The original data we are provided with is arranged in a very specific fashion. There are two main sources of data: (i) satellite imagery and (ii) best track data, which we briefly describe below. Details on the data sources can be found in the official Digital Typhoon website.
Satellite images are stored individually as HDF5 files with the following naming convention:
An example of this is
1998091906-199808-GMS5-1.h5, which is an image
taken the 19th of September 1998 at 6AM with GMS5 satellite and belonging to
In addition, they are arranged in folders according to the typhoon sequence they belong to. Below we illustrate this hierarchy.
path/to/image/data |- 197801/ | |- image1.h5 |- image2.h5 : |- 197901/ : |- 197902/ | : :
Consecutive images in a sequence are usually 1 hour apart, but some exceptions may arise. The images are of shape 512x512 and the pixel value encodes the temperature (K).
Best Track data¶
Best Track data contains different parameters from typhoon sequences. A direct match between this data and satellite images can be established provided that sample dates coincide. Some of the parameters included in this dataset are the wind speed, centre pressure, latitude and longitude (full list can be found here).
Note: This data is obtained every 6h, hence there is an observation frequency miss-match with satellite imagery data. Therefore, spline interpolation has been used  to align this data with the image counterparts. As a consequence, only 16.6% of Best Track data is actual real data.
The data comes in .TSV format per typhoon sequence with name convention
path/to/best/data |- 197801.tsv |- 197901.tsv |- 197902.tsv :