In 3 consecutive years from 2016 to 2018, extreme ocean warming events, or marine heatwaves (MHWs), occurred during boreal summers in the East China Sea (ECS) and South Yellow Sea (SYS), which is unprecedented in the past four decades based on the satellite record. In this study, we used a high‐resolution hydrodynamic model based on Finite Volume Community Ocean Model (FVCOM) to simulate the evolution of these warming events. An upper ocean temperature budget (0–20 m) analysis based on the model results shows that the shortwave radiation and the ocean advection anomalies jointly contributed to the anomalous warming in the three successive summers (June–August) in the SYS and the north part of the ECS. In addition, the reduction of surface wind speeds during the 2016 and 2017 summers further weakened the vertical mixing, thereby enhancing the anomalous warming in the north part of the ECS adjacent to the SYS. During the three summers, the increases of shortwave radiation were closely related to the East Asian Summer Monsoon (EASM) variability, which reduced the cloud cover in the ECS and SYS, whereas the advection anomalies were mostly associated with regional wind anomalies. In summer 2018, upper ocean heat was transported into the central trough of the SYS, accumulated in an anticyclonic eddy generated by the anomalous wind stress curls. Therefore, despite the primary driver of the MHWs is the EASM variation, regional processes are critical to driving the spatial pattern of the MHW intensity in the ECS and SYS.

Surface wind is significant for ocean state climate, ocean mixing, and viability of wind energy techniques. However, surface wind simulated from the regional climate model generally features substantial bias from observation. For the first time, this study compares the performance of five bias correction techniques, (1) linear scaling, (2) variance scaling, (3) quantile mapping based on empirical distribution, (4) quantile mapping based on Weibull distribution, and (5) cumulative distribution functions transformation, in reducing the statistical bias of a regional climate model wind output, which was downscaled from a global climate model CNRM‐CM5 during 1991–2000. The surface wind of JRA55 reanalysis data is used as reference. Results show that all bias correction methods are consistent in reducing the climatological mean bias in spatial patterns and intensities. The linear scaling method always performs the worst among all methods in correcting higher‐order statistical biases such as skewness, kurtosis, and wind power density. The other four bias correction methods are generally similar in reducing the statistical biases of different measures based on spatial distribution maps. However, when it comes to spatial averaged mean of statistical measures over CORDEX‐East Asia in January and July, the quantile mapping based on Weibull distribution generally shows the best skills among all methods in bias reduction.