The promised plots.
I'll show data since 2003 only because only then did real measurements really kick off. Before that, measurements were so spotty with 1 in January in one year, then 3 years nothing, then 5 measurements in for example August and so on.
Data source for measurements is #NOAA https://www.ncei.noaa.gov/access/world-ocean-database-select/dbsearch.html

The other plot is from a #ReAnalysis called EN4.2 https://www.metoffice.gov.uk/hadobs/en4/download-en4-2-2.html#c14_analyses
It is based on the same NOAA data, and goes back to 1900. EN4.2 has better quality assurance than my download. And EN4.2 is able to calculate the likely values in adjacent coordinates and adjacent months.

Among other Reanalyses, EN4.2 has also been used in the newest preprint by the Dutch team around #vanWesten. It sees AMOC tip in 2065 in RCP8.5 and in 2085 in RCP4.5. But it is based on biased Reanalyses and known-to-be too stable #CMIP6: https://arxiv.org/pdf/2407.19909

Whereas the first new preprint by van Westen's team, which tips AMOC 2037-2064, uses only real measurements of the 3 dedicated monitoring arrays in the North, tropical and South #Atlantic. https://arxiv.org/pdf/2406.11738
Here, the caveat is: very short data series.

With measurements being so spotty, I only show monthly salinity 2003ff in down to 10m depth. And only from the area in the South Atlantic. Averaged on a 2 by 2 degree grid from Argentina's coast to longitude -48, and latitudes 40-48 South.

Sorry, not sorry: neither dataset shows a smoking gun.
#FridaysForFuture #ChartsForFuture

YES! Found the issue. The pipeline was processing #CMIP6 models and it turned out I had downloaded multiple versions of the same model. Filtering to keep only one version solved the issue!

I hate CMIP6.