The experts might know about this already, but the following paper just "hit the streets":

T. Bolton, L. Zanna, "[Applications of deep learning to ocean data inference and subā€grid parameterisation](", _Journal_ _of_ _Advances_ _in_ _Modeling_ _Earth_ _Systems_, 4 January 2019.

This is (at least) the second paper which applies machine learning to the deep technical problem of convective fluids. The first I know of is

P. A. O'Gorman, J. G. Dwyer, "[Using machine learning to parameterize moist convection: potential for modeling of climate, climate change and extreme events](", _Journal_ _of_ _Advances_ _in_ _Modeling_ _Earth_ _Systems_, 3 October 2018.

To me this is very heartening, and I'm having a solid chuckle because a well-known expert in such systems announced at the Lorenz-Cheney symposium last year in response to a question that such techniques offered no promised at all for work in this area.