Untangling the effects of seasonality and stream channel erosion on the runoff composition in a previously burned mountain catchment

Stream channel incision and deposition are common after wildfire, and after these geomorphic changes occur, they may impact runoff mechanisms and the composition of pre‐event and event water in runoff. To investigate this, we monitored discharge and electrical conductivity at six nested sites within a 15.5 km2 watershed in the northern Colorado Front Range that had burned several years prior, and then experienced large flooding and well‐documented and significant channel erosion and deposition in the following years. Over the study period, which occurs 3 years after the fire, and 2 years after the beginning of significant geomorphic changes, the watershed experienced seven precipitation events. For each hydrograph, we separate baseflow from runoff using a new method to characterize and account for the strong diurnal signal in the baseflow. Electrical conductivity is used as a tracer in a two‐component end‐member mixing analysis to separate the event hydrographs into event and pre‐event water. Correlation coefficients were computed between key variables of the hydrologic response (such as runoff ratio, volumes of event and pre‐event water) to storm and basin characteristics (including stream channel erosion/deposition, fraction of high/moderate burn severity, precipitation intensity and antecedent precipitation). The strength and significance of correlations was found to vary seasonally. In the early season, event and pre‐event volumes did not vary significantly with basin or storm characteristics. In the late season, antecedent precipitation correlated with a decrease in event runoff (R2 = 0.34) and total runoff (R2 = 0.40), increased precipitation intensity correlated with an increase in event runoff (R2 = 0.48) and local erosion correlated with an increase in pre‐event runoff (R2 = 0.60) and total runoff (R2 = 0.53). We hypothesize that the difference in correlations between early season and late season is due seasonal variations in the groundwater table gradient. These findings indicate that seasonality and postfire stream channel erosion influence the makeup of runoff response.


| INTRODUCTION
Wildfires impact a wide range of ecological, hydrologic and geomorphic processes (Ebel & Moody, 2013;Santi & Rengers, 2020).Studies investigating the immediate effects of wildfires on hydrologic response and event runoff have shown that fire and soil heating can affect the infiltration and hydraulic connectivity of soils by creating hydrophobic soils, (Ebel, 2012;Ebel & Moody, 2013;Ebel et al., 2018;Hoch et al., 2021) or soil sealing (Larsen et al., 2009).Infiltration rates and porosity can also be affected (Imeson et al., 1992), both of which can impact whether runoff during and after storms is composed of event water or pre-event water.Furthermore, destruction of vegetation by wildfire has been found to increase runoff ratios (Stoof et al., 2012) and when compared to removal of vegetation by clear cutting, the water repellency of soil following the wildfire had the greater effect on generation of event water during storm runoff (Scott, 1997).Work on wildfires in Mongolia (Kopp et al., 2017) has shown that increased soil water repellency following a wildfire can interact with other geologic and topographic conditions of the basin, such as a seasonally melting permafrost, to alter the expected hydrologic runoff response.
The effects of wildfire go beyond the immediate impacts of hydrophobic soils and burned vegetation, however.Geomorphic processes such as sediment delivery to channels, hillslope erosion, and erosion and deposition within the channel network have been shown to accelerate in the years after wildfire, especially when the fire is followed quickly by significant runoff events (Brogan et al., 2017;Brogan, MacDonald et al., 2019;Brogan, Nelson, & MacDonald, 2019;Guilinger, 2021;Guilinger et al., 2020;Kampf et al., 2016;Rengers et al., 2021;Wohl, 2013;Wohl & Scott, 2017).These long-term postfire changes can compound the fire's effects on basin hydrology and have been shown to affect the timing of damaging debris flows (Hoch et al., 2021;McGuire & Youberg, 2019;Raymond et al., 2020;Tang et al., 2019aTang et al., , 2019b)).However, it is not clear how these geomorphic changes may affect watershed hydrologic response over the longer term.
The relative contribution of event and pre-event water in runoff is impacted by watershed characteristics such as the degree of urbanization (Pellerin et al., 2008) and deforestation (Scott, 1997).Studies on forested mountain catchments have found some consistency in response to some storm and basin characteristics and event/preevent water compositions using various approaches for hydrograph separation, including isotopes and electrical conductivity (Klaus & McDonnell, 2013).In particular, event and pre-event water have been found to be correlated with precipitation intensity (Blume et al., 2007;Matsubayashi et al., 1993;McDonnell, 1990;Norbiato et al., 2009) and the boundary state of soil wetness, or antecedent precipitation (Brown et al., 1999;Inamdar et al., 2013;Litt et al., 2015;von Freyberg et al., 2018).However, the effects of postfire geomorphic changes on the event/pre-event water composition have not been investigated.
Electrical conductivity (EC) is especially useful for studying the hydrologic response in systems with high-intensity, short-duration storms, because EC measurements are readily collected at a high temporal resolution.Electrical conductivity has been used for hydrograph separation for decades (e.g.Pilgrim et al., 1979) and has become more popular due to the low cost and ease of installation, particularly in alpine and other difficult to access locations (Cano-Paoli et al., 2019;Laudon & Slaymaker, 1997).While EC is a nonconservative tracer compared with other methods of hydrograph separation such as measuring stable isotopes, some studies have shown that EC can produce comparable results and can be used as a reliable method of separating event water from pre-event water, especially when precipitation and groundwater have significantly different EC signatures (Klaus & McDonnell, 2013;Lott & Stewart, 2016;Matsubayashi et al., 1993;Pellerin et al., 2008).
Here, we investigate the characteristics of hydrologic response of a mountain catchment that experienced moderate-to-severe wildfire three years prior to the study and a major geomorphically effective flood two years prior to the study (Brogan et al., 2017).We use field measurements of precipitation, stream depth and electrical conductivity distributed throughout the catchment to characterize runoff response during and after summer storms.We develop event hydrographs from stream depth measurements for the period of the study and use EC as a tracer in a mass balance equation to separate the hydrographs into event water and pre-event water.
We use our observations to address the following questions:

