Over the past several decades, many efforts have been devoted to increasing the simulation performance of climate models, but significant biases remain that hinder the performance of coupled systems. Hence, bias correction is regarded not only as a useful tool for improving climate simulations but also as an important step before data assimilation, which depends on the hypothesis of unbiasedness. In this study, using sea temperature climatological data, a new ensemble‐based approach is proposed for correcting the biases of the sea temperature in CAS‐ESM‐C. Through analyzing the results of the proposed bias correction method with various intensities and time windows, its performance in suppressing the simulation biases of ocean fields is evaluated. The simulation biases of atmospheric variables are also reduced via air‐sea interactions, which will improve the ocean simulation performance. Additional benefits can be realized by applying the bias correction method. For example, a superior simulation of climate variabilities in a coupled model, such as ENSO (El Niño‐Southern Oscillation), is realized due to the improvement of climatological fields. The ability to assimilate various ocean observations is also significantly improved with a better background mean state.

In this study, based on an ocean data assimilation system for the coupled climate model CAS‐ESM‐C, how to reasonably assimilate altimetry data are explored. In sea surface height (SSH) assimilations, the mean dynamic topography (MDT) is an important factor that can coordinate the observed sea level anomalies with the modeled SSH. The SSH assimilation results are first compared through assimilation experiments using three different MDTs, including the observed reference height, the MDT from model control run, and the MDT from the assimilation experiment in CAS‐ESM‐C, with the climatological World Ocean Atlas (WOA) temperature assimilated into the coupled model. The results show that the third one can significantly improve the ability of the ocean data assimilation system to assimilate the SSH observations. Using this MDT, the SSH assimilation scheme of the ocean data assimilation system was established for the CAS‐ESM‐C, and a long‐term SSH assimilation experiment from 1994 to 2017 was carried out. The results show that the SSH assimilation performs much better than the SST assimilation in reproducing the ocean states and seasonal‐interannual variability. The improvements of SSH assimilation compared with SST assimilation may be due to the different properties of the two ocean variables. While SST is a thermodynamic variable used to evaluate the thermal condition of the ocean surface layer, SSH is a dynamic variable linked to the dynamical information of the whole ocean layer. Thus, SSH assimilation can constrain the ocean model with observed dynamic information, which is more important to the state estimation and temporal evolution of the ocean.

Currently, several ocean data assimilation methods have been adopted to increase the performance of air–sea coupled models, but inconsistent adjustments between the sea temperature with other oceanic fields can be introduced. In the coupled model CAS-ESM-C, inconsistent adjustments for ocean currents commonly occur in the tropical western Pacific and the eastern Indian Ocean. To overcome this problem, a new ensemble-based bias correction approach—a simple modification of the Ensemble Optimal Interpolation (EnOI) approach for multi-variable into a direct approach for a single variable—is proposed to minimize the model biases. Compared with the EnOI approach, this new approach can effectively avoid inconsistent adjustments. Meanwhile, the comparisons suggest that inconsistent adjustment mainly results from the unreasonable correlations between temperature and ocean current in the background matrix. In addition, the ocean current can be directly corrected in the EnOI approach, which can additionally generate biases for the upper ocean. These induced ocean biases can produce unreasonable ocean heat sinking and heat storage in the tropical western Pacific. It will generate incorrect ocean heat transmission toward the east, further amplifying the inconsistency introduced through the tropical air–sea interaction process.

In this study, high spatiotemporal resolution of five latent heat flux (LHF) datasets are compared over the Northwest Pacific Ocean for the period 1998–2007. Overall, the datasets show encouraging agreement in spatial and temporal characteristics. However, there are substantial discrepancies, especially in magnitude, both spatially and temporally. The ERA-Interim dataset gives the maximum LHF while the SeaFlux dataset gives the minimum. These consistencies and discrepancies of the LHF in the five datasets are mainly due to the differences in bulk variables, specifically the sea–air specific humidity difference (Qs-a) and wind speed. Generally, large Qs-a leads to large LHF. However, wind speed and Qs-a play different roles in different regions. The maximum LHF in the Kuroshio western boundary current region arises from high wind coupling with large Qs-a, and the minimum LHF at low latitude is mainly due to weak winds whilst at high latitude it is mainly due to the rapid poleward decrease of Qs-a.