Fig. 8. PS of observed SST (solid black), observed SST – GHG MMM (dotted-dashed black), observed SST – ALL MMM (dotted black) and associated 95% confidence intervals (black shading) in NA (a), GT (b), and NARI (c), compared to the PS of piC simulations. Similar to Figure 4, mean PS by model are colored by average SST, where blue is colder than observed, grey is observed, and red is warmer than observed.
However, deficiencies in simulating SST cannot explain the difference in simulated externally forced precipitation variability between CMIP5 and CMIP6. The only notable difference in simulated SST between the two ensembles is that CMIP6 warms NA (and therefore NARI) less than CMIP5 in the GHG simulations (Figure 7j and l). As in simulated Sahel precipitation, warming of NA and NARI in CMIP6 ALL simulations is larger than the smoothed sum of simulated SST change in the individual-forcing simulations (burgundy dashed curve), which, aside from volcanic eruptions, remains below the confidence interval for the CMIP6 MMM (dark blue shaded area) from 1950 onward (this discrepancy is, again, robust to differences in model availability for the different sets of forcing agents). Thus, a non-linear interaction between forcing agents in CMIP6 balances the additional SST warming in CMIP5 in the ALL simulations, and the difference in coupled simulations of Sahel rainfall between CMIP5 and CMIP6 must derive from changes in the fast response to forcing, SST teleconnections, or both.
e. The NARI teleconnection in Coupled Simulations
Now that we have examined SST in the coupled simulations, we may determine whether the teleconnection strength estimated from amip-piF simulations is consistent with coupled simulations. This is verified by the fact that the amip-piF teleconnection strength falls within the range of teleconnection strengths calculated from individual piC simulations in CMIP5 (\(0.5\pm 0.6\)) and CMIP6 (\(0.4\pm 0.6\)), but the ranges are large (possibly because the increased presence of atmospheric and oceanic IV and decreased variance of NARI in the individual piC simulations obscures the teleconnection). As a second test, we compare the confounded teleconnection strength in the amip-hist simulations (\(0.93\pm 0.41\)) to that of bootstrapped MMMs in the coupled ALL simulations in CMIP5 (\(0.66\pm 0.28\)) and CMIP6 (\(1.5\pm 0.3\)). The confounded teleconnection strength in amip-hist simulations is consistent with the confounded estimate in CMIP5, but is smaller than and inconsistent with the confounded estimate in CMIP6. This may be because NARI variability in the coupled simulations is smaller relative to the magnitude of external radiative forcing than it is in the amip-hist simulations. If this is the cause for the apparent inconsistency, we may still confirm the NARI teleconnection strength in CMIP6 simulations by showing that the implied fast response to forcing is consistent with the fast response from the amip-hist simulations.
f. Fast and Slow Responses to Forcing in Coupled Simulations (\(\overrightarrow{f}\) and\(F\rightarrow SST\rightarrow P\))
Under the assumption that the dominant simulated path of SST influence on the Sahel is captured by a linear relationship with NARI, we estimate the slow response to forcing in coupled simulations as the simulated NARI MMM scaled by the teleconnection strength derived from uncoupled simulations (\(0.87\frac{\text{mm}}{day\ C}\), Section 4.c), so that a warm (cold) NARI predicts a wet (dry) Sahel. In Figure 9, simulated NARI (as in Figure 7, right column) is displayed on the left ordinates in light blue (CMIP6) and turquoise (CMIP5). The right ordinates are scaled by the teleconnection strength so that, when read on the right ordinates, simulated NARI represents the estimated slow component of the precipitation response to forcing. Also on the right ordinates are the total simulated precipitation responses to forcing (as in Figure 2) in CMIP5 (right column) and CMIP6 simulations (left column), colored by forcing agents. The simulated precipitation responses to forcing (colors) match the estimated slow response to forcing (turquoise) reasonably well: the main differences appear after about 1970 in CMIP5 and 1990 in CMIP6.