Spatial modelling of concentration in topsoil using random and systematic uncertainty components: comparison against established techniques

Bettencourt da Silva, Ricardo J N, Argyraki, Ariadne, Borges, Carlos, Palma, Carla and Ramsey, Michael (2022) Spatial modelling of concentration in topsoil using random and systematic uncertainty components: comparison against established techniques. Analytical Letters. pp. 1-21. ISSN 0003-2719

[img] PDF - Accepted Version
Restricted to SRO admin only until 23 March 2023.
Available under License Creative Commons Attribution-Non-Commercial.

Download (2MB)
[img] PDF - Published Version
Restricted to SRO admin only

Download (3MB)

Abstract

The assessment of contaminated land often requires the collection and analysis of soil samples and the discrete or continuous modeling of the contamination. The continuous modeling involves interpolating analyte concentrations between sampled positions. The contamination can be expressed either as a map showing the spatial distribution across the site, or as a frequency distribution, both of which can be compared with a threshold. The characterization of land contamination is affected by random and systematic uncertainty components of the sampling, chemical analysis, sampling positioning and modeling. This work describes the application and comparison of three techniques for developing continuous models of the contamination of a site supported by the quantification of none, some or all relevant uncertainty components. The 7.3 ha site was characterized using 100 soil samples and the lead contamination modeled by “inverse distance weighting” (IDW), “ordinary kriging”(OK) and a new Monte Carlo simulation method (MCM). The IDW only uses the positions and concentrations of the samples without their uncertainty. The OK also requires the “measurement error” and other parameters to select the variogram. The MCM uses the measurement uncertainty (including random and systematic effects arising from both sampling and analysis) and ‘GPS coordinates uncertainty’. The measurements of Pb concentration across the site were log-normally distributed, and therefore log-transformed prior to modeling. The output of the models was compared against a discrete model by ‘probabilistic block mapping’ (PBM) that also considers measurement uncertainty. The OK produced a smoothed spatial variation of Pb concentration that appears more realistic. However, IDW and OK underestimate the land contamination, while MCM prediction most closely matches that of the measured concentrations including the impact of uncertainty components. The MCM allowed a metrologically sound modeling of the contamination and the linear interpolation of data reduced assessment subjectivity. Both MCM and PBM made realistic estimates of the proportion of the site that was contaminated over the threshold and included all relevant uncertainty components in the continuous or discrete modeling, respectively. The PBM identifies areas where the true value of the contaminant concentration could exceed the threshold value, even though the single measured value did not.

Item Type: Article
Keywords: Spatial modelling, Geostatistics, Monte Carlo Method, Probabilistic modelling, Land contamination, Measurement Uncertainty
Schools and Departments: School of Life Sciences > Evolution, Behaviour and Environment
Depositing User: Michael Ramsey
Date Deposited: 23 Mar 2022 09:35
Last Modified: 23 Mar 2022 09:45
URI: http://sro.sussex.ac.uk/id/eprint/104990

View download statistics for this item

📧 Request an update