Environmental magnetism as a stream sediment tracer: an interpretation of the methodology and some case studies.
Environmental magnetism is increasingly being used in a wide range of environmental studies such as sourcing the sediments in reservoir catchments (Stott 1986); establishing stream arm sediment contributions at river confluences (Caitcheon 1993a, 1993b; Crockford and Starr 1996); sourcing estuarine sediments (Oldfield et al. 1985); characterisation of soils (Maher 1986); tracing overland soil movement (Dearing et al. 1986); identification of fire-induced magnetic oxides in soils and lake sediments (Rummary et al. 1979); and description of bacterial magnetite in lake sediments (Snowball 1994). It was used by Crockford and Willett (1995a) to help identify the chemical changes that occurred during oxidation of a sulfidic soil. Crockford and Willett (1995b) also examined the effects of simulated dam storage on the magnetic properties of 2 soils.
The original purpose of this river sediment study was to determine how much sediment was being contributed by the subcatchments of these rivers and how this was related to land use and soil erosion. Environmental magnetism was chosen as the tracer tool. As the study proceeded, a number of methodological variables emerged that had to be considered. For quantifying the sidestream inputs at river confluences, these variables included the location of the sediments at the river confluences and the particle size/magnetic properties relationships. Associated with the latter is the issue of whether crossplots of magnetic properties are used (Caitcheon 1993a, 1993b; Crockford and Starr 1996), or whether mass magnetic properties are more appropriate (Crockford and Starr 1996). For determination of the actual sources of sediments, the effects of particle breakage and abrasion on magnetic properties (Crockford and Olley in press) present added difficulties.
In this methodological study, environmental magnetism was used in the Molonglo/Queanbeyan river system to assess the inputs of sediment at the major river confluences in order to test the use of environmental magnetism as a tracer in these conditions and to relate these data to the conditions and land use in the subcatchments. This river system has the advantage of having a range of geology within the catchments (Fig. 1). The geology of the Queanbeyan catchment ranges from granite in the upper reaches, to sedimentary and volcanic rocks in the lower reaches where it enters the Molonglo River, whereas, the geology of the Molonglo catchment above the junction is a mixture of sedimentary and volcanic areas. The magnetic properties of the sedimentary and volcanic soils in this area are virtually indistinguishable (Crockford and Starr 1996). The granite and sedimentary/volcanic geologies have, however, quite different particle size/magnetic relationships, which allows a wide-ranging investigation, and influenced our understanding of the variations possible.
[Figure 1 ILLUSTRATION OMITTED]
Twelve confluences were examined, but only 3 are discussed in this paper. These confluences represent extremes of the particle size/magnetics relationships and concentration/magnetic grain size features of the magnetic minerals in the sediments.
The sidestream input estimates, at all confluences in each river, and for all particle sizes, are presented by Crockford and Starr (1996). Also included are the downstream trends in the sediment magnetics, along non-inflow reaches.
Materials and methods Catchment description
The Molonglo/Queanbeyan catchment (Fig. 1) is about 1400 [km.sup.2] in area, situated in south-eastern New South Wales. The Queanbeyan River rises in the south at a level of 1500 m, initially flowing northwards through an undulating granitic plain. After 25 km, the river enters hilly country on folded Ordovician sediments. The next 60 km, to where it joins the Molonglo, is through very hilly to mountainous terrain along a border of Ordovician sediments and volcanics.
The Molonglo River rises to the north-east of the Queanbeyan River catchment and initially flows for 10 km through Ordovician sediments to the Captains Flat dam. From there, for a similar distance, it passes through Silurian volcanics, slates, and sandstone which underlie undulating land. From Ballallaba Creek to the Dairy Station Creek, it flows through a mixture of volcanics and calnozoic sediments. From Dairy Station Creek to the confluence with the Queanbeyan it flows through Ordivician sediments.
Sediment inputs of all named tributaries were assessed by sampling at the confluences.
Collection and preparation of samples
Fig. 2 shows a river confluence. The section above the confluence, upstream, is U, the sidestream is S, and downstream of the confluence is D. The aim was to collect sediment samples from sites U, S, and D, sieve them to an appropriate size range, magnetically analyse them, then calculate sidestream inputs (SI) for all particle sizes from the magnetic data. This was done for all confluences shown in Fig. 1.
