Thomas Black, Michael Baldwin, Keith Brill, Fei Chen, Geoffrey DiMego, Zavisa Janjic, Geoffrey Manikin, Fedor Mesinger, Kenneth Mitchell, Eric Rogers, and Qingyun Zhao
A set of changes is being introduced into the early and Meso Eta systems
which affects its data assimilation system (the EDAS), the forecast model,
and the post-processing. The desirability of and/or need for these changes
has grown quickly so that individual parallel tests of each of the many
separate modifications was not feasible. This is not to say that each change
was not tested inidividually, but rather that a full-blown parallel test
may not have been possible. The changes to the assimilation and forecast
model are described in Section 2, changes to the Eta post-processor are
covered in Section 3, and an assessment of their impact as determined from
a set of parallel forecasts is provided in Section 4.
2. CHANGES AFFECTING THE ASSIMILATION CYCLE AND FORECAST
Since the EDAS involves the running of a series of short Eta Model forecasts
(only one 3-hour forecast in the case of the Meso Eta), any change incorporated
into the forecast model will also necessarily affect the data assimilation
and its evolution. The modifications in the soil package will be considered
first since one change in it strictly affects only the assimilation cycle.
2.1 Soil Package
1. The first guess for the value of the soil moisture at the beginning
of the EDAS is taken from the Global Data Assimilation System (GDAS). When
compared against certain observations (e.g. monthly total precipitation),
it was shown that the soil moisture in the GDAS was typically too large.
In the new version, large values of GDAS soil moisture are reduced if they
exceed a particular threshold and they are never allowed to exceed a maximum
2. One of the specified fields in the soil package is the so-called
green fraction which indicates the fraction of an area that is covered
by green vegetation each month of the year. In the former version, this
field was taken from a NASA dataset that has a 1x1 degree resolution and
is based on a one-year sample from 1987. The new version uses a NESDIS
dataset that has a 0.15 degree by 0.15 degree resolution and is based on
a 5-year sample. The values in the newer data are roughly 10-15% larger
than in the older one and this has led to an increase in surface evaporation
due to the greater amount of moisture available in the plant leaves through
3. One of the factors that determines the amount of evaporation that
takes place from bare soil is the amount of water in the soil at that moment.
As the amount of soil water drops, so does the evaporation. In the former
version, the range of values of soil moisture through which the evaporation
changed from large to small was very narrow. Based on recent observations,
the new version uses a more gradual transition between large and small
evaporation rates as a function of soil moisture.
4. Partitioning of energy during the situation of melting snow has been
refined. Snow melts in the model whenever the temperature of the snow surface
(Ts) is computed to be greater than 0oC. In the former
version, the amount of snow melted was found by using up the energy represented
by the difference in temperature between Ts and 0oC
and at the same time the amount of evaporation from the snow surface was
determined using the Ts (>0oC). The result was
that more energy was used than was actually available in the system and
most notably the magnitude of the evaporation was too large. In the new
corrected version, when Ts is initially found to exceed 0oC,
the surface energy balance is used explicitly to determine the amount of
evaporation and melting but only after resetting Ts to 0oC
since the phase change between ice and liquid water can only take place
at that temperature.
5. Three modifications were made to reduce what seemed to be too large
values of albedo in the presence of snow on land. First, based on the current
literature, the value of the albedo over a pure snow surface was reduced
from 0.60 to 0.55. Second, the depth of snow in the model below which the
albedo of the snow-free surface begins to be considered was increased from
1 cm to 2 cm. Third, the vegetation fraction has been incorporated into
the computation of the albedo such that a larger fraction of the snow albedo
value is now used when the vegetation cover is small while a smaller fraction
of the snow-free albedo value is used when there is more vegetation cover.
The overall effect of these modifications will be somewhat warmer low level
air temperatures during the daytime in regions of relatively shallow snow.
2.2 Cloud/Radiation Interaction
Both of the changes made to the current operational explicit cloud scheme (Zhao et al. 1997) are related to the manner in which the clouds interact with the radiation package.
