Open AccessEstimating Atterberg Limits of Fine‐Grained Soils by Visible–Near‐Infrared Spectroscopy
Author(s) -
Rehman Hafeez Ur,
Knadel Maria,
Kayabali Kamil,
Arthur Emmanuel
Publication year - 2019
Publication title -
vadose zone journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.036
H-Index - 81
ISSN - 1539-1663
DOI - 10.2136/vzj2019.04.0039
Subject(s) - atterberg limits , penetrometer , detection limit , soil water , mathematics , soil test , soil science , environmental science , materials science , statistics
Core Ideas Two methods each to measure each Atterberg limit (SL, PL, and LL) were evaluated. A Vis‐NIRS technique successfully estimated Atterberg limits for fine‐grained soils. The Atterberg limits correlated with typical clay signatures in the Vis‐NIR spectrum Vis‐NIRS model performance was independent of the method used to generate the reference data. The Atterberg limits (shrinkage limit [SL], plastic limit [PL], and liquid limit [LL]) describe the physico‐mechanical behavior of soils and thus are crucial for civil and agricultural applications. Conventional laboratory methods for measurement of these limits are tedious and costly for a large number of samples. Our objective was to develop visible–near‐infrared spectroscopy (Vis‐NIRS, from 400–2500 nm) based reliable models to estimate the Atterberg limits. Two conventional methods for each Atterberg limit were used to generate the reference data: paraffin wax and Hg methods for SL; rolling and motorized devices for PL; and Casagrande cup and drop‐cone penetrometer methods for LL. Calibration models were built on 80% of the data using partial least squares regression and validated with the remaining 20% of the dataset. The Vis‐NIRS independent validation of LL showed very good estimation with standardized RMSE (SRMSE = RMSE/Range) of 0.16 and 0.15, respectively, for LL drop‐cone and LL Casagrande methods. Similarly for PL, the Vis‐NIRS estimation accuracy was quite good with SRMSE values of 0.18 and 0.22, respectively, for PL motorized and PL rolling . Reasonably good estimation was obtained for the SL paraffin and SL Hg with SRMSE of 0.25 for both methods. The results suggest that the Vis‐NIRS calibration models and the accuracy following validation were similar for the pair of methods used for the SL, PL, and LL. Finally, analyses of the model regression coefficients revealed that the important wavelengths to estimate the SL, PL, and LL in the Vis‐NIR regions were present across the entire Vis‐NIR spectrum and were strongly related to clay type and content.