[Figure 2 ILLUSTRATION OMITTED]
The sampling system described here, and used at the reported confluences, is general to the whole catchment. The distance over which the samples were collected varied according to the amount and location of the sediments, but generally they were between 40 and 300 m from the confluence. Particular care was taken with the downstream samples. In order to be confident that they would be an equilibrated mixture of upstream and sidestream sediments, careful examination was made of the confluence in respect to the angle of entry of the sidestream and the location of sediment. Between 10 and 20 samples were taken from each of U, S, and D at each confluence and combined into a smaller number of samples or bulked. The very large proportion of sediments at all confluences could be described as sands in that there was very little organic matter, silt, or clay. Further details on the sampling method are contained in Crockford and Starr (1996).
The samples were kept in sealed plastic bags, either wet or dry, until they could be prepared for sieving and magnetic analysis. In order to minimise particle breakage, the samples were carefully wet-sieved to the following sizes: +A, 2-0-3.6 mm; A, 188.8.131.52 mm; B, 500/[micro]m-1.4 mm; C, 250-500/[micro]m; D, 125-250/[micro]m; E, 63-125/[micro]m; F, [is less than] 63/[micro]m. The sieved fractions were then dried overnight at 50 [degrees] C in an air-draught oven.
Measurement of magnetic properties and terminology
Magnetic susceptibility ([chi]) was measured at low frequency (0.45 kHz; [chi]lf) and high frequency (4.7 kHz; [chi]f) using a Bartington dual frequency sensor. This enables frequency dependence (fd) to be determined
[[chi]fd% = ([chi] lf - [chi]hf)l00/[chi]lf]
The [chi] data presented are the low frequency values. Isothermal remanent magnetisation parameters (IRM20, IRM200, and SIRM) were measured in a Molspin flux-gate magnetometer after magnetisation in a Molspin pulse magnetiser (induced at 20, 200, and 850 mT). For convenience, 850 mT is termed SIRM, although the term more strictly applies to induction at 1000 mT. Anhysteristic magnetisation (ARM) was also measured in the Molspin magnetometer after magnetisation with an anhysteristic attachment. The peak AF field used was 100 mT and the DC bias was 0.04 mT. Details of the methods may be found in Thompson and Oldfield (1986). Throughout the paper, [chi] is expressed as [10.sup.-8] [m.sup.3]/kg, and remanence as mA [m.sup.2]/kg. IRM200 was measured because the difference between SIRM and IRM200 is a measure of the presence of anti-ferromagnetic material. This parameter is not presented in this paper, but IRM200 is measured in the procedural process.
Measurements were done in the following order: susceptibility, ARM, IRM20, IRM200, and SIRM. Remanent magnetisations were induced at 1-h intervals, and read in the magnetometer 15 min after induction. These periods were chosen because (a) time between inductions affected the `next' value, and (b) viscous loss of remanence was small after 15 min. Also, induction of remanent magnetisation prior to measurement of susceptibility will enhance the susceptibility by up to 10%. Mass magnetic values are the mean of the observed values of the property from all tested samples in a given size range. If a relationship is established between 2 properties over the sample range and graphed, it is termed a crossplot (the graphing of 1 parameter against another gives an indication of magnetic grain size).
Estimation of sidestream input
Sidestream input (SI) can be calculated for each of the particle sizes from the equation
(1) SI (Sm as %Dm) = 100[1 - (Dm- Sm)/(Um- Sm)]
where SI is sidestream input, Um is the magnetic value of upstream sediment, Sm is the magnetic value of sidestream sediment, and Dm is the magnetic value of downstream sediment. This equation can be applied to the mass magnetic values, the magnetic ratios, and the slopes of the crossplots.
For a `sensible' input, Dm must be between Um and Sm. If Um = Dm, there is no input by the sidestream, i.e. SI = 0, and SI = 100% if Sm = Dm. If Dm does not lie between Um and Sm, it suggests that either (1) the samples were not truly representative at 1 or more of the sites, or (2) the samples are representative but there is a `pulse' at a site, probably at the upstream (U) or sidestream (S) site. A `pulse' could be deposition due to some unusual flow pattern, such as a sudden diminution of flow rate, causing deposition of certain materials in much greater concentrations than normal. Another possibility is material derived from a source which does not normally contribute, e.g. a bank collapse further upstream. Because the mobility of the various particle sizes is different, such pulses need not apply equally to all particle sizes.
The sampling and variability problem
Are the samples representative? Between 10 and 20 samples were taken from each branch and either bulked or combined into a smaller number of samples. They were collected from the areas of greatest accumulation, 40-300 m from the confluence. The distance varied for the branches because of location of areas of accumulation. Features of the collection sites such as bars and pools were also considered.
