Main Text
Introduction
Quantitative estimates of annual fluvial suspended sediment yield
(hereafter, ‘sediment yield’) are sought by physical scientists as
signals of environmental dynamics, by ecologists for their associations
with water quality and habitat value, and by engineers for
hydro-infrastructure and river system design. Measuring sediment yields
presents a challenge, because sediment transfer is inherently variable
in space and time (Morehead, Syvitski, Hutton, & Peckham, 2003; Orwin,
Lamoureux, Warburton, & Beylich, 2010). Research from glaciated Arctic
catchments indicates that sediment transfer typically reflects
catchment-scale processes (Hodgkins, Cooper, Wadham, & Tranter, 2003),
including meteorological forcing (Lewis & Lamoureux, 2010; Syvitski,
2002), glacial dynamics (Bogen & Bønses, 2003; Gurnell, Hannah, &
Lawler, 1996; Hodson & Ferguson, 1999; Hodson, Tranter, Dowdeswell,
Gurnell, & Hagen, 1997), and other geomorphological processes (Hasholt
et al., 2005). The complexity of meteorological forcing on sediment
transfer in glaciated catchments is accentuated by reports of
substantial variations through time over recent millennia (Saarni,
Saarinen, & Lensu, 2015), and across spatial scales (Striberger et al.,
2011). Sediment transfer can vary significantly on inter-annual,
decadal, and century scales (Bogen & Bønses, 2003; Gurnell et al.,
1996; Lewis, Lafrenière, & Lamoureux, 2012; Lewkowicz & Wolfe, 1994;
Orwin & Smart, 2004a; Richards, 1984; Tape, Verbyla, & Welker, 2011),
and may reflect landscape evolution over glacial-interglacial cycles
(Church & Ryder, 1972; Church & Slaymaker, 1989; Leonard, 1997).
Within a single open-channel season, the majority of annual
catchment-scale sediment yield can be transported during one, or a few,
discrete events (Bogen & Bonses, 2003; Dugan, Lamoureux, Lafrenière, &
Lewis, 2009; Fenn, Gurnell, & Beecroft, 1985; Hasholt et al., 2005;
Lewis et al., 2012; Schiefer et al., 2017; Østrem, 1975). Sediment
availability and exhaustion can greatly affect seasonal sediment
transfer (Bogen & Bønses, 2003; Forbes & Lamoureux, 2005; Hodgkins,
1999; Hodgkins et al., 2003; Hodson et al., 1998; Irvine-Fynn et al.,
2005).
Proglacial instrumentation and sampling programs to directly measure
suspended sediment concentrations (SSCs) at daily, or preferably hourly
or finer sampling intervals (Orwin et al., 2010), over periods longer
than one open-channel season are rare in the Arctic (Hasholt et al.,
2005). Consequently, statistical models are relied upon to estimate
annual sediment transfer from discontinuous samples of SSC. In Arctic
Canada, automated sampling has enabled continuous measurement of SSC
fluctuations and estimation of sediment yields using spline curves
(Cockburn & Lamoureux, 2008; Favaro & Lamoureux, 2014; Lewis et al.,
2012); however, such intensive sampling throughout the full length of
the open-channel season is not always feasible. More typically,
statistical models comprise simple sediment rating curves, using either
discharge (Bogen & Bonses, 2003; Dugan et al., 2009; Fenn et al., 1985;
Forbes & Lamoureux, 2005; Hodgkins, 1996; Hodson et al., 1998;
Horowitz, 2003; Lamb & Toniolo, 2016; Lewkowicz & Wolfe, 1994;
McLaren, 1981; O’Farrell et al., 2009; Rasch, Elberling, Jakobsen, &
Hasholt 2000; Østrem, 1975), or, less often, turbidity (Harrington &
Harrington, 2013) as the single predictor of SSC. Despite their
popularity, failure to adequately account for quasi-autocorrelation has
been identified as a pitfall associated with the use of simple sediment
rating curves (Hodgkins, 1999; Hodson & Ferguson, 1999).
Quasi-autocorrelation arises from shortfalls in formulation of the
regression model, including: an incorrect fit, failing to identify the
presence of lags, changes in response between the dependent and
independent variables, and omitting relevant independent variables (Fenn
et al., 1985; Gao, 2008; Hodson & Ferguson, 1999). SSC may be
underestimated or overestimated by simple sediment rating curves (Gao,
2008; Walling, 1977), with perturbations smoothed and margin for error
reduced as the monitoring period increases (Horowitz, 2003). However,
monitoring spanning more than a couple of months of one or two
open-channel seasons is rare in remote arctic environments (e.g. Bogen
& Bønses, 2003). Statistical methods can be applied to address
quasi-autocorrelation, improving the predictive ability of sediment
rating curves. For example, separating rating curves according to
discrete temporal periods, or stage, have both proven popular, with
varied success (Harrington & Harrington, 2013; Hodgkins, 1996; Hodson
et al., 1998; Horowitz, 2003; Lewkowicz & Wolfe, 1994; McLaren, 1981;
Richards, 1984; Walling, 1977; Østrem, 1975).
