@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix dc: . @prefix skos: . vivo:departmentOrSchool "Arts, Faculty of"@en, "Geography, Department of"@en ; edm:dataProvider "DSpace"@en ; ns0:degreeCampus "UBCV"@en ; dcterms:creator "Richards, George"@en ; dcterms:issued "2009-08-06T00:00:00"@en, "2001"@en ; vivo:relatedDegree "Master of Science - MSc"@en ; ns0:degreeGrantor "University of British Columbia"@en ; dcterms:description """Measuring suspended sediment concentration (SSC) is both costly and labour intensive. Temporal records of SSC are, however, of paramount importance in elucidating issues relating to geomorphology, ecology and water quality. Rating curves, that relate SSC and discharge by a simple linear regression function, are frequently employed by workers to address the problems of recording SSC. Such functions, however, rarely account for more than 50% of the variability in observed SSC. The aim of this thesis is to formulate subseasonal predictive SSC models and to investigate hydrologic controls on proglacial suspended sediment dynamics using data collected from a glacier-fed stream, Coast Mountains, British Columbia. In order to model proglacial SSC, the hydrologic season was initially divided into nival, nival-glacial, glacial and autumn recession periods, according to sudden shifts in the ratio of stream discharge between the glacierised and a neighbouring unglacierised catchment of similar size and aspect. Multiple regression functions, to predict SSC, were then developed for each period. These regression models incorporate a suite of easily measured variables and are shown to reduce significantly, initial problems of autocorrelation, heteroskedasticity and non-linearity of the SSC-discharge relationship. Analysis of the significant parameters in the multiple regression models, the hysteretic relationship between SSC and discharge, and downstream changes in SSC reveal that short-term, within channel, storage of fine sediment may be an important control on proglacial suspended sediment dynamics in this complex glaciofluvial lacustrine system."""@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/11827?expand=metadata"@en ; dcterms:extent "6255961 bytes"@en ; dc:format "application/pdf"@en ; skos:note "HYDROLOGIC CONTROL ON PROGLACIAL SUSPENDED SEDIMENT DYNAMICS By GEORGE RICHARDS B.A. (Hons), University of.Oxford, 1999 THESIS SUBMITTED IN PARTIAL FULFILLMENT THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE i n THE FACULTY OF GRADUATE STUDIES (Department of Geography) We accept t h i s thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA September 2001 © George Richards, 2001 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of fe°6&APK7 The University of British Columbia Vancouver, Canada Date K fcP120 mg/L), approximately 0.7% (0.7 mg/L). Thus measurement errors were small. A n important question is whether the ISCO samples were representative of the stream cross section. Table 3.1 summarises results of a paired sample T-test that was used to test the hypothesis: Ho : p:D = 0 H i : P-D ^ 0, where p,D is the mean of the differences between paired ISCO and right bank DH48 samples. Table 3.1: Results of ISCO - DH48 Paired Sample T-test T-test information Value n 24 Degrees of Freedom (df) 23 T-statistic (T t) -1.206 T-critical for Two Tailed Test (T c ) 2.069 P (1TTI < T c ) 0.240 The T-statistic is less than the critical T-value (Table 3.1) and so there is no evidence to reject Ho at a significance level of 5%, thus indicating that the ISCO sampler provided unbiased estimates of SSC in comparison with the DH48 sampler. Figure 3.3a shows a -scatter plot of ISCO versus DH48 values of SSC and the proximity of the points to the 1:1 line. 30 35 Figure 3.3: (a) ISCO versus DH48 values of SSC for all samples and (b) left versus right bank DH48 values of SSC. 31 To test the hypothesis that there was a difference in SSC between the ISCO sampler and the eight paired DH48 samples, SSC values were analysed by a two-way Analysis O f Variance ( A N O V A ) using the General Linear Model function in S Y S T A T 9.0. The factors were time of sampling (8 levels) and type of sample (3 levels: ISCO, DH48 right bank, DH48 left bank). Results indicated that time of sampling was significant (p<0.001) and type of sample was insignificant (p=0.502). There is, therefore, evidence that sampling location does not have a significant influence on SSC. Again, Figure 3.3b shows a plot of left versus right bank SSC, indicating that all points plot close to the 1:1 line. 3.3 Streamflow and SSC variations 3.3.1 Place Creek discharge and SSC time series A continuous record of discharge was obtained from 12 May to 31 Nov 2000, after which streamflow was minimal and thus not of interest for the purpose of this thesis (Figure 3.4). The SSC series stretched from 18 May to 8 Oct 2000, but was discontinuous. Hourly air temperature (TA) and daily precipitation (P) records spanned the periods 19 May to 26 Sept and 19 May to 30 Nov, respectively. Through May and June, background discharge rose and was punctuated by rainfall events on 20 to 21 May (11.4 mm), 26 to 27 May (18.5 mm), 5 June (13.7 mm), 18 to 19 June (7.4 mm) and 1 to 2 July (12.4 mm), and a snowmelt event on 28 to 29 June (Figure 3.4). Following a lull in discharge during early July, discharge rose again and was characterised by an increasing diurnal oscillatory amplitude. From 27 to 28 July, 31.3 mm of rainfall produced the largest recorded discharge of the season (3.83 m3/s). During early August discharge fluctuated daily, from around 2.0 to 2.8 m Is. Cooler air temperatures then 32 25 4-May jli! U L J _ J i L 3-Jun 3-Jul 2-Aug 1-Sep 1-Oct 31-Oct 30-Nov Figure 3.4: Time series of discharge, SSC, air temperature and precipitation for Place Creek (2000 season). 33 prevailed from 10 to 21 Aug and average daily discharge fell from approximately 2.4 to 1.3 m 3/s. After a period of renewed glacial melt from 22 to 26 Aug, induced by warm air temperatures, background discharge followed a seasonal recession trend and was again punctuated by rainfall and late glacial melt events. From 10 to 30 Nov, discharge remained minimal at <0.1 m 3/s. A tentative relation between SSC and discharge can be noted from Figure 3.4, in which peaks in SSC generally corresponded to peaks in discharge. Following the first two SSC events of the season, during mid-May and at the beginning of June, SSC declined from around 40 to 10 mg/L, despite similar discharge event magnitudes. During July, SSC rose and peaked at 123 mg/L during the storm event on 27 to 28 July. For the first 10 days in August, SSC mimicked diurnal flow cycles, but SSC then decreased in mid-August during a dry period of cooler air temperatures. No values of S S C were recorded from 3 to 25 Sep because of time constraints on access to the field site. The last field campaign of the season recorded SSC during the rain event of 28 to 30 Sep, during which similar peak concentrations (43 mg/L) to those at the same discharge during the rain storm on 29 Aug, were measured (44 mg/L). 3.3.2 Eight Mile Creek discharge and SSC time series The streamflow record for Eight Mile Creek covered the period from 14 May to 31 Nov 2000, again beyond which discharge was minimal and thus not of interest for this investigation (Figure 3.5). Following a period of high discharge during late June and subsequent shift in the stage-discharge rating curve (Figure 2.8b), a break in the discharge series was experienced from 2 July 00:00 to 6 July 15:50, after which the stilling well was 34 4-May 3-Jun 3-Jul 2-Aug 1-Sep 1-Oct 31-Oct 30-Nov Figure 3.5: Time series of discharge, SSC, air temperature and precipitation for Eight Mile Creek (2000 season). 35 re-secured. The record of SSC ranged from 19 May to 3 Sep 2000 and as for Place Creek, the series was sporadic. At the start of the season, discharge rose from 0.6 m 3/s to around 5 m 3/s at the end of June. During this period streamflow responded to the same rainfall and snowmelt events as •a Place Creek did. Following a drop in discharge to just above 2 m /s at the beginning of July, background flow was approximately 2.5 m 3/s, but displayed similar synoptic scale increases as for Place Creek. Despite maximum daily air temperatures of above 30 ° C during 24 to 30 June, streamflow dropped rapidly over this period. Background levels of discharge remained at about 2.5 m 3/s through July. The storm event on 27 to 28 July, which produced the maximum seasonal discharge in Place Creek, did not have such a significant impact in Eight Mile Creek. Discharge then declined rapidly, from 6 to 22 Aug, and then more steadily through to the end of the season. Again, during this latter portion of the season, synoptic scale storm events induced responses in discharge similar to those in Place Creek. Background SSC in Eight Mile Creek was generally <10 mg/L, except during periods of elevated discharge. The maximum recorded SSC over the measurement season was 56 mg/L, which occurred during the first rain event of the season (20 to 23 May). The following three events (same dates as for Place Creek), all of similar peak discharges (around 5 m /s), were associated with progressively declining SSC, from >51 mg/L during the second rain event, to 23 and then 8 mg/L for the third and fourth events, respectively. The only other significant periods of suspended sediment transport occurred during the second of two discharge events, closely spaced over time, on 23 and 28 July. For the remainder of the season, minor fluctuations of SSC above 0 mg/L can be attributed to small rain events and possibly measurement error. 