Using source‐specific models to test the impact of sediment source classification on sediment fingerprinting

Abstract

Sediment fingerprinting estimates sediment source contributions directly from river sediment. Despite being fundamental to the interpretation of sediment fingerprinting results, the classification of sediment sources and its impact on the accuracy of source apportionment remain underinvestigated. This study assessed the impact of source classification on sediment fingerprinting based on diffuse reflectance infrared Fourier transform spectrometry (DRIFTS), using individual, source-specific partial leastsquares regression (PLSR) models. The objectives were to (a) perform a model sensitivity analysis through systematically omitting sediment sources and (b) investigate how sediment source-group discrimination and the importance of the groups as actual sources relate to variations in results. Within the Aire catchment (United Kingdom), five sediment sources were classified and sampled (n = 117): grassland topsoil in three lithological areas (limestone, millstone grit, and coal measures); riverbanks; and street dust. Experimental mixtures (n = 54) of the sources were used to develop PLSR models between known quantities of a single source and DRIFTS spectra of the mixtures, which were applied to estimate source contributions from DRIFTS spectra of suspended (n = 200) and bed (n = 5) sediment samples. Dominant sediment sourceswere limestone topsoil (45 ± 12%) and street dust (43 ± 10%). Millstone and coals topsoil contributed on average 19 ± 13% and 14 ± 10%, and riverbanks 16 ± 18%.Due to the use of individual PLSR models, the sum of all contributions can deviate from 100%; thus, a model sensitivity analysis assessed the impact and accuracy of source classification. Omitting less important sources (e.g., coals topsoil) did not change the contributions of other sources, whereas omitting important, poorlydiscriminated sources (e.g., riverbank) increased the contributions of all sources. In other words, variation in source classification substantially alters source apportionment depending on source discrimination and source importance. These results will guide development of procedures for evaluating the appropriate type and number of sediment sources in DRIFTS-PLSR sediment fingerprinting.This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Temporal variation in suspended sediment transport: linking sediment sources and hydro‐meteorological drivers

“…Estimated contributions from riverbanks are therefore highly uncertain, as additional (unclassified) sources might still be captured within this group. However, the analysis also suggests that the part represented by the riverbank samples is an important source; omitting the riverbank as a source impacts significantly on the estimated contributions of other sources (Vercruysse and Grabowski, ). Therefore, riverbank was retained as a potential source, but the high confidence intervals should be kept in consideration.…”

“…The discriminant analysis demonstrated that urban street dust and material from grassland topsoil in the three lithological areas can be discriminated well based on their DRIFTS spectra, while riverbank has a lower degree of discriminatory power (Vercruysse and Grabowski, ). The difference in discriminatory power can be attributed to the primary origin of the sources.…”

“…The DRIFTS‐based sediment fingerprinting applied in this study is described in detail by Vercruysse and Grabowski (). In summary, the method consists of three main steps: (i) measuring SS and sediment source samples with DRIFTS; (ii) testing whether source samples can be discriminated from each other based on their DRIFTS spectra through a discriminant analysis; and (iii) developing source‐specific partial least squares regression (PLSR) models based on DRIFTS spectra of experimental mixtures (54 mixtures; Vercruysse and Grabowski, ) with known source‐quantities, which can subsequently be applied to estimate SS source contributions from DRIFTS spectra of SS.…”

“…Riverbank material is generally a mixture of floodplain deposits consisting of various primary sediment sources, while street dust originates from distinctly different anthropogenic and natural sources (e.g. weathering of car tires, atmospheric deposition, road construction works) (Taylor and Owens, ; Vercruysse and Grabowski, ). The differences in discriminatory power are subsequently reflected in the uncertainty associated to the source‐specific PLSR models, whereby the 95% confidence intervals range from ±10% for urban street dust and coals grassland, ±12–13% for limestone and millstone grassland, and ±18% for riverbank material (Table , see also Supporting Information).…”

“…non‐conservancy) (Laceby et al ., ). To address the uncertainty related to the number of sources, Vercruysse and Grabowski () performed a sensitivity analysis of the PLSR model results by omitting individual sources from the source classification. The results demonstrate that when an important source (in terms of contribution) is omitted, more contribution is attributed to the least well‐discriminated source (i.e.…”