基于分数阶微分优化光谱指数的土壤电导率高光谱估算

土壤电导率与含盐量具有高度相关性,精准的土壤电导率监测有助于了解区域土壤的盐渍化程度,对区域盐渍化防治与调控,农业可持续发展以及生态文明建设具有重要意义。为寻求预测土壤电导率的最佳高光谱参数,实现土壤盐分信息的高效监测,本研究对土壤样品进行室内高光谱和电导率测定,利用两波段优化算法对简化光谱指数(nitrogen planar domain index, NPDI)进行波段优化,筛选不同高光谱数据(原始高光谱反射率及其对应的5种数学变换)运算下的最敏感高光谱参数,从而建立土壤电导率高光谱估算模型。结果表明:1)NPDIs与土壤电导率之间的相关性显著,在原数据及其平方根、倒数、对数倒数、1.6阶微分变换形式下,优化光谱指数对土壤电导率的敏感程度更强,相关系数绝对值均超过0.80,且基于1.6阶微分变换的(R_(2020nm)+R_(1893 nm))/R_(1893 nm)波段组合相关系数绝对值最高,达到0.888。2)基于1.6阶微分波段优化的预测模型效果最佳,预测精度为R■=0.84,RMSE_(Pre)=2.07mS/cm,RPD=2.94,AIC=158.11。因此,对高光谱数据的适当数学变换有利于优化光谱指数更好地估算土壤电导率,进一步实现土壤盐渍化高精度动态监测。

Hyperspectral estimation of soil electrical conductivity based on fractional order differentially optimised spectral indices

Soil electrical conductivity is highly correlated with salt content. Accurate soil conductivity monitoring helps to determine the degree of salinisation in regional soils, and it is of great significance to the prevention and control of regional salinisation, the sustainable development of agriculture, and the construction of ecological civilisation. In this study, indoor hyperspectral and conductivity measurements were performed on soil samples. For the purpose of determining the best hyperspectral parameters for predicting soil conductivity, the simplified spectral indices (nitrogen planar domain index, NPDI), were carried out by the two band optimisation algorithm. The most sensitive hyperspectral parameters of different hyperspectral data (original hyperspectral reflectance and the corresponding 5 mathematical transformations), were selected to establish the hyperspectral estimation model of soil conductivity, for the realisation of efficient monitoring of soil salinity information. The results showed that the correlation between NPDIs and soil conductivity was significant. With the transformation of the original data, optimised spectral indices were more sensitive to soil conductivity, and the absolute value of the correlation coefficient exceeded 0.80. Among them, the correlation coefficient of the (R_{2020 nm}+R_{1893 nm})/R_{1893 nm} band combination based on the 1.6 order differential transformation was the highest, reaching 0.888. The prediction model based on 1.6 order differential band optimisation was the accurate, and the prediction accuracy was R_pre²=0.84, RMSE_Pre=2.07 mS/cm, RPD=2.94, and AIC=158.11. Therefore, the appropriate mathematical transformation of hyperspectral data was beneficial to optimise the spectral index to better estimate soil conductivity, and further achieve high precision dynamic monitoring of soil salinisation.