Simultaneous prediction of several soil properties related to engineering uses based on laboratory Vis-NIR reflectance spectroscopy

Evaluating the particle densities of subarctic soils using pedotransfer functions and vis–NIR spectroscopy

“…(2019), who concluded that the prediction of ρ s was favorable for soils exhibiting a large range in SOM contents (0.002–0.67 kg kg −1 ). Further, vis–NIR has been used to predict ρ s of 220 soils from Kurdistan, for which the OC ranged from 0.003 to 0.091 kg kg −1 and the predominant soil textural classes were clay, clay loam, and silt loam (Davari et al., 2021). The ρ s was predicted with acceptable accuracy (RMSE of 0.05 Mg m −3 ) in the study of Davari et al.…”

“…These are soil properties with direct spectral signatures, but vis-NIRS was also successfully used to predict the ρ s of 179 Danish and German soils by Manage et al (2019), who concluded that the prediction of ρ s was favorable for soils exhibiting a large range in SOM contents (0.002-0.67 kg kg −1 ). Further, vis-NIR has been used to predict ρ s of 220 soils from Kurdistan, for which the OC ranged from 0.003 to 0.091 kg kg −1 and the predominant soil textural classes were clay, clay loam, and silt loam (Davari et al, 2021). The ρ s was predicted with acceptable accuracy (RMSE of 0.05 Mg m −3 ) in the study of Davari et al (2021), and the ability to predict ρ s from vis-NIR spectra was attributed to spectral responses from clay.…”

“…Further, vis-NIR has been used to predict ρ s of 220 soils from Kurdistan, for which the OC ranged from 0.003 to 0.091 kg kg −1 and the predominant soil textural classes were clay, clay loam, and silt loam (Davari et al, 2021). The ρ s was predicted with acceptable accuracy (RMSE of 0.05 Mg m −3 ) in the study of Davari et al (2021), and the ability to predict ρ s from vis-NIR spectra was attributed to spectral responses from clay.…”

“…Their range in ρ s was, however, significantly higher (1.60−2.66 Mg m −3 ), resulting in an r 2 of .87. that the high accuracy obtained in their model was attributed to their extensive range in SOC (0.001−0.451 kg kg −1 ), which they supported by only achieving an r 2 of .62 for 83 mineral topsoils with a SOC range of 0.001−0.06 kg kg −1 . Conversely, clay content was suggested to be the most important chromophore for ρ s prediction by Davari et al (2021), who obtained an RMSE CV of 0.05 Mg m −3 (r 2 of .62) in a calibration model for 220 Kurdistan topsoils with a ρ s range of 2.38-2.65 Mg m −3 . Since the full dataset in the present study exhibits a relatively large OC range (0.009−0.273 kg kg −1 ) and a moderate clay range (0.020−0.185 kg kg −1 ), the poor performance is likely caused by high variability in the mineralogy and/or the presence of dense sesquioxides of the QA soils, for which the vis-NIR model is unable to account.…”

Artificial Intelligence-Based Prediction of Key Textural Properties from LUCAS and ICRAF Spectral Libraries

“…These methods are efficient for analyzing the large number of samples required for high resolution soil mapping [13]. Over the last few decades, there has been a growing focus on spectral methods, especially with the increasing demand for precision agriculture tools to assist characterization of soils effectively [14,15]. Lately, the application of machine learning (ML) algorithms in soil science has increased.…”