36 3.3.3 Standardised discharge series During the snowmelt and autumn portion of the season, streamflow was greater in Eight Mile than in Place Creek because the basin area of Eight Mile is almost double that of Place Creek and thus collects more snowmelt and rainfall runoff. Time series of standardised discharge (Q s), that is discharge per unit drainage area (m 3 s\"1 km 2), reveal a good correspondence during the major period of snowmelt and spring storm events, from 14 May to June 30 (Figure 3.6). For this period, the linear relation between the two time series is QS(EM) = 0.02 + 0.82QS(PC) (r 2 = 0.949, SEE=0.01), (3.1) where S E E is the standard error of the estimate. From the beginning of July, melt from Place Glacier began to make a significant contribution to streamflow in Place Creek as inferred from the increase in diurnal discharge cycles. In Eight Mile Creek, however, flows generally decreased from July to October. The differences in standardised discharge for this period probably reflected the effects of glacier melt in Place Creek. From the end of October until the end of the season, Eight Mile yielded more discharge per km than Place Creek, which was probably because there are greater volumes of groundwater in the former basin. 3.4 Seasonal hydrograph division 3.4.1 CUSUMplot of Q ratio The use of cumulative sums ( C U S U M S ) provides a powerful technique for identifying regime shifts in time series (Mac Nally and Hart, 1997). For this sub-section, the discharge series for Place Creek will be sub-divided using the time-ordered E M / P C discharge ratio series. The C U S U M S are thus defined as the running tally of the deviations of each value of the E M / P C discharge ratio from the mean of the series, such that 37 0.00 0.30 0.25+ ^0.20 0.15 S g o . i o 0.05 (b) i i i | ! 1 1 ' -400 4-May 3-Jun 3-Jul 2-Aug 1-Sep 1-Oct 31-Oct 30-Nov Figure 3.6: Standardised discharge time series for (a) Place Creek and (b) Eight Mile Creek. Arrows on the (c) C U S U M S plot indicate sub-seasonal transition dates. 38 Sr = Sr_l + Qr (3.2) where Sr is the r* partial sum, Sr.i is the previous partial sum, and Qr is the r t h value of and Qr is the mean of the EM/PC discharge ratio series. The CUSUMS can then be plotted as a time series (Figure 3.6c). Abrupt shifts in the slope of the plot indicate that the ratio between Eight Mile and Place Creek discharges has changed. According to the greatest changes in gradient of Figure 3.6c, the season can be sub-divided into a 'nival' (N) period (start of the season to 17 June), a 'nival-glacial' (NG) transition period (18 June to 23 July), a 'glacial' (G) period (24 July to 2 Oct) and an 'autumn recession' (AR) period (3 Oct to 30 Nov). Discharge and SSC varied amongst the 4 sub-seasons (Table 3.2). Table 3.2: Descriptive Statistics for Place Creek Sub-seasons. SSC Q (mg/L) (mVs) N N G G A R N N G G A R Mean 9.9 16.3 27.7 10.6 0.84 1.71 1.45 0.23 Standard Deviation 11.7 7.6 12.9 2.9 0.50 0.39 0.61 0.19 Minimum 0.0 1.5 8.8 7.4 0.00 0.97 0.52 0.02 Maximum 70.6 42.5 122.6 16.8 2.55 2.54 3.83 1.33 No. Data Points 104 233 388 11 5238 5184 10224 8496 3.4.2 Relations between electrical conductivity and discharge Unfortunately, laboratory restrictions did not permit HCO3\" to be determined and so a charge balance calculation for measuring chemical analysis accuracy could not be performed. Given that total anion concentrations ( 2 \" ) should equal total cation concentrations (Z+), however, the strong relation between EC and S + (Figure 3.7) indicates that EC was a valid measure of Total Dissolved Solids (TDS) for Place Creek. 39 40 To support the validity of the C U S U M S method for defining sub-seasons of the Place Creek discharge series, the relation between E C and discharge for Place Creek, divided into the 4 sub-seasons, can be examined (Figure 3.8). In the literature, hydrograph separation using E C has been successful in separating hydrograph components (Teti, 1979; Laudon, 1995). From Figure 3.8 it appears that there are three distinct groupings of points: a concave-up relation for the nival period, a concave-up relation for the combined glacial and autumn recession periods, and an unstructured group for the nival-glacial transition. These groupings indicate that water sources and/or flow paths varied amongst the hydrologic sub-seasons defined by the C U S U M S technique, supporting their validity. 3.5 SSC-Q relations for Place Creek 3.5.1 Seasonal SSC-Q relation The time series of SSC and Q (Figure 3.4) indicated that there was a tentative relation between SSC and Q. Accordingly, the simplest way to predict S S C is from a S S C - Q rating curve. A common technique for constructing the rating relation is to use Ordinary Least Squares (OLS) regression, following the function SSC, = b0+b1Q,, (3.3) where SSC, is the estimated value of SSC, subscript i represents a particular value at the time of sampling, and bo is the intercept and bi the slope of the function. The parameters for the fitted O L S equation for SSC are shown in Table 3.3. 41 30 26 H • • • Nival A NivalGlacial o Glacial o Autumn Recession 22 A 18H 14H *A A A A^ * A « ^ * A & A * ^ A ^ A * A A A O A A floo %> g o o § 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Q (m7s) Figure 3.8: E C - Q relations for Place Creek. Data points are grouped by sub-seasonal division dates defined in Figure 3.6c. 42 Table 3.3: Summary of O L S S S C - Q Seasonal Regression Model (values in brackets are the p-values for bo and bi). n r 2 SEE F-ratio bo bi (mg/L) (mg/L) mg LV(m 3 s1) 686 0.445 9.8 548.94 -2.36 (0.026) 14.86 (<0.001) Consistent with the relatively low r 2 value, there is considerable scatter about the regression line (Figure 3.9a). Such low model accuracy is widely noted in the literature (reasons outlined in Section 1.2). The significance of the seasonal S S C - Q model and its predictive ability cannot, however, be correctly interpreted unless the assumptions of regression are met. These assumptions are: normality of the error terms, equal variances of the errors, independence between error terms, and linearity. Plots of the residuals of the fitted model (Table 3.3) reveal violations of the assumptions (Figure 3.9b, c, d). As values of SSCt increase, absolute values of e, (that is SSCr SSCt) increase, indicating heteroskedasticity (Figure 3.9b). Beyond SSC, values of 35 mg/L there are no negative values of e„ which provides evidence for a lack of fit in the model (Figure 3.9b). Figure 3.9c illustrates autocorrelation (i.e., time dependence) of the residuals. Over the entire season a trend towards higher positive residuals, and thus SSC underestimation, can be seen. On a shorter timescale, this trend is also punctuated by marked short-term SSC underestimation, particularly during the rainfall induced discharge events described in Sub-section 3.3.1. Violations of the assumptions were tested statistically. To test for normality, the Shapiro-Wilks (W) test (Neter et al., 1996) was used under the hypothesis: 43 44 Ho: error terms normal H i : error terms not normal The probability value for the computed W-statistic was less than 0.001, indicating that at a significance level of 0.05, Ho can be rejected. Thus, it can be concluded the error terms are not distributed normally. Homoskedasticity was tested for by generating a 'pseudo-White's' test statistic (x ) using SAS 6.12. The hypothesis tested was: Ho: equal variances H i : unequal variances The probability value for the %2-statistic was less than 0.001. At a 5% significance level %2 is thus significant and the null hypothesis can be rejected. It can be concluded that there is evidence for heteroskedasticity. For independence, a Durbin-Watson (DW) test was performed under the hypothesis: Ho:p = 0 H i : p * 0 , where p is the lag-1 autocorrelation amongst the error terms. The D W test statistic was 0.269. At a 5% significance level, the critical bounds for the test statistic were 1.65 for the lower and 1.69 for the upper bound (for >100 samples and 1 parameter (QJ). The test statistic is below the lower bound and so Ho can be rejected, thus suggesting the presence of residual autocorrelation. 