Work

PMTB-VSRnet: A video super-resolution network with radiometric consistency guidance for Antarctic passive microwave brightness temperature

Year: 2026

Type: article

Abstract: Monitoring Antarctic surface melt using satellite-based Earth observation relies heavily on passive microwave data for their all-weather capability; however, their coarse spatial resolution limits the detection of fine-scale melt features. To address this limitation, we propose PMTB-VSRnet, a video super-resolution network with radiometric consistency guidance for Antarctic passive microwave data. In PMTB-VSRnet, neighboring frames are first aligned using Pyramid, Cascading and Deformable (PCD) alignment, while low-frequency Normalized Cross-Correlation (NCC)-guided fusion enhances radiometric feature consistency and suppresses unreliable observations. The aligned features are then propagated bidirectionally to exploit temporal information and reconstruct fine-scale brightness temperature structures. Finally, a radiometric high-frequency constraint is introduced to enhance reconstruction accuracy. Extensive experiments on AMSR2 multi-channel data show that PMTB-VSRnet outperforms all compared methods, improving PSNR over the average performance of the competing methods by 1.57, 2.30, 2.47, and 2.81 dB for the 36.5 H, 36.5 V, 18.7 H, and 18.7 V channels, respectively, under a 4 × super-resolution setting. Additional experiments across diverse Antarctic terrains, seasons, and continent-scale regions further demonstrate the robustness and scalability of the proposed framework. The source code is publicly available at https://github.com/jingli1999/PMTB-VSRnet . • Combining PCD alignment and NCC-guided fusion to enhance local radiometric feature consistency. • Integrating local temporal modeling and bidirectional global propagation to improve temporal aggregation and reconstruction robustness. • Introducing a radiometric high-frequency constraint to enhance reconstruction accuracy. • Achieving consistent improvements across regions, seasons, and channels, with scalability to continent-scale observations.

Source: International Journal of Applied Earth Observation and Geoinformation

DOI: 10.1016/j.jag.2026.105371

Authors: Jing Li, Lu An, Zhiwen Yang, Yujie Sun, Zhe Zhou

Institution: Tongji University

Language: English

Topic: Advanced Image Processing Techniques

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

SDG: Climate action

Open Access status: gold

Estimated APC paid: $3,550