CMIP6 models cannot capture long-term forced changes in the tropical Pacific sea surface temperature gradient

The tropical Pacific Ocean plays an outsized role in global climate, affecting, for instance, temperature and precipitation globally, as well as cyclone genesis and location. The equatorial zonal gradient in tropical Pacific sea surface temperatures (SSTs) has strengthened toward a more La Niña-like state in observations over the 20^th and 21^st centuries. Confirming whether climate models are capable of matching the observed gradient strengthening through some combination of a forced response and internal variability is therefore an active topic of research. While some studies have demonstrated that models can skillfully reproduce observed trends in the tropical Pacific SST gradient, others have argued that models fail to simulate these trends. However, these prior papers have focused on different and specific periods in the observational record to perform their assessments, with implications for the nature of the trends identified and the characterization of correspondence between the models and observations. Moving beyond assessments over a single time interval, we perform a comprehensive analysis over all trends in intervals of twenty years or longer from 1870 to 2024. We compare the observed trends from 5 observational datasets with simulated trends from 14 CMIP6 large ensembles (337 ensemble members in total). We demonstrate that models are not able to match many long-term trends in the observed gradient, especially those that end more recently. Models that are able to match these trends do so through excessive internal variability that compensates for their gradient-weakening forced responses. We additionally find that trends in the observed gradient strengthen at an increasing rate with time, a forced response that is in contrast to the behavior of most models.