Speaker: Pierre Gentine
Affiliation: Columbia University
In recent years, we have witnessed an explosion in the applications of machine learning, especially for environmental and climate problems. Yet for broader use, those algorithms may need to respect important physical constraints such as the conservation of mass and energy. In addition, environmental applications (e.g. drought, heat waves) are typically focusing on extremes (i.e. tail of the distribution) and on out-of-sample prediction rather than on interpolation within the training distribution, as well on a shift of the distribution (e.g. climate change). This is a major problem for standard machine learning algorithms, which interpolate well but have difficulties extrapolating. I will here show how a hybridization of machine learning algorithms, imposing physical knowledge within them, can help with those different issues and offer a promising avenue for climate applications and process understanding with applications to convection and turbulence. We will then close the loop by showing how machine learning can also be used to learn new physics on complex problems, by aiding the physical scientist in creating systematic and consistent lower dimensional representation of a complex physical problem.