ORBIT-2 Dataset for Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling

Lu, Dan; Wang, Xiao; Tsaris, Aristeidis; Choi, Jong Youl; Ashfaq, Moetasim

doi:10.13139/OLCF/2589526

Title: ORBIT-2 Dataset for Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling

Dataset
Other Related Research

Abstract

This dataset release corresponds to the work conducted in ORBIT-2: Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling, where large-scale AI methods were applied to improve climate and weather resolution. The collection integrates four widely used, publicly available datasets: ERA5, PRISM, DAYMET, and IMERG. To prepare the data for ORBIT-2 model training and evaluation, we applied a preprocessing pipeline that generates paired low-resolution and high-resolution samples, enabling supervised downscaling experiments. The transformation from coarse to fine scales was performed using bilinear regridding, consistent with the procedures described in WeatherBench2, a community benchmark for weather and climate AI models. This dataset supports the development and evaluation of foundation models designed for weather and climate downscaling at exascale. Additional details on methodology and applications can be found in Wang et al., ORBIT-2 (arXiv:2505.04802, 2025).

Authors:

Lu, Dan; Wang, Xiao; Tsaris, Aristeidis; Choi, Jong Youl; Ashfaq, Moetasim

Oak Ridge National Laboratory

Publication Date:: Fri Oct 10 04:00:00 UTC 2025

DOE Contract Number:: AC05-00OR22725

Research Org.:: Oak Ridge National Laboratory (ORNL), Oak Ridge, TN (United States)

Sponsoring Org.:: Office of Science (SC)

Subject:: 54 ENVIRONMENTAL SCIENCES; 58 GEOSCIENCES; 97 MATHEMATICS AND COMPUTING

OSTI Identifier:: 2589526

DOI:: https://doi.org/10.13139/OLCF/2589526

Citation Formats


                    Lu, Dan, Wang, Xiao, Tsaris, Aristeidis, Choi, Jong Youl, and Ashfaq, Moetasim. ORBIT-2 Dataset for Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling.  United States: N. p., 2025. 
        Web.  doi:10.13139/OLCF/2589526.

Copy to clipboard


                    Lu, Dan, Wang, Xiao, Tsaris, Aristeidis, Choi, Jong Youl, & Ashfaq, Moetasim. ORBIT-2 Dataset for Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling.  United States.  doi:https://doi.org/10.13139/OLCF/2589526

Copy to clipboard


                    Lu, Dan, Wang, Xiao, Tsaris, Aristeidis, Choi, Jong Youl, and Ashfaq, Moetasim. 2025.  
"ORBIT-2 Dataset for Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling".  United States.  doi:https://doi.org/10.13139/OLCF/2589526.  https://www.osti.gov/servlets/purl/2589526. Pub date:Fri Oct 10 04:00:00 UTC 2025

Copy to clipboard


                    
@article{osti_2589526,

  title        = {ORBIT-2 Dataset for Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling},

  author       = {Lu, Dan and Wang, Xiao and Tsaris, Aristeidis and Choi, Jong Youl and Ashfaq, Moetasim},

  abstractNote = {This dataset release corresponds to the work conducted in ORBIT-2: Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling, where large-scale AI methods were applied to improve climate and weather resolution. The collection integrates four widely used, publicly available datasets: ERA5, PRISM, DAYMET, and IMERG. To prepare the data for ORBIT-2 model training and evaluation, we applied a preprocessing pipeline that generates paired low-resolution and high-resolution samples, enabling supervised downscaling experiments. The transformation from coarse to fine scales was performed using bilinear regridding, consistent with the procedures described in WeatherBench2, a community benchmark for weather and climate AI models. This dataset supports the development and evaluation of foundation models designed for weather and climate downscaling at exascale. Additional details on methodology and applications can be found in Wang et al., ORBIT-2 (arXiv:2505.04802, 2025).},

  doi          = {10.13139/OLCF/2589526},

  journal      = {},

  number       = ,

  volume       = ,

  place        = {United States},

  year         = {Fri Oct 10 04:00:00 UTC 2025},

  month        = {Fri Oct 10 04:00:00 UTC 2025}

}

Copy to clipboard

Dataset:

View Dataset

DOI: https://doi.org/10.13139/OLCF/2589526

Save / Share:

Export Metadata

Save to My Library

Similar records in DOE Data Explorer and OSTI.GOV collections:

ORBIT-2 Dataset for Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling

Dataset Lu, Dan ; Wang, Xiao ; Choi, Jong Youl ; ...

