Mitigation of Land Contamination in SMOS L1C Brightness Temperature Data Based on Convolutional Neural Networks

• Published in: IEEE journal of selected topics in applied earth observations and remote sensing Vol. 17; pp. 18666 - 18682
• Main Authors: Chen, Ke, Yang, Qian, Tian, Yuanyuan, Li, Qingxia, Jin, Rong
• Format: Journal Article
• Language: English
• Published: IEEE 2024
• Subjects: Antenna measurements; Aperture synthesis radiometers (ASRs); Contamination; convolutional neural network (CNN); Earth; forward modeling; land contamination mitigation; Microwave imaging; Microwave measurement; Microwave radiometry; Ocean temperature; Pollution measurement; Radiometers; Sea measurements; soil moisture and ocean salinity (SMOS)
• ISSN: 1939-1404; 2151-1535
• DOI: 10.1109/JSTARS.2024.3476470

Due to the Gibbs oscillation effect of microwave aperture synthesis radiometers (ASRs) occurring near land/ocean transitions, soil moisture and ocean salinity (SMOS) brightness temperature (TB) measurements are subject to land contamination in ocean regions (within a few hundred kilometers of the shore). In this article, the magnitude of land contamination in the SMOS L1C TB is demonstrated by comparison with the results of collocated forward modeling. Then, a land contamination correction algorithm for microwave SARs based on a convolutional neural network is presented. The basic process of the algorithm is to design a land contamination mitigation network (LCMN) to learn the land contamination features from the simulated TB dataset of the microwave ASRs and then to use the well-trained network to remap the actual observed TB images to mitigate the land contamination error. Land contamination error correction experiments based on LCMN remapping were executed on SMOS L1C TB data, and the performance of the algorithm was verified by comparison with forward modeling and salinity retrieval. The experimental results show that the proposed LCMN algorithm can effectively mitigate land contamination at the SMOS L1C TB.