size, it's becoming harder to find diagnostics that actually diagnose the mesoscale.

End of Caveats...

Observations / Comments of SM Zubrick (SOO-WFO/LWX) on the new 12km Eta

(not in any particular order...)

1) The obvious: for certain parameters (e.g., 500 hPa vorticity, QPF) gradients are sharper and max/minima are more intense in the 12km Eta than in the 22km Eta.

The ramification is that forecasters used to looking at absolute values of say 500 hPa vorticity will notice a broader range of values (e.g., 30 unit sort vs 40 unit sort in the new 12km Eta).

2) Overall, I detected a tendency in the new 12 km Eta to develop areas of light precipitation (less than 0.1") where the 22 km Eta did not. During my observation period (Oct 19-Nov 6), I would estimate that most of this light precis. was less likely to occur as measured, although some may have fallen in data sparse areas. In viewing local radars, there were often light radar echoes associated with some of the frontal areas of light precis.

Perhaps the 12-km Eta is trying to be like its socialist comrade the Aviation model: a little precipitation for everyone :)

In addition, I sensed that the new 12 km Eta, like it's 22 km parent, tried to overdevelop areas of light precis over the coastal section of the SE and Mid-Atlantic U.S. whenever there was an onshore component to the flow (e.g., an offshore E-W ridge of high pressure off the Mid-Atlantic coast. This happened a lot around Oct 19-20.

3) I was more impressed with the 12km Eta handling of near surface (10-meter) winds over the Mid-Atlantic than the 22-km.

The 12-km Eta winds over the Chesapeake Bay were more realistic that the 22 km Eta and appeared to begin to reflect some of the local wind regimes we notice on the Bay. However, the 12-km resolution seems too coarse to characterize, e.g., the channeling of winds up the bay that occurs in late afternoon/early evening on fair (sunny) weather days.

A case in point...the low level boundary layer wind forecasts valid during the daylight hours on Sat, Oct 20 were interesting. The 22km Eta winds over most of the LWX/CWA had a northerly component while the 12km Eta winds were SW-Srly. In this case, the 12km Eta was correct... there were no northerly winds on Sat-20th. I worked the day shift on Sunday (10/21) and used the 00Z/21th 12km Eta winds to help me do the morning update to our marine forecast. The 12km Eta also helped convince me that there was NOT going to be a cold front (weak) move through our C.A. today (Mon-10/22)...so I adjusted some of the wording in the forecasts...every little bit helps.

4) Timing of frontal features, both surface and aloft, seemed about the same for the 12km Eta as was for the 22 km Eta.

5) The 12km Eta surface temperatures (2-m) seemed warmer than the 22 km Eta. I believe in the instances I noted, the 12 km Eta temps were more likely to be closer to observed than the 22 km temps.

6) The 12km Eta surface moisture (BL) seemed to be more moist, possibly reflecting its ability to "see" more of the Chesapeake Bay than the 22 km Eta. This moistness seemed more noticeable at longer times ranges (>36 hrs) in the forecast.

7) I did not observe any apparent model cycle bias (00Z vs 12Z).

8) With one of the cold fronts that swept across the Ohio Valley, the 12 km Eta model QPF fields did an outstanding job on defining the areas of afternoon convective precipitation (i.e., thunderstorms) that occurred. The 22 km Eta also did a reasonable job, but the 12 km QPF areas where quite remarkable...and the 12 km Eta also picked up a secondary firing of convection late evening convection in the Ohio Valley.

9) The 12 km Eta QPF fields on several occasions correctly anticipated the dissipation of ongoing precipitation in the Ohio Valley and across Western PA/and WV when it reached the western portion of our C.A.

10) The timing and intensity of upper level lows appeared similar in both the 12 km and 22 km Eta runs.

Well...that's about all I have. You know, there really was not a lot of weather to associate with this observation period. So, it's tough to say the 12 km model was better. But...I'd say that the 12 km Eta was more reasonable in handling what few weather features we had compared to the 22 km Eta.

Appendix B: Additional subjective evaluation of the Eta-12

>B.1. Mike Vescio (NWSFO, Dallas/Fort Worth, Texas)

I wish I could report more to you but we have been in a quiet weather pattern during this brief period of evaluation. Basically, there have been little if any differences between the operational Eta and the Eta-12...at least across the southern Plains. For example, the frontal position at 48 hours on the latest run of Eta-12 is nearly identical to the operational model. The same is true for the upper features. During the past week we have noticed that the two models generally produce very similar output through the entire forecast cycle.

From my perspective, it would seem that the increased resolution would be most useful in convective situations when subtle variations in the mass and moisture fields are most critical.

B.2. Kirby Cook (NWS Western Region SSD, Salt Lake City, UT)

I passed on your request to our SOO's. To be honest, I'm not sure who has been looking at the data on the web page. I must also admit that I haven't had the chance to look at the experimental products as much as I should, and wouldn't feel comfortable passing on any comments with such a limited examination. I will say that response from the slides that Eric [Rogers] gave me for the Intermountain Workshop last week as well as some comments from the SOOs have been favorable and always end with "When will we get the high resolution data?"...

B.3. Pete Manousos (NCEP Hydrometeorological Prediction Center)

Limited feedback. The only things we can say are

1. Isn't doing any worse than current Eta 22

2. Has done just as bad as Eta 22 with tropical systems

B.4. Tim Oram (Johnson Space Flight Center, Houston, TX)

We were able to look at the 12km Meso Eta in support of our Space Shuttle and Space Station missions. This provided an opportunity to look at several areas of the country. Unfortunately, the evaluation period was only over^ the past week.

At the large scale, the 12km Meso Eta seemed to match with the operational Eta. I noted no major discrepancies between the two models particularly with regard to the precipitation patterns. I am not a numerical modeler by training, but I felt this result was important since I understand that 12km resolution is in the gray area between convective parameterization and the need for explicit cloud prediction. The good overall agreement from my subjective standpoint and the precipitation bias statistics presented on the web pages give me pretty good confidence in the update.

I also downloaded some of the high resolution grib data and displayed this with gempak. As expected, the influence of terrain was readily apparent and matched my expectations. I would also say surface wind forecasts showed an improvement with the 12km Eta for those locations of the country where I did a detailed comparison with observed winds. Unfortunately, we were under strong easterly flow at Kennedy Space Center and were not able to observe a true sea breeze develop (a major concern). However, a sea breeze appeared to be predicted along the Gulf of California, but verification is a bit problematic.