Potential of Spectroradiometry to Classify Soil Clay Content

ABSTRACT Diffuse reflectance spectroscopy (DRS) is a fast and cheap alternative for soil clay, but needs further investigation to assess the scope of application. The purpose of the study was to develop a linear regression model to predict clay content from DRS data, to classify the soils into three textural classes, similar to those defined by a regulation of the Brazilian Ministry of Agriculture, Livestock and Food Supply. The DRS data of 412 soil samples, from the 0.0-0.5 m layer, from different locations in the state of Rio Grande do Sul, Brazil, were measured at wavelengths of 350 to 2,500 nm in the laboratory. The fitting of the linear regression model developed to predict soil clay content from the DRS data was based on a R2 value of 0.74 and 0.75, with a RMSE of 7.82 and 8.51 % for the calibration and validation sets, respectively. Soil texture classification had an overall accuracy of 79.0 % (calibration) and 80.9 % (validation). The heterogeneity of soil samples affected the performance of the prediction models. Future studies should consider a previous classification of soil samples in different groups by soil type, parent material and/or sampling region.

摘要：漫反射光谱法（Diffuse reflectance spectroscopy, DRS）是一种快速且低成本的土壤黏粒检测替代技术，但仍需进一步研究以明确其适用范围。本研究旨在构建基于漫反射光谱数据预测土壤黏粒含量的线性回归模型，并将土壤划分为三类质地类别，该分类标准与巴西农业、畜牧业和食品供应部一项法规中定义的类别相符。本研究采集了巴西南里奥格兰德州不同地点的412份0.0~0.5 m土层土壤样品，并在实验室中测定了其350~2500 nm波段范围内的漫反射光谱数据。所构建的线性回归模型用于预测土壤黏粒含量，其校正集与验证集的决定系数R²分别为0.74和0.75，均方根误差（Root Mean Square Error, RMSE）分别为7.82%和8.51%。土壤质地分类任务的总体准确率在校正集和验证集分别为79.0%和80.9%。土壤样品的异质性会对预测模型的性能产生影响。未来的研究应考虑依据土壤类型、母质及/或采样区域对土壤样品进行预先分组。