/3q05 0217 Mp 391 Tuesday Aug 06 04:01 PM AMS: Forecasting (September 96) 0217
391
S
EPTEMBER
1996 NCEP NOTES
Changes to the Operational ‘‘Early’’ Eta Analysis/Forecast System at the
National Centers for Environmental Prediction
E
RIC
R
OGERS
,T
HOMAS
L. B
LACK
,D
ENNIS
G. D
EAVEN
,
AND
G
EOFFREY
J. D
I
M
EGO
Environmental Modeling Center, National Centers for Environmental Prediction, NWS/NOAA, Washington, D.C.
Q
INGYUN
Z
HAO AND
M
ICHAEL
B
ALDWIN
General Sciences Corporation, Laurel, Maryland
N
ORMAN
W. J
UNKER
Hydrometeorological Prediction Center, National Centers for Environmental Prediction, Washington, D.C.
Y
ING
L
IN
University Corporation for Atmospheric Research, Boulder, Colorado
6 October 1995 and 29 January 1996
ABSTRACT
This note describes changes that have been made to the National Centers for Environmental Prediction
(NCEP) operational ‘‘early’’ eta model. The changes are 1) an decrease in horizontal grid spacing from 80 to
48 km, 2) incorporation of a cloud prediction scheme, 3) replacement of the original static analysis system with
a 12-h intermittent data assimilation system using the eta model, and 4) the use of satellite-sensed total column
water data in the eta optimum interpolation analysis. When tested separately, each of the four changes improved
model performance. A quantitative and subjective evaluation of the full upgrade package during March and
April 1995 indicated that the 48-km eta model was more skillful than the operational 80-km model in predicting
the intensity and movement of large-scale weather systems. In addition, the 48-km eta model was more skillful
in predicting severe mesoscale precipitation events than either the 80-km eta model, the nested grid model, or
the NCEP global spectral model during the March–April 1995 period. The implementation of this new version
of the operational early eta system was performed in October 1995.
1. Introduction
The National Centers for Environmental Prediction
(NCEP, formerly the National Meteorological Center)
began production of operational forecasts from an 80-
km, 38-level version of the eta model in June 1993,
replacing the Limited-Area Fine-Mesh Model in pro-
viding ‘‘early’’ forecast guidance over North America.
The June 1993 version of the so-called early eta system,
as described by Black et al. (1993), consists of two
components: a regional optimum interpolation (ROI)
(DiMego 1988) analysis using a 6-h forecast of the
NCEP global spectral model from the Global Data As-
similation System (GDAS; Kanamitsu et al. 1991) as
a first guess, followed by a 48-h forecast of the eta
Corresponding author address: Dr. Eric Rogers, Environmental
Modeling Center, National Centers for Environmental Prediction,
WINP22, WWB, Room 204, Washington, DC 20233.
E-mail: wd20er@sun1.wwb.noaa.gov
coordinate model (Mesinger 1984). The early eta sys-
tem was changed in September 1994 (Rogers et al.
1995) to include a more realistic depiction of the model
orography and to improve the ROI analysis in the lower
troposphere.
The purpose of this note is to describe changes to
the early eta system that were implemented in October
1995. This upgrade consists of four components:
1) a decrease in horizontal grid spacing from 80 to
48 km,
2) incorporation of a cloud prediction scheme in the
eta model physical package,
3) use of an eta-based intermittent data assimila-
tion system to provide initial conditions to the eta
model, and
4) use of satellite-sensed total column water data in
the eta ROI analysis.
A description of each component of the early eta
upgrade package and its impact on model performance
/3q05 0217 Mp 392 Tuesday Aug 06 04:01 PM AMS: Forecasting (September 96) 0217
392
V
OLUME
11WEATHER AND FORECASTING
T
ABLE
1. 48-km eta model characteristics.
