ChaosBench

We propose ChaosBench, a large-scale, multi-channel, physics-based benchmark for subseasonal-to-seasonal (S2S) climate prediction. It is framed as a high-dimensional video regression task that consists of 45-year, 60-channel observations for validating physics-based and data-driven models, and training the latter. Physics-based forecasts are generated from 4 national weather agencies with 44-day lead-time and serve as baselines to data-driven forecasts. Our benchmark is one of the first to incorporate physics-based metrics to ensure physically-consistent and explainable models. We establish two tasks: full and sparse dynamics prediction.

🔗: https://leap-stc.github.io/ChaosBench/

📚: https://arxiv.org/abs/2402.00712

Getting Started

Step 1: Clone the ChaosBench Github repository

Step 2: Install package dependencies

cd ChaosBench
pip install -r requirements.txt

Step 3: Initialize the data space by running

cd data/
wget https://huggingface.co/datasets/LEAP/ChaosBench/resolve/main/process.sh
chmod +x process.sh

Step 5: Download the data

# NOTE: you can also run each line one at a time to retrieve individual dataset

./process.sh era5            # Required: For input ERA5 data
./process.sh climatology     # Required: For climatology
./process.sh ukmo            # Optional: For simulation from UKMO
./process.sh ncep            # Optional: For simulation from NCEP
./process.sh cma             # Optional: For simulation from CMA
./process.sh ecmwf           # Optional: For simulation from ECMWF

Dataset Overview

• Input: ERA5 Reanalysis (1979-2023)

• Target: The following table indicates the 48 variables (channels) that are available for Physics-based models. Note that the Input ERA5 observations contains ALL fields, including the unchecked boxes:

  Parameters/Levels (hPa)	1000	925	850	700	500	300	200	100	50	10
  Geopotential height, z ($gpm$)	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
  Specific humidity, q ($kg kg^{-1}$)	✓	✓	✓	✓	✓	✓	✓
  Temperature, t ($K$)	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
  U component of wind, u ($ms^{-1}$)	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
  V component of wind, v ($ms^{-1}$)	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
  Vertical velocity, w ($Pas^{-1}$)					✓

• Baselines:

  • Physics-based models:
    • UKMO: UK Meteorological Office
    • NCEP: National Centers for Environmental Prediction
    • CMA: China Meteorological Administration
    • ECMWF: European Centre for Medium-Range Weather Forecasts
  • Data-driven models:
    • Lagged-Autoencoder
    • Fourier Neural Operator (FNO)
    • ResNet
    • UNet
    • ViT/ClimaX
    • PanguWeather
    • Fourcastnetv2
    • GraphCast

Evaluation Metrics

We divide our metrics into 2 classes: (1) ML-based, which cover evaluation used in conventional computer vision and forecasting tasks, (2) Physics-based, which are aimed to construct a more physically-faithful and explainable data-driven forecast.

• Vision-based:
  • RMSE
  • Bias
  • Anomaly Correlation Coefficient (ACC)
  • Multiscale Structural Similarity Index (MS-SSIM)
• Physics-based:
  • Spectral Divergence (SpecDiv)
  • Spectral Residual (SpecRes)

Leaderboard

You can access the full score and checkpoints in logs/<MODEL_NAME> within the following subdirectory:

• Scores: eval/<METRIC>.csv
• Model checkpoints: lightning_logs/