干旱半干旱区土壤含盐量和电导率高光谱估算

以新疆南疆地区渭干河-库车河三角洲绿洲不同盐渍化程度的土壤为研究对象, 将土壤样品的含盐量与电导率数据和使用ASD FieldSpec Pro FR光谱仪测得土壤样品高光谱数据作为本研究原始数据. 首先, 对原始土壤光谱反射率曲线进行Savitzky-Golay滤波以消除光谱曲线噪声可能引起的误差, 对得到的光谱曲线进行对数、 倒数等15种光谱变换; 其次, 对土壤样品含盐量与电导率关系进行分析, 并通过两者与土壤光谱反射率不同光谱变换形式的相关性比较分析, 遴选出一阶微分、 对数倒数一阶微分、 连续统去除和连续统去除一阶微分四种相关性较好的变换形式; 最后, 以此变换形式对含盐量和电导率进行建模, 并对二者进行了高光谱估算精度的比较. 结果表明: 渭干河-库车河三角洲绿洲的含盐量与电导率的相关性较高, 达到0.9975; 电导率与四种不同光谱变换形式之间的相关性要优于含盐量, 特别在一些土壤盐渍化的敏感波段尤为突出; 无论是含盐量还是电导率, 多元逐步回归模型均优于一元线性模型, 且以电导率建立的两种回归模型的决定系数均高于含盐量. 研究表明, 土壤电导率对高光谱信息的反应比含盐量更敏感, 同时, 以电导率建模的估算精度比含盐量更高. 因此, 以电导率替代含盐量进行土壤盐渍化高光谱估算研究是一种精度更高、 速度更快的方法, 可为提高土壤盐渍化高光谱估算提供理论依据.

Estimating the soil salt content and electrical conductivity in semi-arid and arid areas by using hyperspectral data

The objective of this study is to ascertain the mechanisms of hyperspectral remote sensing monitoring soil salinization, which is of great importance for improving the accuracy of hyperspectral remote sensing monitoring. The soil samples were collected from the delta oasis between the Weigan River and the Kuqa River in the north rim of the Tarim basin, which were taken to laboratory for measuring the salt content (SC) and electrical conductivity (EC). Hyperspectral images were obtained via ASD FieldSpec Pro FR and were regarded as basic data sources. Hyperspectral images were deal with Savitzky-Golay filtering for noise smoothing, and were transformed via 15 different approaches, including logarithm, inversion, root mean squares, continuum removed and first order differential, etc. Through comparative analysis of correlations with SC, EC and different approaches, it is found that the following four approaches have better correlations: first order differential, log-inverse first order differential, continuum removed and continuum removed first order differential. Relationship between SC and EC were studied. Correlations between hyperspectral indices and SC and EC were analyzed. The estimation accuracy of SC through hyperspectral technique was compared with that of EC. Results show that there is significant correlation between SC and EC in delta oasis between the Weigan River and the Kuqa River. The correlations between EC and first order differential, log-inverse first order differential, continuum removed, and continuum removed first order differential are better than those of SC, which is significant in some sensitive bands with soil salinization. The coefficient of determination (R²) of model is established by using stepwise multivariate linear regression, which is better than that established by using simple linear regression. Therefore, the responses of high spectral information to EC are more sensitive than those of high spectral information to SC. Accuracy of EC predicted from high spectral data is higher than that of SC estimated from high spectral data. This study will provide a theoretical basis to improve hyperspectral remote sensing estimating accuracy of soil salinization.