Dynamics
Horizontal Arakawa E-grid, rotated spherical coordinates
Lorenz vertical grid; model top
Å
25 mb; step-mountain (eta) vertical coordinate (Mesinger 1984; Mesinger et al. 1988), 38 layers
Atmospheric prognostic variables: u, V, T, q, q2 (turbulence velocity squared), m (cloud water–ice mixing ratio), surface pressure
Split explicit time differencing (Janjic´ et al. 1995) with an adjustment time step of 120 s
Space differencing: conserving Arakawa-type scheme (Janjic´ 1984)
Arakawa (Arakawa and Lamb 1977) vertical advection except for moisture; piecewise-linear (e.g., van Leer 1977) vertical
moisture advection
Gravity wave coupling scheme (Mesinger 1973; Janjic´ 1979)
Silhouette-mean orography (Mesinger 1995)
Energy conservation in transformations between kinetic and potential energy, in both space and time differencing (Mesinger et al.
1988; Janjic´ et al. 1995)
Physics
Modified Betts–Miller scheme for deep and shallow convection (Betts 1986; Betts and Miller 1986; Janjic´ 1994)
Explicit prediction of grid-scale cloud water/ice mixing ratio (Zhao et al. 1995), with predicted clouds used by the radiation
scheme
Mellor–Yamada (1982) level-2.5 model for free atmosphere vertical turbulent exchange above lowest model layer
Geophysical Fluid Dynamics Laboratory radiation scheme (Fels and Schwarztkopf 1975; Lacis and Hansen 1974)
Viscous sublayer over water surfaces (Janjic´ 1994)
Surface layer: similarity functions derived from Mellor–Yamada level-2.0 model (L
Q
obocki 1993)
is presented in section 2. The results of an objective
verification of the upgrade package versus the 80-km
version of the early eta, a subjective evaluation of the
48-km eta model, and two forecast examples are given
in section 3. A summary follows in section 4.
2. Changes to the early eta system
This section describes changes to both the model and
analysis components of the early eta system. A general
description of the eta coordinate and the structure of
the vertical and horizontal grids can be found in Black
(1994). Table 1 lists the dynamical and physical pack-
age in the 48-km version of the eta model as imple-
mented in October 1995. For more details concerning
eta model dynamics and physics refer to the references
listed therein.
a. Horizontal resolution
The main purpose for decreasing the horizontal grid
spacing of the operational eta model is to provide fore-
casters with improved early guidance over all of North
America. Since the operational 29-km mesoscale eta sys-
tem (Black 1994), which NCEP has run twice daily since
28 March 1995, does not include Alaska, the upgrade of
the early eta system to 48-km grid spacing satisfies
NCEP’s commitment to produce high-resolution fore-
casts for that state. As shown in Fig. 1a, the 48-km do-
main is nearly identical to the current 80-km domain (see
Fig. 2a of Black et al. 1993). Although a smaller grid
spacing such as 40 km would be preferable, a 40-km grid
covering the domain depicted in Fig. 1a has 23% more
grid points, which would tax NCEP’s computer re-
sources. For this reason, 48-km grid spacing was chosen.
The vertical resolution of the upgraded early eta sys-
tem remains at 38 levels, with the pressure at the top
of the model raised from 50 to 25 mb. The vertical
distribution of these layers is shown in Fig. 1b. The
highest resolution is seen in the boundary layer, where
the lowest layer is defined to be exactly 20 m deep for
the standard atmosphere. A secondary maximum ap-
pears near the tropopause for the purpose of better re-
solving jet stream features.
An obvious consequence of the increase in horizon-
tal grid spacing is greater detail in the model orography,
shown for both 80- and 48-km grid spacing in Fig. 2
over the western United States, eastern United States,
and Alaska. While both grids are able to resolve the
two highest peaks in the Alaskan region (Fig. 2c), Mt.
McKinley (elevation 6194 m) and Mt. Logan in ex-
treme southwestern Yukon (elevation 5867 m), the
model elevation at both locations is 800–1000 m
higher at 48-km grid spacing.
