Changelog

All notable changes to the dataset will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

`v1.1.0` - 04.12.2024

Overall, the structure of the input data now looks like:

from datasets import load_dataset

dataset = load_dataset('AGBD.py',trust_remote_code=True,streaming=True)
for sample in dataset['train']:
  _in = sample['input']
  break

np.array(_in).shape # (24, 15, 15)

Where the shape follows the convention below:

- 12 x Sentinel-2 bands
- 3 x Sentinel-2 dates (s2_num_days, s2_doy_cos, s2_doy_sin)
- 4 x Latitude/longitude (lat_cos, lat_sin, lon_cos, lon_sin)
- 3 x GEDI dates (gedi_num_days, gedi_doy_cos, gedi_doy_sin)
- 2 x ALOS bands (HH, HV)
- 2 x CH bands (ch, std)
- (N + 1) x LC (lc_encoded, lc_prob), where N = 14 for onehot, N = 5 for cat2vec, and N = 2 for sin/cos
- 3 x Topographic bands (slope, aspect_cos, aspect_sin)
- 1 x DEM (dem)

Added

New features:
- slope : the percentage of the slope where each pixel is located; it is a float between $0$ and $1$
- aspect_cos, aspect_sin : the sine and cosine encoded aspect, or orientation of the slope, which is measured clockwise in degrees from $0$ to $360$; it is a float between $0$ and $1$
- s2_num_days : the date of acquisition of the Sentinel-2 product, calculated as the number of days since the start of the GEDI mission (April 17th, 2019); it is an int
- s2_doy_cos, s2_doy_sin : the sine and cosine encoded corresponding day of the year (DOY); it is a float between $0$ and $1$
- gedi_num_days : the date of acquisition of the GEDI footprint, calculated as the number of days since the start of the GEDI mission (April 17th, 2019); it is a uint
- gedi_doy_cos, gedi_doy_sin : the sine and cosine encoded corresponding day of the year (DOY); it is a float between $0$ and $1$
The load_dataset() function now provides the following configuration options:
- norm_strat (default = 'pct') : the strategy to apply to process the input features. Valid options are: 'pct', which applies min-max scaling with the $1$-st and $99$-th percentile of the data; and 'mean_std' which applies mean/variance standardization`.
- encode_strat (default = 'onehot') : the encoding strategy to apply to the land classification (LC) data. Valid options are: 'onehot', one-hot encoding; 'sin_cos', sine-cosine encoding; 'cat2vec', cat2vec transformation based on embeddings pre-computed on the train set.
- input_features (dict) : the input features to be included in the data, the default values being:
```
{'S2_bands': ['B01', 'B02', 'B03', 'B04', 'B05', 'B06', 'B07', 'B08', 'B8A', 'B09','B11', 'B12'], 
 'S2_dates' : False, 'lat_lon': True, 'GEDI_dates': False, 'ALOS': True, 'CH': True, 'LC': True, 
 'DEM': True, 'topo': False}
```
- additional_features (list, default = []) : the metadata to include in the data. Possible values are:
```
['s2_num_days', 'gedi_num_days', 'lat', 'lon', 'agbd_se', 'elev_lowes', 'leaf_off_f', 'pft_class', 'region_cla', 'rh98', 'sensitivity', 'solar_elev', 'urban_prop']
```
  This metadata can later be accessed as such:
```
from datasets import load_dataset

dataset = load_dataset('AGBD.py',trust_remote_code=True,streaming=True)
for sample in dataset['train']:
  lat = sample['lat']
  break
```

Fixed

Statistics: there was in bug in the computation of the percentiles, and the statistics were fixed accordingly.
Latitude and longitude: while the latitude and longitude in the raw .h5 files are absolutely correct, there was a bug in the generation of the .parquet files for HuggingFace, resulting in offsetted lat and lon in the additional features. We've now corrected this.

Removed

The load_dataset() function does not provide the normalize_data configuration anymore, as it was replaced by the norm_strat configuration, giving users more flexibility regarding the processing of the input feature.

Changelog

v1.1.0 - 04.12.2024

Added

Fixed

Removed

`v1.1.0` - 04.12.2024