45 Lastly, for lack of fit, the residual plot was assessed visually and concluded to be non-linear at the upper end. The linearity assumption must, therefore, also be rejected. In an attempt to reduce heteroskedasticity and normalise the variable distributions a log e transformation of the discharge and SSC data was performed. The log e transformation linearised the S S C - Q relation for high values of discharge, but a large number of low values of SSC then exhibited a non-linear relation with streamflow (Figure 3.9d). Additionally, the variances clearly became heteroskedastic. Thus, transforming the variables did not appear to provide a solution to the problems of non-linearity and heteroskedasticity. 3.5.2 Sub-season SSC-Q relations This section tests the hypothesis that the nature of the S S C - Q relation for Place Creek differs among the hydrologic periods defined in Sub-section 3.4.1. Given that the autumn recession period only has 11 data points, which are all spaced between 3 to 8 Oct, this sub-season will be ignored. Further analyses and reference to the \"Entire season\" thus refer to the other three periods. Given that the data set has been sub-divided into three groups it is not possible to compare r 2 values, either between sub-seasons or against the whole season, because the data in each group are different. By introducing a dummy variable into a S S C - Q O L S model for the first three sub-seasons, however, a single value of r can be generated and compared to the value for the Entire season. Beginning with an O L S model for the Entire season (Table 3.4), a dummy variable for each sub-season was introduced, leading to the expression SSC, = b 0 + bi Qt +b2Su + b3Si2 + bAQiSu + b5QiSi2, (3.4) . 46 where Su = 1 if nival period, otherwise 0; and 5,2 = 1 if nival-glacial period, otherwise 0. Equation 3.4 includes interaction terms for both the slope and intercept of the original Entire season rating function (Table 3.4). The R value for the dummy-season model is 0.648, an increase of 0.202 from the Entire season model, and the S E E has also decreased from 9.8 to 7.8 mg/L (Table 3.4). Thus the dummy-season model has improved the accuracy of SSC predictions. Statistical analysis revealed significantly different slopes and intercepts amongst subseasons (details provided in Appendix A) . Table 3.4: Summary of S S C - Q O L S Regression Models (values in brackets are the p-values for bo and bi). Sub-season n r 1 SEE (mg/L) F-ratio b 0 (mg/L) bi mg I/V(m3 s1) Entire season 675 0.446 9.8 540.83 -3.20 (0.004) 15.31 (<0.001) Dummy model 675 0.684 7.8 246.70 - -Nival 104 0.583 7.6 142.70 -18.10 (<0.001) 31.32 (<0.001) Nival-glacial 233 0.393 5.9 149.57 -4.38 (0.012) 11.86 (<0.001) Glacial 338 0.516 9.0 358.43 -1.33 (0.367) 15.67 (<0.001) Comparing the residual plot for the sub-seasonal model against the Entire season model (Figure 3.10), it appears that the residuals are more homoskedastic and linear in the former, which indicates that seasonal division has also improved the validity of the regression assumptions. From Figure 3.11, however, it is clear that although the different slopes and intercepts of the sub-season models account for general seasonal differences in hydrologic control on proglacial suspended sediment dynamics, over 30% scatter remains to be explained. In order to account for remaining scatter, a number of other explanatory variables were introduced into each sub-season rating function. 47 E OT 0) DC E, 75 3 •g 'in v rr 10 15 20 25 30 35 40 45 50 Predicted SSC (mg/L) 55 60 Figure 3.10: (a) Seasonal versus (b) sub-seasonal S S C - Q relation residual plots. 48 £ 40 H 30 H 20 H 10 H x Nival A Nival-Glacial o Glacial 0.0 Glacial Nival-Glacial 4.0 Q (m7s) Figure 3.11: Sub-seasonal S S C - Q rating relations (note that S S C scale has been reduced so that the grouping of points can be more readily distinguished). 49 3.5.3 Multiple regression models A suite of variables from which to select for Multiple Linear Regression ( M L R ) was derived from published literature and consideration of local factors that may have influenced proglacial suspended sediment transport in Place Creek. Reasons for selecting these variables are outlined below. Each variable has been drawn from relevant literature unless otherwise indicated. 1. Q 2 was selected to account for the non-linearity present in Figure 3.11. Having fitted a non-linear function to the data set, in the form S S C = aQ b , using the non-linear regression tool in S Y S T A T 9.0, an associated r 2 value of 0.454 did not show great improvement over the value of 0.446 for the Entire season model (Table 3.4). 2. Q M . By lagging Q at different intervals (current time (t) - 0.5 hr, t-1 hr, t-1.5 hr ... t-250 hr) and then separately adding each variable into regression function 3.3, the optimum lag for Q was selected as t-1 hr, based on R 2 values. The reason for lagging discharge was to account for the residual autocorrelation noted in Figure 3.9c (Gurnell at al., 1994). As stated in Section 2.4, the SSC sampling interval was irregular. Such irregularity complicates attempts to correct for residual autocorrelation (e.g. using A R L M A modelling). By lagging the diurnal phasing of discharge to the best-match fit with the diurnal phase of SSC, it is possible, however, to account for some degree of the residual autocorrelation in the Entire season O L S function (Table 3.4). 3. Z Q . Cumulative discharge over the monitoring period (m xlO\" ) was incorporated into the suite of predictor variables because it accounts for seasonal changes in suspended sediment supply. The assumption of E Q is, therefore, that processes governing seasonal 50 changes in suspended sediment supply are represented by the increasing flux of snowmelt, glacial melt and storm runoff that passed through the drainage basin. Specifically, Z Q acts as a surrogate for suspended sediment supply exhaustion (Hodson and Ferguson, 1999). 4. A Q 3 represents the rate of change of discharge over 3 consecutive hours and was obtained by subtracting from current values discharge, values recorded 3 hr previously. It was thus positive during the rising hydrograph limb and negative during the falling limb. Recent literature suggests that SSC is more dependent on A Q , rather than the current discharge, Q (Willis et al., 1996). The optimum time-interval selection for A Q was obtained using the same method outlined for Q t . i , with AQ0.5, A Q i , AQ1.5 . . . AQ250-5. SSC(ref)- Many SSC M L R models published in contemporary literature adopt various lagged values of S S C as autoregressive predictors (e.g. Hodson and Ferguson, 1999). Such variables, however, limit the predictive capacity of the model because an intense sampling programme must be conducted to obtain lagged values of SSC. A n alternative approach would be to include a measured, reference value of SSC for each day. If successful, this approach could provide an effective trade-off between minimising costs of monitoring while maintaining accuracy of predicted SSC values. Inspection of diurnal-scale S S C time series for Place Creek indicated that peak values generally occurred at approximately 18:00. By using the daily value of SSC, measured at 15:00, to predict values between 9:00 that day and 8:00 the following morning (approximate timing of beginning and end of diurnal phase of SSC), a low maintenance sampling programme could be used to 'tune' the overall model to the physical state of the sediment 51 supply system for each day. This is a variable that has not, to the knowledge of the author, been used in previous studies similar to this one. 6. P24 was calculated by summing precipitation (mm) over the 24 hour period previous to the time of each SSC sample. Evidence from the literature suggests that precipitation is an important determinant of SSC during high discharge storm events (Hodgkins, 1999). 