土壤全氮含量与碳氮比的高光谱反射估测影响因素研究

基于土壤高光谱反射特征可以实现土壤全氮(TN)含量与碳氮比(C∶N)等土壤属性的快速、无损测定,但其估测模型受土壤颗粒粒径水平与光谱指数(预处理)等因素影响。通过研磨准备2、0.25和0.15 mm共3个水平颗粒粒径的土样,分析了原始(RAW)及多次散射校正MSC(Multiple Scattering Correction)、一阶微分FD(First Derivative)、连续统去除CR(Continuum Removal)等预处理的土壤反射光谱与TN含量、碳氮比变化之间的关系,发现土壤研磨可以提高反射光谱对TN含量变化的响应,而FD、CR与MSC等光谱预处理能够明显缩小不同颗粒粒径水平土样的光谱反射-TN含量、碳氮比相关性差异。结果表明:0.25 mm颗粒粒径土样的FD预处理光谱在2 250 nm和2 280 nm处分别与TN含量、碳氮比变化存在最大相关,但最大相关单波段线性回归模型的TN含量、碳氮比估测精度不如全波段光谱PLSR模型。其中,0.25 mm土样RAW光谱全波段PLSR模型估测TN含量的表现最佳(RPD=3.49,R²=0.92,RMSEP=0.1 g/kg);而碳氮比的估测结果并不十分理想,其最优估测模型(0.25 mm土样FD预处理的全波段PLSR模型)的RPD仅为1.21,可能与土样的碳氮比变化范围较小有关,在以后的研究中可以尝试采集更多的样本数量或土壤类型,使训练样本具有较大的变量范围,以取得较好的估测效果。

Effects on Estimating Soil Nitrogen Content and Ratio of Carbon to Nitrogen Using Hyperspectral Reflectance

Hyperspectral reflectance features can be used as quick and nondestructive assessing method for soil properties,like soil total nitrogen content (TN),the ratio of carbon to nitrogen (C∶N),and so on.However,the prediction models are affected by many factors,such as particle size level of soil samples,spectral pretreatment termed spectral indices.In this paper,three-leveled particle size of 2,0.25 and 0.15 mm soil samples were prepared and Multiple Scattering Correction (MSC),First Derivative (FD) and Continuum Removal (CR) were taken as spectral pretreatment and compared to raw spectra (RAW).Correlation between TN and C∶N spectral reflectance show that the response of spectral reflectance to TN content can be improved by soil grinding,and that difference between TN and content,C∶N spectral reflectance of various particle sizes can be obviously reduced by FD,CR and MSC pretreatment.This study indicates that maximal correlation between TN content and C∶N spectral reflectance is located at 2250 nm and 2280 nm of FD-pretreated spectra of 0.25 mm soil samples.However,assessment accuracy of linear predicting models based on these single wavebands is less than that of Partial Least Square Regression (PLSR) derived from full wavebands.The highest RPD (3.49) and R^2 (0.92),and least RMSEP (0.1 g/kg) were yielded by PLSR model for predicting TN content based on RAW spectra of 0.25 mm soil samples,however,which is not satisfying that the accuracy of predicted C∶N conducted by PLSR models derived from FD-pretreated spectra of 0.25 mm soil samples.The worse prediction of C∶N is perhaps related to less range of C∶N of collected soil samples,therefore,much more samples or types of soil should be collected so as to get larger range of C∶N for improved assessing capability.