Japanese efforts on coupled data assimilation: Lesson learnt from the coupled 4DVAR development and recent challenges in Japan Meteorological Agency (JMA).

In the 2000s, the Japanese K7 research group developed a coupled atmosphere-ocean four-dimensional variational (4DVAR) data assimilation (DA) system. The system targeted estimates of seasonal-to-interannual variations on time scales longer than 10 days. It usually used nine-month DA windows, and optimized the coefficients of the air-sea bulk formula, as well as the initial conditions of ocean variables. The system showed that strong constraint on a coupled system using both ocean and atmosphere observation data can be realized by a coupled adjoint implementation based on 4DVAR approach. As a result, ENSO forecasting experiments using the optimized parameters successfully improved the forecast skill for some past El Nino events. Despite these advantages, the system was not applicable to coupled predictions and reanalysis in operational centers due to the need for long-term observation data and huge computational resources, and its inability to reproduce daily weathers.

However, their work inspired Japan Meteorological Agency (JMA)/ Meteorological Research Institute (MRI) to initiate coupled DA studies. First, MRI developed a system to assimilate only ocean observation data into a coupled model. In this study, they found that reconstruction of the negative feedback between sea surface temperature (SST) and precipitation by a coupled model was essential to reproduce tropical cyclone activities and global-scale atmospheric circulation. Then, they developed a weakly coupled DA system that combined JMA's operational atmosphere and ocean DA systems. Although the system showed improved relationships between SST and atmospheric elements (precipitation, surface air temperature, etc.) over its uncoupled counterpart, the atmospheric improvement was small due to the constraint by atmospheric DA, and the ocean fields were adjusted to the atmospheric fields instead. The result indicated that weakly coupled DA systems have difficulty improving the atmospheric fields more effectively than uncoupled ones. In order to improve atmospheric fields through coupled DA in the future, it is desirable to introduce coupled 4DVAR so that the atmospheric fields can be modified to be consistent with the modification of the ocean fields even for the shorter time-scales than those of the K7 system.