New authoritative report has data on when pacific oxygen may run out

This is one of the figures in an important new article in the Proceedings of the National Academh of Sciences:
https://www.pnas.org/content/early/2019/06/25/1906556116
This is an important clue to the question I still wonder about: how soon will oxygen start to run out in the deep Pacific Ocean, causing a release of poison gas enough to kill every human on earth (as has happend five to ten times before in earth history)? Will it all start in 30 years, or never, or even sooner in case of an Antarctic surprise? No one on earth knows; in my video interview,  https://vimeo.com/342606790, I explain the best I knew just a few weeks ago, but I keep looking for new information.
The curves above give an important new clue. The melting part of the top curve (pieces of downward slope) is not a threat to the oxygen system; it is the upwards pieces which represent freezing of oxygenated salty sea ice. this figure is for the Ross Sea, where the main oxygen supply for the Pacific Ocean is located (the primary "polynya"). The top curve looks just fine, just as much upslope in recent years as before. But that raises a question: WHY the decline observed in the NOAA/Purkey data figure, copied over in my paper www.werbos.com/Atacama.pdf? My best guess is that rising fresh water is the main reason... how else could the same amount of freezing be followed by less oxygen in the depths?
The bottom curve suggests that the warming trend of the more recent years really is sending the overall levels down, enough so that 30-50 years from now, the top curve will not be so nice. Combined with the freshening effect, it starts to look again as bad as it did when I wrote the Atacama paper. Maybe. But yes, unproven. (People do buy insurance for much lower risks!)

How could we narrow down the HUGE uncertainty here? Lots of  people have built giant climate models, but I know from my work at DOE in the 1980's that megamodels can be out of touch with time-series data. And so, why not apply more sophisticated neural networks to understand where this is going? https://comp-astrophys-cosmol.springeropen.com/articles/10.1186/s40668-019-0029-9 might be relevant, though I think I could do better. The PNAS article shows that great time-series data do exist for ice cover. the NASA Aquarius salinity database is, sadly, a lot weaker, but some of us have known for decades how to use certain types of neural network for data fusion with missing data and even for multiscale modeling (essential here since higher resolution would be desirable in the Ross Sea polynya). Salinity, ice cover and surface temperature at multiresolution... I wonder what that would tell us? A lot more than an aggregate time series...