Scholarlink

<p>Due to the limited spatial coverage of Global Navigation Satellite System (GNSS) receivers, Total Electron Content (TEC) maps often exhibit spatial incompleteness, particularly in regions with sparse observations or severe ionospheric disturbances. While global TEC maps provided by the International GNSS Service (IGS) attempt to fill these gaps through interpolation and assimilation, the process remains time-consuming and requires substantial manual effort. Thus, an efficient TEC completion model is in demand. In our previous work, we constructed a TEC completion model using pix2pixhd based on generative adversarial networks, performing badly around the edges of the ionospheric peak regions (TEC ≥ 50 TECU). To address this challenge, we present an improved model, TEC-EDGE, based on IGS TEC maps from 2003 to 2018, which incorporates prior knowledge in the form of edge maps derived from the 50 and 20 TECU contours. The results demonstrate that TEC-EDGE consistently outperforms the model without prior knowledge (TEC-WO), achieving a reduction in average RMSE from 2.4341 to 2.3019 TECU and an improvement in average SSIM from 0.7411 to 0.7455 on the test set. Notably, TEC-EDGE demonstrates significant improvements in edge regions and performs better under both quiet and disturbed geomagnetic conditions, as well as during solar maximum compared to solar minimum periods. Our work demonstrates a new possibility of applying deep learning with prior knowledge to space weather research, particularly for problems of data lack.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

TEC map completion using a deep learning approach with prior knowledge

  • 期刊论文

  • 发表日期：

    2025年8月7日

查看全文
  • Ding Yang,
  • Licheng Liu,
  • Erxiao Liu,
  • Lei Qiao,
  • Chaofan Duan

摘要

Due to the limited spatial coverage of Global Navigation Satellite System (GNSS) receivers, Total Electron Content (TEC) maps often exhibit spatial incompleteness, particularly in regions with sparse observations or severe ionospheric disturbances. While global TEC maps provided by the International GNSS Service (IGS) attempt to fill these gaps through interpolation and assimilation, the process remains time-consuming and requires substantial manual effort. Thus, an efficient TEC completion model is in demand. In our previous work, we constructed a TEC completion model using pix2pixhd based on generative adversarial networks, performing badly around the edges of the ionospheric peak regions (TEC ≥ 50 TECU). To address this challenge, we present an improved model, TEC-EDGE, based on IGS TEC maps from 2003 to 2018, which incorporates prior knowledge in the form of edge maps derived from the 50 and 20 TECU contours. The results demonstrate that TEC-EDGE consistently outperforms the model without prior knowledge (TEC-WO), achieving a reduction in average RMSE from 2.4341 to 2.3019 TECU and an improvement in average SSIM from 0.7411 to 0.7455 on the test set. Notably, TEC-EDGE demonstrates significant improvements in edge regions and performs better under both quiet and disturbed geomagnetic conditions, as well as during solar maximum compared to solar minimum periods. Our work demonstrates a new possibility of applying deep learning with prior knowledge to space weather research, particularly for problems of data lack.