摘要
以桂东北寨底峰丛洼地土壤为研究对象,利用二阶微分和去除包络线二阶微分方法对土壤光谱进行处理,筛选出3种光谱指数与土壤有机碳(SOC)相关系数最高的特征波段,通过比较偏最小二乘回归、多元线性回归与多元逐步回归等模型的精度,确定SOC最佳估测模型。结果表明:(1)研究区土壤样品有机碳质量分数最小值为0.20%,最大值为6.06%,变异系数为63.28%,具有中等强度的空间异质性;(2)二阶微分光谱指数建立的多元线性回归模型精度优于原始光谱反射率及包络线二阶微分的模型;(3)二阶微分、包络线二阶微分光谱指数建立的偏最小二乘回归预测模型均比通过原始数据建立的模型精度高出0.3;(4)基于二阶微分所建立的多元逐步回归模型具有较高的预测精度(R^(2)=0.75,均方根误差RMSE=4.83和较大的剩余估计偏差RPD=2.00)。
The study is conducted to overcome the difficulties in hyperspectral prediction of soil organic carbon(SOC)content in crested depressions and to improve the accuracy of SOC content prediction.In this paper,the soil of the Zaidi peak cluster depressions in northeastern Guangxi is taken as research prediction.The three characteristic bands with the highest correlation coefficients between the spectral indices and soil organic carbon are selected by second-order derivative and removal of envelope second-order derivatives of soil spectra.The model accuracy is compared by partial least squares regression,multiple linear regression and multiple stepwise regression to determine the optimal estimation model for SOC.The results show that,(1)The minimum value of organic carbon mass fraction of soil samples in the study area is 0.20% and the maximum value is 6.06%,with a coefficient of variation of 63.28% and a moderate intensity of spatial heterogeneity;(2)In the multiple linear regression model,the accuracy of the multiple linear regression model established by the second-order differential spectral index is better than that of the original spectral reflectance and the second-order differential of the envelope;(3)In the partial least squares regression model,the spectral index of the second-order differential and that of envelope second-order differential to build the partial least squares regression prediction model are all 0.3,more accurate than the model built from the original data;(4)In the multivariate stepwise regression model,the multiple stepwise regression model based on second-order differentiation has a high prediction accuracy(R^(2)=0.75,root mean square error RMSE=4.83 and large residual estimation bias RPD=2.00).This study can provide a reference for the optimization of the prediction model of soil organic carbon content in the crest depression and provide a basis for the restoration of the regional ecological environment.
作者
唐梅蓉
杨奇勇
唐海涛
TANG Meirong;YANG Qiyong;TANG Haitao(Institute of Karst Geology,CAGS,Guilin,Guangxi 541004,China;China University of Geosciences(Beijing),Beijing 100083,China;Northeast Agricultural University,Harbin,Heilongjiang 150030,China)
出处
《中国岩溶》
CAS
CSCD
北大核心
2021年第5期876-883,共8页
Carsologica Sinica
基金
国家重点研发计划项目(2016YFC0502404)
广西重点研发计划(桂科AB20159022)。
关键词
峰丛洼地
高光谱预测
有机碳
多元线性回归
偏最小二乘回归
peak-cluster depression
hyperspectral prediction
organic carbon
multiple linear regression
partial least square regression
