Assessment of soil texture from spectral reflectance data of bulk soil samples and their dry-sieved aggregate size fractions

Abstract

Diffuse reflectance spectroscopy (DRS) continues to emerge as a rapid approach for estimating several soil properties including soil texture. To improve the accuracy of DRS approach for soil texture assessment, a data fusion approach was tested in which the reflectance spectra of bulk soil and its aggregate size fractions obtained by dry sieving approach were combined with three chemometric models such as partial-least-squares regression (PLSR), a locally-weighted PLSR (PLSRLW) and PLSR with feature selection (PLSRFS). This newly developed approach was tested using three different soil (Vertisols, Alfisols and Mixed) spectral libraries containing a total of 1013 soil samples each with 9 different aggregate size fractions. Because soil aggregation is influenced by the nature of clay minerals in addition to organic carbon, iron and aluminum oxides, XRD analyses were carried out in selected aggregate fractions. Results showed that there was reasonable improvement in the estimation accuracy of sand, silt and clay contents in KVS and KAS when an aggregate size fraction spectrum was added to bulk soil spectrum. For example, consideration of PLSRFS or PLSRLW models along with best-performing aggregate spectra could reduce RMSE values in the validation datasets in the order of 18% for clay to 31% for silt in KVS samples, 32% for silt to 49% for clay in KAS samples, and 6% for sand to 15% for silt in WBODS samples compared to when PLSR model was used alone in the F0 spectra. Correlation analysis showed that textural fractions were more strongly correlated to spectra of macroaggregates (aggregate size >0.20 mm) than those of microaggregates (aggregate size <0.13 mm). Correlation analysis between soil separates and the amount of soil mass within each aggregate size fraction showed that the microaggregates contained more information for soil textural components than macroaggregates. The proposed approach of using bulk soil spectra and aggregate fraction spectra may be a step forward in texture analysis through DRS approach. Although extra time is required for collecting soil aggregate fraction spectra, the method has advantage of allowing the measurement of geometric mean diameter using traditional methods.