Balu Nadiga is a scientist at the Los Alamos National Laboratory. His research interests include climate dynamics and modeling, fluid turbulence, dynamical systems and reduced order modeling and machine learning. Under his climate research portfolio, he currently heads the predictability group of a US Department of Energy (DOE) project that focuses on high-latitude climate, leads an effort to develop and use machine learning techniques for climate modeling, and co-leads a project on the next-generation of the DOE climate model Energy Exascale Earth System Model (E3SM) that resolves non-hydrostatic dynamics in the atmosphere.
Machine learning as a tool for climate predictability studies
Because of the chaotic nature of the dynamics underlying many complex systems such as weather and climate, evolution of ensembles of trajectories have to be considered in order to produce future predictions of such systems. An analysis of the dynamics of such ensembles provides insights into mechanisms that make the system predictable. Irrespective of whether we are dealing with the actual climate system or with its surrogates---comprehensive climate models, reduced order dynamical models are a necessary tool in conducting such predictability studies. We develop such a reduced order model using machine learning and show that it comprehensively out performs the state of the art Linear Inverse Model in terms of predictive skill. In comparison to the state of the art, the new method not only performs much better in situations where training data is limited, but it also captures the longer term climatological behavior much better. As such, this method has very wide applicability. Furthermore, given the nonlinear nature of the new method, it has the potential to provide new insights into predictability of climate.