When estimating the sidestream input values, both concentration of magnetic minerals and the magnetic grain size (MGS) have to be considered, because magnetic grain size affects the response to the various applied magnetic fields. This is shown in Fig. 3, which is based on data presented by Maher (1986). MGS can be inferred from the ratios of mass magnetic values; but such an MGS is a composite of a range of crystal sizes of the magnetic minerals maghemite and magnetite, which may be located in different parts of the sediment particles and may therefore respond somewhat differently to the applied magnetic fields.
[Figure 3 ILLUSTRATION OMITTED]
If the samples are truly representative, the SI calculated for any particle size from the U, S, and D data will be similar for each magnetic property used ([chi], IRM20, ARM, and SIRM) if D is a mixture of U and S, in respect to both concentration and magnetic grain size. This is rarely the case. In the Molonglo catchment the SI values derived from the 4 mass properties sometimes varied by as much as 50%.
The variation of SI through the mass magnetic values can be due to the magnetic grain size of the D samples not being a genuine mixture of the U and S samples. Even if the concentrations of magnetic mineral in samples are the same, differences in MGS cause the mass values to differ, and to different extents, depending on how different the MGS differences are. This can be influenced by the position on the MGS response curves (Fig. 3) where the differences occur, e.g. if a change occurs in the 0.02-0.04 [micro]m area, a smaller magnetic grain size causes SIRM to decrease sharply and [chi] to increase, and visa versa. However, when the change is on the right-hand side of the SIRM peak, a similar-sized change will cause a much smaller SIRM/[chi] difference. The situation worsens again as the 1 [micro]m size is approached, because [chi] continues at a steady level and SIRM approaches zero.
Examination of various ratios of magnetic mass values enables the magnetic grain size and differences in magnetic grain size to be associated with the range of SI values shown by the mass properties. This leads to a better understanding of the calculated SI values and may justify a mass property being ignored. If, for example, the magnetic grain size was such that a small change could cause a large difference in property M, and little change in the others, property M can be ignored. This consideration of magnetic grain size also applies to the use of crossplots (below).
Use of mass magnetic values and magnetic ratios for measuring sidestream input
Eqn 1 can be applied to each of the mass magnetic properties averaged from all samples of a given particle size. These values of SI are then in turn averaged to give the value for that particle size. This is shown in Table 1 for the case study material.
Table 1. Sidestream input values (S as percentage of downstream, D) calculated from mass magnetic values for the Ballalaba/Molonglo confluence --, D not between U (upstream) and S
Particle size [chi] SIRM IRM20 ARM Mean CV A 100 73 87 100 90 10 B 100 94 100 100 98 3 C 68 61 73 89 73 17 D 34 25 43 55 39 33 E 34 36 78 89 59 48 F -- -- -- -- --
As indicated in the previous section, differences in MGS between the upstream and sidestream material may confound some of the values of SI using mass magnetic properties. Magnetic ratios can be used to assess this and indicate which properties should be ignored or down-weighted. The values of the ratios themselves can also be used in Eqn 1 to derive values of SI, as can the slopes of the parameter crossplots. Examples are shown in Table 2.
Table 2. Sidestream input values (S as percentage of downstream, D) calculated from ratios and crossplots for the Towneys Ck/Queanbeyan River confluence Crossplots: Zero-S, intercept = zero by subtracting z and y intercepts I = Zero, regression put through the origin WDM, weight distributed mean
Particle WDM Zero- S I=Zero Mean size ratios crossplot crossplot SIRMv. [chi] A 26 24 33 28 B 29 25 27 27 C 86 84 85 85 D 63 56 57 59 E 27 27 21 25 F 55 7 42 35 IRM20 v. [chi] A 33 33 102 56 B 21 17 19 19 C 73 78 76 76 D 63 66 58 62 E 44 48 36 43 F 69 ? 40 36 IRMZO v. SIRM A 65 59 45 56 B 79 73 79 77 C 100 92 103 98 D 90 140 86 105 E 53 50 49 51 F 53 69 55 59 Particle CV Uncertainty size A 14 37 B 6 19 C 1 29 D 5 35 E 11 28 F 58 95 A 58 88 B 9 12 C 3 17 D 6 38 E 12 56 F 78 35 A 14 45 B 4 23 C 5 28 D 23 18 E 4 22 F 12 64
In consideration of which method to use, the relative differences in the mass values and ratios are important. For example, if there is little difference in the magnetic grain sizes (as indicated by the ratios) of U and S, but the mass values are substantially different, then the input estimate of the mass values is acceptable. If, however, the magnetic grain sizes of U and S are more consistently different than the mass values, then the ratio contributions are used. The case for crossplots is more complicated as shown below.