| Study area
Skin Gulch is a 15.5 km 2 tributary basin to the Cache la Poudre River in north central Colorado with two primary branches (Figure 1).This study focusses on the western branch, which is 8.9 km 2 in area and occupies elevations ranging from 1890 to 2580 m.The precipitation patterns in the basin are typically characterized by snowfall from November to May, and short, high-intensity convective storms through the summer months.In higher elevations of the basin, snowpack exists throughout the year, but most of the catchment snow cover is intermittent.Underlying geology in the basin consists of primarily Precambrian metamorphic schists, gneiss and igneous rocks with Redfeather sandy loam soils (Abbott, 1970).The basin lies on the Stove Prairie Fault line and the western branch that this study focusses on coincides with a shear zone, which other studies have shown may influence flow locations and channel head initiation (Martin et al., 2021).
In June 2012, the High Park Fire burned more than 350 km 2 in the Colorado Front Range, including the Skin Gulch watershed.Over 50% of the Skin Gulch study area was classified as high burn severity, with an additional 20% classified as moderate burn severity (Figure 1).
Several rainfall events following the fire caused significant changes in the landscape and geomorphic processes in the basin.The convective storms in the months following the fire caused large amounts of hillslope material transport and deposition in the lower reaches of the watershed (Brogan et al., 2017;Kampf et al., 2016).
In September 2013, 1 year after the fire, a high-volume, longduration storm caused widespread flooding and damage to the area of the Colorado Front Range.This flood had significant geomorphic impacts on the Skin Gulch watershed, flushing sediment and causing significant erosion through many sections of the channel (Brogan et al., 2017;Brogan, Nelson, & MacDonald, 2019).Differencing of repeat airborne LiDAR digital elevation models collected over a 4-year period following the fire and storms found that sediment volume changes in the stream corridor were correlated with contributing area, channel width and per cent high or moderate burn severity, among other factors (Brogan, Nelson, & MacDonald, 2019).Data from that analysis are used in this paper to characterize geomorphic variables for the site.

