农田土壤有机碳高光谱特征及定量监测研究

Hyperspectral Characteristics and Quantitative Monitoring of Soil Organic Carbon in Farmland

为实现对农田土壤有机碳含量的快速定量监测,采用9种常规预处理分析方法,优化原始光谱信息并分析各预处理光谱与小麦土壤有机碳含量之间的关系,利用连续投影算法(SPA)提取土壤有机碳的光谱特征,并建立基于全谱和光谱特征波段的两类土壤有机碳光谱监测模型。研究表明,与原始光谱相比,预处理光谱可以显著提高其与小麦土壤有机碳的相关性。同时,采用SPA方法提取并证实了光谱区域400~450 nm、510~620 nm、1010~1060 nm、2000~2400 nm的土壤有机碳含量等重要信息。对比两类模型表现可知,相同预处理光谱条件下,连续投影算法-多元线性回归(SPA-MLR)模型优于偏最小二乘回归(PLSR)模型,其中,基于多元散射校正(MSC)预处理光谱条件下的农田土壤有机碳含量估算模型整体表现最好(R_(v)^(2)=0.726、RMSE_(v)=0.109、RPDv=1.956),且具有实践上的应用潜力。本研究证实,光谱预处理在一定程度上可以提高光谱反射率与小麦土壤有机碳含量的相关性,且影响监测模型表现,同时模型构建方法可能对模型估算精度产生更为积极的效果。本研究结果可以为农田土壤有机碳含量检测提供一定的理论依据和实践探索。

In order to realize the rapid and quantitative monitoring of soil organic carbon content in farmland,nine conventional pretreatment analysis methods were used to optimize the original spectral information and analyze the relationship between each pretreatment spectrum and wheat soil organic carbon content,and the continuous projection algorithm(SPA)was used to extract the spectral characteristics of soil organic carbon,and two types of soil organic carbon spectral monitoring models based on full spectrum and spectral characteristic bands were established.The results showed that the correlation between the pretreatment spectrum and soil organic carbon in wheat could be significantly improved compared with the original spectrum.In the meantime,the SPA method was used to extract and confirm the important information of soil organic carbon content in the spectral regions of 400~450 nm,510~620 nm,1010~1060 nm and 2000~2400 nm.Comparing the performance of the two types of models,it can be seen that the continuous projection algorithm-multiple linear regression(SPA-MLR)model is better than the partial least squares regression(PLSR)model under the same pretreatment spectra,and the overall performance of the soil organic carbon content estimation model based on multivariate scattering correction(MSC)pretreatment spectra is the best(R_(v)^(2)=0.726,RMSE_(v)=0.109,RPD_(v)=1.956),and it has practical application potential.This study confirms that spectral pretreatment can improve the correlation between spectral reflectance and soil organic carbon content in wheat to a certain extent,and affect the performance of the monitoring model,and the model construction method may have a more positive effect on the accuracy of model estimation.The results of this study can provide a theoretical basis and practical exploration for the detection of soil organic carbon content in farmland.