在基于可见-近红外反射光谱的土壤养分速测技术中, 不同类型土壤间的模型转移是目前亟需解决的关键问题和难点。 以土壤全氮为研究对象, 探讨了两种不同土壤间的模型传递方法及其效果。 以青岛李村河畔土壤为主样品, 通过分段直接矫正结合斜率/截距修正(PDS-S/B)、 分段直接矫正结合线性插值(PDS-LI)、 典型相关性分析结合斜率/截距修正(CCA-S/B)、 典型相关性分析结合线性插值(CCA-LI)、 直接矫正结合斜率/截距修正(DS-S/B)、 直接矫正结合线性插值(DS-LI)等算法, 进行模型转移, 实现对青岛浮山山麓土壤全氮含量不同程度的预测。 其中, PDS-S/B的模型转移效果最好, 均方根误差、 平均相对误差、 最大相对误差均最小, 分别为0.04, 6.6%, 19.0%。 主、 从样品经遗传算法提取特征变量后再进行模型转移, 相比无任何前处理的模型转移, 均有不同程度的提高, 其中LI相关的模型转移方法比S/B相关的方法提高的程度更大。 研究了不同样品在同一仪器、 相同测试环境下的土壤养分的模型传递问题, 初步探讨了同一仪器共享一个土壤养分光谱模型的可能性, 这将从根本上提高速测效率, 有利于光谱技术在土壤养分速测中的推广应用。

Model transfer among different soils is the key point and obstacle and obstacle for rapidly determining soil nutrients by Vis-NIR spectroscopy. Here, we studied the methods and results of model transfer for total nitrogen(TN) between two types of soils in Qingdao, China. A main spectral model was firstly set up using soils sampled from Licun River. Then, by using piecewise direct standardization combined slope/bias algorithm (PDS-S/B), PDS combined linear intercept algorithm(LI), canonical correlation analysis (CCA) combined S/B (CCA-S/B), CCA-LI, direct standardization (DS) combined S/B (DS-S/B), and DS-LI, the concentrations of total N in soils sampled from Fushan Montain were predicted by the main spectral model with different accuracy. Results of model transfer by PDS-S/B was the best, whose root mean square error (RMSE), mean relative error, and maximum relative error were 0.04, 6.6%, and 19.0%, respectively. Pretreatment before building the spectral model could influence the transfer results. Here, the main spectral model built after extracting the diagnostic spectra genetric algorithm had better results than those built without any pretreatment. Transfer methods could also affect the transfer results. The transfer methods related to LI had a larger increase in accuracy than those related to S/B. The best model transfer was from PDS-S/B, suggesting that PDS-S/B was the better method for this study. This study resolved a specific model transfer for TN between two different types of soils under the same conditions by the same instrument, different from other studies which studied the model transfer of the same soil under different instrument and work conditions. This study explored the possibility that a spectrometer shared a same spectral model, which will improve the efficiency and promote the use of soil nutrient rapid determination by spectroscopy.