https://doi.pangaea.de/10.1594/PANGAEA.993969Khan, TaimurTaimurKhan0000-0001-7833-5474Helmholtz Centre for Environmental Research - UFZPANGAEA2026ForestMultispectral Image DataphenologyTimeseriesUAVDateRaster graphic, GeoTIFF formatFile contentCoordinate reference systemUTM Easting, Universal Transverse MercatorUTM Northing, Universal Transverse MercatorLongitude, westboundLatitude, southboundLongitude, eastboundLatitude, northboundPixel sizeImage pixel widthImage pixel heightCommentDescriptionFeature countsDate/time startDate/time endDurationSolar elevation angleSolar azimuth angleIndicatorDrone, DJI, Mavic3M2025-03-06T00:00:00/2025-11-05T00:00:00Dataset10.21105/joss.08056492 data pointstext/tab-separated-valuesCreative Commons Attribution 4.0 InternationalWe present HeideBench, a very-high-resolution multispectral uncrewed aerial vehicle dataset for forest crown phenology collected over a forest patch in Dölauer Heide, Halle (Saale), Germany. Dölauer Heide is currently dominated by pine plantations (Kiefernforste), which cover the largest area but are increasingly affected by dieback, while its potential natural vegetation is sessile oak–hornbeam forest rich in small-leaved lime (Albrecht et al., 1993). In addition to these pine stands, the area contains near-natural mixed deciduous forests with oaks, birches, and beeches, making it a particularly relevant setting for observing seasonal canopy development under contrasting forest structures and ongoing ecological transition. Against this background, HeideBench provides repeated observations of the same forest patch through the growing season. The dataset contains 18 georeferenced multispectral GeoTIFF orthomosaics acquired between 6 March 2025 and 5 November 2025, spanning a 244-day seasonal period from early spring to late autumn. The acquisitions have a median revisit interval of 14 days, with intervals ranging from 4 to 27 days, and an average ground sampling distance of 5.53 cm per pixel. The valid imaging footprint covers approximately 32.1 ha and is bounded by 11.902653–11.911325°E and 51.499959–51.508576°N.
Data were collected using a DJI Mavic 3M Enterprise uncrewed aerial vehicle equipped with four multispectral cameras measuring green (560 nm), red (650 nm), red-edge (730 nm), and near-infrared (860 nm) reflectance, in that order. Flights used a real-time kinematic (RTK) positioning module for centimeter-level geolocation, and all data are provided in coordinate reference system EPSG:25832. Imagery was processed with Agisoft Metashape 2.3.1 to generate calibrated multispectral orthomosaics. The dataset further includes 5,885 crop-safe individual tree crown instance segmentations over the same footprint, extracted with the DeepTrees software package (Khan et al., 2025). HeideBench is intended to support crown-centric analyses of seasonal canopy development, temporal representation learning, phenology-aware feature extraction, and the evaluation of tree crown delineation under seasonal change.
HeideBench is a result of the Dynamic Platform Project titled "PhenoEmbed: Multispectral UAV AI Embeddings for phenology-aware tree crown delineation" of the Integration Platform 1: "Sustainable future land use" (IP1) at the Helmholz Centre for Environmental Research (UFZ) in Leipzig, Germany.11.90265311.91132551.49995951.508576