# ClimateLearn: Benchmarking Machine Learning for Weather and Climate Modeling

**Authors:** Tung Nguyen, Jason Jewik, Hritik Bansal, Prakhar Sharma, Aditya Grover  
**arXiv:** [2307.01909](https://arxiv.org/abs/2307.01909)  
**Date:** 2023-07-04  
**Summary page:** https://www.summarizepaper.com/en/paper/2307.01909v1/

## Key points

- ClimateLearn is an open-source PyTorch library for training and evaluating machine learning models in climate science.
- It provides holistic pipelines for dataset processing using popular sources like ERA5, CMIP6, and PRISM.
- Implements state-of-the-art deep learning models such as Transformers and ResNets specifically tailored for weather and climate modeling tasks.
- Offers comprehensive quantitative and qualitative evaluation techniques to assess model performance.
- Provides extensive documentation, contribution guides, and quickstart tutorials to enhance accessibility.
- Fosters community growth through collaboration among researchers in weather/climate modeling and machine learning.
- Conducted comprehensive forecasting and downscaling experiments to showcase its capabilities.
- The first large-scale open-source effort that combines research in weather/climate modeling with cutting-edge machine learning techniques.

## Summary

ClimateLearn is an open-source PyTorch library that simplifies the training and evaluation of machine learning models for data-driven climate science. It provides holistic pipelines for dataset processing using popular sources like ERA5, CMIP6, and PRISM. Additionally, it implements state-of-the-art deep learning models such as Transformers and ResNets specifically tailored for weather and climate modeling tasks. It also offers comprehensive quantitative and qualitative evaluation techniques to assess model performance. The library goes beyond its technical functionalities by providing extensive documentation, contribution guides, and quickstart tutorials to enhance accessibility while fostering community growth through collaboration among researchers in weather/climate modeling and machine learning. To showcase the capabilities of ClimateLearn, comprehensive forecasting and downscaling experiments have been conducted which demonstrate its potential impact on advancing research in weather/climate modeling using modern machine learning systems. As far as current knowledge goes, ClimateLearn stands out as the first large-scale open-source effort that bridges research in weather/climate modeling with cutting-edge machine learning techniques. Researchers can access it publicly on GitHub at https://github.com/aditya-grover/climate-learn.