不同类型土壤总氮的近红外光谱技术测定研究

该研究从江苏、河南、山西、河北、吉林采集部分表层土壤,并在山西河北两地采集表下层土壤,经前处理后通过近红外光谱仪扫描得到光谱图,用传统开氏法测得其总氮,运用OPUS软件建立了土壤总氮和光谱图之间的数学模型,并初步探讨了模型的应用范围。结果表明:各地区表层土壤建立的模型良好,交叉检验的均方根误差均在0.01%以内,相关系数平均在0.85以上,并且该模型能够很好地定量分析同一采样范围内表层和表下层的土壤总氮,定量分析的均方根误差基本在0.01%以内,相关系数在0.80左右;可能受土壤类型的影响,该模型在地区之间的运用具有一定的局限性;从各地区随机取出部分数据作为一个新的集合建立的综合模型良好,其交叉检验的均方根误差和相关系数分别为0.010 2%和0.985 6,并且该模型能很好地预测各地区的土壤总氮。

Study on the Determination of Total Nitrogen(TN) in Different Types of Soil by Near-Infrared Spectroscopy (NIS)

In the present paper, the feasibility of determination of total nitrogen (TN) content in different types of soil based on NIS technology was studied. Some surface and subsurface soil samples were collected from Jiangsu, Henan, Shanxi, Hebei and Jilin province in China, on behalf of paddy, calcareous, loess, coastal alluvial and black soil types separately. After being air-dried, milled and screened, the scanned spectra were obtained by near-infrared spectrometer using the sample with diameter be-low 1 millimeter, meantime the content of TN was measured with the traditional Kelvin method. A model between TN content and spectrum was established and also it was modified through some spectrum pretreatment using the OPUS software. Then a best model was chosen with root mean squares of crosscheck (RMSECV) being the lowest. Finally the stability and application of the model was discussed through quantitative analysis. The results showed that a good model can be separately established in each province using the surface soil from the same area on behalf of the same soil type, whose RMSECV were within 0. 01% on average, and the correlation efficient was above 0. 85 basically. The model established using the surface soil can be well used for analysis of TN content in other surface and subsurface soil from the same area whose TN was within the range of the model The root mean squares of prediction (RMSEP) and correlation efficient of the quantitative analysis were at about 0. 01% and 0. 80 separately. However, maybe affected by soil types, the model established in one area had certain limitations in the application to other areas even as its TN content was within the range of the model. Whenas, choosing some data from each area randomly to form a new sample group can establish a good new integrated model with RMSECV and correlation efficient being 0. 010 2% and 0. 985 6 respectively and that can be well used to predict TN content in soil from each area.