土壤盐渍化，一个涉及农业、资源、环境及生态的重要问题，其覆盖面积每年都在快速增加，在全球范围内越来越备受关注，我国同样面临着严峻的土壤盐渍化问题。同时，随着我国社会经济的发展，对耕地的需求量越来越大，盐渍化土地作为一种后备的耕地资源，其土壤全盐含量及盐离子含量等详细信息的获取就显得愈加重要。传统的土壤盐渍化信息采集，已经无法满足我们对于盐渍化土地实时监测、改良、治理等一系列需求。随着遥感科学与技术的发展，高光谱遥感为实现土壤全盐含量及盐离子含量反演成为可能。

本研究以宁夏北部引黄灌区表层土壤为例，选取土壤盐渍化情况最严重的冬季雪融化后春季播种灌溉之前的返盐期这一特定时间，从室内控制实验高光谱数据、野外土壤高光谱数据以及影像数据三种高光谱数据源出发，在系统测定并提取全盐及盐离子光谱特征基础上，提出耦合室内控制实验光谱特征方法（CLSCM），针对全盐及不同盐离子含量构建系列反演方法，在影像数据上完成全盐及盐离子含量反演后，构建离子平衡约束下土壤盐化类型判断方法（IBCJM），实现土壤盐化类型科学有效判断。研究内容及结果如下：

（1）设计多层次室内控制实验，系统测定并提取了全盐及盐离子光谱特征。

针对缺乏对土壤中主要盐离子的光谱特征系统分析与提取的问题，根据研究区土壤及土壤盐渍化情况，选取五种盐离子（氯离子、硫酸根离子、钠离子、镁离子和钙离子）构成的六种类型盐结合三种主要的土壤类型，按照不同类型土壤、不同盐及不同的盐含量设计了多层次室内控制实验，共计模拟了270组土壤盐渍化情况，并测量了其高光谱数据；在分析不同全盐含量及不同盐离子的光谱特征基础上，将原始光谱数据进行了12种常用的光谱形式变换和9种指数变换，结合三种特征提取方法（特征重要性分析、相关分析及连续投影算法）系统地提取了三种土壤类型下全盐含量及盐离子含量的原始光谱及变化光谱的光谱特征。研究表明，提取特征光谱较为集中的区间为：土壤全盐含量，较为集中在950nm-1200nm、1800nm-2100nm和2300nm-2500nm；氯离子含量，较为集中在1400nm-1600nm、1900nm-2000nm和2200nm-2500nm；硫酸根离子含量，较为集中在950nm-1200nm、1600nm-1700nm和1800-1900nm；钠离子含量，较为集中在1800nm-2200nm和2300nm-2500nm；镁离子含量，较为集中在1800nm-2500nm；钙离子含量较为集中在1900nm-2000nm和2200nm-2500nm。

（2）提出耦合室内控制实验光谱特征方法（CLSCM），针对全盐及不同盐离子含量构建系列反演方法。

为获取高精度的土壤全盐及各离子含量反演方法，同时为解决将室内控制实验数据获取的精细的光谱特征有效应用于实际情况中问题，提出耦合室内控制实验光谱特征方法（CLSCM），将在室内控制实验中获取的全盐及各盐离子相关特征，与相同波段位置的土壤野外光谱数据分别耦合，使土壤野外光谱数据具有突出表达相关信息的能力。土壤野外原始光谱数据在经过CLSCM方法耦合后，进行全盐及各盐离子含量反演时，反演精度均得以提升，尤其是盐离子含量反演精度，使得R²从低于0.5提高到0.7及以上，RPD能从低于1.5提高到2及2以上。在验证CLSCM有效提高全盐及盐离子反演精度基础上，将室内控制实验光谱数据多种变化形式下的光谱特征与野外数据一一耦合，结合三种特征提取方法后，应用综合性能最优的随机森林回归，构建全盐及盐离子含量系列反演方法，并获取了全盐及各盐离子含量反演精度最高的方法。

（3）耦合光谱特征到影像数据，在反演土壤全盐及各盐离子含量基础上，构建离子平衡约束下土壤盐化类型判断方法（IBCJM）。

针对当前研究鲜有实现基于影像数据的土壤盐化类型科学有效的判断问题，将经过CLSCM耦合的土壤野外光谱数据特征耦合到影像数据中，参照土壤野外光谱全盐及盐离子含量反演方法，构建并获取基于耦合光谱特征的影像数据的全盐及盐离子含量反演最优方法，实现全盐及盐离子含量反演。根据盐离子平衡构造约束方程，在综合考虑各盐离子之间相互关系优化盐离子含量反演结果的基础上，实现土壤盐化类型及土壤中主要阳离子类型科学有效的判断。研究表明，应用IBCJM后能有效提高土壤盐化类型及土壤中主要阳离子类型判断精度，其中精度提高最多的是对土壤盐化类型的判断，应用IBCJM后准确度提高了0.13，精确度提高了0.18，召回率提高了0.17，F1分数提高了0.22。

本文围绕着基于多源高光谱数据的表层土壤全盐及盐离子含量反演方法研究，充分挖掘室内控制实验高光谱数据优势，在野外土壤高光谱数据上获取全盐及盐离子含量高精度反演的系列方法，实现了基于影像数据的土壤全盐含量反演、盐离子含量反演及盐化类型判断。为后续应用高光谱遥感手段进行土壤盐渍化实时监测、详细信息获取进行了尝试并提供了一系列的方法及研究基础。

Soil salinization is a serious problem involving agriculture, resources, environment and ecology. Its coverage area is increasing rapidly every year, and it has attracted more and more attention worldwide. China is also facing severe soil salinization problem. At the same time, with the development of China's social economy, the demand for cultivated land is increasing. As a reserve cultivated land resource, the acquisition of detailed information such as the soil salt content and salt ion content of salinized land becomes more and more important. The traditional collection of soil salinization information has been unable to meet our series of needs for real-time monitoring, improvement, and governance of salinized land. With the development of remote sensing, hyperspectral remote sensing has become possible to realize the inversion of soil salt content and salt ion content.

