| 摘要: |
| 【目的】 提高地—气能量交换参数地表温度（Land Surface Temperature，LST）、地表发射率（Land Surface Emissivity，LSE）、大气水汽含量（Water Vapor Content，WVC）和近地表空气温度（Near Surface Air Temperature，NSAT）的反演精度。【方法】 文章提出了基于人工智能（Artificial Intelligence，AI）的热红外遥感多参数一体化反演范式理论和技术，通过物理逻辑推理证明深度学习输入和输出参数之间能够构造闭合的物理方程组使得AI遥感多参数一体化反演具有物理意义和可解释性，即输入变量能够唯一确定输出变量。对于输入变量和输出变量之间具有强相关性的情况，参数可以高精度地直接反演；对于只存在弱相关性的情况，加入强相关的先验知识可以提高反演精度。【结果】 物理逻辑推理表明热红外遥感多参数反演至少需要4个热红外波段构建4个辐射方程组以确保输入变量唯一确定输出变量。根据输入变量和输出变量之间的因果关系，确定了一体化反演的两种参数反演技术模式，即“直接同步反演”与“迭代反演”。利用MODIS数据5波段（27、28、29、31和32波段）对4个参数进行一体化反演应用示范。反演结果显示，LST的平均理论误差在0.5 K以下，发射率在0.008以下，WVC误差在0.1 g/cm2以下，NSAT反演应用平均误差在2.0 K以下。【结论】 直接同步反演和迭代反演的合理应用可最大化多参数的反演精度，同时可以优化卫星传感器设计，因此基于AI的热红外遥感多参数一体化反演理论提出和技术实现对推动热红外遥感参数反演研究具有里程碑意义。 |
| 关键词: 人工智能 热红外遥感 多参数反演 一体化反演 |
| DOI：10.12105/j.issn.1672-0423.20240305 |
| 分类号: |
| 基金项目:风云卫星应用先行计划“风云全天候陆表温度时空融合数据集研制及应用”（FY-APP-2022.0205）；第二次青藏高原综合科学考察研究“地表温度和土壤水分参数AI反演与融合算法研究”（2019QZKK0206XX-02） |
|
| Thermal infrared remote sensing multi-parameter AI integrated retrieval paradigm theory and technology |
|
Mao Kebiao1,2,3, Wang Han1, Yuan Zijin1, Shi Jiancheng4, Qin Zhihao5, Wu Shengli6
|
|
1.Institute of Agricultural Resources and Regional Planning，Chinese Academy of Agricultural Sciences/State Key Laboratory of Efficient Utilization of Arid and Semi-arid Arable Land in Northern China，Beijing 100081，China;2.School of Physics and Electronic-Electrical Engineering，Ningxia University，Yinchuan 750021，NingXia，China;3.State Key Laboratory of Remote Sensing Science，Aerospace Information Research Institute，Chinese Academy of Science，Beijing 100094，China;4.National Space Science Center，Chinese Academy of Sciences，Beijing 100190，China;5.Nanning Normal University，Nanning，530100;6.National Satellite Meteorological Center，Beijing 100081，China
|
| Abstract: |
| Purpose To improve the inversion accuracy of land-atmosphere energy exchange parameters such as Land Surface Temperature（LST），Land Surface Emissivity（LSE），Water Vapor Content（WVC），and Near-Surface Air Temperature（NSAT）.Method This study proposed an AI-based thermal infrared remote sensing multi-parameter integrated inversion paradigm theory and technology. Through physical logical reasoning，it was demonstrated that a closed set of physical equations can be constructed between the input and output parameters of deep learning，which ensured the physical meaning and interpretability of AI remote sensing multi-parameter integrated inversion. This meant that the input variables could uniquely determine the output variables. For cases with strong correlations between input and output variables，the parameters could be directly inverted with high accuracy. For cases with weak correlations，incorporating strongly correlated prior knowledge could improve the inversion accuracy.Result Physical-logical reasoning indicated that thermal infrared remote sensing multi-parameter inversion required at least four thermal infrared bands to construct four radiation equation sets，which ensured the unambiguous determination of output variables by input variables. Based on the causal relationship between input and output variables，two parameter inversion techniques inversion were identified for integrated inversion："direct synchronous retrieval" and "iterative retrieval". The integration of MODIS data from 5 bands（27，28，29，31 and 32）was applied for the inversion of the four parameters. The inversion results showed that the average theoretical error of LST was less than 0.5 K，the emissivity error was less than 0.008，the atmospheric water vapor error was less than 0.1 g/cm2，and the average error of the NSAT inversion application was less than 2.0 K.Conclusion The reasonable application of direct synchronous inversion and iterative inversion can maximize the inversion accuracy of multiple parameters，making the AI-based thermal infrared remote sensing multi-parameter integrated inversion a milestone in the history of thermal infrared remote sensing parameter inversion. |
| Key words: Artificial Intelligence thermal infrared remote sensing multi-parameter retrieval integrated retrieval |
