Lightning Talk: Atlantic Meridional Overturning Circulation (AMOC) Tipping Point: A GAN Test Case

Jay Brett, David Chung, Jennifer Sleeman, Anand Gnanadesikan

Keywords: oceanography, generative adversarial network, surrogate modeling, climate change

Abstract: Using the four-box AMOC model described by Gnanadesi-kan (1999), we aim to identify underlying causes of the inter-model differences in projected AMOC strength under global warming. Using outputs from 1850-2100 under SSP3-70, we fit vertical and isopycnal mixing rates and the parameter for resistance (as in Brett et al. 2022) to overturning for two global climate models, the Community Earth System Model (CESM2) and the Goddard Institute for Space Studies model (GISS-E2.1). Human exploration of the thus-fitted surrogates for these two global climate models shows that the AMOC shutoff occurs under different strengths of the hydrological cycle and winds over the Southern Ocean. Using a multi-gen-erator tipping-point GAN (generative adversarial network, TIP-GAN, Sleeman et al. 2022) will allow for the identifica-tion of the separatrix in the 10-dimensional space of the sur-rogates, rather than the 2D space explored by hand. The re-sults of the continuing work will feed back into the global models by allowing scientists to better understand inter-model disagreement and the risks of the AMOC tipping point in a warming climate.