7. T A was used as a surrogate for glacial melt and thus accounted for diurnal fluctuations in discharge during the dominant summer ablation period (Aizen et al., 1996). Values of variables 1 to 7 were entered into a S Y S T A T 9.0 spreadsheet for each of the 675 values of SSC. To construct M L R models, the data set was again divided into its respective sub-seasons and variables were selected by backwards stepwise interactive selection. The selection was interactive because SSC(ref) and A Q 3 were 'forced' into each model. After analysing similar M L R models from the literature, it was expected that SSC(ref) and A Q 3 should, physically, provide the most explanation for the scatter in Figure 3.11. Selection criteria were an Alpha-to-Enter level of 0.049 and an Alpha-to-Remove level of 0.050. The resulting sub-season M L R models are given in Table 3.5. The number of data points used to derive each model is less than those used for corresponding sub-season models (Table 3.4) because on occasions SSC was not measured at 15:00 and thus a value of SSC(ref) could not be used with other values of SSC that same day. 52 < 1—1 O 9 § ON <^ CM o 9 § o d in CN CO O 9 § \"i 8 ° d cn CN o d i o CN CN O o o CO CO c OO CO NO o o o ^ d CO CO oo CN CN CN / — V o O cn o o in o l - H d Sf1 >n o CN O cn o o o m o NO O cn o ^ © O u cn 055) r--NO (000 00 CN (000 CN i o 00 O ON O 00 P - H O ^O o vq o oo o CN o • ON in ON cn O CN CN NO O CN NO cn cn o NO ON d o oo o oo cn ON o CN CN O cn fi o vi US CU CO I .a fi C3 > _c3 O cd I > 5 3 2 'o 03 3 In Sub-section 3.5.1 it was shown that the original O L S model (Table 3.3) failed to meet all of the regression assumptions. Under the same hypotheses and using the same tests as those described in Sub-section 3.5.1, Table 3.6 summarises the sub-seasonal M L R model assumption test statistics. Table 3.6: Summary of M L R Assumption Test Statistics (PRESS is the Prediction Sum of Squares). Sub-season Pr%2 DW PRESS Nival 0.1126 0.2536 1.176 1839 Nival-glacial 0.1325 0.0257 1.139 2779 Glacial <0.001 0.0304 0.863 11860 For the nival period, the normality and equal variances tests were passed. For the nival-glacial sub-season, the data were normally distributed, but slightly heteroskedastic. For the glacial period, the data were again slightly heteroskedastic, but not normally distributed. The upper and lower bounds of the Durbin Watson test statistic were, for each sub-season, 1.78 and 1.57, respectively. There is, therefore, evidence of minor autocorrelation in each model. The validity of the models can be assessed from the values of the PRESS statistic, which are calculated by summing the squares of the differences between each observed value of SSC and its predicted value based on a regression function developed from the remaining n - 1 values of SSC. The values of PRESS are thus a useful measure of how well the use of the fitted values for a subset of each sub-seasonal M L R model can predict the observed values of SSC (Neter et al, 1996). For the Entire season model (Table 3.4), the PRESS statistic (not shown) was 65242. Although PRESS values cannot be directly compared between models, since they are all based on different samples, the values for the M L R nival 54 and nival-glacial sub-seasons (Table 3.6) are much lower than for the Entire season model, indicating that both M L R models have small prediction errors. For the glacial period the prediction errors are larger than for the other two periods, but still represent a great improvement in the predictive capacity over the O L S Entire season model. 3.5.4 SSC-Q hysteresis relations Several types, strengths and scales of S S C - Q hysteresis can be noted according to seasonal, synoptic and diurnal changes in dominant hydrologic sources. To allow process inferences to be made from the hysteresis loops in Place Creek, examples from different sub-seasons and synoptic-scale events will be compared. It must be noted that quantitative measures of hysteresis have not been developed and so qualitative analysis must suffice. To ensure minimum subjectivity, the direction of the loop will describe its type and relative enclosed area its strength. Figure 3.12 gives examples of hysteresis loops with their associated discharge and SSC time series plots. For storm runoff events, hysteresis was strong and clockwise early in the season (Figure 3.12a), but later in the season was weaker and anticlockwise (Figure 3.12b). For snowmelt events, both strong and weak clockwise hysteresis occurred within sequential days during late June (Figure 3.12c). On the first day, discharge and SSC peaked at approximately the same time (27 June at 00:00). Although they again peaked at similar times the following evening (27 June at 21:00), discharge did not begin to fall for another 6 hr, therefore inducing stronger clockwise hysteresis. Further complexities can be noted in Figure 3.12d, in which the latter part of the loop is vertical and anticlockwise, indicating a negative S S C - Q relation. Another interesting loop on 20 to 21 July (Figure 3.12e) depicts a 'figure 8' formation 55 12:00 0:00 12:00 0:00 12:00 0:00 20-22 May if V 2.0 1.8 12:00 0:00 29-30 Aug 1.6 S, o 1.4 1.2 12:00 2.2 12:00 0:00 12:00 0:00 26-28 June 12:00 Figure 3.12: Examples of S S C - Q hysteresis for: ((a) and (b)) storm runoff events; (c) a snowmelt event; and ((d) and (e)) more complex S S C - Q relations. Arrows on S S C - Q plots indicate direction of hysteresis (filled and open arrows divide discharge cycles/events). Time plots (right) correspond to hysteresis loops. 56 1.8 2.0 2.2 2.4 2.6 12:00 18:00 0:00 6:00 12:00 Q (m3/s) 20-21 July — Q SSC Figure 3.12 (continued) 57 (Williams, 1992), combining initially clockwise and then anticlockwise sub-loops. The former occurred during high discharge and the latter at lower discharge. During the glacial season, the dominant diurnal S S C - Q hysteresis was strong and clockwise (Figure 3.13), indicating that SSC peaked before discharge. On a synoptic scale, hysteresis was also clockwise during mid to late August, as indicated by the clockwise progression of the loops themselves. To aid the comparison of hysteresis loops in Figure 3.12, Figure 3.14 illustrates the pattern of S S C and its proportion that was organic material, for four discharge events. A l l four plots indicate that organic SSC was approximately constant throughout the events, but from event a through d, a progressive decrease in organic SSC occurred. Given that total SSC rose and fell in each event, while the percentage organics remained approximately constant, the latter had a negative relation with SSC (Figure 3.14). This percentage was much lower (10-20%) during the fourth event than for the previous three events (20-100%). Since this last event was dominated by snowmelt, while the others were principally rainfall induced, Figure 3.14d confirms that sediment sources differed for different sources of runoff to the main channel. 58 Figure 3.13: Diurnal and synoptic scale S S C - Q hysteresis during the main ablation period (arrows indicate direction of synoptic scale hysteresis). A l l individual loops have a clockwise direction. 59 60 3.6 Upstream-downstream comparisons of SSC This section focuses on changes in SSC between Place Lake outlet and the valley site during two glacial melt cycles, one at the beginning and one at the end of August (Figure 3.15). As a result of time constraints, an adequate stage-discharge rating curve could not be established at the lake outlet. Data from 4 to 6 Aug clearly show that SSC at the lake outlet was approximately constant over the sampling period, at around 55 mg/L. In the valley, however, SSC followed a diurnal increase-decrease pattern, with a peak value of 60 mg/L at 20:00 and minimum value of about 30 mg/L at 10:00 on 5 Aug. Respectively, these times are also similar to those of the maximum and minimum values of valley discharge. At the end of August the SSC data series were complicated by a rain event, which appears to have caused an increase in SSC at the lake outlet, from 59 to 75 mg/L, during the period of rain (01:00 to 19:00 on 29 Aug). During this period, sediment runoff was observed from morainal slopes at the lakeside. In the valley, minimum S S C values were similar to values recorded in early August, but maximum values were approximately 40 mg/L and thus less than upstream values. For the case of approximately constant discharge at the lake outlet (i.e., early August upstream-downstream field campaign), an approximate suspended sediment budget was constructed, assuming that the total outflow over a diurnal cycle is the same at Place Lake outlet and Place Creek at the valley site. The sediment fluxes can thus be expressed as (3.5) (3.6) 61 Figure 3.15: Upstream-downstream comparison of lake outlet SSC, and valley discharge and SSC for (a) early August and (b) late August 2000. 62 where SSFPLo and SSFV are the suspended sediment fluxes over a daily discharge cycle at Place Lake outlet and the valley bottom, respectively, SSCPLO is the average hourly SSC at the lake outlet over a daily discharge cycle, and Q(PC) is average hourly discharge at the valley bottom. Equations 3.5 and 3.6 can be approximated as where At is the change in time (seconds). For the lake outlet, the total daily sediment flux for 5 Aug 12:00 to 6 Aug 11:59 was 12300 kg (12.3 t) and for the valley bottom was 9300 kg (9.3 t). Approximately, the suspended sediment budget at the basin outlet was -3000 ± 2 0 0 kg (-3 ± 0 . 2 t). Errors for suspended sediment fluxes were calculated considering errors in SSC ( ± 2 % ) and discharge (±5%) measurement. 