Research Article 2014 WRR Burdekin sediment budget | Page 7

Water Resources Research
10.1002/2013WR014386
Table 2. Catchment-Specific Suspended Sediment Yield Contributions (tonnes km 22 yr 21 ) and Mean Annual Concentration (MAC) (mg L 21 ) During the Five Monitored Water Years
(2005–2010). Sample Size for Each Site/Water Year is Shown in Italics
Sediment Yield (t km 22 yr 21 ) and Sample Size (n) for Each Water Year
Major Subcatchment
Upper Burdekin
Cape
Belyando
Suttor
Burdekin Falls Dam Overflow
(capturing above catchments)
Bowen
Burdekin River (end-of-
catchment)
a
Upstream
Area (km 2 ) 2005/2006 n 2006/2007 n 2007/2008 n 2008/2009 n 2009/2010 n Mean MAC Range
2005–2010 (mg L 21 )
36,140
15,860
35,055
10,870
114,260 60
2
5
9
3 12
7
12
4
31 85
12
4
9
15 14
8
9
7
55 130
32
6
65
27 15
30
33
35
97 415
30
3
13
43 26
13
8
8
102 47
10
5
21
4 8
115
93
63
63 147
17
5
23
18 680–795
205–360
55–650
120–370
81–260
7,110
129,600 35
7 49
23 370
55 48
52 1035 a
115 43
53 670 a
85 0
52 540 a
19 0
47 530
56 1780–3600
320–730
Note lower confidence in the Bowen River loads (and therefore sediment yields) in the latter years with wide CV related to lack of monitoring data in these wet seasons.
1. linear and quadratic terms for streamflow;
2. the concept of higher TSS concentrations during a ‘‘first flush’’ and the characterization of TSS concentra-
tions on the rise and fall of an event;
3. a discounted flow term that captures historical flows and the exhaustion of sediment supply over the
flow period.
The addition of terms such as a rising-falling limb and flow discounting strengthen the predictive capability
of the model, as clearly demonstrated by the improved explanatory power achieved by shifting from a sim-
ple rating curve style approach to the LRE model that includes these additional terms (supporting informa-
tion Table A2). The discounting flow term provided the greatest increase in the explanatory power of the
model, contributing 25–40% of deviance explained for each site. Additional terms (vegetation ground cover
and ratio of flow from above and below the BFD [Kuhnert et al., 2012]) were included in the LRE model for
the end-of-river (Inkerman) site to accommodate its size and complexity. These terms were not relevant to
the subcatchment sites.
The LRE characterizes the loads through a regression modeling relationship for each site that takes into
account concentration data collected over multiple water years, with the capacity to predict loads for years
that have limited data. We have higher confidence in the loads calculated for well-sampled water years,
with associated uncertainty ranges <5–10%. In this regard, preceding wet season TSS data sets from the
Bowen River (Myuna; 2002/2003–2004/2005, 40 samples) and Burdekin River (Inkerman; 1986/1987–2004/
2005, 465 samples) were utilized in the calculation of sediment loads for the water years included in this
study. Importantly, the number of samples collected in our study increased throughout the monitoring pro-
gram (Table 2), which coincided with larger streamflow events (see FS01), with the LRE (GAM) model devel-
oping a strong relationship for each site (with the exception of the Bowen River: see discussion). This, in
turn, allowed reasonable confidence in the loads to be generated for the Belyando and Suttor subcatch-
ments of the Burdekin despite limited sampling carried out in the 2005/2006 water year, further highlight-
ing the benefits of applying the LRE model [Kuhnert et al., 2012]. The method can detect changes in annual
sediment loadings due to catchment condition as the LRE model can characterize the pattern in TSS con-
centration using the relationship with flow and additional model explanatory terms such as seasonal/annual
changes in ground cover over the entire timeframe for modelling [see Kuhnert et al., 2012].
The LRE model quantifies the uncertainty in the load estimate, which is reported in this paper as 80% confi-
dence intervals [Kuhnert et al. 2012]. This envelope takes into account uncertainty and variability in TSS con-
centrations associated with the surface grab sampling field method (i.e., variations in TSS concentrations
across the stream profile, subsampling), errors associated with the laboratory analysis, as well as potential
errors associated with opportunistic stream gauge positioning and sampling error. See Kuhnert et al. [2012]
for further detail on the LRE, including input data used to quantify the two errors in flow.
3.4. Catchment-Wide Discharge and Sediment Load Budgets
Catchment-wide discharge and sediment load budgets were constructed for each of the five monitored
water years using streamflow and suspended sediment load data from the gauged study sites. The four
BAINBRIDGE ET AL.
C 2014. American Geophysical Union. All Rights Reserved.
V
9073