论文部分内容阅读
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0.15mm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(MultipleScattering 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:0.25mm颗粒粒径土样的FD预处理光谱在2 250nm和2 280nm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,0.25mm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=0.1g/kg);而碳氮比的估测结果并不十分理想,其最优估测模型(0.25mm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。
Based on the hyperspectral reflectance characteristics of soils, rapid and non-destructive determination of soil properties such as total nitrogen (TN) content and carbon / nitrogen ratio (C: N) can be achieved. However, the estimation model is affected by soil particle size and spectral index ) And other factors. By grinding and preparing the soil samples with 3 horizontal particle sizes of 2, 0.25 and 0.15 mm, the primary and secondary Scatter Correction (MSC), First Derivative (FD) (Continuum Removal) and other pretreatment soil reflectance spectra and TN content, carbon and nitrogen ratio changes found that soil grinding can improve the response of the reflectance spectrum of TN content changes, and FD, CR and MSC spectral pretreatment can be significant The differences of spectral reflectance-TN content and C / N ratio of soil samples with different particle sizes were reduced. The results showed that there was the highest correlation between the TN content and C / N ratio at 0.255 mm and 2.82 nm for the 0.25 mm particle size soil samples. The TN content, C / N ratio The estimation accuracy is not as good as the full-band spectral PLSR model. Among them, the PLSR model of full spectrum of 0.25mm soil sample shows the best TN content (RPD = 3.49, R2 = 0.92, RMSEP = 0.1g / kg); while the estimation of C / N ratio is not very satisfactory, The RPD of the optimal estimation model (full-band PLSR model with 0.25mm soil sample FD pretreatment) is only 1.21, which may be related to the smaller range of C / N ratio of soil samples. In the later study, we can try to collect more The number of samples or soil types, training samples have a larger range of variables in order to obtain a better estimation effect.