Observing-System Research and Ensemble Data Assimilation at JAMSTEC - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 503

Table 21.1 The root-mean square difference of the 5-day geopotential height forecast at 500 hPa
averaged over the Northern Hemisphere for the different versions of AFES forecast from the JMA
analysis
Version

Resolution RSMD m Remarks

1.22 T159L48 55.4 Original

2.2 T159L48 52.1 ALERA

2.7 T119L48 51.3

2.7 with MATSIRO T119L48 49.5

3.6 T119L48 48.9 ALERA2

21.2.1 The Forecast Model

AFES integrates the primitive equations of winds, temperature, specific humidity,
cloud water and surface pressure using the spectral transform method and Eulerian
advection, and it has physics schemes common to many forecast and climate
models. As in the version of AFES used in ALERA, the radiative fluxes are
parameterized using mstrnX ( Sekiguchi and Nakajima 2008 ), and the cumulus
convection is represented by the Emanuel scheme ( Emanuel 1991 ; Emanuel and
Zivkovic-Rothman 1999 ; Peng et al. 2004 ) without discrimination between shallow
and deep convection. Updated physics schemes of AFES include cloud ( Kuwano-
Yoshida et al. 2010 ) and land-surface schemes ( Takata et al. 2003 ). The new cloud
scheme with moist turbulence improves the representation of the boundary-layer
clouds in the eastern oceanic basins. The land-surface scheme MATSIRO improves
the modeling of the hydrological cycle.
Reduction in the forecast error of AFES has been achieved continuously.
Tab le 21.1 shows the geopotential height error averaged in the Northern Hemisphere
.> 30

in the forecast experiments for August 2004 with different versions of
AFES. We conducted a 5-day forecast from the analysis at 12 UTC on each day in
August 2004. The root-mean square difference from the JMA analysis was regarded
as an error. AFES 2.2 used in ALERA has an error of 52.1 m, a 3.3 m (approximately
6 %) decrease from the original version ( Ohfuchi et al. 2004 ). Better estimations of
the cloud water ( Bony and Emanuel 2001 ) and of the saturation specific humidity
allow the error to decrease from AFES 2.2 to 2.7 despite the use of a somewhat
coarser resolution. The introduction of the new land-surface and cloud schemes
contributed to a further reduction in the forecast error. The RSMD of AFES 3.6
used in ALERA2 is reduced by approximately 6 % from AFES 2.2 used in ALERA
and by more than 10 % from the original version.

N /

21.2.2 Analysis Scheme

ALEDAS2 uses the LETKF for analysis as in ALEDAS but with distance-based
covariance localization. Before illustrating this improvement, the formulation is
briefly described following Hunt et al. ( 2007 ). LETKF is a deterministic ensemble