基于连续小波变换的表层土壤有机碳含量的高光谱估算

Hyperspectral estimation of organic carbon content in surface soils based on continuous wavelet transform

土壤有机碳含量的高光谱估算,可快速、准确监测土壤肥力,为农业生产进行合理施肥提供科学依据。以博斯腾湖西岸湖滨绿洲为研究区,应用ASD FieldSpec3光谱仪测定表层土壤的高光谱反射率,采用重铬酸钾-外加热法测定表层土壤有机碳(SOC)含量;运用连续小波变换(CWT)分别对土壤高光谱反射率(R)及其一阶微分变换(R′)进行尺度分解,分析不同尺度分解后的数据与表层SOC含量的相关性,筛选敏感波段,分别建立偏最小二乘回归(PLSR)、随机森林(RF)和支持向量机(SVM)3种模型估算表层SOC含量。研究结果表明,土壤高光谱反射率与SOC含量呈负相关,经过一阶微分变换后,通过极显著性检验(P<0.01)的波段数由1689个降低为227个,最大相关系数绝对值(|r|)由0.39提高至0.54;土壤高光谱数据CWT处理后,与表层SOC含量的相关性随分解尺度的增加呈现先增后降的趋势。R′-CWT-SVM模型估算效果最优,建模集和验证集R 2分别为0.83和0.80,RMSE分别为5.24和3.56,RPD值为2.12,能够有效估算研究区表层SOC含量。

Hyperspectral estimation of soil organic carbon content can rapidly and accurately monitor soil fertility and provide scientific basis for rational fertilization in agricultural production.Taking the west lakeside oasis of Bosten Lake as the study area,the ASD FieldSpec3 spectrometer was applied to collect hyperspectral reflectance of surface soil samples,and the organic carbon(SOC)content of surface soil was determined by the potassium dichromate-external heating method.The continuous wavelet transform(CWT)was used to decompose the soil reflectance(R)and its first-order differential transform(R′)respectively,and the data after decomposition at different scales were analyzed and correlated with the surface SOC content.The correlation between the decomposed data and the surface SOC content was analyzed using the continuous wavelet transform,and three models,namely partial least squares regression(PLSR),random forest(RF)and support vector machine(SVM),were developed to estimate the surface SOC content.The results showed that soil hyperspectral reflectance was negatively correlated with surface SOC content.After the first-order differential transformation,the number of bands passing the highly significant test(P<0.01)decreased from 1689 to 227,and the absolute value of maximum correlation coefficient increased from 0.39 to 0.54.After continuous wavelet transform,the correlation between soil hyperspectral data and surface SOC content increased first and then decreased with the increase of decomposition scale.The R′-CWT-SVM model had the best estimation effect,the R 2 of the modeling set and validation set were 0.83 and 0.80,the RMSE were 5.24 and 3.56,and the RPD value was 2.12,which could effectively estimate the surface soil organic carbon content in the study area.