Detecting soil properties in agricultural lands using field spectroscopy and regression models
Date
2020
Authors
Zahinda, Franck Mugisho
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Reflectance spectroscopy can be used to non-destructively characterize materials for a wide range of applications. In this study, visible-near infrared (Vis-NIR) spectroscopy was evaluated for the prediction of diverse soil properties (clay content, SOC, TN, and pH) related to different soil samples from the Eastern Cape Province in South Africa. Soil samples were scanned by a portable spectrometer at 1 nm wavelength resolution from 350 to 2500 nm. Calibrations between soil properties obtained from digital soil maps and reflectance spectra were then developed using cross-validation under partial least squares regression (PLSR) and support vector machine regression (SVMR). Raw reflectance and Savitzky-Golay first derivative data were used separately for all the samples in the data set. Key wavelengths to predict clay content, SOC, TN, and pH were identified using the variable importance projection (VIP) and Boruta algorithms. Data were additionally divided into two random subsets of 70 and 30% of the full data, which were each used for calibration and validation. The results indicated that Vis-NIR spectroscopy can be successfully used to predict soil clay content, SOC, TN and pH. For clay content, SOC, and pH, the best results were obtained by SVMR with first derivative data (RPD = 2.05, Rp2 = 0.83, RMSEP = 1.95% for clay content; RPD = 2.40, Rp2 = 0.87, RMSEP = 2.48 g.kg-1 for SOC; and RPD = 2.87, Rp2 = 0.89, RMSEP = 0.16 for pH). In contrast, PLSR with raw data outperformed SVMR models for TN prediction (RPD= 2.15, Rp2 = 0.77, RMSEP = 0.20 mg.kg-1). Key wavelengths to predict the four properties were identified mostly around 400-700 nm and 2200-2450 nm. In conclusion, Vis-NIR spectroscopy was variably good in estimating clay content, SOC, TN and pH in laboratory conditions, and showed potential for substituting traditional wet laboratory analyses or providing inexpensive data
Description
A research report submitted to the Faculty of Science, University of the Witwatersrand, in partial fulfilment of the requirements for the degree of Master of Science in Geographical Information Systems and Remote Sensing, 2020
Keywords
Citation
Zahinda, Franck Mugisho (2020) Detecting soil properties in agricultural lands using field spectroscopy and regression models, University of the Witwatersrand, Johannesburg, <http://hdl.handle.net/10539/31000>