| Discharge and EC measurements
Decagon CTD-10 sensors were placed at six nested locations on 10 June 2015 within the Skin Gulch watershed, as shown in Figure 1.
The CTD sensor employs a vented pressure transducer, a four-probe electrical conductivity transducer and a thermistor to measure depth, EC and temperature, respectively.Water depth is automatically corrected for changes in barometric pressure through venting.Electrical conductivity is automatically temperature corrected based on the procedure outlined in the US Salinity Labs Handbook 60 (USDA Labratory Staff, 1954).Depth and EC were measured continuously at each sensor location in 1-min increments from June through October 2015.Because the CTD-10 sensor cannot operate in freezing conditions, measurements were taken throughout the freeze-free period only, which generally occurs from June to October for the lower elevations of the basin.
F I G U R E 1 Skin Gulch basin area overlain with burn severity levels.Rain gage sensors are labelled SU (upper elevations) SM (for middle elevations) and SL (for lower elevations).CTD-10 sensor sites are labelled 1, 3, 4, 5, 8 and 9.The inset table shows contributing area, burn severity percentages and erosion (NVC) data for each site (NVC <0 corresponds to net erosion, NVC >0 corresponds to net deposition).
Depth-discharge rating curves were developed for each sensor location based on seven manual discharge measurements made at each site throughout the season.In the early season when flow depths were adequate, a Sontek Stream Tracker acoustic Doppler velocimeter was used to measure stream velocity along a cross section.Velocity and depth measurements were taken at 0.1 m increments across the stream, and the velocity-cross sectional area method was used to develop a discharge rating for each site.In the latter part of the season, when shallow depths made the use of the velocimeter difficult, discharge was measured using a slug injection method (Day, 1976;Moore, 2005).Studies have shown this is a comparable substitute for traditional stream gaging methods (Weijs et al., 2013).Per methodology outlined in the literature (Hudson & Fraser, 2005), a salt solution with 2 kg per 1 m 3 /s of discharge was added to the stream.For Skin Gulch, this equated to 1 L of solutions of table salt at 100 g/L applied at each slug injection.This solution was injected 50 m (approximately 25 stream widths) upstream of each sensor.During the slug injections, the CTD-10 sensors were set to measure depth and EC every 10 seconds and changed back to 1-min intervals after the slug injection was completed.These measurements did not coincide with any of the storms measured in this analysis, and no impact to the EC was noted after the slug injection.Due to the recession of baseflow throughout the season, measurements during high baseflow conditions at the beginning of the season were higher than almost all of the individual storm-event discharges.

| Storm identification, baseflow separation and removal of diurnal fluctuations
Seven discharge events were registered by the sensors and were isolated for analysis.A total of 34 hydrographs were developed from six sites and seven storms, with eight site/storm combinations being excluded due to sensor malfunction or stream discharge dropping below the depth of the sensor.
Outside these events, baseflow stream discharge receded throughout the season to between 20 and 30% of the baseflow discharge at the time of the first event in June (Figure 2).It should be noted that the 1 July storm (Storm 4) produced discharge volumes high enough to significantly influence the baseflow recession trend.At all sensor sites, baseflow discharge exhibited significant diurnal fluctuations throughout the season, shown in Figure 2.
In order to determine the seasonal trend of baseflow, a line was fitted to the discharge record at the minimum flow value for each day in the segments of the hydrograph outside the identified storm events, which is a method that has been used by other studies.This line was used as the baseflow trend for the entire season (black line in Figure 2b).However, when separating the baseflow from an individual storm event, this trendline underpredicted the baseflow due to significant diurnal fluctuations (see the solid black line in Figure 4).Therefore, we developed a new method to account for the diurnal fluctuations in the baseflow.A more detailed line that captures the daily variation in discharge was fitted to the segments of the hydrograph between storm events using a Bayesian spline fit procedure (D'Errico, 2021); one of these sections is shown in detail in Figure 2c.This line was used to calculate the difference between the diurnal fluctuations and the seasonal baseflow trend.This variation was normalized to the time of day, and an average daily time series of variations, shown in Figure 2d, was calculated for each recession segment of the hydrograph.These averages were added to the baseflow trends during the individual storm events to create a more accurate representation of the baseflow during that event.
Electrical conductivity may also have exhibited similar diurnal fluctuations during the nonstorm recession periods.However, the magnitude of these fluctuations did not exceed the normal variability in the data, and therefore, baseflow EC for each storm event was assumed to be a constant equal to the average EC for the 3 h preceding each storm event.

