Subject No. 7: Advanced Four-Dimensional Data Assimilation System Using a Coupled Model Toward Construction of High-Quality Reanalysis Datasets

Introduction:
Our research goal is to construct an innovative four-dimensional coupled data assimilation system capable of providing a high-quality comprehensive dataset. This is stimulated by the recent remarkable progress in the earth observing system and numerical models and by the emergence of a gigantic computer, "the Earth Simulator" For this purpose, we have been constructing the four-dimensional variational data assimilation system (4D-VAR) on the coupled atmosphere-ocean-sea ice-land model. Since the beginning of our project, we have (1) improved the coupled model, (2) completed each adjoint model of the assimilation system, and (3) started experiments by the coupled data assimilation. (April 2004)

Coupled Simulation
We use the Coupled Atmosphere-Ocean-Sea Ice model for the Earth Simulator (CFES). The atmospheric component is based on the CCSR/NIES model. The ocean and the ice components are from the GFDL/MOM3 and the IARC Ice model respectively. We also integrated the land parameterization scheme MATSIRO (Minimal Advanced Treatments of Surface Interaction and Runoff) into the CFES instead of the current one layer model so called "bucket model"
We added many modifications and tunings to the CFES to improve its performance. One example of such tunings is shown in Figure 1. It is well known that in many coupled models the precipitation over the tropical Pacific Ocean is too symmetric about the equator (so called "Double ITCZ" as is shown in Figure 1 (a). Our tuning to the cloud parameters, however, lessened the Double ITCZ structure.
Figure 1. Precipitation over tropical Pacific Ocean from the CFES. (a) Before tuning. (b) After Tuning.

After tunings, the CFES reproduces important climatological phenomena reasonably well. This is the good base for further perfection of the datasets by the assimilation system. Figure 2 confirms that the interannual variation pattern of the sea surface temperature (SST) of the CFES is close to the one of observation.
Figure 2. Liner regression of SSTs at each grid box against Nino 3.4 SST anomalies. (a) From CFES. (b) From the observation (ERSST).
Assimilation of each component:
We have constructed the assimilation system of each component. Figure 3 shows the assimilation results of the atmospheric model to the SSM/I observed wind speed. The difference of the observed 10m wind speed between the simulation and the observation (Figure 3(a)) decreased well after the assimilation to the 1-month averaged data (Figure 3(b)). This figure demonstrates that the assimilation to the climatological (beyond weather) data is possible by our system.
Figure 3. The difference of the 10m wind speed. (a) Between the simulation and the observation. (b) Between the assimilation and the observation.
Figure 4 shows that the SST along the Equator of the 1997/1998 El Niño is improved by the assimilation system of the ocean model.
Figure 4. The SST along the Pacific equator from January 1996 to June 1999. (a) Simulation. (b) Observation. (c) Assimilation.
Coupled Assimilation:
We have started the experiments of the assimilation on the coupled model. Our preliminary result shows that the cost function to the observational data successfully decreases by controlling the bulk coefficient for the air-sea fluxes.
Our 4DVAR coupled data assimilation system is expected to be better able to represent the nature of climate variability and thereby offer greater information and forecast potential than do existing models and data alone.
 
See the Annual Report of RR2002 Project