Magnetic signature of overbank sediment in industry impacted floodplains identified by data mining methods

Chudaničová, M and Hutchinson, SM ORCID: 2016, 'Magnetic signature of overbank sediment in industry impacted floodplains identified by data mining methods' , Geophysical Journal International, 207 (2) , pp. 1106-1121.

PDF - Accepted Version
Download (2MB) | Preview


Our study attempts to identify a characteristic magnetic signature of overbank sediments exhibiting anthropogenically induced magnetic enhancement and thereby to distinguish them from unenhanced sediments with weak magnetic background values, using a novel approach based on data mining methods, thus providing a mean of rapid pollution determination. Data were obtained from 539 bulk samples from vertical profiles through overbank sediment, collected on seven rivers in the eastern Czech Republic and three rivers in northwest England. K-means clustering and hierarchical clustering methods, paired group (UPGMA) and Ward’s method, were used to divide the samples to natural groups according to their attributes. Interparametric ratios: SIRM/χ; SIRM/ARM; and S-0.1T were chosen as attributes for analyses making the resultant model more widely applicable as magnetic concentrations values can differ by two orders. Division into three clusters appeared to be optimal and corresponded to inherent clusters in the data scatter. Clustering managed to separate samples with relatively weak anthropogenically induced enhancement, relatively strong anthropogenically induced enhancement and samples lacking enhancement. To describe the clusters explicitly and thus obtain a discrete magnetic signature, classification rules (JRip method) and decision trees (J4.8 and Simple Cart methods) were used. Samples lacking anthropogenic enhancement typically exhibited an S-0.1T < c. 0.5, SIRM/ARM < c. 150 and SIRM/χ < c. 6000 A . m-1. Samples with magnetic enhancement all exhibited an S-0.1T > 0.5. Samples with relatively stronger anthropogenic enhancement were unequivocally distinguished from the samples with weaker enhancement by an SIRM/ARM > c. 150. Samples with SIRM/ARM in a range c. 126-150 were classified as relatively strongly enhanced when their SIRM/χ > 18,000 A . m-1 and relatively less enhanced Downloaded from by guest on August 30, 2016 when their SIRM/χ < 18,000 A . m-1. An additional rule was arbitrary added to exclude samples with χfd% > 6 % from anthropogenically enhanced clusters as samples with natural magnetic enhancement. The characteristics of the clusters resulted mainly from the relationship between SIRM/ARM and the S-0.1T, and SIRM/χ and the S-0.1T. Both SIRM/ARM and SIRM/χ increase with increasing S-0.1T values reflecting a greater level of anthropogenic magnetic particles. Overall, data mining methods demonstrated good potential for utilisation in environmental magnetism.

Item Type: Article
Schools: Schools > School of Environment and Life Sciences > Ecosystems and Environment Research Centre
Journal or Publication Title: Geophysical Journal International
Publisher: Oxford Journals
ISSN: 0956-540X
Related URLs:
Funders: Various
Depositing User: Dr Simon M. Hutchinson
Date Deposited: 01 Sep 2016 10:12
Last Modified: 16 Feb 2022 18:01

Actions (login required)

Edit record (repository staff only) Edit record (repository staff only)


Downloads per month over past year