Crossplots as a means of measuring sidestream input
The crossplots procedure involves plotting one mass magnetic property against another, for a number of samples from each of U, S, and D. This procedure relies on differences in MGS of the U and S samples, i.e. the U and S samples will be located in different parts of the MGS-response spectrum shown in Fig. 3.
An example of the crossplots procedure is shown in Fig. 4. Consider hypothetical properties M and N. If there is a significant difference in slopes, and D lies between U and S, the difference in slopes is used to calculate inputs. This procedure relies on the samples from each branch (U, S, and D) having a big range of mass values but very similar ratios of the 2 chosen properties, i.e. there will be a considerable range of M and N values from each branch but the MIN ratios for the samples from each branch will be similar, i.e. the magnetic grain sizes will be similar. This combination results in high regression coefficients, which enables estimates of SI to be confidently made, even when the MIN slopes of U, S, and D are not very different.
[Figure 4 ILLUSTRATION OMITTED]
In Fig. 4 all 3 lines pass through zero. In practice, this is rarely the case, the reasons being magnetic grain size and sample variations. The method used here involves forcing the U, S, and D lines through zero then doing an input estimate using the gradients (Caitcheon 1993a). The U and S lines are forced through the origin by adjusting each value by the value of the intercept for that parameter, i.e. the lines are raised or lowered so as to pass through zero, and the `new' lines are parallel to old ones. For convenience this process is called `intercept subtraction'. For the D line, however, the regression is put through the origin, which causes a change in its slope, called `slope adjustment'. The input equation used for mass inputs is then applied to the slopes, and the SI calculated. The confidence limit of the estimate is then calculated from the standard deviation of the individual ratio values from each site.
A potential problem with this method is that it may not always be reasonable to force the lines through zero. It will depend on which parameters are being plotted, e.g. with SIRM and [chi] it is possible to have [chi] but no SIRM (Fig. 3), causing the line to have a positive intercept with the susceptibility axis. This an occur when the magnetic grain size is in the FV-SP range, or approaching the multidomain area (see Fig. 3); the former can be recognised by the [chi]fd value.
It is not possible, however to have a positive intercept on the SIRM axis, i.e. SIRM but no susceptibility. The best procedure may be to force all 3 lines through zero by slope adjustment and sacrifice the standard deviation to some extent. Similar arguments can apply to other crossplots. In most cases, however, the spread of results is such that forcing the lines through zero creates only small differences in the SI values.
With any crossplots `impossible intercepts' can occur when samples from a site have very different ratios, i.e. different MGSs. If such a case occurs using, say, 5 or 6 samples, a more sensible relationship will almost certainly be established with a larger number. It is quite common for an `impossible intercept' to be caused by 1 or 2 samples having MGSs substantially different to the others.
The sort of regression lines acquired depends on the range of MGSs and mass magnetic values (concentrations). A reasonable line requires an appropriate combination of magnetic grain sizes and concentrations. The best of course is a wide range of concentrations with identical ratios for the individual samples from sites U, S, and D, i.e. [R.sup.2] = 1.0 for each site.
If the range of concentrations at a branch is small, even small variations in the ratio cause the points to form a cluster. This often presents as a line of a most unlikely slope and a low regression. A cluster at branch U, S, or D eliminates the possibility of calculating SI from the slopes. In this case the SI could be calculated from the means of the mass magnetic values (see previous subsection) which can also be used when the slopes of the lines are not different or ordered enough.
Choice of parameters to be used in crossplots
Crossplots will be influenced by the range of mass magnetic values for both parameters plotted for each of the 3 sites, and the variation of the magnetic grain size of the individual samples from each site.
Selection of appropriate crossplot parameters is important. Although SIRM v. [chi] is commonly used (Caitcheon 1993a, 1993b), other combinations are sometimes better; IRM20 v. [chi] is a good example of this. This ratio works best in the larger PSD-MD area (Fig. 3), i.e. in the area where there is little or no FV (fine viscous) material, normally measured as [chi] fd%. If the parameters are behaving similarly in their part of the magnetic grain size spectrum, ratio variance will be minimised.
