Rapid On-Site Determination of the Total Petroleum Hydrocarbon Content of Soils by Handheld Fourier Transform Near-Infrared Spectroscopy: Development of a Global, Site- and Scanner- Independent Calibration Model

Authors: Hoelen T , Miao T , Pfeifer F , McDaniel C , Gea Neves MD , Wise BM and Siesler HW

DOI: 10.23880/ppej-16000409

Volume 9, Issue 2

Abstract

In a previous publication it was demonstrated that hand-held near-infrared (NIR) spectroscopy is an excellent method for the rapid, quantitative determination of total petroleum hydrocarbons (TPHs) in soils and thus contributes significantly to rapid decisions in environmental site assessment projects and exploration work [1]. The soils investigated for these previous investigations were collected at two different US sites with TPH reference values in the range of 350-30,000 ppm and the Locally Weighted-Partial Least Squares (LW-PLS) regression method was applied to develop a calibration model for the prediction of site-independent, unknown samples. Root-Mean-Square errors for calibration/cross validation (RMSEC/ RMSECV) of 687/959 ppm TPH were obtained and the average absolute prediction error of test samples not included in the calibration was 703 ppm TPH. The challenges to the objectives of the present work were twofold: 1. To extend the calibration model from a geographically restricted to a globally applicable one and 2. The calibration model should become independent of hardware fluctuations in the production of the NIR scanner used. To achieve these goals, a total of 314 samples from six different sites in three different countries/continents (USA, Kuwait, Australia) with TPH levels between 6 - 39,000 ppm were analysed. In order to reach independence from the geographical origin of the samples, the LW-PLS algorithm was applied for the calibration model development in analogy to the previous publication. To our knowledge, hardware production variations of the NIR scanner used were compensated for the first time in this study by using eight auxiliary scanners of different serial numbers for the additional measurement of selected, representative calibration and test spectra (> 40% of the total number of spectra). The final calibration based on 412 calibration spectra (286/126 measured with the master/auxiliary scanners) and 46 test spectra (28/18 measured on the master/auxiliary scanners) had RMSEC/RMSECV values of 1336/1983 ppm TPH and the average absolute prediction error of the test samples was 1424 ppm TPH. This promising result ensures that in future it will be possible to equip worldwide-distributed sites with different scanners and use a global master calibration for predicting new, unknown soil samples in any site with an average absolute prediction error < 1500 ppm TPH.

Keywords: Quantitative determination of total petroleum hydrocarbons in soil; Rapid on-site analysis by handheld NIR spectrometer; Development of a global, site and scanner-independent calibration model

View PDF