Assessing urban forest biodiversity through automatic taxonomic identification of street trees from citizen science applications and remote-sensing imagery

Abstract

The positive impact of urban forests and trees on the well-being of urban residents worldwide is well known. Resistance to pests, diseases, and extreme weather events are among the most critical characteristics of resilient cities, closely related to species richness and, consequently, to the diversity of street trees. However, urban forest inventories are currently scarce worldwide. For this reason, urban trees' biodiversity and capacity to provide these ecosystem services are not developed enough. Three state-of-the-art species identification applications were tested, Plant.id, Pl@ntNet and Seek (iNaturalist) to identify a large number of tree families, genera, and species automatically. Two individual Google Street View images were queried for each tree in the study area, adjusting the Field of View and pitch parameters. The predictive capacity of the three apps was compared, and a biodiversity analysis was performed for different geospatial scales within the study area (i.e., at the whole study area, neighborhood, and street levels, respectively). Notably, our research contributes in an innovative way to the assessment and monitoring of the ecosystem services provided by street trees and sheds light on the great potential of combining remote sensing, citizen science and artificial intelligence for urban forest biodiversity assessments at multiple spatial and temporal scales.

Keywords

• Urban forest inventory
• Geospatial analysis
• Artificial intelligence
• Citizen science
• Species identification apps
• Remote sensing