珠海一号高光谱遥感的表层土壤有机质含量反演方法

Inversion of Topsoil Organic Matter Content by Hyperspectral Remote Sensing of Zhuhai-1

表层土壤有机质含量影响土壤光谱特性且在空间分布上呈异质性。采用光谱分辨率高、波段连续性强的高光谱遥感影像反演区域表层土壤有机质空间分布状况,可为精准农业提供科学管理依据。针对以往方法很少基于高光谱影像大尺度反演土壤表层有机质含量,以安徽省淮南市舜耕山以南的三和镇、曹庵镇为研究区,探索珠海一号高光谱遥感反演表层土壤有机质含量的方法。研究结果表明,研究区表层土壤有机质含量与珠海一号高光谱影像原始光谱反射率最大相关波段为656 nm(r=-0.680);采用小波包分解原始光谱后,低频分量和高频分量与表层土壤有机质的最大相关性均有所提高,低频分量最大相关波段为656 nm(r=-0.797),高频分量最大相关波段为700 nm(r=-0.804)。采用多元线性回归对原始光谱、小波包分解低频分量、小波包分解高频分量建立土壤有机质预测模型取得了良好的效果,R 2分别为0.747、0.770、0.789。依据小波包分解的低频分量、小波包分解的高频分量建立的基于高斯核变换的支持向量回归模型预测效果优于多元线性回归模型,预测值与实测值更接近。研究结果为开展大尺度遥感反演表层土壤有机质工作提供了新方法、新思路。

The content of topsoil organic matter affects soil spectral characteristics and presents heterogeneity in spatial distribution.The inversion of regional spatial distribution of surface soil organic matter using hyperspectral remote sensing images with high spectral resolution and strong band continuity can provide scientific management basis for precision agriculture.Since there are few inversion methods of soil surface organic matter content based on a large scale hyperspectral images,this paper takes Sanhe town and Caoan town,south of Shungeng mountain,Huainan city,Anhui province,as the research area to explore the method of retrieving surface soil organic matter content by hyperspectral image of Zhuhai-1.The results show that the maximum correlation band between the surface soil organic matter content and the spectral reflectivity of the hyperspectral image of Zhuhai-1 is 656 nm(r=-0.680).After the wavelet packet decomposition of the original spectrum,the maximum correlation between the low-frequency component and the high-frequency component and the topsoil organic matter is improved,and the maximum correlation band of the low-frequency component is 656 nm(r=-0.797),and the maximum correlation band of the high-frequency component is 700 nm(r=-0.804).Multiple linear regression is used to establish the soil organic matter prediction model for the spectrum,wavelet packet decomposition of low-frequency component and wavelet packet decomposition of high-frequency component,and good results are achieved,where,R 2 is 0.747,0.770 and 0.789,respectively.The prediction effect of the support vector regression model based on Gaussian kernel transformation is better than that of the multiple linear regression model,and the predicted value is closer to the measured value.The results provide new methods and ideas for large-scale remote sensing inversion of surface soil organic matter.