基于机器学习的卫星温室气体遥感反演方法研究

doi:10.3964/j.issn.1000-0593(2025)10-2983-09

摘要
参考文献
相关文章 (15)

全文: PDF (18726 KB)
输出: BibTeX | EndNote (RIS)

摘要：温室气体卫星遥感为气候变化研究提供了重要的数据支撑，精确获取温室气体浓度的时空分布是有效开展碳排放核算的关键。本研究旨在建立基于机器学习的卫星温室气体遥感反演模型，实现温室气体垂直干空气混合比柱浓度（XCO₂和XCH₄）的快速高精度反演。首先，基于大气辐射传输模型和全球总碳柱观测网络（TCCON）的观测数据构建训练样本集；其次，采用一维卷积神经网络（1DCNN）机器学习方法，并融合自适应矩估计（Adam）和贝叶斯优化（BO）算法训练构建卫星温室气体浓度反演模型，并与随机森林（RF）和反向传播神经网络（BPNN）的训练结果进行对比，结果表明基于1DCNN的反演模型对测试样本集反演XCO₂和XCH₄的相关系数分别可达0.953 1和0.957 3，性能优于RF和BPNN。最后，利用国产高分五号B卫星（GF-5B）搭载的主要温室气体监测仪（Greenhouse gases Monitoring Instrument，GMI/GF-5B）高光谱观测数据，反演了2022年至2024年全球XCO₂和XCH₄，通过与TCCON 站点观测浓度数据对比验证结果表明：卫星反演XCO₂和XCH₄与9个地面站点观测数据的相关性分别优于0.938 5和0.959 8，其中，XCO₂误差整体优于2 ppm（99.15%的验证样本误差优于1.5 ppm），XCH₄误差整体优于10 ppb。

关键词：温室气体；卫星遥感反演；机器学习；GMI/GF-5B

Abstract：Greenhouse-gas satellite remote sensing provides vital data support for climate-change research. Accurately obtaining the spatiotemporal distribution of greenhouse gas concentrations is key for effective carbon emission accounting. This study aims to develop a satellite-based greenhouse-gas retrieval model using machine-learning methods, enabling rapid and high-precision inversion of column-averaged dry-air mixing ratios (XCO₂ and XCH₄). First, a training dataset was constructed using an atmospheric radiative-transfer model combined with measurements from the Total Carbon Column Observing Network (TCCON). Next, a one-dimensional convolutional neural network (1D CNN) was employed as the machine-learning method. The model leveraged Adaptive Moment Estimation (Adam) and Bayesian Optimization (BO) during training. Its performance was compared to that of Random Forest (RF) and Backpropagation Neural Network (BPNN) models. Results showed that the 1D-CNN model achieved correlation coefficients of 0.953 1 for XCO₂ and 0.957 3 for XCH₄ on the test dataset, outperforming both RF and BPNN. Finally, high-spectral-resolution observations from the Chinese GF-5B satellite's Greenhouse gases Monitoring Instrument (GMI/GF-5B) were used to retrieve global XCO₂ and XCH₄ for 2022—2024. Comparisons with data from nine TCCON stations demonstrated strong agreement: the correlation coefficients exceeded 0.938 5 for XCO₂ and 0.959 8 for XCH₄. Overall retrieval errors were within 2 ppm for XCO₂ (with 99.15% of validation samples showing errors below 1.5 ppm) and within 10 ppb for XCH₄.

Key words：Greenhouse gases; Satellite remote sensing retrieval; Machine learning; GMI/GF-5B

收稿日期: 2025-02-17 修订日期: 2025-06-11

中图分类号:

X511

基金资助: 国家重点研发计划课题（2023YFF0714703），国家自然科学基金项目（42271403），安徽省高校协同创新项目（GXXT-2022-001）资助

通讯作者: 邹铭敏 E-mail: zoumm@ahu.edu.cn

作者简介: 盛书丽，女，1996年生，安徽大学物质科学与信息技术研究院硕士研究生 e-mail: 17611081211@163.com

引用本文:

盛书丽，邹铭敏，刘天奇，程泳萍，陈子郑，王旭文. 基于机器学习的卫星温室气体遥感反演方法研究[J]. 光谱学与光谱分析, 2025, 45(10): 2983-2991.
SHENG Shu-li, ZOU Ming-min, LIU Tian-qi, CHENG Yong-ping, CHEN Zi-zheng, WANG Xu-wen. Research on Satellite Greenhouse Gas Remote Sensing Retrieval Methods Based on Machine Learning. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(10): 2983-2991.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2025)10-2983-09 或 https://www.gpxygpfx.com/CN/Y2025/V45/I10/2983

[1] YANG Xiao-yu, WANG Zhong-ting, PAN Guang, et al（杨晓钰, 王中挺, 潘光）. Journal of Atmospheric and Environmental Optics（大气与环境光学学报）, 2022, 17（6）: 581.
[2] Liu Yi, Lu Daren, Chen Hongbin, et al. Remote Sensing Technology and Application, 2011, 26（2）: 247.
[3] Lopez F P A, Zhou G, Jing G, et al. Remote Sensing, 2022, 14（11）: 2622.
[4] WANG Lei, HUAN Ke-wei, LIU Xiao-xi（王磊, 宦克为, 刘小溪）. National Analytical Chemistry Bulletin（分析化学）, 2022, 50（12）: 1918.
[5] PENG Hao-jie, ZHOU Yang, HU Xiao-fei（彭豪杰, 周杨, 胡校飞）. Journal of Remote Sensing（遥感学报）, 2023, 27（2）: 430.
[6] MIAO Yun-fei, ZOU Ming-min, SHENG Shu-li, et al（缪云飞, 邹铭敏, 盛书丽, 等）. China Environmental Science（中国环境科学）, 2023, 43(S1): 20.
[7] Li K, Bai K, Jiao P, et al. Remote Sensing of Environment, 2024, 304: 114039.
[8] LI Jing-bo, ZHANG Ying, GAI Rong-li（李静波, 张莹, 盖荣丽）. China Environmental Science（中国环境科学）, 2023, 43（4）: 1499.
[9] Zhao Z, Xie F, Ren T, et al. Journal of Quantitative Spectroscopy and Radiative Transfer, 2022, 278: 108006.
[10] Zhang W, Wang Z, Li T, et al. Atmosphere, 2025, 16(3): 238.
[11] Huang X, Deng Z, Jiang F, et al. Geophysical Research Letters, 2024, 51(8): e2023GL107536.
[12] Gong X, Zhang Y, Fan M, et al. Atmosphere, 2024, 15(1): 118.
[13] YE Han-han, WANG Xian-hua, WU Shi-chao, et al（叶函函, 王先华, 吴时超, 等）. Journal of Atmospheric and Environmental Optics（大气与环境光学学报）, 2021, 16（3）: 231.
[14] Liu Shuanghui, Li Xiaoying, Cao Xifeng. Remote Sensing Technology and Application, 2022, 37（2）: 436.
[15] LI Yun-fei, LI Jun, HE Lin（李云飞, 李军, 贺霖）. Journal of Remote Sensing（遥感学报）, 2022, 26（8）: 1614.
[16] TANG Yong-sheng, CHEN Zheng-guang（唐永生, 陈争光）. Spectroscopy and Spectral Analysis（光谱学与光谱分析）, 2019, 41（3）: 892.
[17] Saberioon M, Gholizadeh A, Ghaznavi A, et al. Computers and Electronics in Agriculture, 2024, 227: 109494.
[18] Wu Lijun, Wang Baoxing, Zhang Lei, et al. Journal of Near Infrared Spectroscopy, 2020, 28（3）: 153.
[19] Yoshida Y, Kikuchi N, Morino I, et al. Atmospheric Measurement Techniques, 2013, 6（6）: 1533.
[20] GAO Xiao-hong, YANG Yang, ZHANG Wei, et al（高小红, 杨扬, 张威, 等）. Remote Sensing Technology and Application（遥感技术与应用）, 2015, 30（5）: 849.
[21] Zhao R, Liu L, Liu X, et al. Nuclear Science and Techniques, 2025, 36（2）: 29.
[22] LI Qin-qin, WANG Xian-hua, YE Han-han, et al（李勤勤, 王先华, 叶函函, 等）. Acta Optica Sinica（光学学报）, 2020, 40（6）: 26.