4 Discussion

4.1 Modeling streamwater chemistry in burned watersheds

Multiple lines of evidence indicated that stream NO3- concentrations had greater spatial variability and weaker spatial structuring relative to Na+. First, semivariance was greater for stream NO3- than Na+ across all flow-connected distances (Figure 3) which suggests higher variability in stream NO3-concentrations across all measured scales (Isaak et al., 2014). Secondly, the nugget effect was orders of magnitude greater for stream NO3- than Na+ (0.385 and 0.001 respectively) which indicates unmeasured fine-scale variability in stream NO3-concentrations (Cooper et al., 1997). Finally, Na+semivariance increased with lag distance and stabilized around 3,700 m (Figure 3). This strong positive autocorrelation indicates that downstream hydrologic transport was the primary driver of spatially distributed Na+ concentrations. In contrast, the empirical semivariogram for NO3-exhibited irregular trends in semivariance that did not stabilize across the measured range in spatial scales (Figure 3).
SSN model improvements varied with the solute of concern and network position. For Na+, the SSN model reduced the AIC by 61%, RMSPE by 20%, and unexplained variance by 96% compared to the MLR model (Table 4). In contrast, the NO3- SSN model only reduced the AIC by <1%, the RMSPE by 7%, and the unexplained variance by 54% (Table 4). SSN model improvements tend to be smaller where spatial autocorrelation is lower (Isaak et al., 2014) such as with NO3- at our sites. Additionally, SSN models improved predictions more in downstream positions whereas MLR prediction error was relatively consistent across network positions (Figure 4C-D). Moving downstream, SSN models are informed by an increasing number of upstream data points. Conversely, SSN predictions in headwater locations rely more on watershed attributes than upstream data, much like MLR models.

4.2 Post-fire vegetation is a dominant driver of stream NO3-patterns

Large high severity fire has the potential to shift ecosystems from forest to grass and shrubland which can have implications for watershed N cycling. Even decades after the Hayman and nearby fires, 75% of high severity plots had no conifer regeneration and it is possible that forest density will never return to pre-fire levels in these areas (Chambers et al., 2016). Beyond our field sites, there is broad evidence of declining post-fire tree regeneration due to increasing climate aridity and fire activity which can shift previously forested systems into alternative stable states dominated by grassland and weedy, herbaceous vegetation types (Coop et al., 2020; Stevens-Rumann et al., 2018; Tepley et al., 2017; Walker et al., 2018). Forest cover is often a primary mechanism for terrestrial N retention (Dunnette et al., 2014; Vitousek et al., 1979) and changes from forest to grass and shrub cover can impact ecosystem N retention (Lovett et al., 2002). For example, conifers will more strongly regulate N cycling than grasses and forbs given their underlying nutrient use efficiencies (Chapman et al., 2006). Therefore, post-fire watersheds with little tree regeneration will likely be leakier with respect to N cycling.
Spectral vegetation indices were the strongest predictors of stream NO3- in this and other studies. For example, reduced post-fire plant cover, measured as NDVI, explained the persistence of elevated post-fire stream N (Rust et al., 2019). In this study though, the strongest predictor of stream NO3- concentration was mean NDMI (Table 3), a vegetation index that considers both canopy cover and the water stress of that vegetation. NDMI is more sensitive to burn severity, forest type, and forest loss and recovery than NDVI which is broadly sensitive to the amount of photosynthetically active vegetation (Morresi et al., 2019). The strong inverse relationship between NDMI and stream NO3- demonstrates that vegetation cover was a primary control on watershed N retention across spatial scales and the loss of forest cover lead to elevated stream NO3-. This is consistent with earlier work demonstrating that stream NO3-concentrations were inversely related to riparian vegetation exposure (Rhoades et al., 2019).
Rapid in-stream uptake and processing contribute to variability in stream NO3- concentrations (Bernhardt et al., 2003). Nitrate uptake lengths in nearby Wyoming streams ranged from hundreds to thousands of meters (Hall et al., 2009), so uptake is likely to influence NO3- patterns across the range of scales in our study (<9,000 m). However, headwater streams with elevated ambient inorganic N concentrations have a limited ability to moderate downstream transport of inorganic N (Covino et al., 2021b) because nutrient delivery to streams is often orders of magnitude greater than in-stream production or removal (Brookshire et al., 2009). Our previous work at the Hayman Fire demonstrated that in-stream biotic N demand increased after the fire, but N supply from burned uplands exceeded the increase in stream N demand (Rhea et al., 2021). While in-stream uptake likely contributed to spatial variability in stream NO3-, our work demonstrates strong post-fire vegetation controls on the spatial patterns of stream NO3-concentrations.

4.3 Burned headwaters are susceptible to elevated stream NO3-

Patterns of vegetation cover interact with watershed structure to drive spatial distributions of stream NO3-concentrations. Terrestrial inputs of water and dissolved solutes comprise a large portion of streamwater composition in headwater positions, making these areas particularly sensitive to disturbance in the surrounding uplands (Gomi et al., 2002; Likens & Bormann, 1974; Lowe & Likens, 2005). Thus, the vegetation cover of large convergent hillslopes should have stronger proportional influence on stream NO3- concentration in headwater positions relative to locations lower in the network. We found that convergent hillslopes in the headwaters of Brush Creek were associated with low NDMI (Figure 5E) and aligned with locations of high stream NO3- (Figure 5E). Proportional inflows declined downstream and were associated with higher NDMI. Stream NO­3- also declined downstream in Brush Creek, likely due to a combination of reduced proportional influence of hillslope inputs, streamflow dilution, and in-stream N uptake. In the unburned headwaters of Pine Creek, convergent hillslopes were associated with high NDMI (Figure 5F) and likely high terrestrial N demand. Stream NO3concentrations remained low throughout the headwaters with only slight downstream increases where hillslopes were sparsely vegetated (Figure 5F).
This investigation demonstrates that convergent hillslopes in headwater positions are particularly sensitive to wildfire-induced vegetation mortality and can impact both local and downstream water quality. Headwater attributes have been shown to predict downstream water chemistry (i.e., NO3-, PO43-, Ca2+, and Sr2+) at distances > 500 km (French et al., 2020). The sampled stream networks were only 5,520 - 8,289 m, so headwater attributes could feasibly influence downstream chemistry throughout the entire stream networks. Indeed, the watershed with burned headwaters (i.e., Brush), sustained higher stream NO3- concentrations throughout its stream network compared to the watershed with unburned headwaters (i.e., Pine, Figure 5E-F). These findings may help prioritize post-fire watershed rehabilitation efforts aimed at increasing plant cover and nutrient demand to reduce stream NO3-concentrations. More specifically, our findings highlight the potential value for post-fire regeneration in convergent headwater locations to enhance N retention and reduce downstream NO3- export.