Urban tree species benchmark dataset for time series classification
EasyData
We present a benchmark dataset for urban tree species classification based on multi-source optical satellite image time series (SITS). The dataset provides, on the city of Strasbourg (France), surface reflectance values extracted from coregistered Sentinel-2 and PlanetScope imagery on public trees.
creationApr 24, 2025revisionApr 28, 2025publicationApr 28, 2025
Temporal CoverageJan 1, 2022Sep 30, 2022
Overview
Lineage
The first GeoPackage dataset (raw dataset) contains surface reflectance values acquired in 2022 from Sentinel-2 (S2 – L2A) and PlanetScope (PS – L3B) satellites associated to a polygon corresponding to the point location of the urban tree with a buffer of 1m. Time series values are stored in Real format and structured by sensor and spectral band. A total of 45,084 trees representing the 20 most common species are included in the dataset. Sentinel-2 images and PlanetScope images are co-registered with sub-pixel alignment.
The dataset is formatted for time series classification tasks, featuring temporally aligned observations, and patch-level sampling around individual tree locations, enabling seamless integration into deep learning frameworks. The second dataset contains processed model outputs from three deep learning models including predicted species, confidence scores, and correctness flags.
The dataset is formatted for time series classification tasks, featuring temporally aligned observations, and patch-level sampling around individual tree locations, enabling seamless integration into deep learning frameworks. The second dataset contains processed model outputs from three deep learning models including predicted species, confidence scores, and correctness flags.
Links
Downloads
- benchmark_dataset-time_series.gpkg
- README_UrbanTreeBenchmark.txt
- processed_dataset-model_outputs.gpkg
Geolinks
- No links available
Others
- Link to Original Record [easydata.earth] This metadata comes from easydata.earth
- Wenger, R., Bressant, C., Roettele, L., Forestier, G., & Puissant, A. (2024) Improving Urban Tree Species Classification with High Resolution Satellite Imagery and Machine Learning. In IGARSS 2024 – 2024 IEEE International Geoscience and Remote Sensing Symposium (pp. 4679–4682). https://doi.org/10.1109/IGARSS53475.2024.10642787
- Latil, M., Wenger, R., Michéa, D., Forestier, G., & Puissant, A. (2025) Urban trees species classification using Sentinel-2 and Planetscope satellite image time series. In 2025 Joint Urban Remote Sensing Event (JURSE)
- Urban-tree-classification This repository contains the code used for the research paper "Urban trees species classification using Sentinel-2 and Planetscope satellite image time series", presented at JURSE 2025 in Tunis, Tunisia (May 4-7, 2025).
- Interactive t-SNE visualization The visualization shows t-SNE embeddings of tree species classified by a DualInceptionTime model trained on time series from Sentinel-2 and Planet imagery. Correct predictions are shown as circles, misclassifications as crosses. Colors correspond to species.
- Digital Object Identifier (DOI)
Contacts
Data Terra
Keywords
Constraints
License
unrestricted
Other restrictions
CC-BY-4.0