Table 7.

A summary of key results.

	Two-cluster	Three-cluster	Four-cluster	Geology classification 1	Geology classification 2
Sediment sample screening	For sample Bed 5, R, G, HRGB and IRGB fell outside of the range of values found in the source samples. For sample Bed 3 Back IRM fell outside of this range.
Source group classification	Cluster 1: Predominantly contained lower catchment topsoils, Cluster 2: Lower catchment channel banks and upper catchment topsoils.	Cluster 1: Predominantly ironstone samples, Clusters 2 and 3: divide the middle and upper catchment into two sources which appear unrelated to geology but appear spatially grouped.	Comparable to the three-cluster solution, however, it identified an additional cluster of only eight samples with its samples primarily located in the centre of the catchment.	Group 1: Ironstone and Group 2: Sandstone, Limestone, Clays and Channel Banks.	Group 1: Ironstone, Group 2: Sandstone, and Group 3: Limestone, Clays and Channel Banks.
Misclassified samples				Sample S1 (sandstone) was identified as potentially misclassified and was a better fit to the ironstone group so was deleted as it did not fall close to the area of the catchment over ironstone, the Ironstone samples I18 and I19 were also identified as potentially misclassified and fit better as sandstone, clay or limestone samples and reclassified as they were close to the boundary of two geologies
Mean variability ratios	2.1	3.8	4.6	3.1	2.5
Maximum variability ratio	HRGB, 3.7	χlfd, 11.86	χlfd, 15.71	χlfd, 6.63	χlfd, 7.81
Tracers failing to achieve the variability ratio threshold values	χlf, χlfd, χlARM, SIRM, BackIRM, HIRM, R	R	None	R, IRGB	R
Bi-plot conservatism testing	For sample Bed 5 most colour tracers fall outside of the relationships found in the source samples. For sample Bed 3 SIRM and BackIRM fell outside of the relationships in the source samples.
Range test	All tracers passed the range test for source classifications by tracer values in 40% of sediment samples falling within the median +/− one MAD range of the source groups and in 80% of sediment samples falling within the minimum to maximum range of the sources.
Mapped differences between source and sediment tracer concentrations	Ironstone source samples in the lower catchment are very dissimilar to the mean tracer values of the sampled sediments, BackIRM has more variability in the middle and upper catchment whilst XARM shows little variability, Blue is able to differentiate between samples throughout the entire catchment, but with a different trend to χARM
Distributions of tracers in source groups	With the mineral magnetic tracers there was a large difference between the percentile distribution of values in the source groups/clusters representing ironstone and the other source groups. In contrast non-ironstone sources were poorly separated. Colour tracers separated the non‑ironstone sources more effectively; however, all tracers placed the source groups into the same highest to lowest value order, suggesting that problems of equifinality may be present in model outputs when a large number of source groups are used.
Source discrimination (percent correctly classified) (basic, conservative, high variability fingerprints)	90.2%, 95.9%, 90.1% (only contains colour tracers)	89.6%, 91%, 90.5%	89.6%, 89.3%, 87.3%	97.6%, 96.6%, 97.1%	83.8%, 82%, 74.6%
Bi-plots of sources and sediments	Cluster 2 likely dominates contributions to the bed sediment, discrimination appears good.	A combination of clusters 2 and 3 likely dominates contributions to three of the sediment samples and cluster 3 appears to dominate contributions to two samples. Discrimination is good however, discrimination between clusters 2 and 3 is only achieved using DF2, which represents 8.79–9.13% of the total discriminatory power	Clusters 1 and 2 appear to dominate contributions to three samples and inputs from cluster 4 dominate contributions to two of the samples. DF2 representing 20% of total discrimination, is able to discriminate clusters 1 and 4 from clusters 2 and 3. Discrimination between clusters 2 and 3 is limited to a small amount by DF1, therefore equifinality related uncertainties are likely in model outputs.	Sediment provenance is dominated by non‑ironstone sources and source discrimination is good.	Ironstone contributes significantly to one sediment sample. The other sediment samples are likely composed of a combination of sandstone, limestone clays and channel banks. Discrimination between ironstone topsoils and other sources is good, discrimination between the sandstone topsoil and limestone, clays and channel banks group is poor and is only provided by DF2, which accounts for ~5% of total discriminatory power.
