PEFTEC 2017 - Abstract

Abstract Title: Can PCDD/Fs concentration in soil be predicted from FP-XRF metals determination? The case of Brixia, Italy
Abstract Type: Oral
Presenter Name: Dr Andrea Manni
Co-authors:Mrs Maria Grazia Bonelli
Company/Organisation: Chemical Research 2000 Srl
Session Choice: Environmental Analysis of Soil

Abstract Information :

At present large areas devoted to agricultural purposes are polluted by either heavy metals and organic micropollutants such as PCDD/Fs and PCBs. Rapid detection of heavy metals in soil can been obtained by field portable metal analyzer such as X-Ray Fluorescence (FP-XRF). EPA Method 6200 describes the FP-XRF procedure: it allows identifying hot spots where to focus the collection of soil samples to be analyzed by traditional techniques. Instead, the analysis of the organic micropollutants are still very expensive and time consuming posing a serious challenge in the determination of the pollution in large areas such as agricultural soils due to the large number of sample to be analyzed.

The aim of the present study has been to find an analytical procedure to estimate unknown concentrations of PCDD/Fs on the grounds of statistical relationships between the organic micropollutants and metal values measured by FP-XRF by means of an Artificial Neural Network (ANN) data modelling.

The initial phase of the method involves the analysis of a consistent number of samples by ICP/MS (metals) and HRGC/HRMS (organics): their concentrations are used to train an ANN model and to calibrate the FP-XRF at the specific sample preparation conditions adopted. In a secondary phase, a large number of samples are analysed by FP-XRF only while the above-mentioned ANN model estimates the organic compounds. If their concentrations identify a hot spot, confirmatory analysis will be performed by HRGC/MS. The results of the confirmatory tests will be used to train again the neural network, improving model accuracy at each repetition of the procedure.

As an example, the procedure has been applied to an Italian agricultural soil in Brixia, contaminated by Cu, As, Zn, Pb, Mn and PCDD/Fs. The trained network was able to correctly predict 85,6% of the PCDD/Fs values, with a MAE ( Mean Absolute Error) of 0,215 and a RMSE ( Root Mean Squared Error) of 4,26 in the validation set.