1 Introduction

Wildfires are a natural part of many forested ecosystems, but the frequency and severity of wildfires has been increasing across the Western US (Abatzoglou et al., 2017; Westerling, 2016). Elevated wildfire activity can threaten the function of critical forested watersheds that supply clean water to much of the Western US (Brown et al., 2008). Nitrogen (N) typically limits plant growth so N export often indicates ecosystem disturbance and shifts in nutrient supply and demand (Chapin et al., 2011). Short-term (<5 years) increases in stream nitrate (NO3-) have been documented following wildfires across the Western US (Rust et al., 2018; Smith et al., 2011) due to elevated soil N mineralization and leaching (Smithwick et al., 2009; Turner et al., 2007; Wan et al., 2001). In some cases, stream NO3- can remain elevated for decades and has been shown to decrease with post-fire vegetation cover (Rhoades et al., 2019; Rust et al., 2019) and increase with burn extent (Rhoades et al., 2019). These results suggest that a lack of vegetation recovery is likely a dominant driver of persistent post-fire NO3- export, but this relationship remains poorly understood.
The interaction of vegetation cover, watershed structure, and stream network geometry regulates watershed solute export (Abbott et al., 2021; Covino et al., 2021; Creed & Beall, 2009; Likens & Bormann, 1974; Lovett et al., 2002; Shogren et al., 2021; Zarnetske et al., 2018). Watershed structure is the spatial arrangement of divergent and convergent hillslopes across the landscape (Baiamonte & Singh, 2016; Jencso et al., 2010). Divergent hillslopes are convex and contribute little flow to the stream, whereas convergent hillslopes concentrate hydrologic flowpaths and contribute large inputs to channel networks (Detty & McGuire, 2010). In headwater positions, water and solutes are primarily derived from shallow groundwater contributions from adjacent hillslopes (Covino et al., 2021; Gomi et al., 2002; Likens & Bormann, 1974) whereas upstream sources increasingly dominate water composition in lower network positions (Vannote et al., 1980). Therefore, headwaters are particularly sensitive to disturbance in the surrounding uplands (Lowe & Likens, 2005) and contributions to the stream in these locations have the potential to exert strong control on downstream solute concentrations (Abbott et al., 2018; Alexander et al., 2007; Wohl, 2017).
To better understand the spatial patterns in post-fire water chemistry, we consider both conservative and reactive solutes. Conservative solutes, such as sodium (Na+), have low biological demand (Dingman, 2015; Stream Solute Workshop, 1990) and thus are primarily driven by physical transport processes (Webster & Valett, 2006) and watershed geophysical properties (Brennan et al., 2016; French et al., 2020; McGuire et al., 2014). In contrast, biologically active solutes such as NO3- are controlled by interactions between hydrologic transport and biological uptake (Bernhardt et al., 2003, 2005). In particular, forest cover can be a primary control on NO3- export at the watershed scale (Bormann & Likens, 1967; Likens et al., 1970).
Statistical models can be used to partition the spatial variance in stream Na+ and NO3-among landscape (i.e., topographic, vegetation, and fire predictors) and stream network (i.e., flow-connected distance) characteristics. Multiple linear regression (MLR) modeling can be used to determine the relative influence of specific landscape characteristics on spatially distributed solute concentrations (Cho & Lee, 2018; McManus et al., 2020), but this approach assumes independence of sampling locations. Geostatistical modeling approaches, such as spatial stream network (SSN) models, are better suited to differentiate landscape from stream network attributes since they account for spatial autocorrelation of flow-connected samples and the dendritic and unidirectional nature of stream networks (Ver Hoef et al., 2014; Isaak et al., 2014; Peterson & Ver Hoef, 2010). We paired spatially distributed water chemistry sampling with terrain analysis and vegetation and fire mapping to address the following objectives: 1) examine the degree to which topographic, vegetation, and fire variables predict stream Na+ and NO3- across spatial scales and 2) evaluate the performance of MLR and SSN models in predicting stream solute concentrations. To our knowledge, this study is the first to use geostatistics to investigate the drivers of elevated post-fire stream NO3.