Machine Learning Testbed

Objectives: Participants engaging in the DCMIP 2025 ML hands-on session will:

• gain familiarity with running three ML weather forecast emulators: Nvidia's SFNO, Google's GraphCast, and Huawei Cloud's PanguWeather;

• run idealized simulations with these two models to probe aspects of the models' physical fidelity; and

• explore and intercompare model responses as the model inputs stray further and further from their training dataset

Prerequisites: To meet the above objectives in the five half-day hands-on sessions, participants will need background in the following:

• proficiency in unix command line environments

• proficiency with python and with weather/climate data visualization

• basic familiarity with machine learning

• access to NCAR supercomputing systems: specifically casper

See the DCMIP2025-ML github repository for a detailed description of the three test cases.