1. Previously the radiation package required cloud information input
with respect to only three general levels: low, middle, and high. This
meant that the cloud information in all of the model's layers had to be
condensed down into just these three levels. In its new form, the radiation
scheme is allowed to interact explicitly with the predicted cloud water/ice
in each of the model's layers which permits a far more realistic interaction
between the clouds and the radiation.
2. The second of the two changes is the use of a new formulation for
the so-called "cloud fraction" (CF) that exists in each grid
box. CF is an important component in the computations regarding the radiative
effects within the atmosphere. Previously, that quantity was defined as
where RH is the relative humidity in the box and RH0 is a critical value required for large scale condensation (0.75 over land; 0.85 over water). It was determined that this formula led to an underprediction of low level cloud. The new formula is taken from Randall (1995) and is given by
where m is the cloud water/ice mixing ratio. The new relation leads to a cloud fraction that is more consistent with the moisture field and cloud water/ice content and more low level clouds are presented to the radiation package. Tests have indicated that these two cloud/radiation changes tend to produce slightly cooler daytime maximum temperatures and slightly warmer nighttime minimum temperatures under cloudy conditions.
2.3 Shortwave Radiation
As of the beginning of the convective season in the spring of 1996,
some field forecasters noted that there seemed to be excessive warming
and drying in the lowest 200-300 mb of the atmosphere over parts of the
U.S. A primary reason was found to be excessive shortwave solar radiation
reaching the ground which led to an inordinate increase in low level vertical
mixing. Three changes were made to the radiation computations to address
1. Ozone absorbs a significant amount of shortwave radiation in the
Earth's atmosphere yet it was discovered that essentially no absorption
by ozone was taking place inside the Eta Model. Now the entire column of
climatological and seasonally varying ozone is represented in the model
along with the appropriate absorption.
2. Previously the Earth's orbit was assumed to be circular (eccentricity
of 0) which meant that the solar constant (the total amount of incoming
solar radiation intercepted by the Earth at the top of the atmosphere)
was taken as a constant. However, the orbit is actually an ellipse with
an eccentricity of 0.0167 which means that the Earth is actually 3.4% closer
to the sun at perihelion in January than it is at aphelion in July. That
difference in distance leads to there being 6.9% more solar energy intercepted
by the Earth in January than in July. Using a constant average value for
the solar constant in the model meant that it saw somewhat too little solar
radiation over North America in winter and somewhat too much in summer.
The orbit's eccentricity has now been incorporated into the predictions.
3. In addition to ozone, aerosols also have a significant impact on
the amount of radiation reaching the surface due to their absorptive and
scattering properties. On advice from the individuals who are developing
the radiation package used in the Eta Model, the total energy entering
the atmosphere was reduced by 3% in order to simulate the aerosol effects.
In the next update of the radiation scheme following this one, aerosols
will be accounted for explicitly.
2.4 Other Modifications
1. In order to better simulate the way in which the atmosphere interacts
the topography, a form drag scheme has been added (Mesinger, et al. 1996).
The scheme creates an effective roughness length that is dependent on the
wind direction and on the number and elevation of subgrid terrain obstacles
within a grid box. This effective roughness length can be much greater
than the analogous value would be over flat terrain and can attain a magnitude
of 10 m or more. In individual tests, improved prediction of cyclones on
the lee of the Rockies was seen when the effective roughness length was
2. An improved positive definite advection scheme for specific humidity
and cloud water was introduced. Strong gradients should be handled somewhat
better than in the previous scheme.
3. Under conditions of extreme stability, a minimum amount of turbulent
mixing had been specified using an exchange coefficient of 0.01 m2
s-1. Given that the true value of this quantity cannot be determined
accurately while at the same time the Eta forecasts have tended to show
too cool low level temperatures under such stable situations, the minimum
value of the turbulent exchange coefficient under the most stable conditions
has been increased to 0.1 m2 s-1 to produce slightly
4. Computational efficiency is a critical aspect of any operational
model. The Eta Model source code was carefully examined and found to be
lacking in vectorization in some sections and in parallelism in many of
its parts. The code was optimized and a decrease in wallclock time of a
factor of 3 was realized. This optimized code has already been implemented
in the Meso Eta and has now become the basis of the Early Eta source code.
