Machine Learning in Earth System Modeling
Kelly Kochanski, University of Colorado
Recent trends in Earth system modeling, climate data collection, and computing architecture have opened new opportunities for machine learning to improve ESMs. First, new and cheaper satellites are generating large volumes of observational data (e.g. Arctic and Antarctic DEMs), and massive climate modeling projects are generating large volumes of simulated climate data (e.g. CMIP5, CMIP6, CESM-LE). Second, machine learning applications are driving the design of next-generation computing architectures that will accelerate applications like neural nets without ameliorating the computational bottlenecks (ref: NOAA HPC position paper) that limit existing climate models. Third, the climate science community is becoming increasingly familiar with machine learning techniques. Here, I summarize opportunities for Earth scientists to use machine learning techniques to improve Earth system and Earth surface models. I present preliminary results showing how these techniques can be used to optimize sea ice models with SVMs.
Abstract Author(s): Kelly Kochanski, Ghaleb Abdulla