【作者】 刘朔；

【导师】 余涛； 左小敏；

【作者基本信息】 中国地质大学 ， 地球物理学， 2025， 博士

【摘要】 电离层同化方法是通信、导航和空间天气预报等关键领域的重要工具,通过融合观测数据和物理模型,提供对电离层状态的最优估计。该方法依赖数据同化技术,其中误差协方差矩阵用于描述模型状态误差与观测误差之间的关系,从而优化数据融合的精度。电离层的水平和垂直相关性是构建误差协方差矩阵的核心要素,准确捕捉这些相关性能够显著提升电离层预测的准确性和可靠性。然而,传统建模方法在处理复杂的非线性电离层现象和大规模数据时存在局限。为此,近年来引入了深度学习技术,凭借其强大的非线性建模和大数据处理能力,深度学习能够从观测数据中自动提取复杂模式,显著提高模型的预测精度,并有效应对电离层的复杂动态变化。本研究聚焦于电离层的水平与垂直相关性,结合TEC(Total Electron Content,TEC)数据、COSMIC-2 GIS(Constellation Observing System for Meteorology,Ionosphere,and Climate-2 Global Ionospheric Specification)数据和GNSS(Global Navigation Satellite System)观测数据,并利用深度学习技术,系统分析电离层的空间相关特性,构建高精度的空间相关性模型,以优化误差协方差矩阵,并为提升电离层扰动预测能力提供理论支撑。首先,基于TEC数据,系统分析了电离层的水平方向变化趋势。研究发现,电离层的水平相关性在地方时12:00至14:00之间达到峰值,反映了太阳辐射在此时对电子密度的强烈驱动作用。此外,研究表明,电离层的纬向相关性强于经向,尤其在磁共轭点和赤道电离异常(EIA)区域,空间相关性显著增强。这一发现为构建电离层误差协方差矩阵提供了重要参考。在水平相关性研究后,针对数据同化在垂直方向上对误差协方差矩阵的需求,进一步利用COSMIC-2 GIS数据,探讨了电离层在不同高度层的相关性特征,特别是F层(300公里)和E层(120公里)的差异。结果显示,F层的水平相关性与TEC数据较为一致,呈现稳定的变化趋势;而E层在日出和日落时段表现出显著的双峰结构,特别是在日落时相关性明显增强。这一现象表明,E层对太阳辐射变化更加敏感,展现出其复杂的空间结构特征。基于对水平相关性的深入研究,本文还分析了电离层的垂直相关性,以满足数据同化中对垂直误差协方差矩阵的需求。研究表明,电离层的垂直相关性随高度变化显著。在低高度区域(100-200公里),垂直相关性较弱;而在中高度区域(200-300公里),垂直相关性显著增强。随着高度继续增加至高层(300公里以上),垂直相关性呈现出复杂的波动,尤其在极地和低纬度地区,表现出显著的地方时和季节效应。该研究为构建更为精确的电离层垂直误差协方差矩阵提供了理论依据。我们通过分析GPSTEC、COSMIC-2 GIS数据,探讨了电离层在不同地理区域的水平相关性,揭示了其随地方时、纬度和空间尺度的动态变化特征,为后续的垂直相关性分析奠定了基础。随后,基于COSMIC-2 GIS数据,进一步研究了电离层的垂直相关性,构建了描述不同高度层电子密度耦合关系的模型,为电离层数据同化中的误差协方差矩阵优化提供了新视角。然而,传统误差协方差矩阵建模在应对电离层复杂动态变化时具有局限性,特别是在精细化和实时变化需求方面。因此,本文引入生成对抗网络(Generative Adversarial Network,GAN)等深度学习方法,实现了电离层数据的端到端建模,提升了对电离层时空演变规律的捕捉,并生成了更为精确的误差协方差矩阵。在此基础上,本文还应用GAN模型对ROTI(Rate of Total Electron Content Index变化率指数)地图进行了建模研究。通过深度学习模型的引入,成功捕捉到了电离层扰动的时空演化特征,尤其在等离子体泡等异常区域表现显著。该模型不仅在处理复杂电离层数据上具备强大能力,还展现了在捕捉电离层扰动特征方面的潜力,为未来电离层扰动建模和空间天气预报提供了更为灵活且精确的技术支持。综上所述,本研究通过结合TEC和COSMIC-2 GIS数据,并借助深度学习模型,系统研究了电离层的水平与垂直空间相关性,构建了精确的误差协方差矩阵模型。这些研究成果为电离层数据同化、空间天气预报以及卫星通信和导航系统的可靠性提供了理论依据和技术支持。更多还原