| Hydrograph separation
Individual storm events were extracted from the overall time series of discharge (Figure 2a) and EC for each site, and lines were fit to both the EC and discharge during each event to filter out measurement variability (Figure 3) using the Bayesian spline fit procedure (D'Errico, 2021).Storm event durations were trimmed to the time when the storm event discharge (total discharge minus baseflow) reached 10% of the peak storm discharge (Figure 4).
Hydrograph separation was performed using the two-component mass balance equations below to calculate event and pre-event runoff: where q (L/s) is the instantaneous discharge at each timestep (1-min) and C (mS/cm) is the instantaneous EC level at each timestep during the storm event.The subscripts refer to: the total value or the value measured by the sensor (t), the baseflow (bf), the pre-event water (pe), and the event water (e).Pre-event water discharge and baseflow discharge are both flowpaths that originate from the subsurface; however, baseflow occurs with or without a storm event, whereas pre-event water is water that is forced into the stream through precipitation.This procedure has been outlined in multiple studies (Kronholm & Capel, 2015;Lott & Stewart, 2016;Pellerin et al., 2008), and several assumptions about the EC of end members are necessary for this analysis.Event water was assumed to have similar EC as the precipitation, as has been used in other studies (Pellerin et al., 2008;Buttle et al., 1995).The National Atmospheric Deposition Program (NADP, 2021) collects weekly data of various water quality metrics of precipitation, including EC, at sites throughout the country.Electrical conductivity for precipitation and event water was based on the weekly data from NADP site CO19 located in Rocky Mountain National Park, 37 km southwest of our project site.These data were compared to other nearby NADP sensors to check for similarity during the storm events to ensure no outlier data were being used.All precipitation EC values during storm events range from 0.005 to 0.017 mS/cm, which are an order of magnitude less than EC values observed in Skin Gulch during baseflow periods.Precipitation EC differing significantly from background or baseflow is a necessary assumption of using EC as a tracer for event/pre-event water (Kronholm & Capel, 2015;Pellerin et al., 2008).
Baseflow and pre-event EC (C bf , C pe ) were assumed to be equal to the average EC observed during the 3 h before any increase in discharge (values range from 0.11 to 0.19 mS/cm).Additionally, the discharge of the baseflow throughout the storm was calculated based on the baseflow separation methods described above.Therefore, Q bf and C bf are known variables in Equations 1 and 2. With these assumptions, Q e and Q pe can be solved for using Equations 1 and 2. Figure 4 shows an example of the resulting event hydrograph separation.Based on these hydrograph separations, the hydrologic variables shown in Table 2 were calculated.Table 1 lists volumes of total discharge (V t ), nonbaseflow or storm runoff (V st ), event-water (V e ) and pre-event water (V pe ), which were calculated by integrating under the curves of the respective hydrographs during the storm period (Figure 4).

| Precipitation data
Precipitation data were collected at the five sites shown in Figure 1, using RainWise tipping bucket rain gages.Peak 15-min rainfall intensity and cumulative depth of rainfall were calculated for each of the rain gage sites.Grided rasters of these data were created using inverse distance weighting (IDW) between the rain gages and across the study area.volume that occurred during the storm event within the contributing area of the respective sensor location.Due to the nested nature of these basins, the results show the expected correlations between basin area and total precipitation volumes; however, by dividing by the contributing area, some of this bias in the data has been removed.

| Geomorphic data
Previous studies on Skin Gulch (Brogan, Nelson, & MacDonald, 2019) used analysis of repeat airborne LiDAR data sets and raster differencing

| Analysis
Ordinary least-squares linear regression was used to compare key variables of the hydrologic response (runoff ratio (RR), volume of event water per precipitation (V e /P), volume of pre-event water per precipitation (V pe /P), volume of pre-event water per volume of storm discharge (V pe /V st ), peak event water discharge per peak total discharge T A B L E 1 Acronyms and definitions for variables used in the analysis of results.Precipitation characteristics AP 10 10-day antecedent precipitation Volume of precipitation occurring in the 10 days before the storm event.
AP 5 5-day antecedent precipitation Volume of precipitation occurring in the 5 days before the storm event.
I 15 15-min storm intensity Peak 15-min intensity of event precipitation.

P Precipitation volume
Volume of precipitation, expressed in depth per contributing area.

Runoff characteristics
V t Volume of total sischarge Total volume of the discharge that occurs during the event.
V st Volume of storm runoff Volume of runoff: Total discharge volume, minus volume of baseflow during the event.
V e Volume of event water runoff Volumetric portion of the storm runoff that has an EC signature similar to precipitation.
V pe Volume of pre-event water runoff Volumetric portion of the storm runoff that has an EC signature similar to the baseflow.

Q t
Peak total discharge Peak flow of total discharge that occurs during the event.
Q bf Baseflow discharge Average discharge of baseflow just before the event.

Q e
Peak event discharge Instantaneous discharge peak of the event water runoff.
Q pe Peak pre-event sischarge Instantaneous discharge peak of the pre-event water runoff.

Analysis variables
V e /P Volume of event runoff per precipitation Volume of event water divided by the precipitation volume.
V pe /P Volume of pre-event runoff per precipitation Volume of pre-event water divided by the precipitation volume.

