Mesoscale Parallel Experiment Log

Experiment Name

Eta-12 with modified cloud microphysics, assimilation of GOES cloud top pressure, assimilation of NEXRAD radial wind velocity data

Parallel Slot

ETAX

Control Slot

Ops Eta-12

Start date of parallel experiment

12Z 2002/08/14 ; restarted from prod at 12Z 9/19/02

End date of parallel experiment

00Z 2003/07/08

Environmental Modeling Center scientists

Brad Ferrier, Ying Lin, Dave Parrish, Manuel Pondeca, Eric Rogers

Abstract (including Motivation, Hypothesis and Method)

The Eta-12 is being used to test the impact modifications to the cloud microphysics, assimilation of NEXRAD radial wind data in the EDAS, and assimilation of GOES cloud top pressure data in the EDAS
ETA MODEL CHANGES

I. Model changes : Cloud Physics

- 1. New grid-scale cloud properties are written and read in the restart files, allowing for their cycling in the forecast system. Old arrays from the Zhao et al. (1997) microphysics were removed from all code.
- 2. Arrays defining convective cloud-base and cloud-top properties used in radiation are now properly reinitialized after being written to the model restart files. Code was restructured to have convective cloud arrays used in radiation, then written to restart files, and then reinitialized to no-cloud values using a single set of arrays. Additional arrays that separate shallow from deep convection have also been added to the model restart files.
- 3. Code is restructured so that cycling will produce the same forecast as a longer forecast. For example, the forecasts will be the same if the model is run for two 3-h forecasts (e.g., two EDAS cycles with no assimilation of observations) compared to one 6-h forecast. In other words, the model forecast does not depend on the reading of arrays that are read in at the end of the restart files when restarting the model from the middle of the forecast (i.e., a nonzero start time).
- 4. Whole array operations were found to be inefficient and were replaced with explicit do-loop indexing.
- 5. Lower limits were placed to turbulent kinetic energy (Q2) throughout the code, set to values of 0.2. This change affects the writing of Q2 to restart files but does not affect the model integration, since lower limits of Q2 are internally enforced in TURBL.
- 6. Longwave radiation is updated every hour along with shortwave radiation.
- 7. Inline compiler directives have been added at the beginning of numerous subroutines, which allow for easy debugging of code. These directives override compiler directives specified in the make files, and they are included in those subroutines that call MPI routines. Now the code can be run with full trapping, checks on array bounds, and checks on argument lists by just manipulating the makefile.
- 8. The lower limit for all hydrometeor mixing ratios is set to 1.e-12 for all model routines (parameter EPSQ).
- 9. Convective precipitation rates used to define optical properties of convective clouds are now modified in the precipitation assimilation code.
- 10. Some of the values in array ZEFFIJ along the eastern boundary were uninitialized. A small patch of code was added in GOSSIP to assign values at these locations to values at the next grid point west (in from) the eastern boundary.
- 11. Tunable parameters in the microphysics defined based on grid resolution in subroutine GSMCONST are restructured more clearly. These quantities are assumed to vary linearly as a function grid resolution, but this functional dependence can be easily changed in the future.
- 12. Probablistic freezing of rain as a function of temperature, as parameterized by Lin et al. (1983) following Biggs (1953), has been correctly coded in the grid-scale microphysics in GSMCONST and in GSMCOLUMN. What is currently coded in operations is producing underflow errors and is effectively not activated, however, this process is an extremely slow process and is not expected to have much impact on precipitation forecasts.
- 13. A small bug was fixed in SFCDIF, in which on rare occasions subscript ML for array ZEFF was out of range (i.e., larger than the upper limit of 4). No discernible impact on the forecast is expected.
- 14. No dependence of cloud type (convective or grid scale) was seen in the GFDL radiation driver, so the array ITYP was removed from RADFS.
- 15. The following changes were made in the representation of clouds in radiation.
a. Convective cloud fraction as a function of precipitation rate of Slingo (1987) was increased by 20%, allowing for shallow, nonprecipitating convective clouds to have an assumed cloud fraction of 10%.
b. Grid-scale cloud fraction is now parameterized following Xu and Randall (1996), replacing the Randall (1994) formulation. Also, relative humidity is now calculated using the same approach as in the EGCP01 grid-scale microphysics. One change was made with respect to Xu and Randall, however, in that a relative humidity of 100% or more will not produce a cloud fraction of 1.0. Grid-scale cloud must be present in order for nonzero cloud fractions to be represented. Grid-scale cloud is defined as the sum of the mixing ratios of cloud water and total ice (i.e., ignoring rain).
c. The radiative effects from low-level clouds in the lowest 100 mb and upper-level clouds above the tropopause are no longer ignored.
d. Absorption coefficients for convective clouds is assumed to be 0.16 at temperatures <= -10C assuming ice is the dominant hydrometeor category as in the grid scheme, 0.08 at warmer temperatures assuming water is dominant at warmer temperatures. The values for these coefficients is based on Harshvardhan et al. (1989). Absorption coefficients for grid-scale cloud water is 0.08*min(1., Qw/Q0), where Qw is the cloud water mixing ratio and Q0=0.1e-3 kg/kg. Absorption coefficients for grid-scale ice is equal to 500*Qi, where Qi is the mixing ratio for all ice particles (kg/kg). This value was derived based on the optical properties for snow provided by Q. Fu (U. Washington, personal communication) for ice particles. Operational code currently assumes the absorption coefficients vary as functions of temperature for convective and grid-scale clouds.
e. Many of the algorithms for calculating cloud fraction were simplified and numerous comments were added to allow future modifications to be made much more easily and effectively.

