Professor Frank Giraldo’s work on atmospheric modeling is in a recent article in Science Magazine as part of an ambitious project backed by prominent technology philanthropists. In the fall of 2018, members of the Scientific Computing group in the Department of Applied Mathematics at the Naval Postgraduate School will begin a project on Data-Driven Earth System Models. The project will be executed by a consortium of institutions comprised of Caltech (lead institution), MIT, and the Naval Postgraduate School. The goal of the project is to create a new Earth System Model for better climate predictions. What makes this effort different from others is the use of machine learning to teach the climate models to behave more like nature by using a multitude of Large-Eddy Simulations (LES) to better represent what is computable and to use vast amounts of satellite data to represent what is non-computable. The idea is that such an approach will allow the entire global Earth modeling system (atmosphere, ocean, etc.) to produce better results than stand-alone computer models. For this strategy to be successful it will require the computer codes to be scalable on modern computing architectures (e.g., graphics processing units or GPUs).
Profs. Frank Giraldo, Jeremy Kozdon, and Lucas Wilcox from the NPS Department of Applied Mathematics department will be in charge of developing the new atmospheric model (the Machine Learning Atmospheric Model or MLAM). The other institutions will be in charge of other aspects of the climate system (e.g., ocean model, etc.). The project will be designed from the beginning to: be open source, use modern programming languages, scale on GPU hardware, and use current software engineering practices.
Unlike typical science projects, this work will be funded not only by the National Science Foundation but also by philanthropies. The Science article can be found here: http://www.sciencemag.org/news/2018/07/science-insurgents-plot-climate-model-driven-artificial-intelligence