Tomographic SAR Reconstruction for Forest Height Estimation

TL;DR

The paper tackles scalable forest height estimation for biomass mapping using SAR by replacing the full tomographic spectral estimation with a covariance-based deep learning pipeline applied to SLC data. It trains a 2D U‑Net to predict canopy height from covariance features derived from TomoSense P-band data (28 SLCs across two headings and multiple polarizations) and systematically evaluates how the number of input SLCs and polarization affect accuracy. Key findings show meaningful MAE reductions when increasing SLCs (notably from 3 to 7) and identify VV polarization as the most informative, achieving a best-case CHM MAE around $4.17$ m; however, the partial method incurs a $16$–$21$ ext{%}$ increase in MAE relative to full TomoSAR, indicating the continued importance of complete tomographic processing. The study provides practical guidance for ESA's Biomass mission data collection strategies and highlights avenues for enhancing the partial pipeline with additional geometric or hyperspectral information to close the gap to full tomographic performance.

Abstract

Tree height estimation serves as an important proxy for biomass estimation in ecological and forestry applications. While traditional methods such as photogrammetry and Light Detection and Ranging (LiDAR) offer accurate height measurements, their application on a global scale is often cost-prohibitive and logistically challenging. In contrast, remote sensing techniques, particularly 3D tomographic reconstruction from Synthetic Aperture Radar (SAR) imagery, provide a scalable solution for global height estimation. SAR images have been used in earth observation contexts due to their ability to work in all weathers, unobscured by clouds. In this study, we use deep learning to estimate forest canopy height directly from 2D Single Look Complex (SLC) images, a derivative of SAR. Our method attempts to bypass traditional tomographic signal processing, potentially reducing latency from SAR capture to end product. We also quantify the impact of varying numbers of SLC images on height estimation accuracy, aiming to inform future satellite operations and optimize data collection strategies. Compared to full tomographic processing combined with deep learning, our minimal method (partial processing + deep learning) falls short, with an error 16-21\% higher, highlighting the continuing relevance of geometric signal processing.

Figures (6)

• Figure 1: Test MAE by Heading, 3 SLC Inputs.
• Figure 2: Test MAE by SLC and Pol. Channel.
• Figure 3: CHM Reconstruction L- Prediction, R- Ground truth
• Figure 4: Exp. 3 Full data results, P-band, VV
• Figure 5: Model Architecture