Cross-Continental Forest Height Mapping via Bayesian Deep Learning on TanDEM-X Single-Pass InSAR Data

Kurzfassung

Forests are among the planet's most critical ecosystems, shaping climate patterns and sustaining biodiversity. Increasingly threatened by human activities, forest preservation demands policies informed by reliable global monitoring of biophysical parameters such as canopy height. In this scenario, deep learning-based approaches applied to satellite Interferometric SAR (InSAR) data have advanced the state of the art, yet remain geographically constrained and often overlook uncertainty quantification, which is essential for reliability and long-term assessment. This work introduces a Bayesian deep learning framework delivering accurate height predictions together with calibrated uncertainty estimates. The framework demonstrates cross-continental generalization capabilities and paves the way for reliable large-scale products for forest monitoring.