3 Results
3.1 Stream Na+ and
NO3-concentrations
Observed stream Na+ concentrations ranged from
3.9-13.1 mg/L (Supplemental Figure 1), with an average concentration of
7.3 mg/L which is similar to the pre-fire average of 6.1 mg/L reported
in these granitic basins (Rhoades et al., 2011). Kelsey had the highest
and Brush had the lowest mean stream Na+ concentration
whereas Brush had the highest and West Turkey had the lowest coefficient
of variation (Table 1). Observed stream
NO3- concentrations varied by three
orders of magnitude (0.005 – 6.2 mg/L) and average stream
NO3- concentration was 0.91 mg/L which
is five times greater than pre-fire concentrations (0.18 mg/L) (Rhoades
et al., 2019). Brush watershed had the highest (3.06 mg/L) and Gunbarrel
the lowest (0.16 mg/L) mean NO3-concentration whereas Fourmile had the greatest and West Turkey the
lowest coefficient of variation (Table 1). The coefficient of variation
was consistently higher for stream
NO3- (Table 1) indicating greater
within-watershed variability in stream
NO3- compared to
Na+.
3.2 Landscape controls on
Na+ and NO3-
Topographic, vegetation, and fire predictor variables were weakly
correlated (≤0.33) with log[Na+] (Table 3). Linear
mixed model selection identified watershed area, slope, riparian extent,
TWI, and tree and shrub cover as the best predictors of
log[Na+] (Supplemental Table 1). Stream
Na+ was related positively to watershed area, slope,
riparian extent, and tree cover, and negatively to TWI, and shrub cover
(Figure 2). All watershed predictors were significant in the
Na+ MLR model and together explained 54.4% of the
variance in log[Na+] (Table 4). Predictor
variables explained 45% of the variance in
log[Na+] in the Na+ SSN model
and all predictors except watershed area were significant (Table 4).
Table 4: Summary of spatial stream network (SSN) and multiple
linear regression (MLR) models that predict log-transformed stream
Na+ and NO3-concentrations. Parameter estimates represent the regression
coefficient, which is the change in the response variable based on a
1-unit change in the predictor variable while holding all other
variables constant. Statistical significance of predictor variables is
denoted with symbols *=0.05, **=0.01, ***=0.001. Variance decomposition
assigns variance in Na+ or
NO3- to watershed predictor variables,
flow-connected autocorrelation, and unexplained variance. MLR models do
not account for flow-connected autocovariance. Model performance metrics
come from iterative leave-on-out cross-validation.