Spectro-pedotransfer functions for estimating Atterberg limits in soil

Abstract

Pedotransfer and spectro-transfer functions are frequently used for estimating Atterberg limits in soil. Here, we employed the Boruta feature selection approach to identify key soil and spectral features to develop hybrid spectro-pedotransfer functions (SPTF) for estimating Atterberg limits. Developed SPTFs were tested using 754 samples collected from 5 different states (Arunachal Pradesh, Sikkim, Bihar, Odisha, and Rajasthan) of India representing a wide range of agroecological conditions. Collected samples were analyzed for sand and clay contents, soil organic carbon (SOC) contents, pH, electrical conductivity (EC), Atterberg limits, and spectral reflectance data over the visible to near-infrared (VNIR) region (wavelength: 350–2500 nm). Results showed that SPTFs developed with sand and SOC contents and absorption depth near the 1900 nm wavelength of VNIR spectra (D₁₉₀₀) may be used to estimate Atterberg limits with the coefficient of determination (R²) values in the range of 0.59–0.75 for pooled data. Because these parameters may be quickly estimated (less than an hour) under laboratory conditions, SPTFs may be used as a rapid method of estimating Atterberg limits. As a scaling-up strategy, we also showed that the normalized burn ratio 2 estimated from Sentinel-2 multispectral imaging data may be used as a proxy for D₁₉₀₀ to estimate Atterberg limits with R² values of 0.56–0.79, which requires large-scale validation. Results also showed that high SOC contents in soils suppress the effects of clay contents on Atterberg limits suggesting that sand content is a better predictor of Atterberg limits than clay contents specifically in high carbon soils.