单 位：
东北林业大学林学院;河南中医药大学药学院
关键词：
高光谱;全磷(TP);全钾(TK);全氮(TN);多元线性回归;偏最小二乘法;MODIS影像
摘 要：
本文旨在研究高光谱遥感分析技术对大面积土壤全氮(Total Nitrogen TN)、全磷(Total Phosphorus TP)、全钾(Total Potassium TK)实时监测的可行性。采用野外调查与室内高光谱(350～2 500 nm)数据测定相结合的研究方法,分析黑龙江省大庆市区和四县(肇源县、肇州县、杜蒙、林甸县)土壤的光谱特征和土壤养分指标,计算出土壤光谱反射率的变化:一阶微分(R)′、倒数(1/R)、倒数一阶微分(1/R)′、对数(log R)等形式;利用一元线性回归(ULR)、多元线性回归(SMLR)和偏最小二乘法(PLSR)建立全氮(TN)、全磷(TP)、全钾(TK)含量的估算模型。并用独立样本对校正模型进行验证。结果显示:土壤全氮(TN)、全磷(TP)、全钾(TK)的最佳吸收波段分别为TN(802 nm,1 108 nm),TP(784 nm,1 085 nm),TK(1 079 nm,1 578 nm);偏最小二乘法(PLSR)的建模精度高于多元线性回归模型和一元线性回归模型,分别为N(R2=0.831,RMSE=2.506 g/kg),P(R2=0.687,RMSE=0.844 g/kg),K(R2=0.832,RMSE=0.097 g/kg)。全氮(TN)、全钾(TK)预测精度较高,全磷预测结果相对较低,但也可用来粗略估算。同时利用MODIS影像对土壤全氮、全磷、全钾含量专项制图。该研究证实,利用高光谱技术结合特定算法能够较好预测差异性较大的土壤全氮、全钾、全磷含量,并可以实现预测信息可视化。这对于实时快速监测大面积土壤环境变化、预测土壤信息变化趋势、监测生环境、建立我国土壤养分数据库和降低土壤成分监测成本具有重要的现实意义。
译 名：
Estimates of Soil N,P,K Concentration by Using Hyperspectral Remote Sensing Technology
作 者：
QIAO Lu;CHEN Lixin;DONG Chengming;College of Pharmacy, Henan University of Traditional Chinese Medicine;College of Forestry, Northeast Forestry University;
单 位：
QIAO Lu%CHEN Lixin%DONG Chengming%College of Pharmacy, Henan University of Traditional Chinese Medicine%College of Forestry, Northeast Forestry University
关键词：
Hyperspectra;;TN;;TP;;TK;;multivariate linear regression(MLR);;partial least squares regression(PLSR);;MODIS image
摘 要：
The purpose of this paper is to study the feasibility of real-time monitoring soil total nitrogen(TN), total phosphorus(TP) and total potassium(TK) in large area by using Hyperspectral Remote Sensing technology. Based on the combined methods of field investigation and indoor data from hyperspectral measurement(350-2 500 nm), the soils in four counties(zhaoyuan county, zhaozhou county, dumeng county, and lindian county), and daqing urban area, Heilongjiang province were studied and analyzed for the characteristics of soil spectrum and soil nutrient index. The variation in soil spectral reflectance was calculated in the form of first-order differential(R)′, reciprocal(1/R), reciprocal first-order differential(1/R)′, logarithm(log R) and so on. Then the estimating models of TN, TP and TK were established by unary linear regression(ULR), multivariate linear regression(MLR) and partial least squares(PLSR), and calibration model is verified by independent samples. The results showed that the optimum absorption bands of TN, TP and TK were TN(802 nm, 1 108 nm), TP(784 nm, 1 085 nm), TK(1 079 nm, 1 578 nm), respectively, the precision of partial least squares(PLSR) was higher than that of Multivariate linear regression(MLR) model and unary linear regression model(ULR), which were TN(R2 = 0.831, RMSE = 2.506 g/kg), TP( R2 = 0.687, RMSE = 0.844 g/kg), TK(R2 = 0.832, RMSE = 0.097 g/kg). The prediction accuracy of TN and TK was higher, while the prediction result of TP was relatively lower, but it can also be used for rough estimation. Meanwhile, MODIS images were used to map the concentrations of TN, TP and TK in soil. This study confirmed that hyperspectral technology combined with specific algorithms can better predict the concentrations of TN, TP and TK in soils with larger differences, and the visualization of predictive information can be realized. It is of great practical significance for real-time and rapidly monitoring large-scale changes in soil environment, predicting the trend of soil information changes, monitoring ecological environment, establishing soil nutrient database in China and reducing the monitoring cost of soil composition.
