Datasets and Codes for MMamba

This database contains the MWSID dataset we generated and the core codes of MMamba.

The MWSID dataset includes:
- Train input set: x1.npy
- Test input set: x2.npy
- Train label set: y1.npy
- Test label set: y2.npy

- Train info: info1.npy
- Test info: info2.npy

For details on each sub-dataset (data channels, dimensions, etc.), refer to Data introduction.txt.

The core code is in codes.zip. After extraction:
- In the get_dataset folder, run get_dataloader.py to access MWSID sub-datasets. Other codes in this folder are for generating the sub-datasets.
- The models folder has core code for comparison models.
- In the mmamba folder, mmamba.py is our model's core code, params_flops.py calculates model complexity, get_MI.py computes mutual information, and the rest are support files.

Note: The database includes a table Dates of MWSID.png, showing the 230-day sampling period for MWSID. These days were chosen as there was at least one complete hurricane observed globally each day.

MWSID integrates five observational datasets and a forecast dataset:
- SMAP: https://data.remss.com/smap/wind/L3/v01.0/daily/FINAL/
- AMSR2: https://data.remss.com/amsr2/ocean/L3/v08.2/daily/
- ASCAT-A: https://data.remss.com/ascat/metopa/bmaps_v02.1/
- ASCAT-B: https://data.remss.com/ascat/metopb/bmaps_v02.1/
- ASCAT-C: https://data.remss.com/ascat/metopc/bmaps_v02.1/

- NCEP GFS: https://rda.ucar.edu/datasets/d084001/  (DOI: 10.5065/D65D8PWK)