3.7 Chapter summary Sub-division of the seasonal discharge hydrograph for Place Creek was achieved by comparison of its standardised discharge with that for Eight Mile Creek. The C U S U M S technique revealed four clear sub-seasons (nival, nival-glacial transition, glacial and autumn recession). For each sub-season, four distinct groups of points on an EC-discharge plot supported the method of hydrograph separation. Relations between SSC and discharge were highly scattered, but showed improvement when data were divided into nival, nival-glacial and glacial sub-seasons. Multiple regression models further improved the accuracy of SSC SSFPLO = SSCPLO • £ (<2 • At) (3.7) SSFv = ^(Q-SSC-At), (3.8) 63 prediction and the ability of the models to meet the regression assumptions. Hysteresis loop characteristics for diurnal scale SSC-Q relations provided additional evidence of the influence of different sources of runoff on suspended sediment transport within glacier-fed rivers. Finally, comparison of upstream-downstream SSC suggested that SSC at the lake outlet was approximately constant, but that at the valley bottom SSC fluctuated with diurnal discharge in the glacial period. Further evidence of the significance of in-channel sediment storage on the downstream delivery of sediment in Place Creek was provided by a negative daily sediment budget at the basin outlet. 64 Chapter 4 Discussion Discussion in this chapter will be focused on Place Creek and Eight Mile Creek will only be used as a reference basin. 4.1 Hydrologic control on SSC 4.1.1 Seasonal scale From the groups of points in the E C - Q relation (Figure 3.8) it can be hypothesized that there are three main hydrologic sources of streamflow within Place Creek: (1) water from Place Lake with an E C of about 10 to 12 uS/cm (measured at the lake outlet during early and late August 2000 by the author); (2) water from a baseflow reservoir with values of E C around 40 uS/cm (as measured on 16 Jan 2000); and (3) high-EC water that is generated during the nival period (25 to 30 u,S/cm). Unfortunately, this hypothesis cannot be addressed with the current data. The distinct groupings of points in Figure 3.8, however, suggest that the relative contributions of these sources vary amongst sub-seasons. The significance of the dummy variables and their interaction with Q in function 3.4 indicates that seasonal changes in the hydrologic sources of discharge (i.e., snow-melt, glacial-melt and a nival-glacial combination) alter the simple S S C - Q rating relation. Different areas of a catchment deliver different quantities and quality of sediment to a stream channel. In the initial part of the nival period, the slope of the nival S S C - Q regression function was steeper than for the nival-glacial and glacial functions (Figure 3.11), indicating that sediment entrainment efficiency was greatest early in the season, as previously found by Chikita (1993). The decrease in the regression slope during the nival-glacial transition period may reflect the retreat of the 65 snowpack to the upper portion of the basin above the tree-line (Figure 2.3), where soils were likely thinner and provided less sediment for entrainment. Additionally, during this transition period, the glacier was only making a small contribution to streamflow, as indicated by the similarity of standardised discharges between Place and Eight Mile Creeks (Figure 3.6), and thus suspended sediment supply from Place Lake was also minor. The regression slope for the glacial period was similar to the slope for the nival-glacial sub-season, but the intercept was greater for the former period. It could be argued, therefore, that the sediment discharge mechanisms during these two periods were analogous, but that during the latter, the increase in glacier-melt discharge from Place Lake supplied more suspended sediment than for the same discharge during the former sub-season. As a result of differences in the S S C - Q relation through sub-seasonal changes in water sources, the need to sub-divide seasonal hydrographs in glacierised basins is of fundamental importance to any investigation of suspended sediment transport within such catchments. Previous studies were highlighted in Section 1.2 as using subjective methods for hydrograph division. In this study, the comparison of Place Creek with a similar basin that had no current glacial effect provided a better means of dividing the seasonal hydrograph for Place Creek. 4.1.2 Synoptic scale During the first two discharge events of the season (20 to 23 May and 4 to 7 June respectively), SSC peaked at 47 and >71 mg/L in Place Creek. The third event, on 18 to 19 June, produced discharge values similar to those for the second event, but S S C peaked at only 43 mg/L. Although streamflow during both the second and third events was similar, rainfall during the former was almost double that during the latter. It is possible, therefore, 66 that the latter was a rain-on-snow event because air temperature was higher on 18 to 19 June than on 5 June. Intuitively, it would seem that snowmelt runoff should carry less sediment to the channel than does storm runoff because the snowpack acts as a protective layer, thus reducing runoff access to sediment sources, whilst raindrops provide an impact mechanism to loosen soil particles. Further evidence of this claim is provided by the fourth, snowmelt, event (28 to 30 June), during which little rain fell. Again, peak discharge during this event was similar to that reached during the second and third events. Assuming from the minimal rainfall that the bulk of discharge was induced by snowmelt runoff, the associated peak recorded SSC of 23 mg/L suggests that snowmelt runoff only carried a relatively minor sediment load to the stream channel. Figure 3.14 also showed that snowmelt events transported a smaller percentage of organic sediment as the nival season progressed. This finding supports the speculation made in Sub-section 4.1.1 that snowmelt runoff entrains less sediment towards the end of the nival period because of the rising elevation of snowmelt to areas where soils and vegetation cover are thinner. Another explanation for the progressive decrease in event-discharge SSC from the second to fourth events would be synoptic scale sediment supply exhaustion, in which, for example, most of the small grains, loosened by freeze-thaw cycles during the winter, or deposited during the previous autumn, were washed into the stream by the first and second events. Additionally, on daily timescales within the fourth synoptic period, increasing diurnal snowmelt discharge cycles were associated with progressively smaller corresponding diurnal SSC cycles. Such a negative relation between SSC and discharge could be evidence of intra-event sediment exhaustion. 67 4.1.3 Diurnal scale - glacial period In order to discuss the hydrologic control on SSC on a diurnal timescale, focus will be made on the glacial period when predominantly fine-grained material, the product of glacial erosion, was transported down Place Creek. Figure 3.15a (4 to 6 Aug) suggests that there was a constant supply of fine sediment evacuated from Place Lake during diurnal glacial melt cycles. Since S S C in the valley mimicked the pattern of discharge, it is unlikely that inputs of water between the two sites induced the diurnal fluctuation of S S C in early August. Assuming no significant measurement errors, a more plausible explanation is that short-term storage and re-mobilisation of fine sediment between Place Lake and the valley bottom occurred during the 24 hr cycle. On the falling hydrograph limb, large within-channel boulders may have provided a surface onto which clay and silts were deposited. During the rising hydrograph limb, some of this fine sediment could have been re-mobilised as the water level rose up around the boulders. Estimates made in Section 3.6 revealed that between 5 Aug 12:00 and 6 Aug 11:59, approximately 12,000 kg of sediment left Place Lake, but only 9000 kg passed by the valley-bottom site. Thus the value of -3000 ± 2 0 0 kg for the approximate daily suspended sediment budget at the mouth of the basin may provide evidence for a net loss of sediment to channel storage during diurnal glacier melt cycles. It must be noted, however, that the sediment budget is an estimate and to validate the difference in upstream-downstream fluxes would require lake outlet discharge to be measured. At the end of August (Figure 3.15b), suspended sediment supply from the Place Lake was complicated by rainfall runoff, as evidenced by the undulating SSC series at the lake outlet, which was probably induced by periodic inputs of sediment from the slopes surrounding the lake. The only time at which valley SSC approached values measured at the 68 lake outlet was at the glacial melt discharge peak on 30 Aug. It can be deduced, therefore, that either transient stores of fine glacial sediment were not re-mobilised during the storm event, or that the rainfall component of discharge, generated downstream of the lake outlet, had a relatively low S S C and thus served to 'dilute' the glacial component of total SSC. The former is possible because streamflow was approximately 1 m /s less in the late August measurement period than in early August. Thus maximum discharge on 29 Aug would not have risen to heights on within-channel boulders that minimum streamflow reached on 5 Aug. The latter is plausible because at the synoptic scale it was shown in Section 4.1.2 that storm event sediment supplies were progressively exhausted from event to event, unless discharge rose beyond values previously reached in the season (i.e., during the largest event of the season, on 27 to 28 July). 