Here is the global OLR (annual mean).

In this plot, thick line = mean; shading = mean +/- 1 standard error; coloured lines = individual members,
where "standard error" is the std error in the ensemble mean, i.e. sqrt[ (sum(X^2) - sum(X)^2/N) / (N(N-1))] over ensemble members (N=4)

Likewise, here is the AMOC transport at various latitudes (in SI units, i.e. 1e-6 Sv, again as annual means) -- NB this is integrated over the top 13 model levels, i.e. to a depth of approx 1250m (the last included level is centred on 995m; the first excluded level is centred on 1500m).

Particularly at 45N there is some obvious decadal variability in the MOC, above the noise.

It is possible that the increase from 1880 - 1910 may be the impact of Krakatoa and other volcanos during the 1880s and 1890s, and possibly then the decrease from about 1910-1940 might be recovery from this (rather than, say, a global warming signal -- certainly it's a bit early for most of the increase in GHG loading).

So we are repeating the transient runs from 1882 onwards with the eruptions during the 1880s and 1890s turned off in the forcing (this is the only change) - run IDs ABZEa to ABZEd. Although the main ensemble is generated using small SST perturbations in 1860, we do not use SST perturbations in 1882 but instead each no-Krakatoa ensemble member is started from the corresponding ensemble member from the main transient run, in order to preserve the ensemble spread.

Here is the change in the forcing. NB the aerosol forcing is represented in terms of four "quarters" of the globe. Here is the aerosol optical depth for the last "quarter" -- I haven't checked where this one corresponds to geographically, but it is pretty similar to the timeseries for the other three that are not shown here. Red is for the original ensemble, yellow is for no-Krakatoa, so where these differ is the suppressed eruptions.

NB this image (and the remainder of the images on this page) is inlined at reduced size as a sort of thumbnail; click to view at its proper size.

At time of writing, this experiment has got to about 1916.

Here is the global OLR from that experiment. (Dark blue line = mean of ordinary transient ensemble, and light blue shading = mean +/- 1 standard error (both as above). Red thick and thin lines = the same but for the no-Krakatoa ensemble. Individual ensemble members not shown.)

So it seems that manipulating the forcing has worked fine.

And now the MOC:

Here are some initial comments based particularly on the signal at 45N.

It certainly seems from the difference between the two ensembles that the MOC increase from 1880 is largely attributable to volcanic aerosol. Our no-Krakatoa experiment does seem to have an MOC increase starting after 1900, which is consistent with the fact that we turn the volcanos back on again then -- interestingly despite the fact that these are rather smaller eruptions (see the forcing or the OLR response above).

By the end of the experiment so far (~1916), there is sign that the MOC is beginning to lose memory of the volcanoes that were switched off, in that (at 45N) the smaller volcanoes post-1900 have by then produced about the same increase - at least given the temporary dip in the blue curve at that time. But it's a bit early to be sure. The new runs are continuing; let's see what they do over the following decade or two. And if necessary we can run some more ensemble members -- maybe a bit expensive, but particularly at the lower latitudes improving the signal-to-noise might help.