Chapter 6 Sediment source variability and behaviour in the Manawatu River Catchment,
6.6.3. Mixing Model Comparison
Comparison between the sediment source estimates from the two model solutions produced in this analysis along with the solution presented from Vale et al. (2016b) shows the influence individual geochemical tracers have on sediment source proportion estimation (Table 35). The S.D. < 35 % and S.D. < 40 % model solutions produce similar values to each other (Table 35; Fig 55; Fig 56) with only minor variations in relative sediment source group contributions. The standard deviations of the model outputs decrease from Vale et al. (2016b) from S.D. < 40 % to S.D. < 35 % indicating lower variation in the model outputs as a result of tracers with less variation (standard deviation) being entered into the model. The objective functions produced form the S.D. < 40 % and S.D. < 35 % (Fig. 57) were both significantly lower than the objective functions produced from Vale et al. (2016b) indicating that the models were able to produce a solution with less residual difference between the actual sample and the modelled parameters. Again, this is likely due to the lower variation input into the model. Important sediment source group differences are seen in comparison to the Vale et al. (2016b) model. The main differences were observed within the Mudstone, Gravel Terrace and Loess sediment
Group Predicted Group Membership
Mudstone Hill Subsurface Hill Surface Channel Bank Mountain Range Gravel Terrace Loess Limestone Mudstone 100 - - - - Hill Subsurface - 88.6 2.9 8.6 - - - - Hill Surface - 5.6 94.4 - - - - - Channel Bank 4.9 4.9 2.4 87.8 - - - - Mountain Range 6.7 - - 13.3 80 - - - Gravel Terrace - - - 75 25 - Loess 25 - - - - 12.5 62.5 - Limestone - - - 100
87.8 % of selected original grouped cases correctly classified
Mudstone 75 - - 25 - - - - Hill Subsurface 2.9 65.7 22.9 8.6 - - - - Hill Surface 5.6 16.7 72.2 5.6 - - - - Channel Bank 17.1 7.3 - 75.6 - - - - Mountain Range - - - 20 80 - - - Gravel Terrace - 25 - - - 50 25 - Loess 25 - - 12.5 - - 62.5 - Limestone - - - 100
sources. Mudstone and Gravel Terrace sediment sources showed significantly higher sediment proportions for S.D < 40 % and S.D. < 35 % solutions (Mudstone = 59.3 % and 61.8 %; Gravel Terrace = 7.3 % and 6.3 %) compared to 37.8 - 46.6 % and 0.2 – 3.6 % respectively (Table 35). The opposite trend is observed for Loess sediment sources whereby contributions of 2.7 % are observed for the S.D. < 40 % and S.D. < 35 % solution compared to 9.1 – 10.2 % (cf. Vale et al., 2016b). This indicates that one or more of the variables removed (i.e. Tm, NiO, Cu, CaO, P2O5, MnO and Cr2O3) from the original Vale et al. (2016b) solution through discriminant function analysis contribute to model sensitivity for those specific source groups by significantly influencing derived sediment source estimates. Whether this provided a greater certainty in the sediment source estimates is unclear, but it does demonstrate the sensitivity that these models have to single variables despite possessing a larger geochemical suite to reduce single variable effects. The remaining source groups displayed minor variations between the estimated solutions and were less affected by the different geochemical suite used.
Sediment source estimates from the S.D. < 40 % and S.D. < 35 % solutions for Hill Surface (11.5 % and 11.3 %), Hill Subsurface (6.6 % and 6.2 %), and Mountain Range (12.0 % and 11.4 %) source groups (Fig 55, Fig 56) displayed slightly lower values compared with Vale et al.
(2016b) estimates of Hill Surface = 12.1 – 16.3 %; Hill Subsurface = 9.2 - 10.8 %; and Mountain Range = 15.9 – 17.5 % respectively. The Channel Bank and Limestone source estimates from the S.D < 40 % and S.D < 35 % remained effectively 0 % for both of these sources as with values published by Vale et al. (2016b).
Table 35: Comparison of sediment source estimates with those of Vale et al. (2016b)
Sediment Source Comparison of Sediment Source Estimates
S.D. < 35 % S.D. S.D. < 40 % S.D. Vale et al. (2016b) S.D. Mudstone 61.8 % 8.8 59.3 % 9.8 37.8 – 46.6 % 10.6 Hill Subsurface 6.2 % 6.5 6.6 % 6.8 9.2 – 10.8 % 6.9 Hill Surface 11.3 % 6.7 11.5 % 7.5 12.1 – 16.3 % 6.5 Channelbank 0.3 % 1.6 0.6 % 2.1 0.0 – 4.3 % 2.7 Mountain Range 11.4 % 4.1 12.0 % 4.2 15.6 – 17.5 % 3.9 Gravel Terrace 6.3 % 6.0 7.3 % 6.0 0.2 – 3.6 % 4.3 Loess 2.7 % 4.6 2.7 % 4.6 9.1 – 15.2 % 7.5 Limestone 0.7 % 0.4 0.1 % 0.4 0.0 – 0.0 % 0.0
Fig. 57: Boxplot showing comparison of Objective Function statistics of the model solutions
6.7
Conclusion
This research showed the significant challenges to account for sediment source estimate uncertainty in sediment fingerprinting research. Notably, geochemical variation between sediment source groups as well as the geochemical variation of individual elements is an important concern pertaining to the uncertainty of sediment source estimates. In this research specific tracers were removed from the sediment mixing model due to their high geochemical variability which increased the proportion of sediment attributed to Mudstone sources by 15 %. Some of the variables which provide good discrimination between sources also have high standard deviation values, which could inhibit valid differentiation of individual sources within the model. Although the general pattern of sediment source estimation from the two mixing models in this paper displayed similar patterns to the research from Vale et al. (2016b), there was significant change in the proportions for some of the sediment sources which evidences the sensitivity sediment fingerprinting models can have to individual tracers. A major limitation however was that there was no comparison with artificial mixtures to know how accurately the models are estimating sediment source group contribution. Geochemical change was observed in sediment samples submerged within the river, but appeared to vary
between the different sediment source groups. The full extent of these changes is inconclusive and requires a substantial investigation in order to fully understand the mechanism and extent of geochemical change occurring in the sediment. Measured information could provide greater certainty by quantifying the geochemical change or informing decisions to exclude individual tracers from source estimation.
6.8
Acknowledgements
We would like to acknowledge funding support from Landcare Research by way of the Murray Jessen PhD Scholarship with which this research is made possible
6.9
Summary
This chapter reviewed the geochemical variability encountered in the Manawatu sediment source data set, finding that some of the sediment source groups i.e. Mountain Range, display a high variability compared to Mudstone sediment which has an overall lower geochemical variability. Additionally, some individual tracers provide good discrimination between sources, yet have high variability between samples. Removal of these elements from the geochemical suite used to characterize the sediment source groups can influence the estimated proportions for some sediment source groups. This has implications for how geochemical tracers are selected and highlights the need to critically assess tracers on an individual basis for their suitability and impact on source proportion estimation.