Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,90 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
+
# ChaosBench
|
| 5 |
+
We propose ChaosBench, a large-scale, multi-channel, physics-based benchmark for subseasonal-to-seasonal (S2S) climate prediction.
|
| 6 |
+
It is framed as a high-dimensional video regression task that consists of 45-year, 60-channel observations
|
| 7 |
+
for validating physics-based and data-driven models, and training the latter.
|
| 8 |
+
Physics-based forecasts are generated from 4 national weather agencies with 44-day lead-time and serve as baselines to data-driven forecasts.
|
| 9 |
+
Our benchmark is one of the first to incorporate physics-based metrics to ensure physically-consistent and explainable models.
|
| 10 |
+
We establish two tasks: full and sparse dynamics prediction.
|
| 11 |
+
|
| 12 |
+
🔗: [https://github.com/leap-stc/](https://github.com/leap-stc/)
|
| 13 |
+
|
| 14 |
+
📚: [https://arxiv.org/](https://arxiv.org/)
|
| 15 |
+
|
| 16 |
+
## Table of Content
|
| 17 |
+
- [Getting Started](#getting-started)
|
| 18 |
+
- [Dataset Overview](#overview)
|
| 19 |
+
- [Evaluation Metrics](#metrics)
|
| 20 |
+
- [Leaderboard](#leaderboard)
|
| 21 |
+
|
| 22 |
+
## Getting Started
|
| 23 |
+
1. Set up the project...
|
| 24 |
+
- Clone the `ChaosBench` Github repository
|
| 25 |
+
- Create local directory to store your data, e.g., `data/`
|
| 26 |
+
- Navigate to `chaosbench/config.py` and change the field `DATA_DIR = <YOUR_WORKING_DATA_DIR>`
|
| 27 |
+
|
| 28 |
+
2. Download the dataset...
|
| 29 |
+
- First we perform initialization:
|
| 30 |
+
```
|
| 31 |
+
cd data/
|
| 32 |
+
wget https://huggingface.co/datasets/juannat7/ChaosBench/blob/main/process.sh
|
| 33 |
+
chmod +x process.sh
|
| 34 |
+
```
|
| 35 |
+
- And finally: (__NOTE__: you can also run each line _one at a time_ to retrieve each dataset)
|
| 36 |
+
```
|
| 37 |
+
./process.sh era5 # For input ERA5 data
|
| 38 |
+
./process.sh ukmo # For simulation from UKMO
|
| 39 |
+
./process.sh ncep # For simulation from NCEP
|
| 40 |
+
./process.sh cma # For simulation from CMA
|
| 41 |
+
./process.sh ecmwf # For simulation from ECMWF
|
| 42 |
+
```
|
| 43 |
+
|
| 44 |
+
## Dataset Overview
|
| 45 |
+
|
| 46 |
+
- __Input:__ ERA5 Reanalysis (1979-2023)
|
| 47 |
+
|
| 48 |
+
- __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:
|
| 49 |
+
|
| 50 |
+
Parameters/Levels (hPa) | 1000 | 925 | 850 | 700 | 500 | 300 | 200 | 100 | 50 | 10
|
| 51 |
+
:---------------------- | :----| :---| :---| :---| :---| :---| :---| :---| :--| :-|
|
| 52 |
+
Geopotential height, z ($gpm$) | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
|
| 53 |
+
Specific humidity, q ($kg kg^{-1}$) | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | | | |
|
| 54 |
+
Temperature, t ($K$) | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
|
| 55 |
+
U component of wind, u ($ms^{-1}$) | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
|
| 56 |
+
V component of wind, v ($ms^{-1}$) | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
|
| 57 |
+
Vertical velocity, w ($Pas^{-1}$) | | | | | ✓ | | | | | |
|
| 58 |
+
|
| 59 |
+
- __Baselines:__
|
| 60 |
+
- Physics-based models:
|
| 61 |
+
- [x] UKMO: UK Meteorological Office
|
| 62 |
+
- [x] NCEP: National Centers for Environmental Prediction
|
| 63 |
+
- [x] CMA: China Meteorological Administration
|
| 64 |
+
- [x] ECMWF: European Centre for Medium-Range Weather Forecasts
|
| 65 |
+
- Data-driven models:
|
| 66 |
+
- [x] Lagged-Autoencoder
|
| 67 |
+
- [x] Fourier Neural Operator (FNO)
|
| 68 |
+
- [x] ResNet
|
| 69 |
+
- [x] UNet
|
| 70 |
+
- [x] ViT/ClimaX
|
| 71 |
+
- [x] PanguWeather
|
| 72 |
+
- [x] Fourcastnetv2
|
| 73 |
+
|
| 74 |
+
## Evaluation Metrics
|
| 75 |
+
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.
|
| 76 |
+
|
| 77 |
+
- __Vision-based:__
|
| 78 |
+
- [x] RMSE
|
| 79 |
+
- [x] Bias
|
| 80 |
+
- [x] Anomaly Correlation Coefficient (ACC)
|
| 81 |
+
- [x] Multiscale Structural Similarity Index (MS-SSIM)
|
| 82 |
+
- __Physics-based:__
|
| 83 |
+
- [x] Spectral Divergence (SpecDiv)
|
| 84 |
+
- [x] Spectral Residual (SpecRes)
|
| 85 |
+
|
| 86 |
+
|
| 87 |
+
## Leaderboard
|
| 88 |
+
You can access the full score and checkpoints in `logs/<MODEL_NAME>` within the following subdirectory:
|
| 89 |
+
- Scores: `eval/<METRIC>.csv`
|
| 90 |
+
- Model checkpoints: `lightning_logs/`
|