基于高光谱的ASTER影像土壤盐分模型校正及验证

快速准确地获取土壤盐分信息是监测和治理土壤盐渍化现象的重要前提.该文以新疆维吾尔自治区典型盐渍化区域——艾比湖流域为研究区,analytical spectral devices(ASD)光谱仪采集的土壤高光谱数据和advanced space borne thermal emission and reflection radiometer(ASTER)影像为数据源,结合实测土壤盐分含量信息,对遥感定量反演土壤盐渍化现象进行研究.再经过光谱反射率数学变换后,结合相关性分析,利用多元回归方法分别建立基于重采样后的高光谱和影像光谱的土壤含盐量估算模型,对遥感影像光谱盐分估算模型进行校正,以提高遥感定量监测盐渍化土壤的精度.结果表明:ASTER影像光谱反射率二阶导数变换和ASD重采样光谱的对数的二阶导数变换所建立的盐分估算模型最佳,决定系数R2分别为0.59和0.82.经ASD重采样光谱模型校正后的ASTER影像光谱的盐分估算模型精度R2为0.91,有效地提高大尺度条件下土壤盐渍化反演精度.研究为大尺度土壤盐分定量遥感监测提供了一种有效方法.

Calibration and validation of soil salinity estimation model based on measured hyperspectral and Aster image

Soil salinization is a major desertification and land degradation that threatens especially the stability of ecosystems in arid land. Either natural factors or human unreasonable use of the soil can cause soil salinization, and also has impact on the sustainable development of resources and the environment. There is an urgent need for intensive monitoring and quick assessment of salinization through hyperspectral remote sensing as a tool for combating soil salinizaiton in such ecosystems. In this paper, estimation the soil salinization of Ebniur lake basin, Xinjiang, China by a multiple regression model was carried out using Analytical Spectral Devices (ASD) data and Advanced Space borne Thermal Emission and Reflection Radiometer (ASTER) image data and soil reflectance spectra. A total of 11 spectral transformation forms of reflectance were used to relate with the measured soil salinity. Spectral reflectance experienced root mean square, logarithmic, and reciprocal transformation, then the first and second derivative of original and transformation forms were calculated. In addition, the first derivative of reciprocal of logarithmic was also calculated. A relationship between the sensitive bands of soil salinity was used to build models of soil salinity. The transformation and model establishment were conducted again after resampling ASD spectral reflectance data. Finally, the model after resampling was correlated with that before resampling to increase the model accuracy. A total of 50 sampling points were obtained and 30 of them were randomly selected for model establishment and the other 20 was used for validation. The results showed that the model with second derivative of reflectance was best withR2 of 0.59 and 0.75, and the root means square of error (RMSE) of 1.29 and 1.24 g/kg. ASD resampling improved the model accuracy with the second derivative of logarithmic transformation as the best model, which yielded theR2 of 0.80-0.82, and RMSE of 0.97-1.05 g/kg. Through the regression analysis, a linear model was established between ASD resampling spectral inversion model and ASTER spectral model withR2 of 0.88. Using the linear model and based on the ASTER spectral model and ASD resampling model, the ASTER spectral model was calibrated. Using measured soil salinity to validate the calibrated model, the resulted showed that the calibrated model improved the model accuracy to 0.91 and reduced the RMSE to 0.96 g/kg. Therefore, the multiple regression method could be used for model calibration for ASTER spectral model based on ASD resampling spectral inversion model, which has great potential for estimation of soil salinity in the arid land. This paper made contribution to the dynamic monitoring of soil salinization, realized the scale transformation from the measured field scale to spectral scale of multi-spectral remote sensing, and also can provide valuable information for future research.