FLAIR-HUB: Large-scale multimodal dataset for land cover and crop mapping

The growing availability of high-quality Earth observation (EO) data enables accurate global land cover and crop type monitoring. However, the volume and heterogeneity of these datasets pose major processing and annotation challenges. To address this, the French National Institute of Geographical and Forest Information (IGN) is actively exploring how to exploit diverse EO data, which require large annotated datasets. IGN introduces FLAIR-HUB, a novel large-scale multi-sensor land cover dataset with very-high-resolution (20cm) annotations, covering 2528km 2 of France. This dataset uniquely integrates six complementary and aligned modalities, enabling more accurate and fine-grained land cover description: aerial imagery, optical multi-spectral Sentinel-2 and SAR Sentinel-1 satellite time series, high-resolution SPOT satellite images, topographic data, and historical aerial images. Extensive benchmarks evaluate multimodal fusion and deep learning models (convolutional neural networks, transformers) for land cover or crop mapping and also explore multi-task learning. Results underscore the complexity of multimodal fusion and fine-grained classification, with best land cover performance (78.2% accuracy, 65.8% mean Intersection over Union) achieved using nearly all modalities. FLAIR-HUB provides a valuable foundation for advancing supervised and pretraining approaches in Earth Observation research, with data and code openly available. • We provide a large-scale benchmark for multimodal land cover and crop mapping • Combines six aligned sources spanning spatial, spectral, and temporal cues • Fine-grained 20 cm annotations support detailed semantic segmentation • Designed for supervised, self-supervised, and transfer learning settings • Extensive benchmarks quantify gains from multimodal fusion with a modern backbone