Predicting bioavailability change of complex chemical mixtures in contaminated soils using visible and near-infrared spectroscopy and random forest regression

Abstract

A number of studies have shown that visible and near infrared spectroscopy (VIS-NIRS) offers a rapid on-site measurement tool for the determination of total contaminant concentration of petroleum hydrocarbons compounds (PHC), heavy metals and metalloids (HM) in soil. However none of them have yet assessed the feasibility of using VIS-NIRS coupled to random forest (RF) regression for determining both the total and bioavailable concentrations of complex chemical mixtures. Results showed that the predictions of the total concentrations of polycyclic aromatic hydrocarbons (PAH), PHC, and alkanes (ALK) were very good, good and fair, and in contrast, the predictions of the bioavailable concentrations of the PAH and PHC were only fair, and poor for ALK. A large number of trace elements, mainly lead (Pb), aluminium (Al), nickel (Ni), chromium (Cr), cadmium (Cd), iron (Fe) and zinc (Zn) were predicted with very good or good accuracy. The prediction results of the total HMs were also better than those of the bioavailable concentrations. Overall, the results demonstrate that VIS-NIR DRS coupled to RF is a promising rapid measurement tool to inform both the distribution and bioavailability of complex chemical mixtures without the need of collecting soil samples and lengthy extraction for further analysis.

Screening Risk Assessment of Agricultural Areas under a High Level of Anthropopressure Based on Chemical Indexes and VIS-NIR Spectroscopy

“…Until now, VIS-NIR spectroscopy was used for total and bioavailable PAHs, alkanes and heavy metals determination [ 27 ], alkanes and PAHs in oil-contaminated soil [ 28 ] or the bioremediation of soils polluted with fuel oil [ 31 ]. In our study, this technique was used for diagnostic risk screening of long-term PAH contaminated soils, and to predict their toxicity based on various risk indexes.…”

“…[ 20 , 23 ], and soil biological parameters such as respiration, microbial biomass, potential mineralization of N and ratio of microbial C to organic C [ 18 ]. VIS-NIR spectroscopy is gradually becoming more popular for the rapid screening of contaminants in soils as more recent studies have demonstrated its efficacy in the detection of heavy metals (Cd, Pb, Cu, Zn) [ 24 , 25 ], total and bioavailable PAHs [ 12 , 26 , 27 , 28 ], individual compounds, e.g., benzo[a]pyrene [ 12 ] or phenanthrene [ 29 ], and total petroleum hydrocarbons [ 30 , 31 ]. The method provides a useful alternative to classical time-consuming chemical methods of soil contamination analysis [ 24 ] and can be used in situ [ 32 ].…”

Exploring relevant wavelength regions for estimating soil total carbon contents of rice fields in Madagascar from Vis-NIR spectra with sequential application of backward interval PLS

“…However, as a linear multivariate calibration, the accuracy of PLS analysis tends to decrease due to the non-linear nature of the relationship between spectral data and the dependent variable (Araújo et al, 2014). As datamining approaches, machine learning techniques, such as artificial neural network (ANN) (Kuang et al, 2015), support-vector machines (SVM) (Morellos et al, 2016), and random forest (Cipullo et al, 2019;Douglas et al, 2018;De Santana et al, 2018) outperformed the PLS analysis for predicting soil properties as it is able to account for the non-linearity associated with the soil spectral responses. More recently, deep learning is a rapidly developing frontier in machine learning that has also been tested for calibrating soil spectra (Ng et al, 2019;Padarian et al, 2019).…”

Comparison of different predictive models on HFMD based on weather factors in Zibo city, Shandong Province, China

“…That is, for any partition feature A, the data was divided into sets D 1 and D 2 by the partition points, calculating the minimum mean square deviation of D 1 and D 2 sets and the corresponding feature and eigenvalue partition points. Random forest regression (RFR) run efficiently on high-dimension data sets, in addition, it is more accurate and robust to noise [ 19 ]. The main step of obtaining an optimal model is to determine two parameters: mtry and decision tree (ntree) [ 29 ].…”

“…Previous studies attempt to apply various traditional models to predict HFMD incidence, such as SARIMA model, Generalised additive model (GAM), as well as combined models [16]. Currently, the application of machine learning provided new ideas for disease warning measures by applying mathematical models after statistical sorting the historical information of data and applied for the prediction of infectious disease [17][18][19], such as HFMD, Dengue [20]. However, no studies have proven that machine learning models have been superior to traditional models in prediction, and traditional predictive models are still used.…”