Machine Learning for Climate Modeling

Climate models are essential tools for understanding Earth system processes and feedbacks. In this talk I will discuss two examples of how machine learning can be applied to further climate science research using climate models.

First, I will discuss a machine learning-approach to emulation and parameter calibration with a land model. Using a perturbed physics ensemble of model simulations, we train an emulator to predict model output given a set of land parameters as input. By comparing the emulated model output with biophysical observations, we can optimize parameter values and help constrain emergent features of the climate system such as the land carbon sink.

Second, I will discuss work using machine learning to automatically detect weather features that produce extreme precipitation events. Here we use a suite of deep learning algorithms to identify fronts and atmospheric rivers in a climate model and evaluate the results using observational and reanalysis products. To study how these features might change with climate change, we compare results between model simulations using present-day and future climate forcing.