为了探索土壤含水量与高光谱植被指数的内在关系，实现土壤含水量的快速且准确监测，以ASD光谱仪测定的研究区植被高光谱数据和环境卫星 HSI 高光谱影像数据为基础数据计算得到26种光谱植被指数，通过灰度关联分析法（grey relational analysis）对不同深度（0～10，＞10～30，＞30～50 cm）土壤含水量与实测光谱指数和影像光谱指数进行分析和筛选，确定了与土壤含水量相关性较高的5个光谱植被指数，采用多元线性回归法（multiple linear regression）分别构建了基于实测数据和影像数据的高光谱植被指数土壤含水量反演模型，并用实测高光谱植被指数模型对HSI影像植被指数模型进行校正。结果表明：2种土壤含水量反演模型对0～10 cm层的土壤含水量均有较高的拟合度，判定系数（R2）均高于0.589，并具有较好的稳定性；实测高光谱植被指数模型精度优于HSI影像植被指数模型，判定系数（R2）分别为0.668和0.589；经过校正的HSI影像土壤含水量反演模型精度有了较大的提高，判定系数（R2）从0.589提升到0.711，均方根误差（RMSE）为0.0014。该研究方法进行土壤含水量监测是可行的，为进一步提高土壤含水量定量遥感监测提供一定参考。
Soil moisture monitoring based on measured hyperspectral index and HSI image index
In order to explore the intrinsic relationship between soil moisture and hyperspectral vegetation indices, and achieve fast and accurate monitoring of soil moisture, 26 vegetation indices were chosen and figured out based on the hyperspectral data measured by ASD spectrometer and the HSI image data in a typical study area. These hyperspectral vegetation indices and soil moisture data measured in laboratory, which were collected from the northeast of delta oasis of Weigan and Kuqa rivers, were analyzed with the gray relative analysis method, and there were 5 indices which had higher correlation with soil moisture among 26 vegetation indices. These indices were selected; using the multiple linear regression, the soil moisture inversion models based on the measured spectral index and the image spectral index were established respectively, and then the inversion model based on the HSI hyperspectral image vegetation indices was corrected with that based on the measured indices. The purpose of this paper was to determine the optimal model through comparing the precision of the model and the correction between the 2 models, and thus could solve the difficulty that hyperspectral vegetation index estimated soil moisture in the oasis of arid area, and improve the precision of estimation. The results showed that: The fittings of 2 kinds of soil moisture inversion models were satisfied, and the 2 models’ determination coefficients (R2) were both higher than 0.589 and the models had better stability; the estimation accuracies of 2 kinds of soil moisture inversion models in the soil depth of 0-10 cm were the best. The accuracy of soil moisture inversion model based on the measured hyperspectral vegetation indices was higher than that based on the HSI image vegetation indices. In the soil moisture inversion model based on the measured hyperspectral vegetation indices, the best combination of vegetation indices included mSR705, SAVI2, mNDVI705, SARVI and VOG3, and the value ofR2 was 0.668, and through the 0.001 significance level inspection, theR2 value of inspection sample was 0.730, the root mean square error (RMSE) was 0.0021. In the soil moisture inversion model based on the HSI hyperspectral image, the best combination of vegetation indices included ARVI, RVI, MSR, NDVI and OSAVI, and the coefficient of determination was 0.589, and through the 0.001 significance level inspection, theR2 value of inspection sample was 0.610, the RMSE was 0.0020. After being corrected, the accuracy of HSI image soil moisture inversion model was greatly improved, and the coefficient of determination was raised from 0.589 to 0.711 and the RMSE was decreased from 0.0020 to 0.0014. The corrected HSI image soil moisture inversion model can improve the accuracy under the regional scale. This paper studies the scale transformation of the soil moisture content high spectral inversion model from the point to the continuous surface of dispersed point and then to the surface of the soil. Through application of this method, it will be feasible to carry out remote sensing monitoring of soil moisture content, and provide reference for further improvement of the accuracy of quantitative remote sensing monitoring of soil moisture content under the regional scale.
国家科技期刊平台
NSTL
万方数据
维普
