Semantically aware tree height estimator for infrastructure monitoring using multimodal satellite images

Accurate tree height estimation is crucial for infrastructure safety monitoring, particularly for assessing risks to power lines, buildings, and transportation networks. This study introduces the Semantically-aware Tree Height Estimator (SaTHE), a multitask deep learning framework specifically designed for high-resolution vegetation monitoring near infrastructure. By integrating Pleiades optical and Synthetic Aperture Radar (SAR) data sources (ICEYE X-band, Sentinel-1 C-band, and ALOS-2 L-band) through an efficient attention mechanism, SaTHE simultaneously performs tree detection and height estimation at sub-meter resolution. Our comprehensive analysis reveals that X-band SAR provides optimal performance at infrastructure-relevant resolutions (0.5 m) while still exhibiting typical canopy penetration limitations in dense vegetation. On the Askvoll dataset, SaTHE outperforms state-of-the-art models in both height estimation accuracy and tree detection precision. The framework’s transferability was validated on the Lillesand dataset, where zero-shot inference achieved satisfying baseline performance, with significant improvements through minimal fine-tuning. Additionally, we provide detailed ablation studies validating our architectural choices and demonstrating the benefits of our multitask approach which establishes SaTHE as an effective solution to include in infrastructure corridor monitoring.