Prithvi-WxC-1.0-2300M

NASA and IBM have teamed up to create an AI Foundation Model for Weather and Climate - Prithvi WxC, using MERRA-2 data. By embracing the principles of open science, both organizations are actively contributing to the global mission of promoting knowledge sharing and accelerating innovations in addressing critical environmental challenges. With Hugging Face’s platform, they simplify model training and deployment, making it accessible for open science users, startups, and enterprises on multi-cloud AI platforms like watsonx. Additionally, Hugging Face enables easy sharing of the pipelines of the model family, which our team calls Prithvi WxC, within the community, fostering global collaboration and engagement. Prithvi WxC has been fine-tuned for several tasks including climate downscaling (released as part of the enterprise-grade IBM granite family of models) and gravity wave parameterization. More details on Prithvi WxC can be found in the joint IBM NASA technical paper.

Current downstreams tasks of PrithviWxC are (please feel free to submit a PR if you want to add yours):
Downscaling (IBM Granite Model) <b><i>>>Try it on Colab<<</i></b> (Please select the T4 GPU runtime)
Gravity Wave

More information:Paper, Code, Model V1 PRs: prithvi-weather-climate-foundation-model (NASA), weather-climate-foundation-model (IBM) nasa-ibm-weather-climate-foundation-model (NASA)

Prithvi WxC is a 2.3 billion parameter model trained on 160 different variables from MERRA-2 data. It has been pretrained on both forecasting and masked reconstruction objectives. I.e.~the model is capable of reconstructing atmospheric state from partial information as well as propagating state into the future. The model takes data from two timestamps as input and generates a single, possibly future, timestamp as output. Currently Prithvi WxC comes in two flavors:

• (This model) prithvi.wxc.2300m.v1 has been pretrained with a 50% masking ratio. The time delta between input timestamps is variable as is the forecast lead time. During pretraining, the input delta was chosen from [-3, -6, -9, -12] hours while the forecast lead time was chosen from [0, 6, 12, 24] hours. We recommend using prithvi.wxc.2300m.v1 for generic use cases that do not focus on forecasting.
• prithvi.wxc.rollout.2300m.v1 has been through further training cycles to be optimzed for autoregressive rollout. Here, we restricted the input delta as well as the lead time to 6 hours. We recommend using prithvi.wxc.rollout.2300m.v1 for forecasting applications.

<div style="display: flex; justify-content: center;"> <b> Zero-shot reconstruction </b> <img src="https://cdn-uploads.huggingface.co/production/uploads/6488f1d3e22a0081a561ec8f/ftaxww2youmdS8XER31RC.png" alt="Reconstruction" width="1024"/> </div> <div style="display: flex; justify-content: center;">

<b>Gravity Wave</b> <img src="https://huggingface.co/Prithvi-WxC/Gravity_wave_Parameterization/resolve/20f8a120752b4e48364a2a606d6d2db26b2aa8b9/prithvi_downstream_gwflux_animation.gif" alt="Gravity Wave" width="512"/>

<b>Hurricane Ida - Zero-Shot Rollout</b> <img src="https://huggingface.co/datasets/Prithvi-WxC/Hurricane/resolve/6edc7c6838d59c1694755508917dce7a203fb9e8/2021C4Ida_2021082700_ground_truth_prediction.gif" alt="Hurricane Ida" width="475"/>

</div>

Citation

If you use this work, consider citing our paper

@misc{schmude2024prithviwxcfoundationmodel,
      title={Prithvi WxC: Foundation Model for Weather and Climate}, 
      author={Johannes Schmude and Sujit Roy and Will Trojak and Johannes Jakubik and Daniel Salles Civitarese and Shraddha Singh and Julian Kuehnert and Kumar Ankur and Aman Gupta and Christopher E Phillips and Romeo Kienzler and Daniela Szwarcman and Vishal Gaur and Rajat Shinde and Rohit Lal and Arlindo Da Silva and Jorge Luis Guevara Diaz and Anne Jones and Simon Pfreundschuh and Amy Lin and Aditi Sheshadri and Udaysankar Nair and Valentine Anantharaj and Hendrik Hamann and Campbell Watson and Manil Maskey and Tsengdar J Lee and Juan Bernabe Moreno and Rahul Ramachandran},
      year={2024},
      eprint={2409.13598},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2409.13598}, 
}