II. Post changes : Cloud Physics

-1. Read in new, expanded restart files.
-2. Post new cloud fields (cloud water, cloud ice, rain, snow, and total condensate) on pressure and Eta surfaces,.as well as original model arrays (F_rain, F_ice, F_RimeF) on Eta surfaces for debugging.
-3. Post total cloud fraction from grid-scale and convective clouds on Eta surfaces using the same algorithms as in the Eta radiation driver (this is not true in the operational Eta post). Also post cloud-base and cloud-top pressures from shallow, nonprecipitating convection, deep convection, and grid-scale convection. The cloud efficiency parameter used in the convective parameterization can now also be posted.
-4. Visibility calculations have been changed to use the new cloud fields, responding to mixing ratios of cloud water, cloud ice, rain, and snow.
-5. The same method is used to calculate relative humidity (RH) throughout the post. This currently is defined as RH with respect to water at all levels, however, it will be very easy to calculate RH with respect to ice(if desired) in the future. There were MANY subroutines that used the ice/water flag from Zhao and Carr (1997) for calculating RH, and the code was poorly structured and not always uniform throughout the post. These blocks of code have been substituted with a simple subroutine call.
-6. The same lower limit for the condensate mixing ratio of EPSQ=1.e-12 is used throughout the post as it is used throughout the model. Small changes were made to the include parameter file "params" that removed the need to use the parameter file "cuparm" in WETBULB. The "cuparm" file is no longer used in the post.
-7. Inline compiler directives have been added to the beginning of numerous subroutines, as in the model, to allow for easy compiling of code with or without traps, checks on array bounds, and checks on argument lists by simply changing lines in the make file.
-8. Bug fixes identified by Tuccillo have been made to the J index in FIXED, and an argument to MPI_ISEND in GRIBIT was changed that solved the problem of the post occasionally failing due to segmentation faults.

III. Brief summary of changes : Cloud Physics

Changes were made allows the new cloud arrays to be written to the model restart files and posted to grib files. Numerous changes were made in the optical properties of grid-scale and convective clouds that should increase the areal coverage of clouds, including modifications in the calculation of cloud fraction and the absorption coefficient of clouds. These changes will result in the radiative properties of the clouds being more consistent with the other physical parameterizations. These changes are also accurately reflected in the post, allowing the end user to review many more aspects of the convective and grid-scale cloud properties simulated in the Eta.

IV. Model changes : Assimilation of GOES cloud top pressure data

Prior to running each 3-hour EDAS segment, a pre-processing program reads in the 3-hour's worth of GOES cloud top data from the PREPBUFR file, and distributes the observations to the appropriate assimilation hour and horizontal Eta grid box.

During EDAS, at each physics time step and on each Eta grid where observed cloud top is available, we zero out model cloud water/cloud ice above the observed cloud top, and set the water vapor mixing ratio to no more than saturation (original value or grid-scale saturation value [w/r/t water if T > -10 C, otherwise w/r/t ice], whichever is less). At the model level closest to the observed cloud top, if the model air is sub-saturated (by an amount of delta(Qv)), the air at this level is then moistened by the amount of delta(Qv)*physDT/3600., i.e. it is moistened at a rate that would bring it to saturation in 1 hour.

The observed cloud top is also used as 'anchor' cloud top in precipitation assimilation, i.e. in the event that we need to create a layer of precipitating cloud, the observed cloud top is used in the cloud creation.

OBSERVATIONAL AND ANALYSIS CHANGES

Radial wind data from NEXRAD are assimilated using the eta 3DVAR analysis. The raw data is converted into super-obs with a spatial resolution of 1 km and 6 degrees of azimuth. All NEXRAD winds within +/- 1.5 hours of the analysis time are used.

Define radar beam vertical extent to increase at the rate of 20m/km. This is about 20% larger than the actual beam size, to allow for beam propagation uncertainty.

Winds at all eta levels covered by the radar beam are adjusted so that the observation is as close as possible to the interval between the minimum and maximum wind at the selected levels. (The min and max are not the actual min and max but those derived from a straight line fit to the winds. This is done to simplify the computation of the gradient of the forward model implied by this process).

All winds, out to the maximum range of the radar can be used with the above defined forward model. Bill Facey increased the distance parameter in the superob code so that superobs are formed out to the maximum radar range (250km).

All VAD wind observation quality marks are collocated in the vertical in 500m bins with the corresponding radar winds. If there is no VAD observation, or if the quality mark is larger than 3, then the corresponding radar winds are not used. This combines the bird algorithm and other checks used on the VAD winds.

If the beam envelope extends below the eta model terrain height, the corresponding radar wind is not used.

If the superob error is larger than 6m/sec, the radar wind is not used.

Gross checks are the same as for conventional winds (residual < 35m/sec)

Experiment changes log

Background links

Evaluation of parallel results

Daily forecast maps

Daily forecast stats

Verification of precipitation and against rawinsonde / surface data

Conclusion

Mesoscale Parallel Home Page

Disclaimer