塔基遥感观测数据作为连接卫星遥感数据和地面生态参数的“桥梁”，发挥着至关重要的作用。其中日光诱导叶绿素荧光（SIF, Sun-Induced ChlorophyⅡ Fluorescence）能够很好的反映植物的生长、病害及受胁迫状态，已成为塔基生态遥感反演的核心参数 之一。然而，SIF 反演强烈依赖于高光谱观测数据的大气吸收波段，对大气辐射传输极 其敏感，塔基传感器与地表之间大气吸收和散射作用会导致观测信号不能表示真实的 地表信息，进而影响 SIF 反演精度。 因此，本文以大满农田生态系统观测站为研究区域，基于长时序塔基观测数据和 模型模拟数据，聚焦塔基平台氧气吸收波段大气辐射传输校正的需求，深入研究了基 于塔基光谱观测数据的关键大气参数反演方法，旨在为塔基平台 SIF和光谱观测提供大 气校正方法支持，对提高塔基生态遥感观测数据的定量化精度具有重要意义。论文主 要研究内容以及结论如下：（1）基于辐射传输模型系统研究了氧气吸收波段的大气透过率敏感性。针对 SIF 反演过程中常用的 O2-B 和 O2-A 波段，通过计算大气透过率相对变化率的方法对各参 数进行敏感性分析。结果表明，各参数影响程度从大到小依次为：太阳天顶角/观测天 顶角（SZA/VZA）、气溶胶光学厚度（AOD）、气压、温度。O₂-A 对各参数的敏感性整 体高于 O₂-B，上行辐射传输过程的敏感性整体高于下行辐射传输过程。（2）建立了大视场角观测模式“等效”辐射传输路径长度的简化模型。在“等效” 辐射传输路径的理论基础上，耦合大气辐射传输过程和地表二向反射特征（BRDF）， 通过加权平均的方式计算等效大气透过率，并建立了简化的等效辐射传输路径长度估 算模型。结果表明：大视场角观测模式的大气透过率对环境因素的敏感性。考虑BRDF 和朗伯体表面时估算的路径长度存在显著差异，最大差距达到21m，强调在模型构建过 程中考虑植被反射特性的重要性。同时，考虑到实际观测过程中地表气压实时波动带 来的影响，建立了地表气压对路径长度影响的校正模型，进一步准确估算“等效”辐 射传输路径长度。（3）基于塔基高光谱观测数据和机器学习方法，构建了AOD反演模型。通过分析太阳辐照度光谱对AOD变化的敏感性及光谱特征，发现通过近红和红光波段辐照度构建双通道比值指数可敏感追踪AOD变化。然后，对高光谱数据进行特征提取和重要性 筛选，与AOD实测值参与机器学习模型训练。最后，系统评估了随机森林，支持向量 机，人工神经网络三种模型的反演精度：R 2分别为0.950、0.936、0.947。并选择表现最佳的随机森林模型，对AOD反演模型预测值的日变化规律和时间扩展性进行了验证分析，均表现出良好的预测性能。 最后，本文基于模拟数据构建了以“等效”辐射传输路径长度和 AOD 为自变量的上行/下行大气透过率查找表。针对上行大气透过率计算问题，系统评估了程辐射在O2- B 和 O2-A 对上行辐亮度的贡献，并构建了简化的程辐射校正模型以提高上行大气透过率的计算精度。在实际观测应用中，通过实时反演获得相关参数并输入查找表，可以快速获取大气透过率，从而对观测数据进行同步大气校正。通过对比分析校正前后表观反射率和 SIF 日变化趋势对大气校正效果进行验证，结果均表现出良好正精度。综上所述，本文系统研究了塔基光谱观测的大气辐射传输过程，构建了塔基光谱观测的关键大气参数反演模型和方法，为塔基平台 SIF和光谱数据的大气校正提供了理论与方法支持，对提高我国塔基光谱观测与参数反演的定量化精度具有重要意义。

Tower-based remote sensing data serves as a critical bridge connecting satellite remote sensing data with ground ecological parameters. Among these parameters, sun-induced chlorophyll fluorescence (SIF) has emerged as a core parameter in tower-based ecological remote sensing inversion due to its strong correlation with vegetation growth, disease, and stress status. However, SIF retrieval heavily depends on atmospheric absorption bands in hyperspectral observations and is highly sensitive to atmospheric radiative transfer. The atmospheric absorption and scattering effects between tower-based sensors and the Earth's surface distort observed signals from true surface information, thereby compromising SIF retrieval accuracy. This study selected the Daman Farmland Ecosystem Observation Station as the research area. By integrating long-term tower-based observations and model simulations, we focused on atmospheric radiation correction requirements for oxygen absorption bands in tower platforms. We systematically investigated inversion methods for key atmospheric parameters using tower based spectral data, aiming to provide atmospheric correction support for SIF and spectral observations while enhancing the quantitative accuracy of tower-based ecological remote sensing data. The main research contents and conclusions are as follows: (1) Systematic investigation of atmospheric transmittance sensitivity in oxygen absorption bands. For the O₂-B and O₂-A bands commonly used in SIF retrieval, sensitivity analysis was conducted through relative transmittance variation rates. Results showed parameter sensitivity in descending order: solar zenith angle/view zenith angle (SZA/VZA), aerosol optical depth (AOD), atmospheric pressure, and temperature. O₂-A demonstrated higher overall sensitivity than O₂-B, with upwelling radiation processes exhibiting greater sensitivity than downwelling processes. (2) Development of a simplified model for equivalent radiation path length under large field-of-view observation. Building on equivalent radiation path theory, we coupled atmospheric radiative transfer with surface bidirectional reflectance distribution function (BRDF) characteristics to calculate equivalent atmospheric transmittance through weighted averaging. The results revealed significant differences (up to 21 m) in path length estimation between BRDF-considered and Lambertian surface assumptions, highlighting the necessity of incorporating vegetation reflectance characteristics. Additionally, we established a correction model addressing real-time surface pressure fluctuations to enhance equivalent path length estimation accuracy. (3) Construction of an AOD retrieval model using tower-based hyperspectral data and machine learning. Analysis of solar irradiance spectral sensitivity identified dual-channel ratio indices (near-infrared/red bands) as effective AOD tracers. After feature extraction and importance screening of hyperspectral data, three machine learning models (Random Forest, Support Vector Machine, and Artificial Neural Network) were evaluated, achieving R² values of 0.950, 0.936, and 0.947 respectively. The optimal Random Forest model demonstrated robust performance in daily variation analysis and temporal scalability verification. Finally, we developed look-up tables for upwelling/downwelling atmospheric transmittance using equivalent radiation path length and AOD as variables. For upwelling transmittance calculation, path radiance contributions in O₂-B and O₂-A bands were systematically evaluated, leading to a simplified path radiance correction model. Practical implementation through real-time parameter inversion and look-up table referencing enabled rapid atmospheric transmittance acquisition and synchronous data correction. Validation through comparative analysis of apparent reflectance and SIF diurnal variations confirmed the method's high correction accuracy. In conclusion, this research systematically investigates atmospheric radiative transfer processes in tower-based spectral observations, develops key atmospheric parameter inversion models and methodologies, and provides theoretical and technical support for atmospheric correction of tower-based SIF and spectral data. These advancements significantly enhance the quantitative accuracy of tower-based spectral observations and parameter retrievals in China.