[A-AS02_29PO1] Data Assimilation in Earth Sciences

[AAS02-P01] Decadal climate prediction using 4D-VAR data assimilation approach

*Takashi MOCHIZUKI¹, Yoichi ISHIKAWA¹, Shuhei MASUDA¹, Toshiyuki AWAJI² (1.JAMSTEC, 2.Kyoto University)

Keywords:decadal prediction, climate prediction, global warming, data assimilation, 4D-VAR

It is very recently that decadal climate prediction experiments have been carried out with initialization. As a first step in decadal prediction, simple initialization approaches have usually been used so far, particularly focusing on ocean states. An advanced initialization technique is a pressing concern toward further enhancing the decadal predictability by obtaining suitable atmospheric and oceanic initial conditions that are compatible with both the model and observations. Here, by employing a 4D-VAR data assimilation approach to initialize the atmosphere-ocean coupled climate model, we attempt to perform ensembles of decadal hindcast experiments in line with the CMIP5 protocol. We perform full-field initialization rather than anomaly initialization and assimilate the atmospheric states together with the ocean states. We can validate the predictive skills in the atmosphere and ocean temperature hindcasts in some areas and, roughly speaking, the spatial patterns of the hindcast skills are similar to those of the multi-model ensembles of the CMIP5 decadal hindcasts. While our assimilation system has been developed originally for the purpose of seasonal-to-interannual climate simulations and we use 9-month assimilation window in these experiments, the hindcast results suggest that the atmosphere and ocean states associated with low-frequency variations beyond annual timescales can also be effectively initialized through the iterations of the forward and backward runs of the 4D-VAR data assimilation.