X-ray fluorescence and visible near infrared sensor fusion for predicting soil chromium content

Abstract

Anthropogenic activities, such as sewage irrigation and application of pesticides and fertilizers, are the main cause of chromium (Cr) contamination in agricultural soils. Cr contamination reduces soil quality and threatens environmental and human health. Conventional Cr measurement methods, although accurate, involve complex sample processing steps and sophisticated laboratory analysis, which are time-consuming, costly, and often environmentally unfriendly. X-ray fluorescence (XRF) and visible near-infrared (vis–NIR) spectroscopy have been recognized as alternatives to measure soil heavy metal contamination in a cheap, fast, non-destructive, and environmentally conscious manner. In this study, 301 paddy soil samples from Fuyang, Zhejiang Province, China were used to explore the feasibility and effectiveness of XRF and vis–NIR spectra separately and in combination for estimating the soil Cr content. Two strategies, including outer-product analysis (OPA) and Granger–Ramanathan averaging (GRA), were used to combine the spectra and spectral models, respectively, from the two instruments (sensor fusion). Partial least-squares regression (PLSR) was used to train the models using a single sensor (XRF or vis-NIR spectra) and OPA fused spectra. Fifty boot straps were used to assess the uncertainty of the predictions for the aforementioned models. The results indicated that XRF spectra performed better than vis–NIR spectra for predictions of Cr content, with a Lin's concordance correlation coefficient (ρc) of 0.83, a root mean square error (RMSE) of 8.80, and a ratio of prediction derivation (RPD) of 1.75. Sensor fusion by OPA gave the highest prediction accuracy with a ρc of 0.90, RMSE of 6.80, and RPD of 2.30. The sensor fusion by GRA gave similar results with a ρc of 0.88, RMSE of 7.40, and RPD of 2.13. The predictions using both methods (OPA and GRA) were acceptable when considering the standard deviation of differences (SDD = 4.23). This suggests that OPA and the GRA sensor fusion methods are efficient and accurate for rapid measurement of Cr and provide a way forward for using these technologies for fast, sensor-based soil characterization.