Runoff ratio
Volume of event sischarge per precipitation volume (V st /P).Note: RR = V e /P + V pe /P.
Q e /Q t Peak event discharge per peak total discharge Event discharge normalized to total discharge for comparison.(RR, V e /P, V pe /P).This method of examining hydrologic response variables has been advocated for in other studies (Blume et al., 2007;Von Freyberg et al., 2018) and helps account for both basin size and storm size when considering hydrologic response.Similarly, peak discharge of event water and pre-event water have been normalized by the peak total discharge of the storm in which they occur to allow comparison of event and pre-event water peaks between storms.Correlation p-values were calculated for each comparison to account for varying number of data points and correlations with p < 0.05 considered significant for this analysis.Gulch (Wilson et al., 2021) as producing the highest amount of hillslope erosion and sediment movement for the 2015 rainfall season.

| Hydrologic response
Total storm discharge volumes (V st ) ranged from 60 m 3 to 3971 m 3 and the fraction of discharge that was pre-event water (V pe /V st ) ranged from 0.3 to 0.99.Total discharge peaks (Q t ) ranged from 9.5 L/s to 91.7 L/s and the pre-event peaks per total peak (Q pe /Q t ) ranged from 12% to 58%.Runoff ratios (RR, volume of storm discharge per volume of precipitation) ranged from 0.2% to 2% over the study period.Volume of pre-event water per precipitation (V pe /P) ranged from 0.04% to 1.6% and volume of event water per precipitation (V e / P) ranged from 0.03% to 0.8%.

| Correlations with hydrologic response
Correlations between hydrologic response volumes and peak discharges were compared with the basin and storm characteristics   shown in Table 3. Results were separated into early season (Storms 1, 2 and 3) and late season (Storms 5, 6 and 7) shown in Figure 2.
Storm 4 was excluded from this data segregation as it occurred at the transition of high and low baseflow and had a noted effect on the overall baseflow for several weeks after the event.The entire period of record included 34 hydrographs, the early season included 16 hydrographs and the late season 13 hydrographs.
During the entire period of record, pre-event water per precipitation (V pe /P) was highly correlated with runoff ratio (RR) (R 2 = 0.94).
Correlations for all variables are shown in Table 2. Event water per precipitation (V e /P) also had a significant correlation with the overall runoff ratio (R 2 = 0.39), but with a much weaker signal.The ratio of peak event water to total peak discharge (Q e /Q t ) was moderately correlated with peak 15-min storm intensity (R 2 = 0.32).
During the early season, volume of antecedent precipitation (AP 5 and AP 10 ) had no significant correlation with the overall volumes of runoff (RR), event water (V e /P) and pre-event water (V pe /P).In the late season, however, antecedent precipitation (AP 10 ) was negatively correlated with volume of event water per precipitation (V e /P) (R 2 = 0.34), but did not have a significant correlation with the volume of pre-event water (V pe /P).Antecedent precipitation (AP 10 ) also correlated (negatively) with total runoff volume per precipitation (RR) (R 2 = 0.41), indicating that preceding precipitation led to a decrease in the overall amount of runoff from precipitation as well as a decrease in the event water per precipitation.
Likewise for 15-min precipitation intensity (I 15 ), storms with higher intensity in the early season correlated with only the relative peaks of both event water (Q e /Q t ) and pre-event water (Q pe /Q t ) (R 2 = 0.31 and 0.53), both increasing with greater precipitation intensity.However, overall volumes of event water (V e /P) and pre-event water (V pe /P) did not increase with greater storm intensity in the early season.In the late season, precipitation intensity (I 15 ) was very highly correlated (R 2 = 0.91) with an increase in relative peak of the event water (Q e /Q t ).In addition, in the late season, higher precipitation intensity correlated with an increase in the volume of event water per precipitation (V e /P) (R 2 = 0.49) but did not significantly correlate with a change in pre-event water (V pe /P) or total runoff ratio (RR).
Localized erosion had little correlation with any significant runoff response during the early season.However, during the late-season storms, net volume change of in-channel sediment (NVC) was negatively correlated with both the overall runoff ratio (RR) (R 2 = 0.53) and the volume of pre-event water per precipitation (V pe /P) (R 2 = 0.61).The negative correlation of net volume change can also be expressed as a positive correlation with more erosion.
Contributing areas with a higher percentage of high burn severity (High B.S.) followed similar patterns as local erosion.In the late season, sensor contributing areas that had higher amounts of high burn severity areas correlated with an increase in pre-event water per precipitation (V pe /P).Areas with higher amounts of moderate burn severity area (Mod.B.S.) followed the opposite pattern, correlating with a decrease in pre-event water per precipitation (V pe /P).It should be

