ClimODE

Models precise weather evolution with value-conserving dynamics, learning global weather transport as a neural flow, which also enables estimating the uncertainty in predictions.
https://github.com/aalto-quml/climode

Category: Climate Change
Sub Category: Earth and Climate Modeling

Last synced: about 22 hours ago
JSON representation

Repository metadata

ClimODE: Climate and Weather Forecasting With Physics-informed Neural ODEs

README.md

ClimODE: Climate and Weather Forecasting With Physics-informed Neural ODEs

Yogesh verma | Markus Heinonen | Vikas Garg

The code repository for the paper ClimODE: Climate and Weather Forecasting With Physics-informed Neural ODEs. More information can be found on the project website.

Citation

If you find this repository useful in your research, please consider citing the following paper:

@inproceedings{
verma2024climode,
title={Clim{ODE}: Climate and Weather Forecasting with Physics-informed Neural {ODE}s},
author={Yogesh Verma and Markus Heinonen and Vikas Garg},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024},
url={https://openreview.net/forum?id=xuY33XhEGR}
}

Prerequisites

Data Preparation

First, download ERA5 data with 5.625deg from WeatherBench. The data directory should look like the following

era5_data
   |-- 10m_u_component_of_wind
   |-- 10m_v_component_of_wind
   |-- 2m_temperature
   |-- constants
   |-- geopotential_500
   |-- temperature_850

Training ERA5

Global Forecast

To train ClimODE for global forecast use,

python train_global.py --scale 0 --batch_size 8 --spectral 0 --solver "euler" 

Global Monthly Forecast

To train ClimODE for global monthly forecast use,

python train_monthly.py --scale 0 --batch_size 4 --spectral 0 --solver "euler" 

Regional Forecast

To train ClimODE for regional forecasts among various regions of earth use,

python train_region.py --scale 0 --batch_size 8 --spectral 0 --solver "euler" --region 'NorthAmerica/SouthAmerica/Australia'

Evaluation ERA5

Global Forecast

To evaluate ClimODE for global forecast on Lat. weighted RMSE and ACC use, (Make sure to change the model path in the file)

python evaluation_global.py --spectral 0 --scale 0 --batch_size 8 

Global Monthly Forecast

To evaluate ClimODE for global monthly forecast on Lat. weighted RMSE and ACC use, (Make sure to change the model path in the file)

python evaluation_monthly.py --spectral 0 --scale 0 --batch_size 4 

Regional Forecast

To evaluate ClimODE for regional forecast on Lat. weighted RMSE and ACC use, (Make sure to change the model path in the file)

python evaluation_region.py --spectral 0 --scale 0 --region 'NorthAmerica/SouthAmerica/Australia' --batch_size 8 

Training on a different custom dataset

To train on a custom dataset, you need to follow the below guidelines

  • Data Loading: You might want to change the data loading scheme depending on your data (e.g. seasonal, daily, etc., and with many different input channels), which can be found in utils.py in the data-loading function.
  • Fitting initial velocity: Depending on the data, you need to estimate the initial velocity to train and test the model (For more details, see the manuscript) via the fit_velocity function.
  • Model Function: Depending on the input observable quantities, you might need to modify the number of input channels to model function in model_function.py.
  • Training and evaluation: Depending on your dataset, you might want to fine-tune and change the various hyper-parameters in training and evaluation files. Make sure to make them consistent in both of them. Also, we report CRPS scores for global hourly forecast only, if you want to compute them for every task please include the evaluation_crps_mm function.

Note: We are also constantly updating and revising the repo to make it more adaptable in a general way, and finidng bugs and removing them and modifying certain parts.


Owner metadata


GitHub Events

Total
Last Year

Committers metadata

Last synced: 9 days ago

Total Commits: 59
Total Committers: 1
Avg Commits per committer: 59.0
Development Distribution Score (DDS): 0.0

Commits in past year: 5
Committers in past year: 1
Avg Commits per committer in past year: 5.0
Development Distribution Score (DDS) in past year: 0.0

Name Email Commits
Yogesh Verma y****2@g****m 59

Committer domains:


Issue and Pull Request metadata

Last synced: 2 days ago

Total issues: 10
Total pull requests: 2
Average time to close issues: about 2 months
Average time to close pull requests: N/A
Total issue authors: 8
Total pull request authors: 1
Average comments per issue: 1.8
Average comments per pull request: 0.0
Merged pull request: 0
Bot issues: 0
Bot pull requests: 0

Past year issues: 9
Past year pull requests: 2
Past year average time to close issues: about 2 months
Past year average time to close pull requests: N/A
Past year issue authors: 7
Past year pull request authors: 1
Past year average comments per issue: 1.78
Past year average comments per pull request: 0.0
Past year merged pull request: 0
Past year bot issues: 0
Past year bot pull requests: 0

More stats: https://issues.ecosyste.ms/repositories/lookup?url=https://github.com/aalto-quml/climode

Top Issue Authors

  • tung-nd (2)
  • JimengShi (2)
  • plyu3 (1)
  • Kongwp (1)
  • Hank0626 (1)
  • Yumenomae (1)
  • marvingabler (1)
  • PiComLab (1)

Top Pull Request Authors

  • einzigsue (2)

Top Issue Labels

Top Pull Request Labels

Score: 4.584967478670572