The parameters ARM and SIRM both increase from a magnetic grain size of about 0-5 [micro] m, to about 0.05 [micro]m, where ARM is still increasing but SIRM commences its decrease (Fig. 3). Above about 0.5 pro, ARM is rapidly approaching zero, whereas SIRM is still responding. This suggests that through the grain size spectrum area of 0.05-0.5 [micro]m the possibility of useful results from these parameters is optimised. Below about 0.01/[micro]m they are both decreasing again, possibly in a similar way, but it is a very small part of the grain size spectrum and probably not useful.
A problem with ARM is that it provides a weak signal, sometimes as little as 0.3% of SIRM. For low concentrations of magnetic minerals, the ARM values approach the limits of instrument sensitivity. This same problem applies to ARM v. [chi], but the optimum grain size range is shifted to the left. When the grain size approaches FV, the ARM/[chi] ratio increases, whereas SIRM/ARM and SIRM/[chi] decrease. The parameters IRM20 and ARM are difficult to use as crossplots because of the ARM sensitivity problem and the `bimodal' behaviour of IRM20 (Fig. 3).
Examination of the ratios provides information on the magnetic grain sizes of the samples, which assists in the choice of the most appropriate properties for the crossplot. It also allows informed evaluation of the regression lines and their intercepts.
Different particle sizes may have different `best magnetic parameters' for the estimates of SI. For example, SIRM v. [chi] might be best for size E, but IRM20 v. [chi] might be best for another size. This is not surprising considering their possibly different sources within the catchment, and the very different mobilities of the particles. At collection time, size A may have been deposited under different flow conditions to size E, and therefore at a different time.
For the crossplots method, the slopes, the regressions, and the intercepts data from U, S, and D, in conjunction with the MGS information derived from the ratios, are examined. The choice of crossplot parameters most likely to be appropriate will depend on the dominant magnetic grain size at U and S. For example, SIRM v. [chi] may be the strongest (in terms of the regression line) for U, but another combination such as IRM20 v. [chi] or SIRM v. ARM may be strongest for S. In this case, a compromise crossplot could be chosen after examination of the regression and magnetic grain size data, ie. alternatives could be tried and evaluated.
Two confluences were examined in detail, one from the sedimentary/volcanics area of the catchment and the other from a granite area. These confluences present extremes in the magnetic properties of the sediments. The magnetic properties of the sediments in the sedimentary/volcanics and granitic parts of the catchment are different in the following ways.
(a) The concentration of magnetic minerals changes through the particle size
range quite differently, as shown in Fig. 5, data for which are contained
in Crockford and Starr (1996). For the sedimentaries/volcanics, the
concentration consistently diminishes from the large size A to size E.
It often increases slightly for size F ([is less than] 63/[micro]m). For the
granites, the reverse occurs, but with size F values sometimes being smaller
than size E. The size F values of both systems will be influenced by the
range of particle sizes contained within size F. These patterns are shown by
all measured magnetic properties ([chi], SIRM, IRM20, and ARM), but to
different extents, depending on the magnetic grain sizes.
(b) The magnetic grain sizes are different for each geology. In this catchment the magnetic grain sizes of the granites are much larger than those of the sedimentaries/volcanics. This is shown by the very low values for [chi]fd% recorded by the granites. This parameter measures the fine viscous (FV) area of the magnetic grain size response spectrum (Fig. 3). Another parameter which unequivocally senses magnetic grain size in the 0.06-0.03 [micro] m area is SIRM/ARM. The granite values are more than double those of the sedimentaries/volcanics (Crockford and Starr 1996). In other catchments, the sedimentary-granites differences in magnetic grain size are not always as obvious as in this one, but the concentration-particle size relationships (Fig. 5) are always shown.
[Figures 3 and 5 ILLUSTRATION OMITTED]
Sidestream inputs at the Ballalaba/Molonglo confluence
This confluence is in the upper reaches of the Molonglo river (Fig. 1), with sedimentary/volcanics geology.
Sidestream input calculated from mass magnetic values
The SI values in Table 1 are the means of the individual samples of each particle size, for each magnetic parameter. They are `weight distributed means'. This takes into account different proportions of the particular sizes in the replicate samples, and is therefore identical to a bulked sample.
Apart from size F, all sizes convincingly show substantial contributions for all mass properties. Different sizes within the size F ([is less than] 63 [micro] m) fraction can, and often do, cause impossible inputs by S, i.e. the S values being much greater or much less than U and D. If the magnetic properties of the [is less than] 63 [micro]m subsizes are different, the result will depend on the relative proportions of the subsizes in U, S, and D. For example, if the 40-63[micro]m subsize was, say, 80% of size F at branch U, but S had 80% of the 2-10 [micro]m size and the magnetic properties of these subsizes were different, the resultant SI will be meaningless or misleading.