| DISCUSSION
The results for the entire season show disparities between storms and sensor locations with few significant correlations (Table 2).Some of this is likely due to the 8 July storm (Storm 4), which was significantly different in storm characteristic from the other storms during the period of record.This 8 July storm produced double the runoff volume of the other storms and shifted the seasonal recession trend of the hydrograph at each site.However, seasonal differences in correlations are observed when the data are segregated between the early-season and late-season events.Early-season storms (Storms 1, 2 and 3) occurred at the beginning of hydrograph in June, before the baseflow recedes below 50% of the baseflow seen at the beginning of the season.Late-season storms (Storms 5, 6 and 7) occurred between August and October, after the baseflow as receded below 50% of the early June baseflow.In late-season storms, event water (V e /P) and total runoff (RR) decrease with antecedent precipitation and increase with intensity.Pre-event water and total runoff increase with local erosion and high burn severity.In the early season, pre-event water peaks (Q pe /Q t ) increase with higher storm intensity and event water peaks (Q e /Q t ) are weakly correlated with intensity, burn severity and erosion.However, volumes of runoff response (RR, V e /P, V pe / P) in the early season are not significantly correlated with net volume change, burn severity, storm intensity or antecedent precipitation.
changing the amount of stored pre-event water that becomes runoff as a result of precipitation.In the late season, the water table has been depleted through sustained baseflow with limited precipitation throughout the season and the gradient from the water table and the stream level is low.With these conditions, the amount of pre-event water that is displaced during storm events to become runoff is significantly less per unit of precipitation in the late season than the early season.Because of this, in the late season, in situations when more of the precipitation is able to infiltrate to the water table, both the volume of event water per precipitation and the total runoff ratio decreases.However, when the stream channel erodes, it lowers the elevation of the stream's intersection with the groundwater

| Antecedent precipitation
Antecedent precipitation, characterized by the 5-day and 10-day volumes of precipitation that preceded the storm event (AP 5 , AP 10 ), affects the runoff characteristics of the basin in disparate ways depending on when in the season the storm occurs.Conceptual models in other studies show that antecedent moisture correlates with an increase total runoff ratio as well as the amount of event-water per precipitation (Detty & McGuire, 2010;Litt et al., 2015;Sidle et al., 1995;Von Freyberg et al., 2014).Our results suggest that in Skin Gulch, during the early season, the peaks of event water (Q e /Q t ) increase with increasing antecedent precipitation, but no significant correlations with the volumes of event or pre-event precipitation exist (Table 2).However, in the late season, our results indicate the opposite of this occurs: antecedent precipitation correlates with a decrease in the runoff ratio (RR), and the amount of event water per precipitation (V e /P).
To explain this, we look first at the effects of antecedent moisture on fire-affected soils, which has been shown to increase the infiltration rate and hydraulic conductivity of hydrophobic postfire soils (DeBano, 2000;Ebel & Moody, 2013;Gilmour, 1968).In our conceptual model shown in Figure 6, antecedent precipitation increases the Conceptual model of how the depth to groundwater table and antecedent precipitation can decrease the event water (V e /P) and total runoff (RR) in the late season.(c) shows late-season conditions with no antecedent precipitation and hydrophobic soils at the surface.(d) shows how when an increased amount of antecedent precipitation occurs the infiltration capacity of the hydrophobic soils event runoff (V e ) is decreased, but due to low groundwater table gradient the pre-event runoff (V pe ) is not increased by a corresponding amount, lowering the overall runoff ratio (RR) (a) and (b) show the same scenario but with early season, high groundwater table conditions allowing excess infiltration to return surface flow, thereby not affecting the volumes of event or pre-event water (V e -V pe ) or the total amount of runoff (RR).
amount of water that is infiltrated to the groundwater, thereby decreasing the amount of precipitation that occurs as event water.
However, although more event water is infiltrated to the groundwater, the gradient of the near-stream groundwater table is low so additional pre-event water is not displaced from the groundwater table at the same rate that event water is decreased.This causes a decrease in the total runoff ratio and event water per precipitation, with an increase in pre-event water, which is supported by our results.
In contrast, in the early season, the near-stream groundwater table gradient is already steep, especially in the alluvial floodplain, because the overall groundwater level is high.In these conditions, saturation excess of the interflow zone may be either preventing more precipitation from infiltrating, or causing interflow zone water to exfiltrate and become surface runoff (Sophocleous, 2002).Consequently, during the early season, the total volumes of pre-event and event water are not significantly altered by the increased permeability of the surface layer and so no significant correlation with volume exists.