【Abstract】 Ionospheric data assimilation is a crucial tool in fields such as communication,navigation,and space weather forecasting.By integrating observational data with physical models,it provides an optimal estimation of the ionospheric state.This method relies on data assimilation techniques,where the error covariance matrix describes the relationship between model state errors and observational errors,thereby optimizing the accuracy of data fusion.The horizontal and vertical correlations of the ionosphere are fundamental to constructing the error covariance matrix,and accurately capturing these correlations can significantly improve the precision and reliability of ionospheric predictions.However,traditional modeling approaches face limitations when dealing with the nonlinear nature of ionospheric phenomena and large-scale datasets.To address these challenges,deep learning techniques have been introduced in recent years.With their powerful nonlinear modeling capabilities and ability to process large volumes of data,deep learning methods can automatically extract complex patterns from observational data,greatly enhancing predictive accuracy and effectively handling the dynamic nature of the ionosphere.This study focuses on the horizontal and vertical correlations of the ionosphere,leveraging Total Electron Content(TEC)data,Constellation Observing System for Meteorology,Ionosphere,and Climate-2 Global Ionospheric Specification(COSMIC-2 GIS)data,and Global Navigation Satellite System(GNSS)observations.By employing deep learning techniques,we systematically analyze the spatial correlation characteristics of the ionosphere,construct high-precision spatial correlation models,optimize the error covariance matrix,and provide theoretical support for improving ionospheric disturbance predictions.First,based on TEC data,we analyze the temporal variations of horizontal ionospheric correlations.The results indicate that ionospheric horizontal correlation peaks between 12:00 and 14:00 local time,reflecting the strong driving effect of solar radiation on electron density during this period.Additionally,the study reveals that zonal correlations are stronger than meridional correlations,particularly in magnetic conjugate points and equatorial ionospheric anomaly(EIA)regions,where spatial correlation is significantly enhanced.These findings provide a valuable reference for constructing the ionospheric error covariance matrix.Following the analysis of horizontal correlation,we further explore the vertical correlation characteristics of the ionosphere using COSMIC-2 GIS data,particularly focusing on the differences between the F-region(～300 km)and the E-region(～120km).The results indicate that the horizontal correlation in the F-region closely aligns with TEC data,showing a stable trend,whereas the E-region exhibits a pronounced double-peak structure during sunrise and sunset,with a notably stronger correlation at sunset.This suggests that the E-region is more sensitive to solar radiation variations,displaying a more intricate spatial structure.Building upon the findings of horizontal correlation analysis,this study also investigates vertical ionospheric correlations to meet the demands of error covariance matrix modeling in data assimilation.The results demonstrate that vertical correlation varies significantly with altitude:weak in the lower ionosphere(100–200 km),increasing substantially in the middle ionosphere(200–300 km),and exhibiting complex fluctuations in the upper ionosphere(above 300 km).In particular,the polar and low-latitude regions display notable local time and seasonal effects.These insights provide theoretical guidance for constructing a more precise vertical error covariance matrix.By analyzing GNSS TEC,COSMIC-2,and GIS data,we examine ionospheric horizontal correlations across different geographic regions,revealing their dynamic variations with local time,latitude,and spatial scale,which lay the groundwork for subsequent vertical correlation analysis.Utilizing COSMIC-2 GIS data,we further investigate vertical ionospheric correlations and construct a model that characterizes the coupling relationships of electron density at different altitudes,offering a new perspective for optimizing the error covariance matrix in ionospheric data assimilation.However,traditional error covariance matrix modeling faces challenges in capturing the complex and dynamic nature of the ionosphere,particularly in terms of high-resolution modeling and real-time variability.To address these limitations,this study introduces Generative Adversarial Networks(GANs)and other deep learning methods to achieve end-to-end ionospheric data modeling.This approach enhances the ability to capture the spatiotemporal evolution of the ionosphere,leading to a more precise error covariance matrix.Furthermore,GAN models are applied to model the Rate of Total Electron Content Index(ROTI)maps,successfully capturing the spatiotemporal characteristics of ionospheric disturbances,particularly in equatorial plasma bubble(EPB)regions.The model demonstrates strong capabilities in processing complex ionospheric data and shows promising potential in capturing ionospheric disturbance features,providing a more flexible and precise technical foundation for future ionospheric disturbance modeling and space weather forecasting.In conclusion,this study integrates TEC and COSMIC-2 GIS data with deep learning models to systematically investigate the horizontal and vertical spatial correlations of the ionosphere.A refined error covariance matrix model is constructed,contributing valuable theoretical support to ionospheric data assimilation,space weather forecasting,and the reliability of satellite communication and navigation systems.更多还原