Assessing the allergenic potential of urban green spaces using orthoimagery and airborne LiDAR data

• Allergenicity is assessed for 52 urban green spaces in the Brussels Capital Region. • Allergenic tree genera are mapped using a convolutional neural network approach. • Allergenicity is quantified for each tree using LiDAR-based tree crown metrics. • Allergenicity indices are corrected for misclassification bias using stochastic simulation. Over the past decades, pollen allergy has become one of the most widespread public health issues. The number of individuals having allergies to pollen has dramatically increased, especially in urban and industrial areas. Quantifying the allergenic potential of urban green spaces and developing allergy sensitive strategies for green space management and planning are therefore becoming increasingly important. Mapping the allergenicity of urban parks requires detailed information on tree species and tree crown volume which for many cities is not available or is not updated on a regular basis. This study assesses the potential of very high-resolution remote sensing for mapping allergenic tree genera and proposes a workflow for quantifying the allergenic potential of urban green spaces (UGS). Using a convolutional network approach six allergenic genera are mapped within 52 urban green spaces across the Brussels Capital Region. The classification model achieves an overall accuracy of 0.86, with precision for the six genera ranging from 0.82 to 0.92. By combining the obtained map with tree crown measures derived from airborne LiDAR data an assessment of the allergenicity of the 52 UGS is made, accounting for misclassification bias in the mapping of tree genera. Smaller, often more centrally located neighborhood parks have the lowest index values. Landscape parks and protected habitats in the periphery of the region have higher allergenicity values.