We release the dataset used in ORBIT-2, an exascale vision foundation model for weather and climate downscaling. The dataset was constructed by preprocessing raw data into paired low-resolution and high-resolution samples for model training and evaluation. Downscaling was performed using bilinear regridding following the WeatherBench2 methodology (https://weatherbench2.readthedocs.io). The dataset integrates four publicly available, open-source sources: ERA5, PRISM, DAYMET, and IMERG, obtained respectively from: • ERA5: https://weatherbench2.readthedocs.io • PRISM: https://prism.oregonstate.edu • DAYMET: https://daymet.ornl.gov • IMERG: https://gpm.nasa.gov/data/imerg Further details on the dataset construction and usage are provided in: Wang, Xiao, et al. “ORBIT-2: Scaling Exascale Vision Foundation Models for Weather and Climatemore »« less
ORBIT-2: Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling

Conference Wang, Xiao ; Choi, Jong Youl ; Kurihaya, Takuya ; ...

Sparse observations and coarse-resolution climate models limit effective regional decision-making, underscoring the need for robust downscaling. However, existing AI methods struggle with generalization across variables and geographies and are constrained by the quadratic complexity of Vision Transformer (ViT) self-attention. We introduce ORBIT-2, a scalable foundation model for global, hyper-resolution climate downscaling. ORBIT-2 incorporates two key innovations: (1) Residual Slim ViT (Reslim), a lightweight architecture with residual learning and Bayesian regularization for efficient, robust prediction; and (2) TILES, a tile-wise sequence scaling algorithm that reduces self-attention complexity from quadratic to linear, enabling long-sequence processing and massive parallelism. ORBIT-2 scales to 10more »« less
Predicting sedimentary bedrock subsurface weathering fronts and weathering rates: Dataset

Dataset Wan, Jiamin ; Tokunaga, Tetsu K. ; Williams, Kenneth H. ; ...

Although bedrock weathering strongly influences water quality and global carbon and nitrogen budgets, the weathering depths and rates within subsurface are not well understood nor predictable. Determination of both porewater chemistry and subsurface water flow are needed in order to develop more complete understanding and obtain weathering rates. In a long-term field study, we applied a multiphase approach along a mountainous watershed hillslope transect underlain by marine shale. Here we report three findings. First, the deepest extent of the water table determines the weathering front, and the range of annually water table oscillations determines the thickness of the weathering zone.more »« less
Dataset for Evaluation of Extreme Weather Impacts on Utility-Scale Photovoltaic Plant Performance in the United States

Dataset Gunda, Thushara ; Jackson, Nicole

This dataset is a fusion of three data types (operations and maintenance tickets, weather data, and production data) that was used to support machine learning analysis and evaluation of drivers for low performance at photovoltaic (PV) sites during compound, extreme weather events. After being processed with machine learning, the data was used in the "Evaluation of Extreme Weather Impacts on Utility-scale Photovoltaic Plant Performance in the United States" manuscript. Additional details are captured in the associated manuscript.
Foundational Dataset for Developing Large-Sample Stream Temperature Models in the Conterminous United States

Dataset Gomez-Velez, Jesus ; Siddik, Md Abu Bakar ; Turner, Sean ; ...

This dataset provides inputs, evaluation results, and trained weights from a large-sample Long Short-Term Memory (LSTM) model designed to predict daily stream temperatures across unregulated river reaches in the conterminous United States (CONUS). It includes dynamic meteorological and hydrologic forcings, static physiographic attributes, and model outputs from cross-validation experiments spanning 300 basins. It supports reproducible modeling, direct application for new basins, and provides data suitable for integration with reservoir and river simulations under current and future climates. It contains two .zip files described below · RQ-AI_runs.zip: Model outputs from 10-fold cross-validation experiments, including observed and predicted daily stream temperatures, alongmore »« less

Similar Records