Skip to main content

Gaussian processes for learning about climate model parameters

Climate models are used to understand the current state of the climate and to predict its future behavior. In this talk I will describe a statistical approach for improving climate model projections of the North Atlantic Meridional Overturning Circulation (AMOC). The AMOC is part of the global ocean conveyor belt circulation and transfers heat between low and high latitudes in the Atlantic basin. The AMOC might collapse in a "tipping point" response to anthropogenic climate forcings. One key source of uncertainty in AMOC projections from a climate model is uncertainty about a climate model parameter called background ocean vertical diffusivity, Kv. Kv cannot be directly measured but can be inferred from two sources of information: (1) physical observations from the oceans (so called tracers) that are related to Kv, and (2) climate model output at various Kv settings. I will describe a Gaussian process-based emulation and calibration approach for Kv inference. This approach accounts for non-linear relationships between the tracers, data-model discrepancies, and various sources of uncertainties and dependence. This approach also uses sparse matrix techniques to overcome the considerable computational obstacles posed by the size of the data.

This is joint work with K. Sham Bhat (Los Alamos National Laboratories), Roman Olson (Geosciences, Penn State University), Klaus Keller (Geosciences, Penn State University).