Near-infrared spectroscopy to estimate the chemical element concentration in soils and sediments in a rural catchment

Abstract

Monitoring the transfer of chemical elements from terrestrial to aquatic environments is essential. To this end, it is necessary to develop fast and low-cost techniques to estimate the concentration of elements in soils and sediments. This study aimed to evaluate multivariate methods and pre-processing techniques to increase the accuracy of estimating the concentration of elements in soils and sediments and to estimate the concentration of elements present in the sediments using calibrated soil models using near infrared spectroscopy (NIR). A total of 316 soil samples and 196 sediment samples were collected in the Guaporé watershed in southern Brazil. These were dried, disaggregated and sieved with a 63 µm sieve size or 230 US mesh standard. Organic carbon was determined by wet oxidation and the total concentration of chemical elements (Al, Ba, Be, Ca, Co, Cr, Cu, Fe, K, La, Li, Mg, Mn, Na, Ni, P, Pb, Sr, Ti, V and Zn) by ICP-OES after microwave-assisted digestion with HCl and HNO3 (3:1). NIR spectra (1000–2500 nm) were recorded for all soil and sediment samples. Multivariate models, for instance partial least square regression (PLSR) and support vector machine (SVM), were tested combined with spectral pre-processing techniques (Detrend, DET; Savitzky-Golay Derivative, SGD; and Standard Normal Variate, SNV) and compared to the raw spectrum (RAW). The SVM model resulted in better predictions for soils, sediments and soils + sediments. In addition, the effects of preprocessing increased the accuracy of the specifications in the order: RAW < SNV < DET = SGD. The best models were obtained by combining the SVM multivariate model with SGD pre-processing. The fit of the models calibrated with soils (R2 = 0.88) and sediments (R2 = 0.89) was superior to the combination of soils and sediments (R2 = 0.85). Prediction of the concentration of elements in sediments through models calibrated with soils had low accuracy, with an average R2 of 0.11 (ranging from 0.00 to 0.46). Therefore, it is necessary to calibrate models separately to estimate the concentration of elements in soils and sediments with greater accuracy. The results showed that it is possible to build accurate spectroscopic models to predict the concentration of elements in soils and sediment samples for all determined elements.