Sidestream input calculated from crossplots
Plots of SIRM v. [chi], IRM20 v. [chi], and IRM20 v. SIRM are presented by Crockford and Starr (1996) for all particle sizes. For almost all crossplots, clustering and spread are obvious for samples from one or more of the branches, and these are often associated with magnetically impossible intercepts. An example is shown in Fig. 6 for size E. The downstream values are rather clustered and form an impossible y-axis (SIRM) intercept. Other examples are contained in Crockford and Starr (1996). The regressions are noted on the Figures. Even where these values are good for each branch, the slopes of the branches can be so similar that no meaningful assessment of SI can be made. An example is Fig. 7, a plot of SIRM v. IRM20 for size B.
[Figures 6 and 7 ILLUSTRATION OMITTED]
For all figures, the regression lines are meaningless for calculation of SI because the U, S, and D slopes are not generally sufficiently different to allow sensible input estimates, some of the intercepts are magnetically impossible, and in many cases D does not lie between U and S.
A summary of the results from this confluence is that ratios and crossplots do not work, but that realistic estimates of sidestream inputs are provided by the mass magnetic values (Table 1). Except for size F, the sidestream inputs calculated from the bulked means of the mass magnetic values through the size range are very consistent and substantial. Thought was given to subsizing the [is less than] 63 [micro] m (size F) fraction to see how the magnetic properties and sidestream inputs varied through these particle sizes. As size F was only 2% of the total at each site, there was too little sample for size fractionation. At each branch, sizes B and C (1.4 mm-250 [micro]m) represented 70-80% of the sediments.
Sidestream inputs at the Towneys/Queanbeyan confluence
The Towneys Ck/Queanbeyan River confluence is in the granites area of the catchment (Fig. 1). As shown previously, the mineral magnetics of this confluence are different to those from the sedimentary/volcanics areas of this catchment. At this confluence, there are substantial differences in the MGS of the sediments at U and S, which enhances the prospects for crossplots as the input assessment procedure. For this reason the inputs calculated for crossplots and ratios will be discussed first.
Sidestream input calculated from crossplots
As stated earlier, an issue that must be kept in mind when using the crossplots procedure is that a particular crossplot may not be best for all sizes. As an example, consider crossplots of SIRM v. [chi] and IRM20 v. [chi] for size C (Figs 8 and 9). The regressions for IRM20 v. [chi] (Fig. 9) are much better than SIRM v. [chi] (Fig. 8) for sites U and S, allowing a more confident estimation of sidestream input.
[Figures 8 and 9 ILLUSTRATION OMITTED]
This is reasonable when one considers the dominant magnetic grain sizes. U is in the PSD to MD area where IRM20 and [chi] are increasing but SIRM is decreasing (Fig. 3), which means that small MGS differences between replicate samples can cause large differences in the SIRM/[chi] ratios. This results in an uncertain slope and a poor regression. If both parameters move in the same direction, the ratios are not so badly affected. Similar responses were shown by other particle sizes (Crockford and Starr 1996).
The following procedures were used for assessment of SI from the ratios SIRM/[chi], IRM20/[chi], IRM20/SIRM, and crossplots of these combinations: (1) bulked sample ratios derived from weight-distributed means of the individual sample values; (2) crossplots where the U and S lines were forced through zero by the `intercept subtractions' method, and D by slope adjustment; (3) crossplots where all 3 lines were forced through zero by slope adjustment.
Table 2 gives the SI values, means, and CVs of the 3 ratios used. Except for size F (the particle size/magnetics trouble, probably), the 3 methods give similar results for the individual particle sizes for each ratio; and the coefficients of variations are reasonable. But SI does vary between the procedures and ratios for some sizes, size A in particular. For this size, the SI value for I = zero for ratio IRM20/[chi] is almost 3 times the values for procedures WDM and ZERO-S, resulting in a CV of 58%. A summary of the mean inputs resulting from the 3 procedures is given in Table 3.
