AlradSpectra: a Quantification Tool for Soil Properties Using Spectroscopic Data in R

• Autor(a) principal: Dotto,André Carnieletto
• Data de Publicação: 2019
• Outros Autores: Dalmolin,Ricardo Simão Diniz, Caten,Alexandre ten, Gris,Diego José, Ruiz,Luis Fernando Chimelo
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Revista Brasileira de Ciência do Solo (Online)
• Texto Completo: http://old.scielo.br/scielo.php?script=sci_arttext&pid=S0100-06832019000100303
• topic: GUI; R environment; multivariate calibration; spectral preprocessing; Pedometrics

ABSTRACT Soil reflectance spectroscopy has become an innovative method for soil property quantification supplying data for studies in soil fertility, soil classification, digital soil mapping, while reducing laboratory time and applying a clean technology. This paper describes the implementation of a Graphical User Interface (GUI) using R named AlradSpectra. It contains several tools to process spectroscopic data and generate models to predict soil properties. The GUI was developed to accomplish tasks such as perform a large range of spectral preprocessing techniques, implement several multivariate calibration methods, generate statistics assessment and graphical output, validate the models using independent dataset, and predict unknown variables using soil spectral data. AlradSpectra has four main modules: Import Data, Spectral Preprocessing, Modeling, and Prediction. The implementation of AlradSpectra is demonstrated by applying visible near-infrared reflectance spectroscopy for soil organic carbon (SOC) prediction. The data contains the value of SOC and Vis-NIR reflectance for 595 soil samples. The prediction statistic assessment of SOC was performed applying all spectral preprocessing and methods. The R 2 considering all models ranged from 0.54 to 0.80. In the partial least squares regression (PLSR) models, the performances were similar to multiple linear regression (MLR) and support vector machines (SVM). The lowest error in the SOC prediction was achieved by PLSR method with standard normal variate (SNV) preprocessing reaching an R 2 of 0.80, the smallest root mean square error (RMSE) of 0.47 %, and ratio of performance to inter-quartile distance (RPIQ) of 3.12. The capacity of performing multiple tasks, being free and open-source, easy to operate, and requiring no initial knowledge of R programming language are features that make AlradSpectra a useful tool to perform different modeling approaches and predict the desired soil variable.