Biased trend amplification ratios in CMIP6 models

Abstract Ratios of linear trends are used to measure amplification ratios for warming and cooling rates in atmospheric layers relative to the surface. This metric is informative as it reveals a fundamental characteristic of the model physics. Excess amplification in CMIP5 model low-latitude regions has previously been noted in the context of excess warming in the tropical troposphere. Determining the statistical significance of the differences in amplification rates is a nontrivial problem. Vogelsang and Nawaz 1 derived confidence intervals for a single trend ratio robust to general forms of autocorrelation and Vogelsang 2 generalized it to the multivariate case. We apply the method to the CMIP6 model archive to test whether amplification rates over 1958–2024 hindcasts are significantly different from observations over the tropics. The results show that CMIP6 models continue to overestimate the warming amplification rate vertically through the 850–100 hPa range. Trend ratios match better in the lower stratosphere. We discuss possible explanations, including potential excess moistening rates under upper atmospheric warming in models.