BigEarthNet - A Large-Scale Sentinel Benchmark Archive

About BigEarthNet v2.0

The BigEarthNet v2.0 dataset was constructed by the Remote Sensing Image Analysis (RSiM) Group and the Database Systems and Information Management (DIMA) Group at the Technische Universität Berlin (TU Berlin). This work is supported by the European Research Council under the ERC Starting Grant BigEarth and by the Berlin Institute for the Foundations of Learning and Data (BIFOLD).

BigEarthNet v2.0 is a benchmark dataset consisting of 549,488 pairs of Sentinel-1 and Sentinel-2 image patches.

BigEarthNet with Sentinel-2 Image Patches

To construct BigEarthNet v2.0 with Sentinel-2 image patches (called as BigEarthNet-S2), 115 Sentinel-2 tiles acquired between June 2017 and May 2018 over 10 countries (Austria, Belgium, Finland, Ireland, Kosovo, Lithuania, Luxembourg, Portugal, Serbia, and Switzerland) of Europe were initially selected. All the tiles were atmospherically corrected by the Sentinel-2 Level 2A product generation and formatting tool (sen2cor v2.11). Then, they were divided into 549,488 image patches. Each image patch was associated with a pixel-level reference map and multiple land-cover class labels (i.e., multi-labels) that were derived from the most recent CORINE Land Cover database of the year 2018 (CLC2018 v2020_u1).

BigEarthNet with Sentinel-1 Image Patches

To construct BigEarthNet v2.0 with Sentinel-1 image patches (called as BigEarthNet-S1), 312 Sentinel-1 scenes acquired between June 2017 and May 2018 that jointly cover the area of all original 115 Sentinel-2 tiles with close temporal proximity were selected and processed. BigEarthNet-S1 consists of 549,488 preprocessed Sentinel-1 image patches – one for each Sentinel-2 patch.

For the details about BigEarthNet v2.0, please see our paper:

K. Clasen, L. Hackel, T. Burgert, G. Sumbul, B. Demir, V. Markl, " reBEN: Refined BigEarthNet Dataset for Remote Sensing Image Analysis ", arXiv preprint arXiv:2407.03653, 2024 .

Downloads

BigEarthNet-S2

(~59GiB) v2.0

🤗 Pretrained S2 Models

BigEarthNet-S1

(~51GiB) v2.0

🤗 Pretrained S1 Models

BigEarthNet-S1 and S2 (S1+S2)

Download BigEarthNet-S2

Download BigEarthNet-S1

🤗 Pretrained S1+S2 Models

Additional Links

Dataset Description Document: This document offers insights into the dataset, the directory structure, naming conventions and additional metadata files.
Reference Maps: The reference maps for the BigEarthNet v2.0 dataset.
metadata.parquet: This parquet file contains the metadata for all BigEarthNet v2.0 patches, except for those that are covered by seasonal snow, clouds or cloud shadows.
metadata_for_patches_with_snow_cloud_or_shadow.parquet: This parquet file contains the metadata for all BigEarthNet v2.0 patches covered by seasonal snow, clouds, or cloud shadows.
bigearthnet-pipeline: This repository contains the code for creating the BigEarthNet v2.0 dataset in a reproducible manner.
ConfigILM: A library that includes an optimized PyTorch Dataset and Lightning Data Module for BigEarthNet (and other RS datasets) to achieve high-throughput deep-learning. ConfigILM is designed for Python developers seeking to rapidly and iteratively develop image and language models within the PyTorch framework.
rico-hdl: A fast and easy-to-use remote sensing image format converter for high-throughput deep-learning.

FAQ

Yes, some Sentinel-2 patches include seasonal-snow. We kept them in the archive to allow tag-based search and retrieval studies independently from seasons. However, if you work on scene classification, content-based image retrieval and search only by using Sentinel-2 image patches, we suggest not to include patches fully covered by seasonal-snow for training and test stages of the machine/deep learning algorithms. For more details, see the dataset description document.

Sentinel-2 tiles in BigEarthNet are associated to tiles with a cloud cover of less than 1%. However, the cloud cover is present in some patches. We kept these patches in BigEarthNet to allow multi-modal learning studies. However, if you work on scene classification, content-based image retrieval and search only by using Sentinel-2 image patches, we suggest not to include these patches for training and test stages of the machine/deep learning algorithms. For more details, see the dataset description document.

BigEarthNet v2.0 uses the most recent CORINE Land Cover database of the year 2018 and the 19-class label nomenclature. As a result some patches were completely or partially unlabeled and not included in the v2.0 dataset. Also, four Sentinel-2 tiles from the v1.0 dataset were dropped due to failing quality indicator checks. For more details, see the official publication reBEN: Refined BigEarthNet Dataset for Remote Sensing Image Analysis .