PyEarthTools: Machine learning for Earth system science
summary
An open source Python framework for building and running machine learning Earth system models, PyEarthTools supplies modular data pipelines (reprojection resampling regridding normalization), prebuilt model hooks and tutorials that run on a laptop or workstation and train with a 4 GB GPU on 3–10 GB of data. It integrates with XGBoost, PyTorch, Apple MLX, separates pipelines from models for reproducibility, and claims published ML architectures that since 2023 match or exceed physics models on specific metrics.
AI-written summary · about
chapters (11)
- 0:00PyEarthTools intro
- 1:09Earth system science
- 3:59Machine learning in Earth science
- 6:28PyEarthTools history
- 8:32What PyEarthTools offers and its design
- 10:37Earth data challenges
- 14:49PyEarthTools strengths
- 16:59Community impact
- 18:43Getting started
- 21:08Roadmap
- 21:34Q&A
related talks
46mThe Python Data Ecosystem: Navigating a fragmented landscape
01h 16mReproducible machine learning and science with python
35mAn End-to-End Machine Learning Ecosystem in a Quarter
57mGood Practices for Applied Machine Learning
04mState-of-the-art Machine Learning research with Google tools
28mHarnessing Machine Learning to Improve Agricultural Resilience in Africa
27mPyTesting the Limits of Machine Learning
48mDeploying Machine Learning Models in Practice