Main content area

Better assessment of the distribution of As and Pb in soils in a former smelting area, using ordinary co-kriging and sequential Gaussian co-simulation of portable X-ray fluorescence (PXRF) and ICP-AES data

Kim, Ho-Rim, Kim, Kyoung-Ho, Yu, Soonyoung, Moniruzzaman, Md, Hwang, Sang-Il, Lee, Goon-Taek, Yun, Seong-Taek
Geoderma 2019 v.341 pp. 26-38
algorithms, arsenic, atomic absorption spectrometry, cutting, data collection, decision making, geostatistics, kriging, lead, polluted soils, prediction, remediation, soil pollution, soil sampling, spatial variation, uncertainty
To design an appropriate remediation plan at polluted sites, the spatial distribution of soil contaminants should be accurately assessed using geostatistical approaches such as co-kriging. In this study, we evaluated whether portable X-ray fluorescence (PXRF) spectroscopy can provide a reliable dataset to improve the spatial assessment of contaminated soils. For this purpose, soil samples were collected at a high density in a former smelting area (70 m by 65 m) that was highly polluted with arsenic (As) and lead (Pb). The samples were analyzed using inductively coupled plasma atomic emission spectroscopy (ICP-AES; n = 153). In addition, at the sampling locations, As and Pb levels were scanned in the field using PXRF spectroscopy (n = 156). As a result, PXRF measurements were highly correlated with ICP-AES data and seemed appropriate as a secondary variable for spatial evaluation with co-kriging. Thus, we conducted ordinary co-kriging (OCK) using a part of ICP-AES data (primary variable) and all PXRF data (auxiliary variable), and compared the interpolation result with that by ordinary kriging (OK) using the same number of ICP-AES data. We assumed that the ICP-AES data represented true values but were less effective in cost and analytical time. A best spatial map of As and Pb levels was constructed by OK using all the training ICP-AES data (n = 123) for performance comparison and the prediction errors were calculated using the test ICP-AES data (n = 30). Comparison results indicated that the spatial distribution of contamination levels estimated by OCK was more similar to the best spatial map than that by OK when the same number of ICP-AES data was used; OCK addressed the spatial heterogeneity despite the small number of training data, while OK lost the heterogeneity. OCK using PXRF data also improved the prediction of spatial distribution even with the cutting of the sample size of ICP-AES data; the number of samples that minimized errors was reduced by 40 and 50% for As and Pb, respectively, in OCK. As a last step, the uncertainty of the spatial extent of soil contamination for remedial action or additional investigation was addressed using sequential Gaussian co-simulation algorithms, which incorporated PXRF measurements as a secondary variable and provided the probability of exceeding the regulatory limits of a soil contaminant. The probability map can help risk-based decision making in contaminated sites within budget and time constraints. The results of this study demonstrate that PXRF provides a reliable auxiliary dataset to improve the spatial assessment of contaminated soils, with reducing the time and cost of investigations.