Evaluating Remote Sensing Technology to Assess Bumblebee Habitat Quality in Open Semi-Natural Grasslands

Bumblebees are very efficient pollinators, important for the survival of many wildflower species and crops. Despite their importance, bumblebees have experienced significant global declines due to factors like habitat loss, underscoring conservation efforts to focus on habitat improvement. Efficient monitoring of bumblebee populations and their habitats is vital to these efforts. Traditional methods are labour-intensive and time-consuming, but emerging technologies like Uncrewed Aerial Vehicles (UAVs) offers more efficient monitoring possibilities.
This PhD research explores the application of UAVs for detecting bumblebee habitat traits to assess their habitat quality, thus informing conservation strategies. Initial traditional habitat surveys on five semi-natural grasslands from March to October identified critical traits affecting bumblebee abundance, specifically flower presence and vegetation structure. These traits guided the subsequent UAV detection efforts using monthly orthomosaics at 4cm of spatial resolution.
Challenges encountered included the precision of classifiers in detecting flower presence, affected by non-floral objects in the imagery during early spring and late autumn, and alignment issues in the location of in-the-field survey plots on UAV imagery. These issues were mitigated by photo-interpreting virtual plots overlaid on the orthomosaics. The assessment of vegetation structure was initially constrained by the use of Digital Surface Models (DSMs) generated with a single RTK setup. This limitation was overcome by developing a vegetation height estimation model using RGB spectral indices derived from the orthomosaics.
This research confirms that high-resolution UAV imagery can effectively and efficiently detect bumblebee habitat traits with high correlations (ranging from 0.50 to 0.93) between in-the-field measurements and UAV-derived data. The most accurate predictor of bumblebee abundance was a model that integrated both in-the-field surveyed habitat traits and UAV-derived variables, capturing aspects such as flower diversity, heterogeneity, and forage abundance. The model relying solely on UAV data still effectively identified the lowest-ranked habitat quality sites. Future research should investigate methods to enhance the prediction of flower diversity using UAV technology, flown at a height that optimizes battery usage while maintaining high resolution and minimizes computational requirements to process the imagery.
This study demonstrates the significant potential of UAV technology for ecological surveys and management of habitats. It establishes a versatile framework that could be adapted and enhanced to effectively support bumblebee conservation efforts and contribute to the ecological health of semi-natural grasslands.