ibm-nasa-geospatial/Prithvi-WxC-1.0-2300M-rollout

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:

• 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.
• (This model) 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.

Zero-Shot Rollout

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}, 
}