GeoBench BigEarthNetV2

Intro

BigEarthNet v2.0 Clasen et al. 2025 is a large-scale multi-modal satellite image dataset for multi-label land cover classification. The dataset contains Sentinel-1 SAR and Sentinel-2 optical imagery with multi-label annotations based on the CORINE Land Cover (CLC) labels. In comparison to v1.0, the v2.0 release provides improved hierarchical multi-label annotations, enhanced data quality, and geospatial splits. It can be used to evaluate multi-label classification and data fusion (SAR + Optical) data fusion schemes.

Dataset Characteristics

• Modalities:

  • Sentinel-1 SAR (VV, VH polarizations)

  • Sentinel-2 Optical (12 spectral bands)

• Spatial Resolution:

  • S1: 10m ground sample distance

  • S2: 10m, 20m, 60m (resampled to 10m)

• Temporal Resolution: Single acquisition per patch

• Spectral Bands:

  • S2: 12 bands (B01, B02, B03, B04, B05, B06, B07, B08, B8A, B09, B11, B12)

  • S1: 2 polarizations (VV, VH)

• Image Dimensions: 120x120 pixels (1.2km x 1.2km patches)

• Labels: 19 CORINE Land Cover classes (multi-label)

• Geographic Distribution: 10 European countries

• Temporal Coverage: June 2017 - May 2018

GeoBenchV2 Processing Pipeline

Preprocessing Steps

1. Split Generation:

   • Used existing train/validation/test splits from original dataset

   • Created additional test subset from training data for extended evaluation

2. Dataset Subsampling:

   • The final version consists of

     • 20,000 training samples

     • 4,000 validation samples

     • 4,000 test samples

References

1. 11. Clasen, L. Hackel, T. Burgert, G. Sumbul, B. Demir, V. Markl, “ reBEN: Refined BigEarthNet Dataset for Remote Sensing Image Analysis “, IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2025. https://arxiv.org/abs/2407.03653

2. Original Dataset Download: http://bigearth.net/

3. CORINE Land Cover: https://land.copernicus.eu/pan-european/corine-land-cover

import os
from pathlib import Path

import torch

from geobench_v2.datamodules import GeoBenchBENV2DataModule
from geobench_v2.datasets import GeoBenchBENV2
from geobench_v2.datasets.normalization import SatMAENormalizer, ZScoreNormalizer
from geobench_v2.datasets.visualization_util import (
    compare_normalization_methods,
    compute_batch_histograms,
    plot_batch_histograms,
    plot_channel_histograms,
)

%load_ext autoreload
%autoreload 2

DATA_ROOT = Path("../../data/benv2")

STATS_SATMAE_PATH = os.path.join(DATA_ROOT, "benv2_stats_satmae.json")
STATS_CLIP_RESCALE_PATH = os.path.join(DATA_ROOT, "benv2_stats_clip_rescale.json")

band_order = GeoBenchBENV2.band_default_order

datamodule = GeoBenchBENV2DataModule(
    img_size=120,
    batch_size=4,
    num_workers=4,
    root=DATA_ROOT,
    band_order=band_order,
    data_normalizer=torch.nn.Identity(),  # we do custom normalization in the tutorial
)
datamodule.setup("fit")

Using provided pre-initialized normalizer instance: Identity
Using provided pre-initialized normalizer instance: Identity

sample_dist_fig = datamodule.visualize_geospatial_distribution()

Dataset Statistics

Image Dataset Statistics

fig = plot_channel_histograms(STATS_SATMAE_PATH)

Target Statistics

Raw Batch Image Statistics

Statistics of a single batch before any normalization

train_dataloader = datamodule.train_dataloader()
raw_batch = next(iter(train_dataloader))

raw_batch_stats = compute_batch_histograms(raw_batch, n_bins=100)

vis_band_order = {
    "s2": [b for b in band_order["s2"] if isinstance(b, str)],
    "s1": [b for b in band_order["s1"] if isinstance(b, str)],
}
raw_figs = plot_batch_histograms(
    raw_batch_stats, vis_band_order, title_suffix=" (Raw Data)"
)

Effect of different Normalization Schemes

zscore_normalizer = ZScoreNormalizer(STATS_CLIP_RESCALE_PATH, band_order)
satmae_normalizer = SatMAENormalizer(STATS_SATMAE_PATH, band_order)

norm_fig, normalized_batches = compare_normalization_methods(
    raw_batch, [zscore_normalizer, satmae_normalizer], datamodule
)

Visualize Batch

fig, batch = datamodule.visualize_batch()