マルチモーダルニューラルネットワークにより大気中CO2センシング性能を向上 （Multimodal Neural Networks Improve Atmospheric CO2 Sensing）

2026-05-15 合肥物質科学研究院（HFIPS）

Time series of daily mean CO₂ column concentrations from MM-LHRNet and NLSM. Circles denote the daily mean values. Shaded bands indicate mean ± 1σ (Image by XIONG Hao).

＜関連情報＞

• https://english.hf.cas.cn/nr/rn/202605/t20260515_1159457.html
• https://www.sciencedirect.com/science/article/abs/pii/S0925400526005514

レーザーヘテロダイン放射測定法を用いた高精度二酸化炭素カラムセンシングのためのマルチモーダルニューラル融合 Multi-modal neural fusion for accurate carbon dioxide column sensing using laser heterodyne radiometry

Hao Xiong, Zhao Chen, Chunyan Sun, Guishi Wang, Kun Liu, Xiaoming Gao
Sensors and Actuators B: Chemical  Available online: 14 April 2026
DOI:https://doi.org/10.1016/j.snb.2026.139973

Highlights

• A novel multimodal neural fusion network is proposed for laser heterodyne sensing.
• The method eliminates dependence on a priori concentration profiles.
• A CO₂ column relative precision of approximately 0.11% is achieved.
• Retrieval speed is improved by more than three orders of magnitude.

Abstract

Laser heterodyne radiometer (LHR) provides high spectral resolution and compact size for atmospheric greenhouse gas column sensing, yet traditional fully physical retrieval algorithms are limited by high computational cost and strong dependence on prior concentration profiles. To address these limitations, we propose a novel multimodal neural network (MM-LHRNet) for CO₂ column retrieval based on LHR observations. In atmospheric CO₂ monitoring experiments conducted in Hefei, China, the proposed method achieved a standard deviation of 0.49 ppm (corresponding to an precision of approximately 0.11%), significantly outperforming the 1.09 ppm (corresponding to an precision of approximately 0.25%) obtained using the traditional nonlinear least-squares method (NLSM). Furthermore, MM-LHRNet eliminates the need for CO₂ prior concentration profiles and improves retrieval speed by more than three orders of magnitude, substantially reducing computational costs. Experimental results show that MM-LHRNet provides a high-precision and efficient retrieval strategy for LHR-based greenhouse gas column measurements, and further offers a novel technical pathway for future real-time atmospheric greenhouse gas retrieval.