Fig. 9. Simulated Sahel precipitation (right ordinates, same as Figure
2) MMMs (solid and dotted curves) and associated 95% confidence
intervals (shaded areas) in CMIP5 (right column) and CMIP6 (left column)
when forced with ALL (blue, top row), AA (magenta, second row), NAT
(brown/red, third row), and GHG (green, bottom row), compared to
simulated NARI (left ordinates, light blue and turquoise, same as Figure
7). The right ordinates are scaled such that a 1°C change in NARI
corresponds to a 0.87 mm/day change in precipitation, given by the
teleconnection strength in the CMIP6 amip-piF simulations (see Section
4.c).
We expect the differences between the simulated Sahel and the rescaled
NARI to estimate the simulated fast response to forcing, but this would
imply a fast response to ALL in CMIP5 (Figure 9e) that is inconsistent
with the uncoupled estimate (purple, Figure 5c): instead of wetting the
Sahel, it consists of a drying response to increasing GHG of\(-0.0042\pm 0.0036\ \frac{\text{mm}}{day*year}\) (Figure 9h). Whether
we should interpret this as a fast response or a non-NARI-mediated
response to SST, this component of the forced response helps delay and
increase the severity of the minimum in precipitation in ALL relative to
the AA simulations.
The estimated fast responses for CMIP6 are displayed in Figure 10 in a
fashion similar to Figure 2, and are compared to the fast response
obtained as the difference between amip-hist and amip-piF simulations
(purple, as in Figure 5c). Unlike the fast response in CMIP5, the ALL
fast response in CMIP6 matches the AMIP fast response significantly
better than noise (r = 0.51, sRMSE = 0.87), giving us confidence that
the NARI teleconnection strength estimated from amip-piF is valid in
CMIP6 coupled simulations. Like the amip-hist fast response, the ALL
fast response in CMIP6 displays wetting after 1980 that is roughly equal
to the sum (burgundy dashed curve) of the fast responses to AA (b,
magenta) and GHG (d, green). The simulated fast wetting after 1980 in
the ALL simulations (a, blue) is smaller than in the AMIP simulations,
as expected if amip-hist is double-counting radiative forcing, but is
still larger than our estimate of the optimal value (0.3 times the AMIP
fast response), consistent with claims that the strength of radiative
forcing is overestimated in the coupled simulations.