https://doi.org/10.1140/epjp/s13360-025-07235-7
Regular Article
Prediction of seismic topographic effects from DEM data via U-Net deep learning model
1
Institute of Geophysics, China Earthquake Administration, 100081, Beijing, China
2
China Institute of Disaster Prevention, 065201, Sanhe, China
Received:
13
August
2025
Accepted:
18
December
2025
Published online:
7
January
2026
Site effects are a critical topic in seismology and earthquake engineering, with topographic effects playing a pivotal role in local seismic response. Traditional models often rely on manually selected topographic parameters, such as slope, curvature, or relative elevation, limiting their effectiveness and interpretability. In this study, seismic ground motion in the Erlang Mountain area of the Sichuan–Tibet region, China, was simulated using the spectral element method (SEM). A U-Net-based deep learning model was then employed to predict seismic topographic amplification, taking digital elevation model (DEM) data as input and peak ground velocity (PGV) amplification factors as output. Multi-layer convolution operations extracted topographic features, while skip connections integrated multi-scale information. Quantified by root-mean-square error (RMSE), the U-Net model outperformed traditional back-propagation (BP) neural networks. Incorporating incident angle and subsurface velocity structures as additional input channels further improved prediction accuracy, highlighting the coupled influence of topography and subsurface conditions. This study provides a data-driven framework for simulating seismic topographic effects.
Copyright comment Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
© The Author(s), under exclusive licence to Società Italiana di Fisica and Springer-Verlag GmbH Germany, part of Springer Nature 2026
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
