• We compared UAV and PlanetScope data for mapping Natura 2000 habitats • UAV achieved higher accuracy than PlanetScope for most habitats • PlanetScope matched UAV accuracy in selected forest and grassland habitats • Spring UAV data outperformed summer data across most habitats • UAV classification of non-forest habitats outperformed that of forest habitats Mapping and monitoring of Natura 2000 habitats (Habitat Directive 92/43/EEC) is essential for effective biodiversity protection in Europe. Remote sensing offers promising tools for this task, yet the high number of habitat types and their heterogeneity challenge the achievable classification accuracy. While recent studies demonstrated good results with very-high-resolution satellite data (e.g. Sentinel-2, RapidEye), spatial resolution in the decimetre range may be required for small patches in fine-scale mosaics. This study primarily investigates the potential of UAV-based mapping to distinguish ten Natura 2000 forest, wetland, and grassland habitats within a ∼20 km² landscape in Central Europe. Using Random Forest classification, we evaluated producer, user, and overall accuracies depending on phenological season (spring, summer), sensor type (multispectral, RGB), predictor type (spectral, textural, object-based), and classification scheme (detailed vs. aggregated habitats). The highest classification accuracy (Cohen’s Kappa 0.71–0.77) was achieved using multispectral UAV data from both seasons. Comparable results (0.67–0.76) were obtained from spring data alone, whereas summer-only data yielded substantially lower accuracy (0.46–0.56). RGB-only imagery also produced satisfactory results (0.65–0.75) when combining spring and summer. Spectral predictors were the most important, yet the inclusion of textural and object-based metrics further improved classification. Non-forest habitats were generally classified more accurately than forests. To explore the usability of alternative data sources, we also compared UAV results with classification using PlanetScope satellite imagery. While UAV mapping outperformed PlanetScope overall, several habitat types reached comparable or even better classification accuracy with satellite data. We conclude that UAV mapping can provide high-accuracy habitat classification when multispectral data from multiple seasons are available. If flight capacity is limited, acquiring multispectral imagery in spring should be prioritized. If only RGB data are available, combining spring and summer can still yield satisfactory results. PlanetScope data offer complementary potential, particularly for some forest and grassland habitat types.
Šímová et al. (Sun,) studied this question.