摘要
【目的】分析不同生境类型下的土壤养分含量垂直分层变化特征,在实地调查基础上探讨应用高光谱技术预测土壤养分含量的可能性,为实现快速评估土壤状况提供参考。【方法】在浙江百山祖国家公园选择4种生境类型,采集0~10、10~20、20~30 cm 3个土层的样品,分别测定其有机碳(SOC)、全氮(TN)、全磷(TP)、全钾(TK)含量。在实测基础上采集每份土壤样品的光谱信息,分别基于原始光谱反射率、一阶微分光谱反射率,建立支持向量机(SVM)、偏最小二乘(PLSR)、随机森林(RF)、一元线性回归(ULR)的4种养分含量反演模型。【结果】1)表层(0~10 cm)土壤的养分含量除全钾以外,有机碳、全氮、全磷含量均显著大于其他层次(10~20 cm、20~30 cm)(P<0.05);2)基于一阶微分光谱反射率建立的模型验证R2基本高于基于原始光谱反射率建立的预测模型,说明一阶转换可以改善建模效果,从而更好地反演土壤养分含量;3)在SVM、PLSR、RF、ULR这4种模型中,SVM模型表现最稳定,预测效果最可靠,基于一阶微分光谱反射率数据的4种养分含量的SVM模型,R2最低也达到0.59,最高达到0.91。【结论】不同生境类型下的土壤养分含量存在显著差异;基于一阶微分反射率的高光谱技术可以较可靠地预测土壤养分含量,未来应进一步探讨如何利用高光谱技术反演土壤养分含量。
【Objective】Analyze the vertical stratified variation characteristics of soil nutrient content in different habitats,and explore the possibility of applying hyperspectral technology to predict soil nutrient content on the basis of field investigations,so as to provide a reference for the rapid assessment of soil conditions.【Method】Four habitat types were selected in Zhejiang Baishanzu National Park,and samples were collected from three soil layers of 0~10 cm,10~20 cm,and 20~30 cm,and the soil organic carbon(SOC),the total nitrogen(TN),total phosphorus(TP),total potassium(TK)content were determined respectively.Based on the measured spectral information of soil sample,four nutrient content inversion models,namely support vector machine(SVM),partial least squares regression(PLSR),random forest(RF)and unary linear regression(ULR),were established based on the original reflectance and the first derivative reflectance.【Result】1)Except for total potassium,the contents of SOC,TN and TP in the surface layer(0~10 cm)were significantly higher than those in other layers(10~20cm,20~30cm)(P<0.05).2)The model established based on the first derivative reflectance verifies that the R2 is basically higher than the prediction model established based on the original reflectance,indicating that the first-order transformation can improve the modeling effect,so as to better invert soil nutrient content.3)Among the four models of SVM,PLSR,RF and ULR,the SVM model has the most stable performance and the most reliable prediction effect.The R2 of the SVM model for the four nutrient contents based on the first derivative reflectance data is 0.59 at the lowest and 0.91 at the highest.【Conclusion】There are significant differences in soil nutrient content under different habitat types;hyperspectral technology based on first derivative reflectance can reliably predict soil nutrient content.In the future,we should further explore how to use hyperspectral technology to invert soil nutrient content.
作者
聂磊超
彭辉
赵欣胜
翟夏杰
李伟
Nie Leichao;Peng Hui;Zhao Xinsheng;Zhai Xiajie;Li Wei(Beijing Key Laboratory of Wetland Ecological Function and Restoration,Institute of Wetland Research,Chinese Academy of Forestry,Beijing 100091,China;Longquan Conservation Center of Qianjiangyuan-Baishanzu National Park,Longquan 323700,China;Institute of Ecological Conservation and Restoration,Chinese Academy of Forestry,Beijing 100091,China)
出处
《陆地生态系统与保护学报》
2022年第3期9-17,共9页
Terrestrial Ecosystem and Conservation
基金
凤阳山森林生态系统(土壤)生态化学计量学项目(ZJDF2021-010)。
关键词
土壤养分
高光谱
预测模型
soil nutrient
hyperspectral
inversion model