In this study, the topsoil of the Yellow River Diversion Irrigation District in northern Ningxia was taken as an example, and the salt-returning period was selected after the snow melted in winter and before sowing and irrigation in spring, which was the most serious time of soil salinization. Starting from three hyperspectral data sources of indoor control experiment hyperspectral data, field soil hyperspectral data and image data, on the basis of systematically measuring and extracting the spectral characteristics of soil salt and salt ions, a Couple Laboratory Spectral Characterization Method (CLSCM) was proposed and s series of inversion methods were constructed for soil salt and different salt ion contents. After the inversion of soil salt and salt ion content was realized on the image data, the type of soil salinization could be scientifically and effectively judged through the Ion Balance Constraint Judgment Method (IBCJM). The research content and results are as follows:

(1) A multi-level indoor control experiment was designed, and the spectral characteristics of soil salt and salt ions were systematically measured and extracted.

In view of the lack of systematic analysis and extraction of the spectral characteristics of the main salt ions in soil, this study selected six types of salts composed of five ions (chloride ion, sulfate ion, sodium ion, magnesium ion and calcium ion) combined with three main soil types according to the soil and soil salinization in the study area. A multi-level indoor control experiment was designed according to different types of soil, different types of salt and different soil salt content. A total of 270 groups of soil salinization conditions were simulated, and their hyperspectral data were measured. On the basis of analyzing the spectral characteristics of different soil salt contents and different salt ions, 12 kinds of common spectral form transformations and 9 kinds of exponential transformations were carried out on the raw spectral data. Combined with three feature extraction methods (Feature Importance analysis, Correlation analysis and Successive Projections Algorithm), the spectral characteristics of the raw spectrum and the change spectrum of soil salt content and salt ion content under three soil types were systematically extracted. Research shows that the extraction characteristic spectrum is more concentrated in the interval: soil salt content, which is more concentrated in 950nm-1200nm, 1800nm-2100nm and 2300nm-2500nm; the chloride ion content is more concentrated in 1400nm-1600nm, 1900nm-2000nm and 2200nm-2500nm; the sulfate ion content is more concentrated in 950nm-1200nm, 1600nm-1700nm and 1800-1900nm; the sodium ion content is more concentrated in 1800nm-2200nm and 2300nm-2500nm; the magnesium ion content is more concentrated in 1800nm-2500nm; the calcium ion content is more concentrated in 1900nm-2000nm and 2200nm-2500nm.

(2) The Couple Laboratory Spectral Characterization Method (CLSCM) is proposed, and a series of inversion methods are constructed for soil salt and different salt ion contents.

In order to obtain a high-precision inversion method of soil salt and each salt ion content, and at the same time to solve the problem of effectively applying the fine spectral characteristics obtained from the indoor control experimental data to the actual situation, the CLSCM was proposed. This method couples the relative characteristics of soil salt and each salt ion obtained in the indoor control experiment with the field spectral data at the same band position, so that the field spectral data has the ability to express relevant information prominently. After the field raw spectral data was coupled by the CLSCM method, the inversion accuracy was improved when the soil salt and salt ion contents were inverted. In particular, the inversion accuracy of salt ion content makes R² increase from less than 0.5 to 0.7 and above, and RPD can be increased from less than 1.5 to 2 and above. On the basis of verifying that CLSCM can effectively improve the inversion accuracy of soil salt and salt ions, the spectral characteristics of the indoor control experimental spectral data in various forms are coupled with the field data one by one. After combining the three feature extraction methods, the random forest regression with the best comprehensive performance was used to construct a series of inversion methods for soil salt and salt ion content. The method with the highest inversion accuracy for the content of soil salt and each salt ion was obtained.

(3) Coupling spectral features to image data, on the basis of realizing the inversion of soil salt content and salt ion content, the Ion Balance Constraint Judgment Method (IBCJM) was constructed.

For the current research, there was little scientific and effective judgment of soil salinization type based on image data. Coupling the field spectral data characteristics coupled by CLSCM into the image data, and referring to the field spectral soil salt and salt ion content inversion method, constructed and obtained the optimal method for soil salt and salt ion content inversion based on the image data coupled with spectral features. And realized soil salt and salt ion content inversion. The IBCJM was constructed, and the soil salinization type and the main cation type in the soil were scientifically and effectively judged on the basis of comprehensively considering the relationship between each salt ion and optimizing the salt ion content inversion results. The research shows that the application of IBCJM can effectively improve the judgment accuracy of soil salinization type and main cation types in soil. Among them, the most improved precision is the judgment of soil salinization type. After applying IBCJM, the accuracy is improved by 0.13, the precision is improved by 0.18, the recall rate is improved by 0.17, and the F1 score is improved by 0.22.

This paper focused on the research on the inversion method of soil salt content and salt ion content in topsoil based on multi-source hyperspectral data, and fully exploited the advantages of hyperspectral data in indoor control experiments, and obtained a series of methods for high-precision inversion of soil salt and salt ion content on field hyperspectral data. And the inversion of soil salt content, the inversion of salt ion content and the judgment of salinization type based on image data were realized. This study provides a certain method and research basis for the subsequent application of remote sensing methods for real-time monitoring of soil salinization and acquisition of detailed information.