3.CHANGES AFFECTING THE POST-PROCESSOR
There are three main areas of changes regarding the post-processor: (1) corrections
to the product labels of GRIB and BUFR output to more accurately identify the
fields and to bring those labels in line with current WMO standards; (2) addition
of new output fields from the soil/surface physics to allow users to examine
more quantities from the soil package; (3) miscellaneous changes to add new
output fields and to correct errors.
1. The change in labeling that will likely affect operational users
is the change of the GRIB Product Definition Section (PDS) for the so-called
"boundary layer" fields which comprise the six lowest layers
above the model ground/ocean surface and are each 30 mb deep. These fields
are available in the AWIPS output files via the OSO server and in other
output files as well. The boundary layer fields are now labeled as PDS
level type #116 which indicates a layer between two levels that each are
described by their pressure difference with that at the ground. These fields
were previously incorrectly labeled as layer type #108 which is a layer
between two sigma levels. (NOTE: As of this time, Data Review Group approval
has been requested but not given for this label change. When authorization
is received, these label changes will be made.)
2. Other changes in labeling will not affect operational users as much
since those fields are not currently available via the AWIPS output files.
The PDS level type for fields on eta surfaces has changed to #119. For
native Eta output grids the central latitude and longitude has been added
to the end of the Grid Definition Section (GDS); this will help in navigating
the native Eta rotated latitude-longitude grids. Turbulent kinetic energy
on eta levels is now correctly labeled as being on layer interfaces rather
than at midlayers. Cloud water fields now contain only cloud water while
cloud ice fields contain only cloud ice. Previously the cloud water fields
contained both liquid water and ice. All precipitable water fields now
include cloud mixing ratio in the computation.
3. In the hourly station output in BUFR, the tables have been changed
to reflect the WMO specification on the number, scale, reference, bit,
and unit for each output variable. Also the latitude and longitude of the
station is now given as that of the model grid point that is used for the
profile rather than the actual latitude/longitude of the station. Finally,
variables from the soil surface physics are set to missing for stations
over water since those variables are undefined there.
3.2 New soil surface fields
Several new output fields that represent quantities from the soil package
are now available via the Eta post-processor.
1. Two-dimensional fields of deep soil temperature (lower boundary condition
for the predictive soil layers), surface exchange coefficient, green vegetation
cover, volumetric soil moisture and temperature from the midpoints of the
two soil layers, total soil moisture, soil moisture availability, ground
heat flux, and plant canopy water are now available in GRIB.
2. In BUFR, the surface runoff, underground runoff, accumulated snowfall, surface exchange coefficient, plant canopy water, and the temperature and moisture of each soil layer are available in the set of files called "class 1".
3. In the BUFR fields from the files called "class 0", the
10-m potential temperature, 10-m specific humidity, and maximum/minimum
temperatures were dropped to make room for the first soil layer temperature,
surface evaporation, surface runoff, and snow water equivalent.
3.3 Other changes
In the GRIB files:
1. Cloud ice on pressure surfaces, cloud top temperatures, instantaneous
precipitation rate, and snow ratio (the fraction of the cloud-scheme precipitation
that is snow) are now available from the explicit cloud scheme in GRIB.
2. Components of radiative fluxes at the surface and top of the atmosphere (longwave up and down, shortwave up and down) are now available separately rather than combined.
3. The snow cover and time-averaged total cloud fraction have been corrected.
4. The albedo and cloud fraction computations in the post-processor are now consistent with those used in the model integration.
In the hourly station output files in BUFR:
1. The calculation of potential evaporation has been corrected.
2. Low, middle, and high cloud fraction as well as the land/sea mask have been added to both classes of BUFR output.
3. The turbulent kinetic energy (interpolated to the middle of the eta layers) has been added to the "class 1" profiles.
3.4 Computational efficiency
As was done with the forecast model source code, an examination of the
post-processor source code was made and significant optimization was effected.
