基于高光谱特征的土壤有机质含量估测研究

Soil organic matter content estimation based on hyperspectral properties

在室内条件下,利用ASD2500高光谱仪测定了潮土和水稻土自然风干土壤样品的光谱。通过系统分析两种不同类型土壤的高光谱特征差异及其有机质含量的敏感波段区位,建立了土壤有机质含量的光谱估测模型。结果表明,具有相同有机质含量的两种类型土壤整体光谱变化趋势无明显差别,但反射率表现出明显差异,一阶导数变换能较好地显现谱图中的肩峰。潮土和水稻土有机质的敏感波段集中在相同区域,原始反射率在685 nm处相关性最高,而一阶导数光谱在554 nm处相关性最高。通过对整体样本的多元逐步回归分析,筛选出两种土壤有机质相同的敏感波段为800 nm、1 398 nm和546 nm。进一步以一阶导数为自变量,基于1 400nm和554 nm两个波段构建了土壤有机质差值指数SOMDI及估测模型,即Y=4.19 12.85×(R_FD554 R_FD1 400)。利用独立的样本对建立的光谱模型进行了检验,预测决定系数均达0.79以上。上述结果表明,利用高光谱技术可实现土壤有机质的快速监测与诊断。

Organic matter（OM） content is a suitable index of soil fertility which is widely used in field nutrient management.This study established spectral indices and derived equations for estimating soil organic matter（SOM） using hyperspectral technology.In the study,visible-NIR spectral reflectance of paddy and fluvo-aquic soils were measured using the ASD2500 device.Then by using dried soil samples from two different soil types,variations in the spectrum characteristics and sensitive wavebands in relation to changing OM content were determined.Then spectral index-based models were established for estimating SOM content.The results showed that under similar SOM content,changing trends of spectrum curves of different soil types exhibited no obvious difference,while their reflectance were different.The first derivative better described the spectrum curve peak.At sensitive wavebands of two soil types existed in similar spectral regions.The original spectral reflectance was negatively correlated with OM in visible-NIR ranges,with the highest significance at 685 nm.The first derivative spectrum had a significant negative correlation at 554 nm.Step-wise multiple regression analysis revealed that for all the calibrated samples,combined spectral bands of 800 nm,1 398 nm and 546 nm well estimated SOM content of two types of soil.Furthermore,estimation model of SOM based on difference index（SOMDI） and the first derivatives of reflectance at 1 400 nm（R_FD1 400） and 554 nm（R_FD554） showed a better prediction performance;with a general equation of Y=4.19 12.85×（R_FD554 R_FD1 400）.The above monitoring models tested with independent datasets from two soil samples gave an R2 = 0.79.This suggested that it was feasible to rapidly estimate SOM using hyperspectral technology.