星载高分五号高光谱耕地主要土壤类型土壤有机质含量估测--以黑龙江省建三江农垦区为例

Estimation of soil organic matter content in different soil types of cultivated land based on hyperspectral data of GF-5 satellite-A case study of Jiansanjiang reclamation area in Heilongjiang province

为了评价国产星载高分五号(GF-5)高光谱影像估测土壤有机质(SOM)含量的潜力,以及不同土壤类型对SOM含量光谱估测精度的影响,本研究以黑龙江省建三江农垦区为研究对象,获取了覆盖研究区域的GF-5高光谱影像和188个土壤样本。对提取的样点GF-5光谱反射率数据进行了反射率倒数、对数、一阶微分等9种光谱数学变换,并采用相关系数法确定了SOM含量的光谱敏感波段。采用偏最小二乘回归(PLSR)线性统计建模方法,对研究区域全部土壤类型以及草甸土、沼泽土、黑土等主要土壤类型,分别构建了光谱全波段和敏感波段的SOM含量估测模型,并进行了精度评价。结果表明,基于GF-5光谱数据的研究区域全部土壤类型的SOM含量估测精度不理想,最优模型精度决定系数(R^(2))为0.265,均方根误差(RMSE)为4.647%,相对分析误差(RPD)为1.135;不同类型土壤在SOM含量光谱估测精度差异较大,草甸土和沼泽土的SOM含量估测精度不高,但黑土的SOM含量估测精度较高,其中全波段光谱反射率对数一阶微分(LnR)′的SOM含量估测精度最高,R^(2)=0.729,RMSE=1.065%,RPD=1.850,SOM含量估测模型可用。按照不同土壤类型构建SOM含量估测模型可以进一步挖掘GF-5高光谱遥感估测SOM含量的潜力。

The purpose of this paper is to evaluate the potential of GF-5 satellite hyperspectral image for estimating soil organic matter(SOM)content in cultivated land,and the effect of different soil types on the estimation accuracy.Taking Jiansanjiang reclamation area of Heilongjiang province as the research object,we obtained GF-5 hyperspectral image covering the study area and collected 188 soil samples.The reflectance data were preprocessed by 9 kinds of mathematical transformation,and the spectral sensitive band of SOM content was determined by correlation coefficient method.Partial least squares regression(PLSR)model was established for SOM content estimation of all soil types,meadow soil,swamp soil and black soil,respectively,and the accuracy of models were evaluated.The results showed that the estimation accuracy of SOM content of all types of soil in the study area did not perform well.The determination coefficient(R2)of the best validation accuracy was 0.265,the root mean square error(RMSE)was 4.647%,and residual predictive deviation(RPD)was 1.135.The estimation accuracy of SOM content of different soil types was quite different.The estimation accuracy of meadow soil and swamp soil was not high,but that of black soil performed well,especially the accuracy of SOM content estimation based on log first-order derivative(LnR)′was the highest,R2 was 0.729,RMSE was 1.065%and RPD was 1.850.SOM content estimation model based on different soil types can further inrestigate the potential of GF-5 hyperspectral remote sensing to estimate SOM content.