GATENet: Spectral-Auxiliary Attention Network for Airborne GNSS-R-Based Topographic Estimation

High-resolution topographic data are critical for geomorphic studies and hazard monitoring. Although LiDAR and InSAR provide high accuracy, their high cost and weather dependence limit large-scale deployment. Airborne GNSS Reflectometry (GNSS-R) offers a low-cost, all-weather alternative, but its vertical accuracy in complex terrain often degrades to the 10-meter level, restricting fine-scale applications. To address this limitation, this study proposes a deep learning–based architecture, GATENet (GNSS-R Adaptive Topography Estimation Network), for GNSS-R elevation inversion. GATENet extracts spatial features from Delay–Doppler Maps (DDMs), integrates flight geometry information, and incorporates surface-type–dependent characteristics to adapt to different terrain conditions. Experiments on the airborne RONGOWAI dataset show that the proposed framework outperforms traditional geometric inversion methods. In ocean environments used as a baseline, the GATENet-based retrieval achieves an RMSE of 3.298 m, representing a 50.7% reduction relative to the baseline. In complex land scenarios, elevation errors generally remain within 5 m. By further incorporating a specular point–based topographic reconstruction strategy, the resulting vertical accuracy reaches up to 2.5 m. These results demonstrate the potential of the proposed framework for practical airborne GNSS-R-based topographic mapping.