The advent of precision agriculture has highlighted the growing demand for high-resolution,
multimodal data capable of capturing the complexity of crop dynamics in the field. As the
agricultural sector increasingly shifts toward data-driven decision-making, integrating aerial
and ground-based measurements has became essential for monitoring plant health, optimizing
resource use, and improving yield forecasting. In this context, drones equipped with multispec-
tral imaging sensors offer unprecedented capabilities for capturing detailed spatial and spectral
information across vast agricultural landscapes.
This study introduces a new multimodal dataset designed for precision agriculture appli-
cations, combining high-resolution UAV imagery with synchronized ground-level biophysical
measurements. Data acquisition was conducted using a DJI Matrice 300 RTK equipped with a
10-band MicaSense Dual Sensor over several agricultural sites in northern France. The resulting
drone imagery, captured at a 5 cm ground sampling distance (GSD), underwent photogram-
metric processing to generate orthomosaics, digital elevation models, and reflectance indices.
Calibration procedures involving a sun sensor and reflectance panel ensured radiometric consis-
tency.
Alongside the aerial data, field campaigns provided synchronized measurements of leaf area
index (LAI), soil moisture, biomass, and irrigation records. Flight parameters were carefully
optimized to balance resolution, coverage, and operational constraints—flying at ˜70–75 m AGL,
8 m/s speed and 80%/75% forward and side overlap. After processing, we generated a range of
useful products—like orthomosaics, elevation models, and vegetation indices.
This dataset provides a crucial link between aerial observations and ground-truth measure-
ments, enabling robust validation and development of data-driven agricultural models.
Keywords: Multimodal learning, UAV, multispectral imagery, soil moisture, crop yield pre-
diction, precision agriculture.