Best-fit Sediment Models
Our best-fit models are statistically significant predictors of SSC in
tributaries to Lake Peters, although there is considerable unexplained
variability (Table 3). Mass wasting independent of discharge (Gao, 2008;
Hammer & Smith, 1983; Hasholt et al., 2005; Walker & Hudson, 2003),
and sediment pulses associated with glacier motion (Hasholt et al.,
2005; Willis et al., 1996), may cause transient flushes of sediment not
accounted for by our models. Compared with Carnivore Creek, the
consistency and accuracy of SSC modeling is lessor in Chamberlin Creek,
which has a smaller sub-catchment size. This is relatable to the
inherent flashiness of smaller catchments (Horowitz, 2003), making
complex sediment transfer processes difficult to quantify, even with
continuous and high-resolution model predictors. NTU-based models
outperformed Q-based models as a predictor of SSC, which is not
surprising given SSC can vary by two orders of magnitude for any one
discharge (Morehead et al., 2003), whereas turbidity is a more direct
surrogate for SSC. In contrast to our models (Table 3), earlier
multiple-regression sediment models developed for arctic rivers have not
incorporated NTU, but have favored alternative discharge variables (ΣQ,
ΔQ, QE, and/or Q2), with positive
and/or negative coefficients depending on the catchment, season, or
sub-season (Hodgkins, 1999; Hodson and Ferguson, 1999; Irvine-Fynn et
al., 2005; Schiefer et al., 2017). Further, these earlier models have
been geographically limited to catchments in Svalbard.
In all cases at Lake Peters, inclusion of additional meteorological or
temporal predictor variables (uncorrelated with NTU or Q) improved
performance of models predicting SSCs (Table 3), and supported our
understanding of sediment transfer processes over two years of
open-channel flow. The best-fit model predictors could not be used
standalone to interpret physical processes, because exclusion of
correlated predictors to avoid overfitting masked some
meteorology–hydrology interactions, and numerous predictor combinations
provided statistically significant model outputs. Therefore, further
analysis of the hydrometeorological data, as well as qualitative
geomorphologic evaluations, informed our understanding of sediment
transfer processes at Lake Peters.