针对以往利用高光谱数据来来反演土壤有机质(SOM) 的可行性与可靠性, 结合微分处理对光谱数据信息提取的高效性, 提出了直接对多光谱遥感影像进行微分处理就可得出SOM建模研究, 旨在为今后SOM速测提供参考。 采用Landsat 8_OLI 多光谱遥感影像数据, 对多光谱遥感影像进行辐射定标、 几何校正、 大气校正、 镶嵌和裁剪, 运用IDL软件对影像进行一阶微分处理和二阶微分处理, 发现一阶微分图像能够更好地表达地物的真实情况, 更好地区别水体与土壤。 原始遥感影像包含大量的信息其中还包括噪声, 通过微分处理后的遥感影像剔出了原始影像中反射率值突兀变化的部分。 在研究区采用五点法采集土壤样品。 室内实验用重铬酸钾氧化-容量法测得SOM数据。 多光谱数据结合地面实测SOM数据, 分析SOM与多光谱数据反射率的关系, 发现一阶微分处理后的遥感数据与SOM含量的相关性存在敏感波段, 说明一阶微分处理可以将原始遥感图像数据在多光谱范围内的一些隐含的土壤有机质信息释放出来。 选取相关性高的数据建立基于原始遥感数据、 一阶微分数据、 二阶微分数据的单波段多光谱线性模型和多波段多光谱线性模型, 选取最优模型来估算和反演土壤有机质含量。 结论如下: (1) 通过对原始影像进行微分处理发现, 微分处理后的影像变化明显, 一阶微分处理的影像噪声降低, 更加突出了影像中土壤有机质隐藏的信息。 二阶微分处理的影像抑制了土壤有机质信息。 (2) 原始遥感影像各波段数据对土壤有机质含量的相关性较低, 一阶微分处理后的遥感影像数据反映出土壤有机质敏感波段即部分波段数据相关性明显高于原始数据, 二阶微分处理后的遥感影像各波段数据对土壤有机质含量的相关性较弱。 (3) 多波段建模效果要优于单波段建模; 一阶微分多波段模型预测精度最优, 其模型的决定系数和模型拟合的决定系数分别为0.898和0.854, 该模型对估算研究区内的SOM含量效果较好; 综合比较了单波段模型和多波段模型的拟合精度, 发现无论在单波段模型还是多波段模型一阶微分处理后的模型都具有更好的预测能力。 (4) 基于一阶微分多波段模型对研究区SOM进行反演, 反演结果与实际情况相符合, 对干旱区SOM含量制图提供了切实可行的方法和参考。

In this paper, according to the feasibility and reliability of using the hyperspectral data to retrieve SOM from hyperspectral data, combined with the high efficiency of differential processing in extracting spectral information, a new method based on differential algorithm for soil organic matter modeling In this study, the content of soil organic matter can be obtained by differentiating the multi-spectral remote sensing images directly, which aims to provide the direction for the future study of soil organic matter rapid measurement is proposed. In this paper, Landsat 8_OLI multi-spectral remote sensing image data is used to perform the radiation calibration, geometric correction, atmospheric correction, mosaic and cropping of multi-spectral remote sensing images. The first order differential and second order differential are processed by IDL software. The image can better express the real situation of the object. The first-order differential image can distinguish the water body from the soil better. The original remote sensing image contains a lot of information, including the noise. The differential image processed by the remote sensing image excludes the original image In the study area, five-point method was used to collect soil samples, indoor potassium dichromate oxidation-volume method to measure soil organic matter data, and multispectral data was used to analyze soil organic matter data from the ground to analyze soil organic matter It is found that there is a sensitive band in the correlation between the first-order differential data and soil organic matter content, indicating that the first-order differential processing can transform the original remote sensing image data in some obscure soil in the multi-spectral range. Organic information is released; select a high correlation number established based on the raw remote sensing data, first-order differential data, single-band multi-spectral data of the second order differential linear and multi-band multi-spectral linear model, and select the best model to estimate soil organic matter content retrieval. The main conclusions are as follows: (1) By differentiating the original image, it is found that the image after differential processing changes obviously and the image noise of first-order differential processing decreases, which further highlights the hidden information of soil organic matter in the image. The second-order differential processing suppresses soil organic matter information. (2) The data of the original remote sensing images have a low correlation with soil organic matter content. The data of the first-order differential treatment reflect the correlation of the soil organic matter sensitive band, that is, the partial band data, and the second-order differential processing after the remote sensing images of each band data on soil organic matter content of the correlation is weak. (3) Multi-band modeling is superior to single-band modeling, and the first-order differential multiband model has the best prediction accuracy. The model’s coefficient of determination and the coefficient of model fitting are 0. 898 and 0. 854 respectively. The soil organic matter content in this region was well estimated. The fitting accuracy of single-band model and multi-band model was compared comprehensively. It was found that both the single-band model and the first-order differential model had better prediction ability. (4) Based on the first-order differential multi-band model, the inversion of SOM in the study area was carried out. The inversion result is in accordance with the actual situation, which provides a practical method and reference for the mapping of soil organic matter content in arid area.