| Precipitation intensity
Precipitation intensity has been linked to increased runoff ratio and increased event water per precipitation in multiple studies (Pellerin et al., 2008;von Freyberg et al., 2018).In Skin Gulch, the runoff response to precipitation intensity changes from the early season to the late season.In the late season, higher precipitation intensity correlates with increased event water per precipitation (V e /P).This response has been seen in other studies (Pellerin et al., 2008;von Freyberg et al., 2018;Waddington et al., 1993) and is generally explained by high rainfall intensity overwhelming infiltration capacity and resulting in greater surface water runoff.In our study, precipitation intensity is also highly correlated with higher event water peaks in the late season, which is consistent with this interpretation.
In the early season, increasing intensity only correlates with the relative peaks of both event water and pre-event water (Q e /Q t , Q pe / Q t ), increasing both, and may be explained simply by the fact that higher precipitation peaks will result in larger runoff peaks.Unlike in the late season though, in the early-season storms the overall volumes of event water and pre-event water per precipitation (V e /P, V pe /P) are not correlated with precipitation intensity.The theoretical difference between late and early seasons is not obvious in this case.
It is possible in reality there is not a different mechanism between late and early season, and the difference in correlations is simply due to the fact that the early-season storms all had similarly high precipitation intensities, making correlations with this variable difficult to identify, whereas in the late season, precipitation intensities were varied from relatively high to low making the correlations more identifiable.
F I G U R E 7 Conceptual model of how the depth to groundwater table and localized erosion can affect runoff in the late season.(c) shows lateseason conditions with no localized erosion and a low gradient groundwater table (d) shows how localized erosion can increase both the total runoff ratio (RR) and pre-event runoff (V pe ) by increasing the gradient of the water table to the stream.(a) and (b) show the same scenario in the early season, when ground water table gradient is already high.In (d) event and pre-event runoff (V e and V pe ) along with total runoff ration (RR) remains the same since the localized erosion does not significantly increase the already steep gradient of the groundwater table.

| Net sediment volume change (NVC)
As far as we are aware, the effect of localized channel erosion on runoff response has not been studied in detail previously.However, other work on stream gains and losses has suggested that groundwater contributions to the stream can vary based on local topography (Harvey et al., 1996;Kasahara & Wondzell, 2003), and we find results here that suggest this is true for localized net sediment volume changes (NVC) in the stream corridor.Similar to the other controls on runoff generation examined in this study, the correlation with localized erosion is disparate based on whether the event occurs in the early or late season.In the early season, sites with local net erosion were only correlated with an increase in the relative peak of event runoff.In the late season, however, locations with local erosion show a strong correlation with an increase in pre-event water per precipitation (V pe /P) and total runoff per precipitation (RR).We hypothesize that this late-season response, and its contrast with the early-season response, can be explained the conceptual model of a lower near-stream groundwater table gradient in the late season, shown in Figure 7.We propose sites with more local erosion lower the elevation of the stream and thereby increase the gradient of the near-stream water table.Therefore, at these sites with more local erosion, precipitation that infiltrates to the groundwater table displaces more pre-event water as runoff than sites with less erosion, and therefore a less-steep groundwater table gradient.In the early season, the gradient between the water table and the stream is already high (as evidenced by the change in baseflow shown in Figure 2) and localized erosion does not significantly increase the gradient to water table.This would result in no significant increase to pre-event water as is supported by the results.
Due to the nested nature of our sensor locations, the fact that this late-season correlation between erosion (NVC <0) and pre-event water is observable shows that this effect is highly localized, and local erosion does not increase pre-event water relative to precipitation at downstream sensor sites, which experienced less erosion or even deposition (NVC >0, Figure 1).This suggests that, in locations where no erosion, or net deposition, occurred, the pre-event water may be removed from the runoff signature in the stream and resupply the groundwater table at that location.
There are also correlations between sensors with a high percentage of the contributing area being high burn severity (High B.S) and an increase in pre-event water (V pe /P).However, this is likely an indirect

| Summary
We have discussed seasonal changes in the correlations between hydrologic response and antecedent precipitation, precipitation intensity, local erosion and burn severity (

