Deriving and validating spectral pedotransfer functions for estimating some soil heavy metal in Vis-NIR range

Background and objectives Direct relationships between the incidences of cancer in people who are exposed to heavy metals, have been investigated and proved in various studies (28). So rapid and periodic monitoring of heavy metals in the areas vulnerable to pollution is important (9). Although conventional methods of soil metals content determination are sufficiently accurate, they are mostly based on wet digestion of soil samples in hot concentrated acids followed by atomic absorption spectrometry (AAS) or inductively coupled plasma (ICP) spectrometry, these methods are time consuming, expensive and require chemical agents and qualified staff (1). Development of visible-near infrared (Vis-NIR) diffuse reflectance spectroscopy provides an alternative to these conventional monitoring methods of the soil heavy metal contamination. Because there are many advantages with using the technique. It is non- destructive, requires a minimum of sample preparation and does not involve any (hazardous) chemicals. The measurements only take a few seconds and several soil properties can be estimated from a single scan. Moreover, the technique allows for flexible measurement configurations and in situ as well as laboratory-based measurements (35). Limited work has been done to predict soil heavy metal content with Vis-NIR through different models or data mining methods in Iran. The aim of this study was to explore the feasibility of ANN in estimating the heavy metal concentration using diffuse spectral reflectance data in the Vis-NIR range,
Materials and methods A total of 57 soil samples were collected from the topsoil of Hormuz Island. The total concentrations of Mo and As elements were measured using inductively coupled plasma (ICP-OES) apparatus. Then reflectance spectra of the collected soil samples were measured using a portable spectroradiometer apparatus (Field Spec 3, Analytical Spectral Device, ASD Inc.) in the Vis-NIR (350-2500 nm) range. Artificial Neural Networks (ANN) method WAS used to predict heavy metal concentration from soil samples reflectance spectra.
Results The results showed that ANN has high capability in estimating the concentration of studied heavy metals using spectral data. Coefficient of determination (R2) for both elements, were desirable and more than 0.9 that represents the correspondence of the observed and predicted data by the neural network model in predicting the Mo and As heavy metals. However, results from other index also indicated that the ability of artificial neural network to predict the concentration of molybdenum was better than arsenic heavy metal, So that the results showed that the coefficient of residual mass was low for this element (CRM = 0.11), the coefficient of Akaike was negative (AIC = -345.8) and modeling efficiency for this element has been close to a 1 (EF = 0.97).
Conclusions In this paper we used hyperspectral reflectance data in visible and near infrared regions (350-2500 nm) to predict concentration of Mo and As heavy Using ANN as calibration model. Overall, results showed that artificial neural networks can be effectively used in deriving spectral-pedotransfer functions and bridging soil spectral reflectance to accurate estimates of molybdenum and arsenic heavy metals in high concentrations.