AgriPotential dataset

 NEWS: the agripotential dataset will be part of ImageCLEF 2026 competition. Please contact us if you’re interested in participating

For more information regarding participation please refer to the following link:

https://www.codabench.org/competitions/12055/ 

NEWS:  an extended version of the dataset is released

Differences:

• 34 timeframes instead of 11 previously
• A more user friendly data format (all .tif with possibility of online data loading)
• Sentinel-2 data is not normalized, leaving the choice of normalization to the user

You can find more information on https://github.com/MohammadElSakka/agripotential

***UPDATE: this work was accepted in the conference IEEE CBMI 2025 that will be held in dublin in late october 2025. A final peer-reviewed version will be published soon***

This is the official page for AgriPotential, a publicly available benchmark dataset designed for assessing agricultural potentials using remote sensing data. It integrates 11  Sentinel-2 satellite images from 2019 across 10 spectral bands, covering agricultural regions in the Hérault Department, Southern France. The dataset includes pixel-wise labels representing agricultural potential levels (from Very Low to Very High) across three crop types: viticulture, market gardening, and field crops. Ground truth annotations are derived from the BD Sol – GDPA database and validated by domain experts. The dataset is stored in HDF5 format and supports a range of machine learning tasks, including ordinal classification, regression, segmentation, and spatio-temporal modeling.

This dataset facilitates scalable, data-driven solutions for land suitability analysis, agricultural planning, and sustainable resource management.

On this page you will find a PDF with supplmentary results for the paper.

A tutorial GitHub is also available: https://github.com/MohammadElSakka/agripotential-dataset 

Raw data that was used to make this dataset can be found here (there isn’t much to do there apart from downloading it): https://zenodo.org/records/15551802