Modeling urban tree canopy height using high-resolution satellite imagery and deep learning

Abstract

Urban forests play a critical role in mitigating urban heat island effects, improving air quality, and enhancing biodiversity, yet accurate and scalable methods for mapping urban tree canopy height remain limited. Traditional approaches rely on either low-resolution global models that lack local specificity or labor-intensive field surveys that are impractical for large-scale applications. This thesis tackles these issues by introducing an innovative framework that leverages high-resolution satellite imagery and deep learning to model urban tree canopy height, emphasizing both scalability and accuracy.

The study leverages diffusion-based deep learning models, a generative approach that outperforms conventional discriminative methods in handling complex tasks. Integrating data from LiDAR, Sentinel-1, and PlanetScope imagery, the proposed model captures fine-grained canopy height variations across diverse urban landscapes. The methodology includes a preprocessing step using a neural network-based vegetation classifier to distinguish tree canopies from non-vegetated surfaces, followed by model training on canopy height data derived from airborne LiDAR in Hong Kong. The trained model demonstrates strong performance, achieving an R2 value of 0.544 and a Root Mean Square error (RMSE) of 2.533 meters on the hold-out samples. Validation in Shenzhen demonstrates the model’s transferability, successfully generating a 3-meter resolution canopy height map for the city. Comparative analyses with existing global and regional models further highlight the advantages of the diffusion-based approach, particularly in urban and hilly areas where traditional methods struggle.

The results offer significant managerial implications for urban forest management, biodiversity conservation, and climate resilience planning. High-resolution canopy height maps enable city planners to identify critical green spaces, optimize urban greening strategies, and assess the ecological impact of urban development. By providing a scalable and accurate tool for urban tree canopy modeling, this research contributes to sustainable urban planning and supports decision-making for enhancing urban environmental quality and resilience to climate change.