Premium
Mapping the Salt Content in Soil Profiles using Vis‐NIR Hyperspectral Imaging
Author(s) -
Wu Shiwen,
Wang Changkun,
Liu Ya,
Li Yanli,
Liu Jie,
Xu Aiai,
Pan Kai,
Li Yichun,
Pan Xianzhang
Publication year - 2018
Publication title -
soil science society of america journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.836
H-Index - 168
eISSN - 1435-0661
pISSN - 0361-5995
DOI - 10.2136/sssaj2018.02.0074
Subject(s) - hyperspectral imaging , partial least squares regression , imaging spectrometer , remote sensing , soil test , soil science , vnir , support vector machine , environmental science , mathematics , spectrometer , soil water , artificial intelligence , computer science , geology , statistics , optics , physics
Core Ideas Vis‐NIR hyperspectral imaging can be used to predict the soil salt content (SSC) in soil profiles. The least squares support vector machine (LS‐SVM) model predicted SSC more accurately than the partial least squares regression (PLSR) model in the field. Hyperspectral imaging is an efficient and nondestructive method for mapping and characterizing the SSC distribution in soil profiles. Recently, visible and near‐infrared (Vis‐NIR) hyperspectral imaging has shown great potential in fine mapping of soil properties in laboratory. Whether it could be used to predict soil salt content (SSC) in the soil profile under field conditions still remained to be determined. In this study, hyperspectral images were acquired in situ from a soil profile with a Vis‐NIR imaging spectrometer, and the optimum SSC prediction model was built to determine SSC of each pixel, and the fine SSC distribution maps were generated. The observed soil profile was located at an experimental station in Dongtai City, Jiangsu Province, China. Hyperspectral images with a spectral range of 397 to 1018 nm were obtained from 21 to 25 May 2015; a total of 140 soil samples were collected. Five spectral preprocessing methods, Daubechies wavelet (Db), LOG 10 (1/Db), Savitzky‐Golay (SG), multiplicative scatter correction (MSC), and standard normal variate (SNV) were applied, and partial least squares regression (PLSR) and least squares support vector machine (LS‐SVM) models were developed. Results showed that the LS‐SVM model predicted the SSC more accurately than the PLSR model, and the highest prediction accuracy was obtained with LOG 10 (1/Db) preprocessed spectra with R 2 p , RMSE p , RPIQ, and RPD values of 0.87, 0.58 g kg –1 , 2.60 and 2.77, respectively. Based on the optimum prediction model, the fine distribution of SSC in soil profiles over 5 d were successfully obtained. This study indicated hyperspectral imaging is an efficient and nondestructive method for mapping SSC distribution in soil profiles and characterizing the vertical transportation of soil salt under field conditions with moderate soil moisture range.