Multiple-regression models, incorporating hydrological, temporal, and
meteorological explanatory variables with optimal lag times, are often
preferable over rating-curve-separation because they account for
processes that can decouple SSCs from contemporaneous discharge or
turbidity fluctuations, and help us understand multifaceted, dynamic
processes driving glaciofluvial sedimentation (Hodgkins, 1999; Hodson &
Ferguson, 1999; Irvine-Fynn et al., 2005; Richards, 1984; Willis,
Richards, & Sharp, 1996). Temporal variables can be used as indicators
of sediment availability, including: hysteresis effects, intra-season
variability, and seasonal variations. Meteorological variables can also
capture temporal variability in SSCs, including: diurnal and longer
cycles of solar radiation and temperature affecting melt-related erosion
and transfer; and rainfall-induced events generating hillslope erosion
and delivery processes, extra-channel erosion, and sediment entrainment
with rising discharge. Further, inclusion of meteorological variables in
sediment modeling can provide useful information for interpreting past
climates from longer sedimentary records, and assessing sensitivity to
climate change through hydroclimatic system forecasting (Forbes &
Lamoureux, 2005; Gordeev, 2006; Lewis & Lamoureux, 2010; Syvitski,
2002). Despite potential advantages, multiple-regression models are
uncommonly used for studying suspended sediment transfer in catchments
above the Arctic Circle (Hodgkins, 1999; Hodson & Ferguson, 1999;
Irvine-Fynn et al., 2005; Schiefer et al., 2017). In Arctic Alaska
(defined hereafter as Alaskan land above the Arctic Circle—66.33°N),
even simple sediment rating curves have rarely been constructed (Arnborg
et al., 1967; Lamb & Toniolo, 2016; Rainwater & Guy, 1961; Trefry,
Rember, & Trocine, 2004). The objective of this paper is to use
multiple-regression models to estimate seasonal sediment yields, and
interpret physical processes driving these yields, at Lake Peters,
northeast Brooks Range, Arctic National Wildlife Refuge, Alaska, a site
selected for hydrological and paleoenvironmental research (Benson,
Kaufman, McKay, Sciefer, & Fortin 2019; Broadman et al., 2019;
Ellerbrook, 2018; Thurston, 2017).
Study Area
Lake Peters (69.32°N 145.05°W) is situated approximately 300 km north of
the Arctic Circle, and 70 km from the Arctic Ocean in the Brooks Range,
Alaska (Figure 1). Lake Peters catchment (171 km2) is
ringed by steep mountains, and glaciated (9%) with some of the largest
valley glaciers in Arctic Alaska. Bedrock comprises low-grade
metasedimentary and sedimentary rocks, primarily southward-dipping
sandstone, semischist, and phyllite, with minor chert and quartzite
(Reed, 1968). Terminal cirque and lateral moraines formed during the
Little Ice Age (ca. 1200-1850 CE) are conspicuous around the
margins of extant glaciers in the headwaters of Lake Peters (Evison,
Calkin, & Ellis, 1996). Interpolated climate data for Lake Peters (1980
- 2009) shows mean annual precipitation of 360 mm, and mean January and
July monthly temperatures of -22.0°C and 10.5°C, respectively (Stavros
& Hill, 2013). Permafrost is known to occur at the bases of hillslopes
and in bottoms of river valleys in the Brooks Range (Kanevskiy et al.,
2011). Soils are sparse, and vegetation largely consists of arctic
grasses, herbs, and riparian shrubs. Above 1300 m, channel-side
vegetation is sparse.
Lake Peters is the primary source of the Sadlerochit River, which
discharges into the Arctic Ocean. Carnivore Creek (128
km2 catchment; 10% glacier coverage based on aerial
photos taken in 2016) and Chamberlin Creek (8 km2catchment; 23% glacier coverage) are the primary inflows to Lake Peters
(Figure 2), although several minor non-glaciated catchments also flow
into Lake Peters over and through large alluvial fans (Figure 1). The
Carnivore sub-catchment covers 75% of the total area of the Lake Peters
catchment. The eastern side of the Carnivore valley is glacierized,
channels are more deeply incised, and side-valley alluvial fans are more
compact, compared to the western side. The lower reach of Carnivore
Creek flows over a shallow slope with plane bed morphology, surrounded
by a terraced floodplain and some periglacial surface features.
Chamberlin Creek’s catchment is comparatively steep, with the summit of
Mount Chamberlin (2750 m asl) only 4.7 km from Lake Peters. Chamberlin
Glacier (1.9 km²) is the third largest glacier in Lake Peters catchment.
In the upper catchment, Chamberlin Creek flows over and through
moraines; in the mid-catchment the channel has incised into a confined
bedrock-controlled valley, with steep step-pools; and downstream of the
alluvial fan apex the creek flows over lower-grade step-pools (Figure
3).
Methodology