4.1.4 Integrated temporal scales The separate S S C - Q rating relations for Place Creek (Figure 3.11) indicated that seasonal suspended sediment transport is principally controlled by two sources of streamflow, glacial and nival, combined with a nival-glacial mixture and rainfall runoff. The points within the three sub-seasons plot in different areas of the S S C - Q plot. Initially, the slope of the relation was steep during the nival period as early season discharge events (both rainfall and snowmelt derived) flushed out sediment that had accumulated over the preceding winter. In the nival-glacial period, the slope decreased because only sand-sized material was being transported and large discharge events did not entrain as much sediment as earlier in the season. Finally, during the glacial period, the average suspended sediment particle size likely decreased from sand- to silt- and clay-sized material. Accordingly, the intercept of the S S C -Q rating function increased as more sediment could be transported at relatively lower 69 discharges, than over the nival-glacial transition period. Thus from a seasonal to sub-seasonal scale, the groups of points in Figure 3.11 moved in an anticlockwise direction, producing anticlockwise hysteresis. At the diurnal scale, Sub-section 3.5.4 highlighted different types and strengths of diurnal scale hysteresis. Changes in diurnal hysteresis imply that different sediment and water delivery processes occur through the season. For storm runoff events, the strength and clockwise direction of hysteresis during the early season suggest that the bulk of suspended sediment was flushed through the basin during the rising limb of storm hydrographs (Figure 3.12a). Later in the season, however, the loops were both tighter and anticlockwise during rain events (Figure 3.12b), indicating that the sediment did not reach the channel until the falling hydrograph limb. The seasonal change in the character of storm-hysteresis probably reflects the decrease in supply of near-bank sediment, which would have been washed into the channel during early season storm events. Thus sediment that is supplied to the channel during late season storms may take a longer time to be transported to the channel by rainfall runoff. During the nival period, complex hysteresis was noted in Figure 3.12d, in which the latter part of the loop was vertical and anticlockwise. This irregular loop illustrates the importance of preceding events, which in this case was a rain event. On the falling limb of the storm hydrograph, maximum daily air temperature rose from 2 0 ° C on 2 July to 2 8 ° C on 3 July. It is plausible, therefore, that although discharge was declining, the storm runoff proportion of streamflow was also declining and being replaced by nival-glacial meltwater which may have initially been carrying higher values of SSC. Another interesting loop on 21-22 July (Figure 3.12e) depicts a 'figure 8' formation. The 'figure 8' loop in Figure 3.12e 70 combined initially clockwise and then anticlockwise sub-loops. The former occurred during higher and the latter at lower discharge, implying that following peak flow, sediment availability was great enough that SSC decreased more slowly than discharge. This figure 8 loop also occurred at the end of the nival-glacial season and so again it may be that diurnal changes in hydrologic sources, from nival to glacial, are reflected in the pattern of S S C - Q hysteresis. During the glacial season, clockwise hysteresis at both diurnal and synoptic scales (Figure 3.13) provided evidence of sediment flushing prior to daily and synoptic peaks in discharge. It is probable that such flushing is of sediment stored within the channel. This sediment may have been derived from its deposition during falling diurnal hydrograph limbs on a daily timescale and, on a synoptic scale, from its deposition during the receding weekly hydrograph limb following high flows during early August. The integration of diurnal, synoptic and seasonal timescales is of fundamental importance in the attempt to understand the proglacial suspended sediment transport system in Place Creek. It has been argued in this thesis that different water sources exert different hydrologic controls on downstream values of SSC in Place Creek. However, to predict accurately how SSC will respond to different hydrologic conditions, effects of processes at different timescales must be integrated. The aim of the next section is, therefore, to show how the M L R models presented in Sub-section 3.4.3 incorporate the ideas of integrated timescales presented in this Sub-section and to analyse the process inferences that can be made from the significant variables in each sub-season M L R model. 71 4.2 Modelling SSC Figure 4.1 summarises the stages followed in this study in order to obtain predictions of SSC, firstly using the C U S U M S technique to sub-divide the seasonal discharge series and then using the M L R models in Table 3.5. Temporal scales are shown above each stage and the final predicted value of SSC represents the integration of the timescales within which each stage operates. Within the sub-seasonal stage, the nival-glacial and then glacial periods are indirectly influenced by the legacies of previous sub-seasons within the entire hydrological season. The diurnal and synoptic scales are combined because each of the predictor variables operates within both scales. Considerable improvement in the accuracy of the sub-season M L R models, over the sub-season O L S models (Table 3.4), is shown by the high adjusted R 2 (R2adj) values (Table 3.5). The R 2 adj for the nival period may be slightly higher because n was smaller and only spanned the period 18 May to 17 June. The prevailing hydrologic conditions and sources of suspended sediment within this time were likely to be more similar than for the entire nival period. For the nival-glacial period, however, the hydrology of the basin was transitional and for the glacial season the period encompassed over two months. During these latter sub-seasons, therefore, it could be expected that model accuracy would be lower than for the nival season because as time increases the stochastic element in the data set may also increase. Such stochastic model elements can, in part, be attributed to short-term SSC autocorrelation and be accounted for by adding a lagged S S C predictor variable to each of the models in Table 3.5 (Hodson and Ferguson, 1999). Having added SSC, lagged by sample interval, as an additional predictor, R 2 adj values improved by up to 0.1. The M L R models 72 SEASONAL Q w SUB-SEASONAL ! SYNOPTIC / DIURNAL NIVAL NIVAL-GLACIAL GLACIAL Q Q2 IQ TA AQ 3 SSC(ref) SSC Figure 4.1: Flow chart model for prediction of SSC from sub-seasonal M L R models. 73 would, however, become redundant for cost-effective SSC prediction because a close-interval sampling scheme would be required to calculate values of such lagged SSC variables. In all of the sub-season M L R models, Q t_i was not significant because its contribution to the model is provided through the combination of the other significant predictor variables. Its insignificance in, for example, the nival season may reflect low mean SSC (9.9 mg/L) for the nival period (Table 3.2) and that when SSC was highest (during rain events), most of the variability in S S C was explained by P24. For the nival period, the predictors are (in order of decreasing significance) Q 2 , P24, A Q 3 , E Q and SSC(ref>. A l l these variables have positive coefficients, indicating that as they increase they cause a collective increase in SSC. From what was reported earlier in this section, it appears that for the nival period, E Q does not, therefore, represent a decreasing supply of sediment. Rather, its positive coefficient possibly reflects the relatively high values of SSC recorded during the rainfall events over the nival period rather than an increasing area of suspended sediment supply. Even though it is insignificant, SSC(ref) is included in the model because it was 'forced' in through the selection process. Its contribution to the overall model may be relatively small because mean SSC for the nival season was low compared to the nival-glacial and glacial seasons (Table 3.2). The M L R model for the nival-glacial period