Causal analyses reveal a strengthening of tropical Pacific ocean–atmosphere coupling in observations but not in climate models

Tropical Pacific sea surface temperatures play a central role in shaping global climate variability. By governing atmospheric circulation patterns such as the Walker circulation and constraining the development of the El Niño–Southern Oscillation through coupled ocean-atmosphere interactions in the Bjerknes feedback, they exert wide-ranging impacts on the global climate system. Understanding how the tropical Pacific mean state evolves under climate change is therefore critical. However, substantial uncertainty remains, as climate models and observations disagree on the direction of recent SST trends. While climate models simulate a weakened zonal SST gradient, observations indicate a strengthening gradient, characterized by warming in the western Pacific and weak cooling in the central and eastern Pacific. This discrepancy has been linked to several systematic model biases. Here, we assess how the strength of ocean–atmosphere coupling has evolved in recent decades in both observations and models, using a causal discovery framework based on the PCMCI algorithm, applied to key variables of the Bjerknes feedback, including trade winds, near-surface temperature, and sea level pressure.

We find that, although climate models qualitatively capture the Bjerknes feedback, they fail to reproduce observed trends in the coupling strength between trade winds and near-surface temperatures in the central tropical Pacific. In observations, the causal effect of trade winds on near-surface temperature has strengthened over recent decades, a trend that lies outside the range of CMIP6 models. This indicates that, for a given wind anomaly, the temperature response has become increasingly sensitive, implying a strengthening of ocean-atmosphere coupling and an enhanced wind-driven cooling effect on SSTs in the central Pacific, which is lacking in CMIP6 models. These results provide a process-based explanation for discrepancies in tropical Pacific SST trends between observations and climate models, and highlight biases in simulated ocean-atmosphere coupling in CMIP6 models in a changing climate.