基于高光谱的土壤含水率预测模型

为定量分析土壤含水率与反射光谱特征之间关系，以便为土壤含水率速测提供理论依据。以安徽省阜阳市临泉县为研究区，以区域典型土壤类型—砂姜黑土为研究对象，将土壤样本分别过10和20目的尼龙筛，并设置土壤含水率梯度实验。采用9点移动平滑法结合一阶微分、反射率对数及其一阶微分三种数学变换方法对光谱曲线进行预处理，分析不同目数、不同含水率下的光谱特性差异；拟合分析变换后的样本光谱数据与含水率相关性，提取特征波段，建立土壤含水率多元线性回归预测模型。结果表明：对4种不同土地利用类型的砂姜黑土样本的反射率数据进行对数一阶微分变换后，土壤含水率和光谱数据的相关性明显提升，根据数学变换后提取的特征波段建立的多元线性回归预测模型的预测精度最好；光谱反射率与水分含量呈负相关关系；土壤光谱法反演水分含量时，基于过10目筛的土壤样本建立的预测模型拟合精度要优于过20目筛的土壤样本，R²最高为0.928。研究结果可以为精准农业管理提供极为关键的参数支撑。

Prediction model of soil moisture based on hyper-spectral technology

Quantitative analysis of the relationship between soil moisture content and spectral reflectance can provide theoretical basis for the rapid measurement of soil moisture. In this paper, lime concretion black soil as the typical soil type from Linquan County, Fuyang City, Anhui Province were passed through 10 and 20 mesh sieves respectively. Meanwhile gradient experiments of soil moisture content under characteristic absorption spectrum were set up. The spectral curves were pretreated by 9-point moving smoothing method, first order differential, reflective logarithm and first order differential of reflective logarithm. Then difference of spectral characteristics were analyzed under different mesh number and soil moisture content. The correlation between the transformed sample spectral data and moisture content was analyzed, diagnostic bands were extracted and a multiple linear regression prediction model for soil moisture content was established. The correlation between soil moisture content and spectral data was significantly improved after logarithmic first-order differential transformation of four types of lime concretion black soil. Based on mathematical transformation of diagnostic bands extracted, the prediction accuracy of multiple linear regression model was optimum. There was a negative correlation between spectral reflectance and moisture content. The fitting accuracy of prediction model based on the 10 mesh soil sample was better than that of the 20 mesh, and the highest R² was 0.928. The research results can provide a very critical parameter support for precision agriculture management.