Virtual mixture source apportionment	Un-mixing models produced the correct provenance of the virtual mixtures. Uncertainties for the mixtures of 100% of each cluster were low; however, with the equal proportions of each cluster they were high.	Mixture apportionment was generally accurate but with a higher associated range of uncertainty than the two-cluster classification. Uncertainty was especially high when apportioning a 100% contribution from cluster 2, with significant estimated contributions from cluster 3 present.	The un-mixing models correctly identified contributions from clusters 3 and 4. However, when apportioning contributions from clusters 1 and 2 uncertainties were high, with significant overlap between the probability density functions for the two sources. The conservative fingerprint failed to identify Cluster 1 as the dominant source when 100% of the mixture was composed of this cluster.	Produced comparable results to the two-cluster groups but with a higher range of uncertainty.	Source apportionment with all three fingerprints for geology-based Classification 2 was unsuccessful. A 100% contribution from clays, limestone and channel banks was not represented in the un-mixing model results and a mixture of equal proportions of the sources produced an output heavily biased towards high sandstone topsoil contributions.
Weightings	A weighting of RI increased the accuracy of mixture apportionment for all three fingerprints.	A weighting of HRGB and CI improved mixture apportionment with the Basic and Conservative fingerprints.	No composite fingerprint improved mixture apportionment. Use of the Conservative fingerprint was discontinued due to its poor performance.	A weighting of BackIRM increased the accuracy of mixture apportionment for the Basic and Conservative fingerprints.	No composite fingerprint improved mixture apportionment. Due to the poor performance of Classification 2, its results were not considered for further analysis.
Goodness of fit	For the cluster analysis derived source classifications, >50% of model iterations exceeded the 0.35 GOF threshold. With the exception of those for sample Bed 5 where in all but four of the models run all iterations failed to achieve a GOF higher than 0.35 and therefore were rejected. The mean GOF of the model iterations passing the threshold was high (>0.75). GOF for geology classification 1 was generally lower than for the cluster-based classifications, the conservative fingerprint for sample Bed 5 had no iterations which exceeded the 0.35 threshold.
Sediment provenance	For sediment samples in the lower half of the catchment and sample Bed 6 in the upper catchment similar contributions were estimated to originate from cluster 1 and cluster 2. Cluster 2 dominated contributions to samples Bed 4 and 5 in the middle catchment. All three composite fingerprints produced similar results although contributions varied by ~20%.	Contributions from cluster 1 are low in all models apart from sample Bed 3 with the basic fingerprint. Topsoil inputs in the lower catchment are primarily from areas which are not over the ironstone geology. Sediment contributions to sample Bed 3 likely originate from localised channel bank inputs. Bed 3 basic fingerprint estimates a much higher contribution from cluster 1 than the other fingerprints, but consistency is reasonable for all other samples. Uncertainties associated with conservative and high variability fingerprints were high for sample Bed 2. Both clusters 2 and 3 are important sediment sources.	There were some large discrepancies between the results of the two composite fingerprints used. Clusters 2 and 3 appear to dominate contributions in to samples Bed 1, however, the basic fingerprint estimated high contributions from cluster 1. For samples Bed 2, 3, and 5 there was either very poor consistency between the composite fingerprints or no model with an acceptable GOF could be produced. For sample Bed 4 cluster 4 which covers a small area in the centre of the catchment dominates contributions, and for Bed 6 cluster 2 dominates.	Ironstone topsoils a minor source in all but sample Bed 3. The basic fingerprint estimates a larger contribution from ironstone than the other fingerprints.	No result produced