Multiresolution Cross-Attention–Transformer for Spatiotemporal Prediction of Permafrost Deformation in InSAR Time Series
IEEE Transactions on Geoscience and Remote Sensing · 2026 · 64 · 5203120-5203120
DOI: 10.1109/tgrs.2026.3662345 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Geoscience and Remote Sensing, 2026. Verfasst von Zexin Xu und 5 weiteren Autor:innen. Die Arbeit nennt 45 Literaturangaben. Gefördert von: National Natural Science Foundation of China, National Railway Group Science and Technology Development Project.
Abstract (Originalsprache)
The Qinghai-Tibet Plateau (QTP) has the largest area of permafrost terrain in the mid- and low-latitude regions of the world. Under the combined influence of global change and human activities, the permafrost on the QTP is undergoing widespread degradation, which will not only have significant impacts on the global environment and carbon cycle, but also result in permafrost-related geohazards. Surface deformation resulting from the freeze–thaw process is an important characteristic of permafrost, and predicting it is of great significance for understanding the development of permafrost. However, intrinsic complexity and multifactorial environmental controls limit the accuracy and generalizability of existing prediction models. In this article, based on the deformation results from the interferometric synthetic aperture radar (InSAR) technique, an integrated algorithm for permafrost deformation prediction is proposed. The algorithm combines wavelet multiresolution analysis and a multiresolution cross-attention (MRCA)–Transformer model, in which both periodic and nonperiodic deformation features, together with environmental variables such as soil moisture and temperature, are incorporated through a multihead cross-attention (MHCA) mechanism for spatiotemporal prediction. The proposed model is trained and evaluated using the InSAR time-series deformation derived from 2017 to 2024 Sentinel-1A data over the QTP permafrost region. The model achieves an average root-mean-square error (RMSE) of 4.97 mm, an average mean absolute error (MAE) of 4.22 mm, and standard deviations of 3.21 and 2.89 mm, respectively. The square of the correlation coefficient and the coefficient of determination are both greater than 0.99 across all time steps. Moreover, the results show that the model demonstrates effective performance across a variety of terrains and observational conditions without additional training. This study provides significant scientific insights and practical value for the monitoring of permafrost evolution, environmental assessment, and early warning of geohazards.
Förderung
- National Natural Science Foundation of China
- National Railway Group Science and Technology Development Project
Zitieren
APA
Xu, Z., Tang, Y., Liu, G., Zhang, H., Hu, Q., & Wang, C. (2026). Multiresolution Cross-Attention–Transformer for Spatiotemporal Prediction of Permafrost Deformation in InSAR Time Series. IEEE Transactions on Geoscience and Remote Sensing, 64, 5203120-5203120. https://doi.org/10.1109/tgrs.2026.3662345
Vancouver
Xu Z, Tang Y, Liu G, Zhang H, Hu Q, Wang C. Multiresolution Cross-Attention–Transformer for Spatiotemporal Prediction of Permafrost Deformation in InSAR Time Series. IEEE Transactions on Geoscience and Remote Sensing. 2026;64:5203120-5203120. doi:10.1109/tgrs.2026.3662345
BibTeX
@article{Xu2026,
author = {Xu, Zexin and Tang, Yixian and Liu, Guiwei and Zhang, Hong and Hu, Qingbo and Wang, Chao},
title = {Multiresolution Cross-Attention–Transformer for Spatiotemporal Prediction of Permafrost Deformation in InSAR Time Series},
journal = {IEEE Transactions on Geoscience and Remote Sensing},
year = {2026},
volume = {64},
pages = {5203120--5203120},
doi = {10.1109/tgrs.2026.3662345}
}
EUVIMED