Sediment Model Results
To develop a useful predictive tool, 2015 and 2016 data was used in one unified model of SSC. To avoid overfitting and spurious relations, our best models do not fit predictors that are significantly correlated. Correlated groups can be generalized as follows: a) logetransformed Q variables and NTU; b) P variables; c) T, GT, and SR variables; and d) coshr and sinhr; with QE and H standalone, although groups specific to each primary predictor were used for modeling. Carnivore Creek NTU is significantly correlated with P12,24 and T12,24 at all elevations, as well as with T2,6 and GT12,24 at 854 m asl, and P6 and GT2 at 1425 m asl. Chamberlin Creek NTU is uncorrelated with most P variables and significantly correlated with most T and GT variables. Carnivore Creek Q is significantly correlated with P12,24 at all elevations, with P6 at 1425 and 1750 m asl, and with all T and GT variables. Chamberlin Creek Q is also significantly correlated with most P variables and all T and GT variables. In some cases, the temporal variables were also correlated with the primary predictors, or with the meteorological or other temporal variables. Compared with simple sediment rating curves, the additional predictor variables (uncorrelated with NTU or Q) improve variability explained (R2) by 15% and 9% in Carnivore Creek, and 9% and 27% in Chamberlin Creek for NTU- and Q-based models, respectively. Best-fit combinations varied between creeks, and with the base predictor–NTU or Q (Table 3). Q was the only retained discharge variable (i.e. ΣQ and ΔQ were not incorporated in any of the best-fit models). Temperature and precipitation variables were excluded from most of the models because of significant correlations with the primary predictors. Exceptions include ground temperature (GT21750 m asl) in the Carnivore Creek NTU-based model, and precipitation (P12 1750 m asl) in the Chamberlin Creek NTU-based model. Sinhr is a significant tertiary predictor in the Chamberlin Creek NTU-based model, indicative of higher SSCs at 1800 than at 0600. In the Q-based model for Carnivore Creek, coshrwas selected, indicative of higher SSCs at midnight than at noon. We found H to be a significant additional predictor in the Chamberlin Creek Q-based model. SSCs output by the NTU-based models reached maximums of 2275 and 2409 mg L-1 in Carnivore Creek, and 261 and 203 mg L-1 in Chamberlin Creek, in 2015 and 2016, respectively (Figure 4).