Table 3. Mean sidestream input values (S as percentage of downstream, D) resulting from the three procedures in Table 2 Particle size SIRM v. [chi] IRM20 v. [chi] IRM20 v. SIRM A 28 56 56 B 27 19 77 C 85 76 98 D 59 62 105 E 25 43 51 F 35 36 59
Consideration of the validity of individual crossplots
The U line for size E for SIRM v. [chi] (Fig. 10) shows a zero slope and a large intercept with the SIRM axis. For reasons stated earlier, the latter is impossible. Examination of Fig. 10 shows that 1 sample (circled) is responsible for the horizontal U line. This is sample No. 1 of the upstream samples. Its magnetic ratios are very different to the others from this site (U) (Crockford and Starr 1996). These ratio differences show that it has a smaller magnetic grain size than the others, higher SIRM/[chi], ARM/[chi], and IRM20/[chi], and lower IRM20/SIRM, SIRM/ARM, and IRM20/ARM, i.e. No. 1 is near the SIRM peak of Fig. 3, while the others are in the PSD area.
[Figure 11 ILLUSTRATION OMITTED]
If this sample is omitted from the U data set, the U slope becomes steeper (0.012 compared with 0.0058) and the standard deviation is smaller. Although the intercepts with the SIRM axis are reduced from 7.2 to 3.1 it is still quite substantial, which suggests that the rest of the sample set may also not be `correct'.
It is noted that if the concentration of magnetics in sample No. 1 is increased by about 60-70%, without changing the ratios, it steepens the slope such that the intercept is about zero. Its ratio differences still label it as an outlier, but it does emphasise the interaction of mass values and slopes (ratios) when using the crossplots method. It is also noted that the input values derived from the bulk sample values assume zero intercepts.
Concerning the labelling of sample No. I as an outlier and its possible exclusion from the data set, it meets the commonly accepted exclusion criterion of its value being at least 3 standard deviations from the mean of the other samples; in fact it is 5-0 times the standard deviation. This sample has higher values (2-3 times) of SIRM/[chi], IRM20/[chi], and ARM/[chi], and lower values (about half) for IRM20/SIRM (Crockford and Starr 1996), which show that it has a smaller magnetic grain size, i.e. slightly further to the left in the PSD area of Fig. 3. But why is it so different to the others? The description of its location was very similar to those of the other sample sites. For this sample, sizes C and D show similar ratios to size E, although the lines are not quite as horizontal (Crockford and Starr 1996).
Sidestream input calculated from mass magnetic values
The sidestream input values calculated from mass magnetic values are given in Table 4. Susceptibility is the only property to show SI values for all particle sizes, i.e. it is the only property for which the D value was consistently between the U and S values. For the other properties, the inputs seem quite random. The likely reason for this difference is the response pattern of the properties through the magnetic grain size range (Fig. 3). The magnetic grain sizes of these samples are within the PSD area where SIRM, ARM, and IRM20 respond quite differently; and small differences in magnetic grain size can cause large differences in the ratios. In contrast, the response curve for x is reasonably flat in this area. For sizes A, B, and E, the SI values derived from x are within the ranges of the bulk ratios and crossplots methods for all particle sizes (Table 2).
Table 4. Sidestream input values (S as percentage of downstream, D) Calculated from mass magnetic values for the Townys Creek/Queanbeyan River confluence
--, D not between U (upstream) and S
Size x SIRM IRM20 ARM Mean A 56 33 -- 22 B 61 31 -- 14 C 30 -- -- -- D 27 -- 100 -- E 49 90 97 29 66 F 23 -- 0 81
In another case study in the Queanbeyan catchment, the Woolpack Creek/Queanbeyan River confluence, it was found that the mass magnetic properties provided better estimates of sidestream input than crossplots or magnetic ratios (Crockford and Starr 1996). This was because `clustering' of concentrations and variable magnetic grain size values caused impossible intercepts, even though the magnetic grain sizes of the upstream and sidestream samples were sufficiently different to provide quite different slopes.
If there is a wide range of mass magnetic values for the samples from each branch and the U and S slopes are sound and sufficiently different, the crossplots procedure for estimation of sidestream input is appropriate, but attention must be paid to the magnetic grain size data in order to choose the most appropriate properties for the crossplots. If, however, the range of mass values is small, even for only one branch, then the mass magnetic values method might be best because, due to clustering, the magnetic parameters slope of this branch may be meaningless.
An interesting point about these alternative methods is that a wide range of mass magnetic values from each site for the chosen parameters will enhance the crossplots procedure but limit the mass values procedure, in that it will adversely affect the statistical confidence values. However, a limited range of mass values reverses this effect, i.e. it enhances the mass magnetic values procedure and disadvantages the crossplots method.
A possible advantage of the mass magnetic values method is that the sidestream input calculated is the mean of all 4 parameters, x, SIRM, ARM, and IRM20, and should therefore take into account, to some extent, the mixture of magnetic grain sizes in the samples.