incorporates SSC(ref), A Q 3 , Q 2 , E Q , T A , P24 and Q in order of decreasing significance. For this model, the increasing importance of SSC(ref) is reflected in the increased variability of SSC through the nival-glacial season and thus the need for a reference value. As glacial melt became a more established component of 74 stream discharge during July, A Q 3 also became more important because the major portion of daily suspended sediment transfer occurred on the rising hydrograph limb in Place Creek. The coefficient of P24 was negative because rainy conditions correlated with subdued snow and glacial melt production. Discharge and T A became significant, but also had negative coefficients, possibly because the individual effects of the variables are confounded by the other predictors. The negative coefficient of EQ implies that seasonal sediment supply exhaustion played a significant role in the temporal control on SSC through the nival-glacial period. During the glacial sub-season, the M L R model contains Q, SSC(ref), EQ, A Q 3 , P24 and T A in order of decreasing significance. Air temperature is the only variable with a negative coefficient and this is likely to be a function of the passage of storm systems with associated pulses of high SSC. A positive coefficient is produced for EQ, indicating that during the ablation season, the supply of suspended sediment (dominantly glacially derived) was not exhausted as commonly noted in the literature. As for the nival-glacial model, SSC(ref) and AQ3 were significant. The former shows that the hydrologic control on proglacial suspended sediment dynamics changed on short timescales. The latter again provides evidence that most sediment is transported during the rising limb of the diurnal glacial melt hydrograph. Overall, the M L R models in Table 3.5 can be relied upon to give accurate predictions of SSC because the PRESS statistics were low (Table 3.6). Additionally, the regression assumptions are generally met for these models. The need for log e transformations and associated problems of bias is thus avoided. Although the glacial M L R model did not meet the regression assumptions, most MLR SSC prediction models proposed in the literature 75 acknowledge that such assumptions are difficult to meet, given white noise in the data set (e.g. Hodgkins, 1999). Frequently, measurements are discarded in the literature without adequate reason, besides being outliers. In this study, no data were discarded because no obvious measurement error was noted for any of the samples. From discussions of transient sediment storage in Section 4.1, it may be that during the glacial period, the relation between SSC and its suite of glacial period predictors is complicated by short term pulses of sediment derived from within channel stores such as the surface of large boulders. 76 Chapter 5 Conclusions 5.1 Summary of key findings 5.1.1 SSC-Q relations in a glacier-fed stream Place Creek is dominated by a nival- and glacial-melt discharge regime, which is punctuated by storm runoff events. A simple S S C - Q rating relation was shown in Sub-section 3.5.1 to be an inaccurate SSC prediction tool. In order to investigate the hypothesis that different sources of water exert different controls on the S S C - Q relation, the seasonal Place Creek discharge series was divided into sub-seasons. Corresponding sub-seasonal O L S rating curves showed great improvement. Coupled with E C data that supported the hypothesis that streamflow within the sub-seasons was derived from different sources (snowmelt, glacier-melt and groundwater), it can be concluded that a fundamental source of explanation for the scatter in the seasonal S S C - Q relation is provided by the transition from nival to nival-glacial and finally to glacial sources of runoff, combined with periodic inputs of storm runoff through the season. 5.1.2 Multiple regression models The use of M L R models to predict SSC from a suite of variables has arguably been extended in the literature to the point at which they are no longer time or cost efficient to evaluate over subsequent seasons. The simple sub-seasonal M L R models presented in Sub-section 3.5.3 can be evaluated from easily measured predictors and thus provide a cost efficient tool with which to predict accurately SSC in Place Creek over future seasons. The M L R models themselves represent an integration of timescales, from seasonal to diurnal. Their improved accuracy from the sub-seasonal O L S models indicates that they provide additional 77 explanation for the scatter in the S S C - Q relation that cannot be accounted for at the sub-seasonal scale. The detailed data set that was used to derive the M L R model parameters in this thesis represents an extensive and accurate record of SSC, something that has rarely been achieved in the literature. 5.1.3 Within-channel transient sediment storage Place Creek appears to be an example of a sediment-transfer system that is dominated by a transient diurnal storage mechanism within the ablation season, possibly provided by the surface of large within-channel boulders. Such storage has not been widely acknowledged in the literature, largely because of a lack of intensive studies of upstream-downstream changes in SSC in streams with minimal tributary or groundwater inflow. Place Lake was shown to be a regulating sink for fine glacially-derived sediment, while diurnal fluctuations of downstream SSC and a negative suspended sediment budget at the basin mouth suggested that the sediment was deposited and re-mobilised within the channel. These data are compatible with the boulder storage hypothesis. 5.2 Future research Clearly this thesis presents a number of claims that require further evidence to validate. Future research projects could (a) generate daily sediment budgets for the entire glaciofluvial lacustrine system; (b) investigate the trapping efficiency of Place Lake and how it varies with discharge; (c) undertake a more extensive upstream-downstream monitoring campaign of SSC and discharge; (d) analyse the caliber of the suspended sediment transported down Place Creek in order to identify spatial and temporal changes in sediment-source areas; (e) collect geochemical data from water samples for diurnal, synoptic and seasonal hydrograph separation with the aim of providing more detailed divisions of water sources that affect the 78 SSC-Q relation, particularly to investigate the hypothesis that the changes in the EC-Q relation reflect seasonal changes in the location of water sources. 79 References Aizen V B , Aizen E M , Melack J M . 1996. Precipitation, melt and runoff in the northern Tien Shan. Journal of Hydrology 186: 229-251. Asselman N E M . 2000. Fitting and interpretation of sediment rating curves. Journal of Hydrology 234: 228-248. Barsch D and Caine N . 1984. The nature of mountain geomorphology. Mountain Research and Development 4(4): 287-298. Benn DI, Evans JA. 1998. Glaciers and Glaciation. London: Arnold: 734 pp. Bogen J. 1980. The hysteresis effect of sediment transport systems. Norsk Geogr. Tidsskr 34: 45-54. Brown G H , Sharp M J , Tranter M , Gurnell A M and Nienow P W . 1994. impact of post-mixing chemical reactions on the major ion chemistry of bulk meltwaters draining the Haut Glacier D'Arolla, Valais, Switzerland. Hydrological Processes 8: 465-480. Brown G H , Tranter M and Sharp M J . 1996. Experimental investigations of the weathering of suspended sediment by alpine glacial meltwater. Hydrological Processes 10: 579-597. Chikita K . 1993. A comparative study on suspended sediment discharge initiated by snow- or glacier-melting. Publication 217, International Association of Hydrological Sciences: Wallingford; 257-267. Collins D N . 1979. Sediment concentration in meltwaters as an indicator of erosion processes beneath an Alpine glacier. Journal of Glaciology 23: 247-257. Collins D N . 1998. Suspended sediment flux in meltwaters draining from Batura glacier as an indicator of the rate of glacial erosion in the Karakoram mountains. Quaternary Proceedings 6: 1-10. Elder K , Kattelmann R, Ferguson R. 1990. Refinements in dilution gauging for mountain streams. Publication 193, International Association of Hydrological Sciences: Wallingford: 247-254. Farley A L . 1979. Atlas of British Columbia. Vancouver: U B C Press: 136pp. Fenn C R . 1989. Quantifying the errors involved in transferring suspended sediment rating equations across ablation seasons. Annals of Glaciology 13: 64-68. Fenn C R , Gurnell A M , Beecroft IR. 1985. A n evaluation of the use of suspended sediment rating curves for the prediction of suspended sediment concentration in a proglacial stream. Geografiska Annaler 61 A{\\-2): 71-82. 80 Ferguson R L 1986. River loads underestimated by rating curves. Water Resources Research 22: 74-76. Ferguson RI 1987. Accuracy and precision of methods for estimating river loads. Earth Surface Processes and Landforms 12: 95-104. Gurnell A M . 