土工试验
土壤科学
偏最小二乘回归
土壤水分
阳离子交换容量
环境科学
土壤有机质
原位
分光计
化学
数学
统计
量子力学
物理
有机化学
作者
Wenjun Ji,Viacheslav I. Adamchuk,Asim Biswas,Nandkishor M. Dhawale,Bharath Sudarsan,Yakun Zhang,Raphael A. Viscarra Rossel,Zhou Shi
标识
DOI:10.1016/j.biosystemseng.2016.06.005
摘要
Mid-infrared (MIR) soil spectroscopy has shown applicability to predict selected properties through various laboratory studies. However, reports on the successful use of MIR instruments in field conditions (in situ) have been limited. In this study, a small portable prototype MIR (898–1811 cm−1) spectrometer was used to collect soil spectra from two agricultural fields (predominantly organic and mineral soils). Both fields were located at Macdonald Campus of McGill University in Ste-Anne-de-Bellevue, Quebec, Canada. In each of the 120 predefined field locations, in situ spectroscopic measurements were repeated three times and one representative soil sample was analyzed following conventional laboratory procedures. For every soil property, a field-specific partial least squares regression (PLSR) model was developed and evaluated using a leave-one-out cross-validation routine. Each soil property was evaluated in terms of the accuracy and reproducibility of model predictions. Among tested soil properties, soil organic matter, water content, bulk density, cation exchange capacity (CEC), Ca and Mg yielded higher model performance indicators (R2 > 0.50 and RPD > 1.40) as compared to soil pH, Fe, Cu, phosphorus, nitrate-nitrogen, K or Na. In most instances, the error estimate representing the prediction reproducibility was found to be as high as 50% of the overall prediction error. This was due to the combination of optical and electrical noise and soil micro-variability causing soil spectra representing the same field location to yield different predictions.
科研通智能强力驱动
Strongly Powered by AbleSci AI