4. RESULTS FROM PARALLEL TESTS
At the beginning of August 1996 the full set of changes described above
was placed into a parallel system that was an analog to the current operational
Early Eta. That system was in place until late September. Below is a sample
of results that illustrates the effects of the changes to the Eta forecasts.
4.1 Station soundings
As stated earlier, the initial catalyst for creating this set of changes
was the observation by some forecasters that the operational Eta forecasts
were showing too much low level heating and drying over parts of the U.S.
as of the start of the 1996 warm season. Much of this problem was traced
to the excessive solar shortwave radiation reaching the ground. Fig. 1
shows both observed and 24-hr forecast soundings from Topeka, Kansas, valid
at 0000 UTC 4 September 1996. The forecast sounding at Topeka from the
operational Eta (Fig. 1a) shows a well-mixed
boundary layer which is warmer and drier than the observed sounding. The
forecast sounding from the parallel test with the above changes (Fig.
1b) is closer to the observed temperature/moisture profile. Additionally,
the parallel forecast reduced the depth of the mixed layer compared to
the operational Eta, with a LCL closer to the observed pressure of 800
Fig. 2 shows observed and 12-h forecast soundings from Topeka valid
at 1200 UTC 4 September 1996. In the operational Eta (Fig.
2a) the excessive boundary layer warming during the previous day has
led to a 5øC error in the low-level inversion. This error
was significantly reduced in the parallel run (Fig.
4.2 Precipitation scores
Bias and equitable threat scores for 24-hr precipitation for the entire
length of the parallel test are shown in Fig.
3a and Fig. 3b. The operational Eta
is denoted by circles with "+" and the parallel Eta by the open
squares. Details on the computation of these scores can be found in Rogers
et al. (1996). The bias score from the parallel Eta was increased at all
thresholds, eliminating a slight dry bias at thresholds above 0.01 in.
Little change was seen in the equitable threat score implying that on average
the areal coverage of forecast precipitation was not changed by the parallel
Eta while the amount of precipitation did.
4.3 Grid-to-surface observation verification
Fig. 4 shows plots of the 10-m wind speed, 2-m temperature, and 2-m
specific humidity for the 0-48 h forecasts from the operational Eta, the
parallel Eta, and the verifying observations. The value plotted is the
average value from surface stations which are collocated with hourly station
profile output from the model forecast. These were computed for 49 12Z
cycles between 7 August and 24 September 1996. The number of accumulated
stations plotted should be divided by the number of cycles to get the number
of verifying surface observations at each forecast hour.
The 2-m temperature plot (Fig. 4a)
shows excessive daytime warming by 1-2 oK in the operational
Eta which was corrected in the parallel run. However, the tendency for
cooler than observed early morning temperatures in the operational Eta
was still seen in the parallel Eta. Although both Eta runs showed a decrease
with forecast hour in the specific humidity (Fig.
4b) the parallel Eta was closer to the observations during the first
36 hours of the forecast. Neither Eta model run was able to simulate the
weaker 10-m wind speeds at night (Fig. 4c).
Mesinger, F., R.L. Wobus, and M.E. Baldwin, 1996: Parameterization of form drag in the Eta Model at the National Centers for Environmental Prediction. Preprints, 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 324-326.
Randall, D. A., 1995: Parameterizing fractional cloudiness produced by cumulus detrainment. Workshop on Cloud Microphysics Parameterizations in Global Atmospheric Circulation Models, Kananaskis, Alberta, Canada, May 23-25, 1995; WMO, Geneva, WCRP-90, WMO/TD No. 713, 1-16.
Rogers, E., T. L. Black, D. G. Deaven, G. J. DiMego, Q. Zhao, M. Baldwin, N. W. Junker, and Y. Lin, 1996: Changes to the operational "Early" Eta analysis/forecast system at the National Centers for Environmental Prediction. Wea. Forecasting, 11, 391-413.
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, in press.