1982. The dynamics of suspended sediment concentration in an Alpine pro-glacial stream network. Publication 138, International Association of Hydrological Sciences: Wallingford: 319-330. Gurnell A M . 1987. Suspended sediment. In Glacio-fluvial Sediment Transfer: an Alpine Perspective, Gurnell A M , Clark M J (eds). Wiley: Chichester: 305-353. Gurnell A M , Fenn C R . 1984. Box-Jenkins transfer function models applied to suspended sediment concentration-discharge relationships in a proglacial stream. Arctic and Alpine Research 16: 93-106. Gurnell A M , Clark M J , Hi l l C T . 1992a. Analysis and interpretation of patterns within and between hydroclimatological time series in an alpine glacier basin. Earth Surface Processes and Landforms 17: 821-839. Gurnell A M , Clark M J , Hil l C T and Greenhalgh J. 1992b. Reliability and representativeness of a suspended sediment concentration monitoring programme for a remote alpine proglacial river. International Association of Hydrological Sciences Publications 210: 191-200. Gurnell A M , Hodson A , Clark M J , Bogen J, Hagen JO, Tranter M . 1994. Water and sediment flux from glacier basins: an arctic and alpine comparison. Publication 224, International Association of Hydrological Sciences: Wallingford: 325-334. Gurnell A M , Edwards PJ, Petts G E , Ward JV. 1999. A conceptual model for alpine proglacial channel evolution under changing climatic conditions. Catena 38: 223-242. Hallet B , Hunter L , Bogen J. 1996. Rates of erosion and sediment evacuation by glaciers: a review of field data and their implications. Global and Planetary Change 12: 213-235. Hammer K M , Smith N D . 1983. Sediment production and transport in a proglacial stream: Hilda Glacier, Alberta, Canada. Boreas 12: 91-106. Hannah D M , McGregor G R . 1997. Evaluating the impact of climate on snow- and ice-melt dynamics in the Taillon Basin, French Pyrenees. Journal of Glaciology 43(145): 563-568. Hannah D M , Gurnell A M . 2001. A conceptual, linear reservoir runoff model to investigate melt season changes in cirque glacier hydrology. Journal of Hydrology 246: 123-141. 81 Hannah D M , Smith B P G , Gurnell A M , McGregor G R . 2000. A n approach to hydrograph classification. Hydrological Processes 14: 317-338. Harbor J, Warburton J. 1993. Relative rates of glacial and non-glacial erosion in alpine environments. Arctic and Alpine Research 25(1): 1-7. Hicks D M , McSaveney M J , Chinn T J H . 1990. Sedimentation in proglacial Ivory lake, Southern Alps, New Zealand. Arctic and Alpine Research 22(1): 26-42. Hodgkins R. 1996. Seasonal trend in suspended-sediment transport from an Arctic glacier, and implications for drainage-system structure. Annals ofGlaciology 22: 147-151. Hodgkins R. 1999. Controls on suspended-sediment transfer at a high-arctic glacier, determined from statistical modelling. Earth Surface Processes and Landforms 24: 1-21. Hodson A , Gurnell A , Tranter M , Bogen J, Hagen JO, Clark M . 1998. Suspended sediment yield and transfer processes in a small High-Arctic glacier basin, Svalbard. Hydrological Processes 12: 73-86. Hodson A J , Ferguson R L 1999. Fluvial suspended sediment transport from cold and warm-based glaciers in Svalbard. Earth Surface Processes and Landforms 24: 957-974. Jansson M . 1985. A comparison of detransformed logarithmic regressions and power function regressions. Geografiska Annaler 67A(l-2): 61-70. Laudon H . 1995. Spatial and temporal variability of the stream water chemistry of an alpine/sub-alpine catchment, in the Coast Mountains of British Columbia, Unpublished MSc. Thesis, Department of Geography, University of British Columbia. Lenzi M A , Marchi L . 2000. Suspended sediment load during floods in a small stream of the Dolomites (northeastern Italy). Catena 39: 267-282. Mac Nally R, Hart B T . 1997. Use of C U S U M methods for water-quality monitoring in storages. Environmental Science and Technology 31(7): 2114-2119. Matthews W H . 1956. Physical limnology and sedimentation in a glacial lake. Bulletin of the Geological Society of America 67: 537-552. Miller D M . 1984. Reducing transformation bias in curve fitting. The American Statistician 38(2): 124-126. Milliman JD, Syvitski J P M . 1992. Geomorphic/tectonic control of sediment discharge to the Ocean: the importance of small mountainous rivers. Journal of Geology 100: 522-544. Moore R D . 1993. Application of a conceptual streamflow model in a glacierized drainage basin. Journal of Hydrology 150: 151-168. 82 Moore R D and McKendry IG. 1996. Spring snowpack anomaly patterns and winter climate variability, British Columbia, Canada. Water Resources Research 32(3): 623-632. Neter J, Wasserman M , Kutner W H , Nachtsheim C J . 1996. Applied Linear Statistical Models. Chicago: McGraw-Hill: 1408pp. 0strem G . 1975. Sediment transport in glacial meltwater streams. In Glaciofluvial and Glaciolacustrine Sedimentation, Jopling A V , McDonald B C (eds). 1975. Society of Economic Palaeontologists and Mineralogists Special Publication 23: 101-122. Picouet C , Hingray B , Olivry JC. 2001. Empirical and conceptual modelling of the suspended sediment dynamics in a large tropical African river: the Upper Niger river basin. Journal of Hydrology 250(1-4): 19-39. Richards K S . 1984. Some observations on suspended sediment dynamics in Storbregrova, Jotunheim. Earth Surface Processes and Landforms 9: 101-112. Richardson J, Rowe D K , Smith JP. 2001. Effects of turbidity on the migration of juvenile banded kokopu (Galaxias fasciatus) in a natural stream. The New Zealand Journal of Marine and Freshwater Research 35(1): 191-196. Rothlisberger H , Lang H . 1987. Glacial Hydrology. In Gurnall A M , Clark M J (eds), Glacio-fluvial Sediment Transfer. New York: Wiley: 207-284. Ryan P A . 1991. Environmental-effects of sediment on New-Zealand streams - a review. The New Zealand Journal of Marine and Freshwater Research 25(2): 207-221. Ryder J M , Thomson B. 1986. Neoglaciation in the southern Coast Mountains of British Columbia: chronology prior to the late Neoglacial maximum. Canadian Journal of Earth Sciences 23(3): 273-287. Singh P, Huebl H , Weinmeister H W . 2000. Use of recession characteristics of snowmelt hydrographs in the assessment of snow water storage in a basin. Hydrological Processes 14: 91-101. Stott T A , Grove JR. 2001. Short-term discharge and suspended sediment fluctuations in the proglacial Skeldal River, north-east Greenland. Hydrological Processes 15: 407-423. Syvitski JP, Morehead M D , Bahr D B , Mulder T. 2000. Estimating fluvial sediment transport: The rating parameters. Water Resources Research 36(9): 2747-2760. Teti P A . 1979. The variability of stream chemistry in a Coast Mountain watershed, British Columbia, Unpublished MSc.Thesis, Department of Geography, University of British Columbia. Walling D E . 1977. Assessing the accuracy of suspended sediment rating curves for a small basin. Water Resources Research 13(3): 531-538. 83 Weirich F H . 1985. Sediment budget for a high energy glacial lake. Geografiska Annaler 67A (1-2): 83-99. Williams GP. 1989. Sediment concentration versus discharge during single hydrologic events in rivers. Journal of Hydrology 111: 89-106. Willis IC, Richards K S , Sharp M J . 1996. Links between proglacial stream suspended sediment dynamics, glacier hydrology and glacier motion at Midtdalsbreen, Norway. Hydrological Processes 10: 629-648. Wren D G , Barkdoll B D , Kuhnle R A , Derrow R W . 2000. Field techniques for suspended-sediment measurement. Journal of Hydraulic Engineering 126(2): 97-104. 84 Appendix A Dummy coding used f o r categ o r i c a l variables i n model. Categorical values encountered during processing are: SEASON (3 levels) 0, 1, 2 Step # 0 R = 0.805 R-Square = 0.648 Ef f e c t C o e f f i c i e n t Std Error Std Coef T o l . df F 'P' In 1 Constant 2 Q 15.673 0.722 0.683 0.53046 1 471.126 0.000 3 SEASON . . . 0.08622 2 23.614 0.000 4 SEAS0N*Q . . . 0.09959 2 21.165 0.000 Out Part. Corr. none Estimates of e f f e c t s SSC CONSTANT 1.332 Q 15.673 SEASON 0 -19.43 6 SEASON 1 -5.716 SEASON 0 Q 15.642 SEASON 1 Q -3.811 Analysis of Variance Source Sum-of-Squares df Mean-Square F - r a t i o P Regression 75547.344 5 15109.469 246.677 0.000 Residual 40977.537 669 61.252 85 "@en ; edm:hasType "Thesis/Dissertation"@en ; vivo:dateIssued "2001-11"@en ; edm:isShownAt "10.14288/1.0090147"@en ; dcterms:language "eng"@en ; ns0:degreeDiscipline "Geography"@en ; edm:provider "Vancouver : University of British Columbia Library"@en ; dcterms:publisher "University of British Columbia"@en ; dcterms:rights "For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use."@en ; ns0:scholarLevel "Graduate"@en ; dcterms:title "Hydrologic control on proglacial suspended sediment dynamics"@en ; dcterms:type "Text"@en ; ns0:identifierURI "http://hdl.handle.net/2429/11827"@en .