# Install the datasets library if you haven't already
!pip install datasets

# Import necessary modules
from datasets import load_dataset
from helper import load_embeddings, process_batch

# User-defined configurations
# 1. Select the features you would like to load (see Dataset Overview for more details)
features = {'S2_bands': ['B01', 'B02', 'B03', 'B04', 'B05', 'B06', 'B07', 'B08', 'B8A', 'B09','B11', 'B12'],
                    'S2_dates' : True, 
                    'lat_lon': True,
                    'GEDI_dates': False, 
                    'ALOS': True,
                    'CH': True,
                    'LC': True,
                    'DEM': True,
                    'topo': True
                   }
# 2. Select the metadata you would like to include
metadata = []
# 3. Select the rest of the configurations
user_config = {'norm_strat': 'pct', # normalization strategy
                'encode_strat': 'cat2vec', # biome encoding strategy
                'input_features': features, # input features
                'metadata': metadata, # metadata features
                'patch_size': 25 # size of the images
               }

# This line will load the embeddings if required
user_config = load_embeddings(user_config)

# Load the dataset, e.g. the train split
ds = load_dataset("prs-eth/AGBD_raw", streaming = True) # load from the raw Parquet files
ds_train = ds['train'].map(process_batch, # this will apply the data processing steps  
                          fn_kwargs = user_config,
                          batched = True, # whether to batch the data processing steps
                          batch_size = 32) # corresponding batch size

# Then, you can iterate over the samples as previously
for batch in ds_train.shuffle(seed = 42, buffer_size = 1024).iter(batch_size = 32):
    features, labels, metadata = batch['input'], batch['label'], batch['metadata']

This code will load the dataset as an IterableDataset. You can find more information on how to work with IterableDataset objects in the Hugging Face documentation.

Note that if you use streaming=True, the iteration process will be pretty slow. However, if you set it to False, it will download the 400GB of data locally.

📊 Dataset Overview

Each sample in the dataset contains a pre-cropped images along with their corresponding biomass labels, and optionally the corresponding metadata. For additional resources, including links to the preprocessed uncropped data, please visit the project page on GitHub.

⚙️ Configurations

We provide the following configuration options:

norm_strat (str) : The strategy to apply to process the input features. Valid options are: 'pct', which applies min-max scaling with the 1st and 99th percentiles of the data; 'mean_std' which applies Z-score normalization; and 'none', which returns the un-processed data.
encode_strat (str) : 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; and 'none', raw LC data.
features (dict) The features to be included in the data.
- S2_bands: Sentinel-2 bands. You can select any combination of the following bands: ['B01', 'B02', 'B03', 'B04', 'B05', 'B06', 'B07', 'B08', 'B8A', 'B09','B11', 'B12'], or none of them [].
- S2_dates: True or False
- lat_lon: True or False
- GEDI_dates: True or False
- ALOS: True or False
- CH: True or False
- LC: True or False
- DEM: True or False
- topo: True or False
metadata (list) A list of additional features the dataset should include. Refer to the documentation below for a description of the variables. 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']
patch_size (int) The size of the returned patch (in pixels). The maximum value is 25 pixels, which corresponds to 250 meters.

🖼️ Features Details

Each sample consists of a varying number of channels, based on the input_features and encode_strat options passed to the load_dataset() function. The channels are organized as follows:

Feature	Channels	Description
Sentinel-2 bands	`B01, B02, B03, B04, B05, B06, B07, B08, B8A, B09, B11, B12`	Sentinel-2 bands, in Surface Reflectance values
Sentinel-2 dates	`s2_num_days, s2_doy_cos, s2_doy_sin`	Date of acquisition of the S2 image (in number of days wrt the beginning of the GEDI mission); sine-cosine encoding of the day of year (DOY).
Geographical coordinates	`lat_cos, lat_sin, lon_cos, lon_sin`	Sine-cosine encoding of the latitude and longitude.
GEDI dates	`gedi_num_days, gedi_doy_cos, gedi_doy_sin`	Date of acquisition of the GEDI footprint (in number of days wrt the beginning of the GEDI mission); sine-cosine encoding of the day of year (DOY).
ALOS PALSAR-2 bands	`HH,HV`	ALOS PALSAR-2 bands, gamma-naught values in dB.
Canopy Height	`ch, ch_std`	Canopy height from Lang et al. and associated standard deviation.
Land Cover Information	`lc_encoding*, lc_prob`	Encoding of the land class, and classification probability (as a percentage between 0 and 1).
Topography	`slope, aspect_cos, aspect_sin`	Slope (percentage between 0 and 1); sine-cosine encoded aspect of the slope.
Digital Elevation Model (DEM)	`dem`	Elevation (in meters).

Regarding lc_encoding*, the number of channels follows this convention:

sin_cos : 2 channels
cat2vec : 5 channels
onehot : 14 channels
none : 1 channel

In summary, the channels are structured as follows:

(Sentinel-2 bands) | (Sentinel-2 dates) | (Geographical coordinates) | (GEDI dates) | (ALOS PALSAR-2 bands) | (Canopy Height) | (Land Cover Information) | (Topography) | DEM

➕ Additional Features

You can include a list of additional features from the options below in your dataset configuration:

"agbd_se" - AGBD Standard Error: The uncertainty estimate associated with the aboveground biomass density prediction for each GEDI footprint.
"elev_lowes" - Elevation: The height above sea level at the location of the GEDI footprint.
"leaf_off_f" - Leaf-Off Flag: Indicates whether the measurement was taken during the leaf-off season, which can impact canopy structure data.
"pft_class" - Plant Functional Type (PFT) Class: Categorization of the vegetation type (e.g., deciduous broadleaf, evergreen needleleaf).
"region_cla" - Region Class: The geographical area where the footprint is located (e.g., North America, South Asia).
"rh98" - RH98 (Relative Height at 98%): The height at which 98% of the returned laser energy is reflected, a key measure of canopy height.
"sensitivity" - Sensitivity: The proportion of laser pulse energy reflected back to the sensor, providing insight into vegetation density and structure.
"solar_elev" - Solar Elevation: The angle of the sun above the horizon at the time of measurement, which can affect data quality.
"urban_prop" - Urban Proportion: The percentage of the footprint area that is urbanized, helping to filter or adjust biomass estimates in mixed landscapes.
"gedi_num_days" - Date of GEDI Footprints: The specific date on which each GEDI footprint was captured, adding temporal context to the measurements.
"s2_num_days" - Date of Sentinel-2 Image: The specific date on which each Sentinel-2 image was captured, ensuring temporal alignment with GEDI data.
"lat" - Latitude: Latitude of the central pixel.
"lon" - Longitude: Longitude of the central pixel.

🤝 Citing

If you use AGBD in a scientific publication, we encourage you to add the following reference:

@article{Sialelli2025,
  title = {AGBD: A Global-scale Biomass Dataset},
  volume = {X-G-2025},
  ISSN = {2194-9050},
  url = {http://dx.doi.org/10.5194/isprs-annals-X-G-2025-829-2025},
  DOI = {10.5194/isprs-annals-x-g-2025-829-2025},
  journal = {ISPRS Annals of the Photogrammetry,  Remote Sensing and Spatial Information Sciences},
  publisher = {Copernicus GmbH},
  author = {Sialelli,  Ghjulia and Peters,  Torben and Wegner,  Jan D. and Schindler,  Konrad},
  year = {2025},
  month = jul,
  pages = {829–838}
}

Downloads last month: 226

Paper for prs-eth/AGBD

AGBD: A Global-scale Biomass Dataset

Paper • 2406.04928 • Published Jun 7, 2024