Linearized Physics for Data Assimilation at ECMWF - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 273

K

60°N 263.2

50 ° W

260

255

250

50°N

40 ° W 245

-5.1

240

235

40°N

4.9

230

225

220

20 ° W 10 ° W

0 °

30 ° W

Fig. 11.8 Adjoint sensitivities of the precipitation cost function defined over the black box (see
Fig. 11.7 ) with respect to 500-hPa temperature at 0000 UTC 9 February 2009, i.e. with a lead time
of 24 h. Sensitivities are shown with white isolines ( solid for positive, dash for negative) and are
expressed in units of 10 4 (mm day 1 )K 1 . The background 500 hPa temperature field valid at
0000 UTC 9 February 2009 is displayed using grey shading and black isolines of mean-sea-level
pressure are also plotted (in hPa)

Tropics. The positive impact is also generally remarkable in the lower troposphere
(e.g. 700 hPa temperature scores or 700 hPa relative humidity scores).
The results presented above only show which impact the linearized physical
parameterizations have on the evolution of the model state from the beginning of
the 4D-Var assimilation window to the time of observations. However, including
physical processes in the linearized model also allows to assimilate observations
that are directly related to the physical processes, such as cloud and precipitation
observations. Therefore further improvement in producing more realistic initial
atmospheric states can be achieved. Since the late 1990s, significant efforts have
been devoted to the assimilation of such observations. This is also the case at
ECMWF, where a 1D C 4D-Var technique has been first used operationally for
the assimilation of precipitation-related observations using microwave brightness
temperatures from SSM/I ( Bauer et al. 2006 ) from June 2005 until March 2009.
This was then replaced by direct 4D-Var assimilation unifying the treatment of clear-
sky, cloudy and precipitation situations, leading to an all-sky approach ( Bauer et al.
2010 ; Geer et al. 2010 ). Direct 4D-Var of rain- and cloud-affected observations
allows a physically consistent adjustment of model dynamics with temperature
and humidity increments, due to the sensitivity of radiance observations to the
atmospheric state through the combined radiative transfer model and the moist-