Modelling urban tree growth under the uncertainty of remote-sensing data: A study across two European cities

Source abstract

Modelling tree size development in urban environments is essential for both landscape planning and studying tree ecosystem services. While conventional methods rely on extensive field measurements, recent advances in remote sensing and data-driven models offer new opportunities to retrieve tree attributes at scale while representing local growth conditions for individual trees. In this study, we present a framework for predicting tree size over different stages of tree growth through the fusion of remote sensing and tree registry data. Using segmented trees from LiDAR and RGB orthophotos, combined with species inventories from municipal and citizen-science datasets, we developed machine learning models to predict tree height and crown diameter across two European cities with different levels of data accessibility. The results of our analysis showed improved prediction of tree height ( R 2 = 0.81) and crown diameter ( R 2 = 0.76) across datasets from Copenhagen and Sofia compared to allometric models. We identified local shading and surface sealing as important predictors of individual tree growth rates, whereas the distance to nearby buildings showed limited impact on model predictions. Incorporating large-scale meteorological variables did not significantly improve model performance across individual cities, although higher-resolution spatial and temporal climate data may improve predictions in future studies. Using tree age as the reference predictor, the proposed framework can support future scenario assessments of tree ecosystem services by accounting for growth across different ages and urban configurations. Overall, our study demonstrates the value of high-quality remote sensing data for urban forest monitoring and supporting assessment tools for urban green infrastructure planning and management.