2.2 Stream sampling

To capture a gradient of disturbance and quantify the spatial variability of post-fire stream Na+ and NO3-, we sampled stream water roughly every 800 meters along the mainstems of our study watersheds (Figure 1). This distance was selected to ensure a consistent sampling interval that maximized the number of samples collected per watershed but would allow us to complete watershed sampling within one day. Low-flow conditions were stable and there were no precipitation events during our sampling period (6/1/2018-6/7/2018). Previous research at the Hayman Fire demonstrated that patterns of elevated stream NO3- in severely burned watersheds persist across flow conditions (Rhoades et al., 2019) so our June sampling date should be broadly representative. All stream samples from a given watershed were collected within a single day in pre-washed 1 L high-density polyethylene bottles moving in the upstream direction. Samples were immediately filtered with 0.45 µm polyvinyl diethylene filters (MilliporeSigma, Burlington, MA) and analyzed for concentrations of stream Na+ and NO3- using ion chromatography (Dionex ICS-3000, Waltham, MA and Waters 580, Sunnyvale, CA). Detection limits for both Na+ and NO3- were 0.01 mg/L; any concentrations below that were replaced with ½ the detection limit.

2.3 Geospatial analysis

We conducted a terrain analysis to characterize the underlying watershed structure. First, flow direction was derived from a 10-m digital elevation model (DEM) (U.S. Geological Survey, 2018) using the multiple triangular flow direction algorithm (Seibert & McGlynn, 2007). Watershed contributing areas were delineated and upslope accumulated area (UAA) was calculated for all sampling points (0.32 – 35.4 km2) using the openSTARS package (Peterson & Ver Hoef, 2014) in R Studio. We summarized topographic, vegetation, and fire variables as means and proportional extents within the contributing areas for each sampling location (Table 3).
Table 3: Watershed predictor variables that were summarized for the contributing area to each sampling point. Pearson correlation coefficients were calculated between each predictor variable and Na+ or NO3-. Vegetation metrics represent current conditions (i.e., June 2018) whereas fire metrics represent immediate post-fire condition (i.e., August 2003). Variables marked with a ˣ were removed prior to linear mixed model selection due to strong correlation (>0.90) with another predictor variable. Coefficients depicted in grey identify variables removed during linear mixed model selection; those in black were retained for subsequent modeling.