Data

Tools

Indexes

Applications

Experiments

Open Source Science

orbit-2

ORIBT-2: Scaling Exascale Vision Transformer for Weather and Climate Downscaling

https://github.com/xiaowang-github/orbit-2

Science Score: 44.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • ✓
    CITATION.cff file
    Found CITATION.cff file
  • ✓
    codemeta.json file
    Found codemeta.json file
  • ✓
    .zenodo.json file
    Found .zenodo.json file
  • ○
    DOI references
  • ○
    Academic publication links
  • ○
    Academic email domains
  • ○
    Institutional organization owner
  • ○
    JOSS paper metadata
  • ○
    Scientific vocabulary similarity
    Low similarity (8.7%) to scientific vocabulary
Last synced: 6 months ago · JSON representation ·

Repository

ORIBT-2: Scaling Exascale Vision Transformer for Weather and Climate Downscaling

Basic Info
  • Host: GitHub
  • Owner: XiaoWang-Github
  • License: mit
  • Language: Jupyter Notebook
  • Default Branch: main
  • Homepage:
  • Size: 0 Bytes
Statistics
  • Stars: 0
  • Watchers: 1
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Created 10 months ago · Last pushed 10 months ago
Metadata Files
Readme Contributing License Citation

README.md

ORBIT-2 Downscaling

Installation and Run Guide

(1) Create and activate your conda environment

(2) Do pip install -r requirements.txt to install related packages in the conda environment

(3) Then do pip install -e . in the parent directory to install the package to the conda environment

(4) Go to examples folder. launch_intermediate.sh is the launch script to run the downscaling code for 9.5 million, 126 million, 1 billion and 10 billion parameters.

(5) To visualize the input, output, and ground truth, run launch_visualize.sh after training. Using only a single node with a single GPU. In visualize.py, do not forget to change the checkpoint path for the model checkpoint that you want to load.

YAML Files for Downscaling Configurations

Both the AI model hyperparameters and dataset configurations are configured in yaml files located in the config folder.

Available training losses include MSE, MAE, latitude weighted MSE, Pearson Score, Anomaly Coefficient. Most recently, hybrid perceptual loss, and bayesian estimation loss with total variation prior. Training losses can be changed in the yaml files.

Available Downscaling Data All the datasets are publicly available.

For ERA5, you can download from https://cds.climate.copernicus.eu/datasets/reanalysis-era5-single-levels?tab=overview

For PRISM, download from https://prism.oregonstate.edu/

For DAYMET, download from https://daymet.ornl.gov/

For your convenience, if you have access to Frontier supercomputer, you can access to the following downloaded datasets for training, validation and testing:

ERA5 5.6 degree "/lustre/orion/lrn036/world-shared/data/superres/era5/5.625deg/"
ERA5 1.4 degree "/lustre/orion/lrn036/world-shared/data/superres/era5/1.40625deg/"
ERA5 1.0 degree "/lustre/orion/lrn036/world-shared/data/superres/era5/1.0deg/"
ERA5 0.25 degree "/lustre/orion/lrn036/world-shared/data/superres/era5/0.25deg/"
PRISM 16 km "/lustre/orion/lrn036/world-shared/data/superres/prism/10.0arcmin"
PRISM 4 km "/lustre/orion/lrn036/world-shared/data/superres/prism/2.5arcmin"
DAYMET 16 km "/lustre/orion/lrn036/world-shared/data/superres/daymet/10.0arcmin"
DAYMET 4 km "/lustre/orion/lrn036/world-shared/data/superres/daymet/2.5arcmin"
DAYMET 3.5 km "/lustre/orion/lrn036/world-shared/data/superres/daymet/2.0arcmin"
DAYMET 800 m "/lustre/orion/lrn036/world-shared/data/superres/daymet/0.5arcmin"

Owner

  • Name: Xiao Wang
  • Login: XiaoWang-Github
  • Kind: user
  • Company: Oak Ridge National Lab

Research Staff at Oak Ridge National Lab

Citation (CITATION.cff) 

# This CITATION.cff file was generated with cffinit.
# Visit https://bit.ly/cffinit to generate yours today!

cff-version: 1.2.0
title: "ClimateLearn: Benchmarking Machine Learning for Weather and Climate Modeling"
message: >-
  If you use this software, please cite it using the
  metadata from this file.
type: software
authors:
  - given-names: Tung
    family-names: Nguyen
    email: tungnd@cs.ucla.edu
    affiliation: 'University of California, Los Angeles'
  - given-names: Jason
    family-names: Jewik
    email: jason.jewik@ucla.edu
    affiliation: 'University of California, Los Angeles'
  - given-names: Hritik
    family-names: Bansal
    email: hbansal@ucla.edu
    affiliation: 'University of California, Los Angeles'
  - given-names: Prakhar
    family-names: Sharma
    email: prakhar6sharma@gmail.com
    affiliation: 'University of California, Los Angeles'
  - given-names: Aditya
    family-names: Grover
    email: adityag@cs.ucla.edu
    affiliation: 'University of California, Los Angeles'
license: MIT
repository-code: "https://github.com/aditya-grover/climate-learn"

GitHub Events

Total
  • Watch event: 1
  • Member event: 1
  • Push event: 3
  • Pull request event: 1
  • Fork event: 1
  • Create event: 36
Last Year
  • Watch event: 1
  • Member event: 1
  • Push event: 3
  • Pull request event: 1
  • Fork event: 1
  • Create event: 36

Dependencies

requirements.txt pypi
pyproject.toml pypi
  • cdsapi >=0.5.1
  • dask >=2022.2.0
  • importlib-metadata ==4.13.0
  • matplotlib >=3.5.3
  • netcdf4 >=1.6.2
  • pytorch-lightning >=1.9.0
  • rasterio >=1.3.7
  • scikit-learn >=1.0.2
  • tensorboard ==2.11.2
  • timm ==0.9.2
  • wandb >=0.13.9
  • xarray >=0.20.2