Remote sensing-based tree species classification for estimating ecosystem services

Kurzfassung

Urban trees play a vital role in enhancing community health and well-being by providing environmental, social, and economic functions, collectively known as ecosystem services. Urban tree growth models have attempted to estimate the extent of these ecosystem services using allometric equations, growth factors and physiological functions. Nevertheless, these models require detailed information on individual tree characteristics, particularly tree species. Traditional data collection methods for tree attributes can be expensive and time-consuming. This study proposes a cost-effective approach utilizing remote sensing products, specifically multi-temporal satellite imagery and a canopy height model with very high spatial resolution, in conjunction with machine learning techniques for single tree classification. Our object-based classification method incorporates various features, including vegetation indices, spectral, geometrical, and phenological attributes to differentiate the most common tree genera in Munich, Germany. To enhance classification accuracy, we adopt a hierarchical approach that considers land use, physiological characteristics, genus, and species. The model was applied over a large-scale area, successfully classifying more than 160,000 trees. Evaluation of the results revealed variations in classification accuracy based on land use and tree genus, achieving an accuracy of up to 88.9% for certain species. Our results are further validated for the estimation of ecosystem services by testing our tree metrics in the CityTree model. This research demonstrates the efficacy of remote sensing and machine learning techniques for accurately classifying urban trees and obtaining detailed tree attributes necessary for ecosystem services models. The findings contribute to a more detailed understanding of urban tree dynamics and provide valuable insights for monitoring urban trees and their ecosystem services.