Seasonal prediction of equatorial Atlantic sea surface temperature using simple initialization and bias correction techniques

Due to strong mean state‐biases most coupled models are unable to simulate equatorial Atlantic variability. Here, we use the Kiel Climate Model to assess the impact of bias reduction on the seasonal prediction of equatorial Atlantic sea surface temperature (SST). We compare a standard experiment (STD) with an experiment that employs surface heat flux correction to reduce the SST bias (FLX) and, in addition, apply a correction for initial errors in SST. Initial conditions for both experiments are generated in partially coupled mode, and seasonal hindcasts are initialized at the beginning of February, May, August and November for 1981–2012. Surface heat flux correction generally improves hindcast skill. Hindcasts initialized in February have the least skill, even though the model bias is not particularly strong at that time of year. In contrast, hindcasts initialized in May achieve the highest skill. We argue this is because of the emergence of a closed Bjerknes feedback loop in boreal summer in FLX that is a feature of observations but is missing in STD. Virtually all state‐of‐the‐art coupled climate models suffer from a severe warm bias in the central equatorial Atlantic that prohibits realistic simulations of the leading mode of SST variability. We use a simple bias alleviation technique to assess whether the bias affects the predictability of equatorial Atlantic SST in seasonal hindcasts. Our results suggest that bias correction enhances the predictive skill of the model. This is illustrated in the figure to the right, where red (blue) curves show the evolution of skill for SST hindcasts when the SST bias develops (is alleviated).