Talk Abstract:
The global meridional overturning circulation is key to climate through its role transporting and storing tracers such as heat and carbon. Climate models suggest that the circulation is changing but the physical drivers are poorly constrained. Here an explicitly transparent machine learning method, Tracking global Heating with Ocean Regimes (THOR), reveals the root mechanisms and assesses changes across a range of climate models. Addressing the fundamental question of the existence of dynamical coherent regions, these are identified and used to understand distinct driving forces modulating the three dimensional overturning. We will focus on the North Atlantic and Southern Ocean, specifically the location of the Gulf Stream, Trans Atlantic Current and downwelling regions, as well as the Antarctic circumpolar current bathymetric interactions and the emergent gyre-like dynamics modulating upwelling. THOR is trained to identify these only using surface fields. Beyond a black box approach, THOR is engineered to elucidate its source of predictive skill rooted in physical understanding. A labeled data set is engineered using an explicitly interpretable equation transform and k-means application to model data, allowing theoretical inference. A multilayer perceptron is then trained, explaining its skill using a combination of layerwise relevance propagation and theory. With abrupt CO2 quadrupling, the North Atlantic overturning circulation weakens due to a shift in deep water formation regions, a northward shift of the Gulf Stream and an eastward shift in the Trans Atlantic Current. In the Southern Ocean, the overturning increases with an expansion of the dynamical regime associated with upwelling. If CO2 is increased 1% yearly, similar but weaker patterns emerge influenced by natural variability. THOR demonstrates a path to progress in oceanographic problems that have resisted classical analysis. Additionally, the method is scalable and allows in-depth dynamical analysis in a range of models using only the ocean depth, dynamic sea level and wind stress,. This reduces the requirement on big data dissemination which can inhibit the analysis of climate model data. With predictions that are physically tractable THOR constitutes a step toward trustworthy ML called for within oceanography and beyond.