It is likely that sidestream contributions will vary through the particle size range even where there is no evidence of `pulse' or sampling problems. This can be caused by (1) difference in the mobility of the various sizes (obviously the finer sizes are more mobile than the larger ones, and where the flow rate and dynamics of S are less than U, it is reasonable to expect smaller inputs by the larger sizes); and (2) very different particle size distributions of the samples from U and S. It is possible to have a substantial input by a particular size, and very little input by other sizes if this size was in greater abundance in S than at U. This would generally be one of the smaller sizes. This is shown by size F and its subsizes at the Primrose/Molonglo confluence (Crockford and Starr 1996).
In evaluating the outcomes, it is useful to examine the particle size distribution of the U, S, and D sediments, and the calculated S inputs for the range of particle sizes. Other information such as streamflow data and land use are also useful in interpretation of the results.
The findings of this work are restricted to river confluences. If the original sources of the sediment are of interest, then possible changes in mineral magnetics occurring during transport have to be considered. In a `simulated' transport breakage and abrasion experiment, Crockford and Olley (in press) found dramatic changes in the magnetics of the particles derived from breakage and abrasion of 2 soils compared with the same-sized particles in the original material. The effects were much greater for the sedimentary soil. These issues must be considered when using mineral magnetics for sourcing sediments. Other tracing techniques such as geochemistry and radiometrics may be similarly affected.
An awareness of the variables discussed in this paper will maximise the use of mineral magnetism as a sediment tracer at river confluences.
Caitcheon, G. G. (1993a) Sediment source tracing using environmental magnetism: a new approach with examples from Australia. Hydrological Processes 7, 349-58.
Caitcheon, G. G. (1993b) Applying environmental magnetism to sediment tracing. In `Tracers in Hydrology'. Proceedings of the Yokohama Symposium. IAHS Publ. no. 215, 285-92.
Crockford, R. H., and Olley, J. M. (in press). The effect of particle breakage and abrasion on the magnetic properties of two soils. Hydrological Processes (In press.)
Crockford, R. H., and Starr, B. (1996). The magnetics of the sediments in the Molonglo/Queanbeyan river system. CSIRO Division of Water Resources, Technical Memorandum 96.19.
Crockford, R. H., and Willett, I. R. (1995a) Drying and oxidation effects on the magnetic properties of sulfidic material during oxidation. Australian Journal of Soil Research 33, 19-29.
Crockford, R. H., and Willett, I. R. (1995b). Magnetic properties of two soils during reduction, drying and oxidation. Australian Journal of Soil Research 33, 597-609.
Dearing, J. A., Morton, R. I., Price, T. W, and Foster, I. D. L. (1986). Tracing movements of top soil by magnetic measurements: two case studies. Physics of the Earth and Planetary Interiors 42, 93-104.
Maher, B. A. (1986). Characterisation of soils by mineral magnetic measurements. Physics of the Earth and Planetary Interiors 42, 76-92.
Oldfield, F., Maher, B. A., Donoghue, J., and Pierce, J. (1985). Particle-size related, mineral magnetic source sediment linkages in the Rhode River catchment, Maryland, USA. Journal of the Geological Society London 142, 1035-46.
Rummery, T. A., Bloemendal, J., Dearing, J., and Oldfield, F. (1979). The persistence of fire-induced magnetic oxides in soils and lake sediments. Annals of Geophysics 35, 103-7.
Snowball, I. F. (1994). Bacterial magnetite and the magnetic properties of sediments in a Swedish lake. Earth and Planetary Letters 126, 129-42.
Stott, A. P. (1986). Sediment tracing in a reservoir-catchment system using a magnetic mixing model. Physics of the Earth and Planetary Interiors 42, 105-12.
Thompson, R., and Oldfield, F. (1986). `Environmental Magnetism.' (Allen & Unwin: London.)
Manuscript received 21 April 1997, accepted 14 October 1997
R. H. Crockford and P. M. Fleming
CSIRO Land and Water, Canberra Laboratory, PO Box 1666, Canberra, ACT, 2601, Australia.
|Printer friendly Cite/link Email Feedback|
|Author:||Crockford, R.H.; Fleming, P.M.|
|Publication:||Australian Journal of Soil Research|
|Date:||Jan 1, 1998|
|Previous Article:||Rainfall erosivity and its estimation for Australia's tropics.|
|Next Article:||Effect of previous additions of superphosphate on sorption of phosphate.|