Data, checkpoints, and analysis code for Tropical Cyclone Attributes in Sparse Autoencoder Features of an AI Weather Model
Authors/Creators
- 1. State Key Laboratory of Hydro-science and Engineering, Department of Hydraulic Engineering, Tsinghua University, Beijing 100084, China
- 2. Department of Risk and Disaster Reduction, University College London, London WC1E 6BT, UK
Description
SAE-cyclone provides the data and code supporting Tropical Cyclone Attributes in Sparse Autoencoder Features of an AI Weather Model. The archive contains analysis and training code, 80 trained sparse-autoencoder checkpoints, and processed inputs for the four main-text figures, nine Supporting Information figures, and Tables S1–S3.
A simplified demonstration notebook is available on GitHub: SAE-cyclone.
The README documents the file structure, data sources, and steps for reproducing the figures and tables. Original project data and SAE checkpoints are licensed under CC BY 4.0, and software under Apache 2.0. Third-party data and model resources retain their applicable upstream terms.
Files
| File(s) | Contents |
|---|---|
README.md | Archive overview, reproduction commands, data sources, and scope. |
sae-cyclone-code.zip | SAE training and cyclone-analysis code, with CPU scripts for reproducing the figures and tables from the processed data. |
sae-cyclone-processed-data.zip | Processed numerical inputs for Figures 1–4, Figures S1–S9, and Tables S1–S3. |
sae-checkpoints-seed-50108.zip to sae-checkpoints-seed-50112.zip | Five archives, one per training seed. Each contains inference weights and scientific metadata for all 16 processor layers: 80 SAE checkpoints in total. Optimizer states are not included. |
checkpoint-index.csv | Maps each checkpoint to its training seed, processor layer, archive member, producer code version, and figure use. |
CHECKPOINTS.md | Checkpoint contents, model configuration, tensor layout, and training and validation periods. |
data-sources.json | ERA5, IBTrACS, and GraphCast source references, including historical provenance limitations. |
LICENSE-CODE and LICENSE-DATA | Apache 2.0 for original project software; CC BY 4.0 for original project data and SAE checkpoints. Third-party materials retain their applicable upstream terms. |
Files
checkpoint-index.csv
Files
(1.3 GB)
| Name | Size | |
|---|---|---|
|
md5:b3cb8ac6c4a4c9da3837a77b7a8c5891
|
12.6 kB | Preview Download |
|
md5:06d4615ee027c0e35c748324aef8103d
|
2.1 kB | Preview Download |
|
md5:b7bb7072b59e61da42de164d1bed3bac
|
4.0 kB | Preview Download |
|
md5:3b83ef96387f14655fc854ddc3c6bd57
|
11.4 kB | Download |
|
md5:70cad6ddeef3e452f02aa4a5d0346d4c
|
20.4 kB | Download |
|
md5:ebfd38768cb326079742f6a58e7c506e
|
4.2 kB | Preview Download |
|
md5:00afb3f54a535de73d7dd308620192ee
|
250.3 MB | Preview Download |
|
md5:de811e1a19d7c4a77489e66d7dfaeee3
|
250.3 MB | Preview Download |
|
md5:3aa32c3b7bdde482a07a47618c519f9b
|
250.3 MB | Preview Download |
|
md5:470b03a256d74dc8a04ea0fb72964b47
|
250.3 MB | Preview Download |
|
md5:b6b5859e5613f0fd9230ea4c6c4a84bc
|
250.3 MB | Preview Download |
|
md5:2f97e91aa0d1874a330572547650f6b9
|
836.8 kB | Preview Download |
|
md5:60fd0c9230191ab60590f7d323bbe43d
|
42.1 MB | Preview Download |
Additional details
Related works
- Is derived from
- 10.25921/82ty-9e16 (DOI)
- References
- https://storage.googleapis.com/weatherbench2/datasets/era5/1959-2022-full_37-6h-0p25deg_derived.zarr (URL)
- https://github.com/google-deepmind/weathernext (URL)
- 10.1002/qj.3803 (DOI)
- 10.1029/2023MS004019 (DOI)