Use of Vis-NIR reflectance spectroscopy for estimating soil phosphorus sorption parameters at the watershed scale

Abstract

Measurement of soil phosphorus sorption parameters (PSPs) provides crucial information on P fertilization and P leaching. Traditional approaches for determining these indices are expensive and time-consuming. To develop rapid indirect methods, this study aims to assess the effectiveness of Vis-NIR spectroscopy ranging from 350 to 2500 nm for estimating various PSPs, including maximum buffering capacity (MBC), the standard buffering capacity (SBC), P sorption maximum (Q_max), soil P buffering capacity (PBC), and standard P requirement (SPR). We collected 100 soil samples in western Iran and related Vis-NIR data to the PSP parameters via Partial least squares regression (PLSR) and artificial neural network (ANN). The observed PSP values showed large variabilities across sites (CV> 48 %), attributed primarily to the wide variation in soil properties controlling PSPs. The PLSR model highlighted that efficient spectral peaks in the band-wise regression coefficients were strongly associated with signature wavelengths of clay minerals, soil organic carbon, and cation exchange capacity, all are key factors influencing the PSP indices. However, the PLSR models had limited predictive power for the PSPs, due to the complex relationships between spectral data and various soil properties indirectly influencing PSPs. Compared to PLSR, the nonlinear ANN model enhanced the prediction accuracy of MBC, PBC, Q_max, SBC, and SPR by 39.25 %, 50 %, 19.28 %, 39.41 %, and 59.32 %, respectively. The best coefficient of determination achieved in validation dataset by the ANN model ranged from 0.65 to 0.85, which is deemed acceptable for practical use on large scale by local farmers and decision-makers for P fertilization strategies.