According to Mesinger and Baldwin (1995), in-
creasing the horizontal resolution of the eta model
tends to improve model accuracy in forecasting precip-
itation, based on the evaluation of quantitative precip-
itation scores. A recent comparison of the equitable
threat score (ETS, defined in the appendix) (Schaefer
1990; Gandin and Murphy 1992; Mesinger and Black
1992) and bias score for 24-h precipitation amount
from both the 80-km early eta and an experimental 40-
km version is presented in Fig. 3. The 40-km eta fore-
cast model differed from the 80-km early eta only in
the horizontal resolution of the model, although the 40-
/3q05 0217 Mp 393 Tuesday Aug 06 04:01 PM AMS: Forecasting (September 96) 0217
393
S
EPTEMBER
1996 NCEP NOTES
F
IG
. 1. (a) The horizontal domain of the 48-km eta model.
F
IG
. 1. (b) The distribution of the 38 layers in the 48-km eta model.
The pressures on the left side indicate the layers’ positions with re-
spect to the standard atmosphere, while the numbers on the right give
an approximate pressure depth of each layer in hectopascals.
km grid domain, while covering most of North Amer-
ica, is smaller than the 80-km domain. Verification is
provided by an analysis of the Office of Hydrology’s
River Forecast Center 24-h observed precipitation data
over the contiguous United States. The analysis is per-
formed on the 80-km eta model grid, and the 40-km
forecasts are remapped to the 80-km grid for verifica-
tion by an area-integral method that conserves total wa-
ter. All precipitation observations inside a grid box are
averaged to produce the analyzed precipitation amount
at each grid point. Figure 3 shows that the higher-res-
olution forecasts have slightly higher ETS at all pre-
cipitation amount thresholds, with greater impact at the
lighter amounts. Both Black (1994) and Mesinger and
Baldwin (1995) reported a similar impact from exper-
imental 40-km eta model forecasts during 1992 and
1993 (not shown). The improved threat scores from
the 40-km forecast are a sharp contrast to that model’s
bias score, which show a significant decrease in bias
compared to the 80-km early eta at thresholds
ú
0.5
in. Figure 3 clearly shows that the increase in resolution
caused a significant underprediction of precipitation
during the 6-week test period in the fall of 1994. Evi-
dence of the underprediction of 24-h precipitation
amounts
ú
0.5 in. by the 40-km eta model was shown
by Black (1994) in precipitation statistics from No-
vember 1993.
b. Prognostic cloud scheme
The June 1993 version of the 80-km eta model used
the diagnostic scheme described by Hoke et al. (1989)
for the production of large-scale condensation and pre-
cipitation. In this method, condensation is allowed to
occur at a model grid point when the relative humidity
exceeds a predefined critical value (95%). The con-
densation is summed layer by layer downward from the
top, with water evaporated into subsaturated layers and
temperature/specific humidity changed accordingly.
Large-scale precipitation is defined as the water that
reaches the ground as the result of this scheme.
Although the 80-km early eta system produced im-
proved precipitation forecasts over other NCEP models
using this simple scheme (Black et al. 1993; Mesinger
and Baldwin 1995), clouds were completely ignored.
At higher model resolutions, the neglect of clouds may
adversely impact the prediction of precipitation, as Fig.
3 suggests. As the important role clouds play in at-
mospheric processes became more apparent (especially
/3q05 0217 Mp 394 Tuesday Aug 06 04:01 PM AMS: Forecasting (September 96) 0217
394
V
OLUME
11WEATHER AND FORECASTING
F
IG
. 2. Surface height for the 80- and 48-km eta model grids over (a) the western United States,
(b) the eastern United States, and (c) Alaska. Contour interval
Å
200 m.
/3q05 0217 Mp 395 Tuesday Aug 06 04:01 PM AMS: Forecasting (September 96) 0217
395
S
EPTEMBER
1996 NCEP NOTES
F
IG
.2.(Continued)
F
IG
. 3. (a) Equitable threat skill scores and (b) bias scores for 24-h precipitation forecasts (all periods) of the 80-km early eta
and the 40-km eta model for the period 7 September–21 October 1994.
