Scalable urban tree canopy modelling for Canadian cities

Abstract

Urban tree cover (UTC) is a key indicator for monitoring climate adaptation, guiding policy goals, and designing equitable cities. However, consistent UTC estimates remain limited. This study proposes a transferable UTC modelling framework using global open-source data to map fractional UTC. The method involves satellite spectral products and high-resolution canopy height data to build city-specific machine learning models of 19 Canadian cities. Model accuracy, assessed through cross-validation, ranged from R² = 0.67–0.88 with RMSE values of 5.5–16.3%. Independent validation against airborne laser scanning in three cities demonstrated strong agreement (Pearson r = 0.82–0.92). Results suggest that UTC distributions are unevenly distributed, typically right skewed, with large parks and riparian corridors accounting for a disproportionate share of UTC. Greenness and moisture are the most influential predictors, while model residuals concentrate in areas with higher UTC. The resulting UTC maps and modelling framework provide a consistent and reproducible framework for UTC distribution assessments, benchmarking across diverse ecozones, and monitoring UTC change at urban scales over time.