Using CMIP6 simulations to assess significance of an AMOC trend seen by the RAPID array

Observations between 2004 and 2020 at the RAPID array suggest a weakening trend in the Atlantic Meridional Overturning Circulation (AMOC). To assess the significance of this trend, trends that one might expect from natural variabilty in a time series of this length are assessed using CMIP6 pre-industrial simulations. The observed trend is not found to be statistically significant relative to this benchmark. Both the observed trend and the standard 
deviation of short-term model trends are found to decrease in magnitude with time. The rate of decrease, however, is faster for the observed trend, further calling into question its significance.

To clarify how variability in short-term model trends is related to power spectra of modelled AMOC strength, a conceptual model is developed. Essentially, trend variance is represented by a random walk in which there is one step for each frequency bin of the power spectrum (with step size determined by the frequency and variance of the bin in question). Most models are found underestimate interannual variability in AMOC strength; however, it is the variability at somewhat longer time scales that most influences model trends. This variability is represented quite differently between the various CMIP6 models. The conceptual model is also used to illustrate how the detectability threshold for trend detection (i.e., the 2 sigma level in a PDF of short-term model trends) is altered by the addition of noise added to make AMOC variance more in line with observations.