| CONCLUSIONS AND FUTURE WORK
Separating storm hydrographs into event water and pre-event water for seven storms across six sites in a mountain catchment affected 3 years prior by fire, and 2 years prior by large rainfall events, revealed correlations with antecedent precipitation, precipitation intensity and net geomorphic channel change.These correlations vary based on whether they occur in the early-season portion of the summer or the late-season portion.We hypothesize that the near-stream gradient of the water table as well as fire-affected hydrophobic soils are part of the explanation behind the disparate results.In the late season, antecedent precipitation decreases total runoff and event water per precipitation, while the pre-event water per precipitation remains the same.We hypothesized that this is due to antecedent precipitation causing higher infiltration but due to the difference in groundwater table gradient the difference in runoff is only observable in the late season.Higher precipitation intensity increases the event water per precipitation as well as the total runoff, but only in the late season.
Finally, our results suggest that locations where localized erosion occurred post-fire plays a role in how much pre-event water per precipitation occurs at a site even 3 years after the fire occurred, potentially by increasing the groundwater table gradient to the stream at that location.
In general, our results show stronger correlations with effects of antecedent moisture, rainfall intensity and net geomorphic volume change during the late season, when we speculate that the gradient of the groundwater table to the stream is low.These results suggest that the gradient of the water table can create conditions where the distribution of event and pre-event water is more sensitive to site and storm characteristics than they would be during early-season conditions with a higher groundwater table.
Our findings could be strengthened with additional seasons of data, and a series of groundwater depth sensors could corroborate the hypothesis that the gradient of the groundwater is behaving as our conceptual models suggest.We acknowledge that these results may be site specific due to confining geology, soil types and other basin characteristics.However, our results suggest that future studies on the runoff characteristics following a wildfire should consider not only direct effects of the fire, such as hydrophobic soils, but also subsequent effects and changes in the geomorphology and topography such as recent channel erosion.Additionally, seasonality may be an important consideration in hydrograph separation studies.This factor may also interact with other watershed characteristics (in this case hydrophobic soils) in unexpected ways, and future studies should consider examining results based on seasonality of the events for differing results from the entire data set.

1.
Is the composition of storm-event runoff several years after a fire affected by postfire channel erosion/deposition or other wildfire/ geomorphic response characteristics?2. What are the storm characteristics that affect the composition of runoff in a disturbed mountain catchment and what underlying mechanisms could explain this response?And through our investigation a follow-up question became necessary: 3. Is event runoff composition affected by water table gradients which vary with local erosion and the season in which the event occurs?

F
I G U R E 2 (a) Cumulative precipitation depth for rain gage SLR1.(b) Discharge record for Site 3 from June to October 2015.Duration of the identified storm events is shown in grey with their corresponding storm number.Storms 1, 2 and 3 were considered early season, and Storms 4, 5 and 6 were considered late season.The fitted seasonal baseflow trend is shown as the black line.(c) The diurnal variation from the baseflow trend in a segment of the hydrograph between storm events with the line that was fitted to the data account for that variation.(d) Daily time series of the diurnal variation and the average of those series, which is applied as the baseflow adjustment in Figure 4.These rasters were used to calculate a spatial average of the 15-min intensities for each storm within the contributing area of each of the depth/EC sensor sites and to calculate the total storm volume of precipitation (P) from cumulative depth across the contributing area of each depth/EC sensor site.The same method of IDW raster generation was also used to calculate precipitation volumes for the 5 and 10 days prior to each storm.Volumes per precipitation shown in Table 1 were calculated by dividing the respective runoff volumes by the precipitation F I G U R E 3 Discharge (a) and EC (c) measurements for Site 8 Storm 3 along with the fitted lines used for calculations.(b) and (d) show the residuals for Discharge and EC, respectively.F I G U R E 4 Hydrograph Separation of Site 8 Storm 3. The solid black line represents the seasonal baseflow trend.The dashed black line shows the adjustment made to the baseflow based on the diurnal fluctuations of the preceding recession period.The red line shows the total discharge, and the blue and green lines show the calculated pre-event and event water discharges respectively.The grey area shows the storm period over which event results were calculated.
to calculate geomorphic and topographic information in 50-m segments of the stream.For this study, geomorphic and topographic variables of net volume change of in-channel sediment, per cent area of high and moderate burn severity, and contributing area, described for each site in Figure1, were examined from the time period of 2012, just after the High Park Fire but before the large flood in 2013, to the beginning of 2015 when this hydrologic information was collected.The geomorphic variable (Net Volume Change, NVC) used for this study was summed from the segment that contained the depth/EC sensor and the segments immediately upstream and downstream of that segment, for total reach length of 150 m around each sensor site.The area around several of the sensor locations experienced significant geomorphic changes in the years following the fire, most of which occurred in response to the large storm events in 2013.Figure5shows the area just upstream of sensor site location 8, which experienced the second largest decrease in volume of sediment in the channel (NVC).
Peak pre-event discharge per peak total dischargePre-event discharge normalized to total discharge for comparison GIESCHEN and NELSON(Q e /Q t ), peak pre-event water per peak total discharge (Q pe /Q t )) to storm and basin characteristics (net volume change of sediment in the stream corridor (NVC), fraction of contributing area with moderate burn severity (Mod.B.S.), fraction of contributing area with high burn severity (High B.S.), 15-min precipitation intensity (I 15 ), volume of precipitation for each gage in the 10 days preceding the storm event (AP 10 ), volume of precipitation for each gage in the 5 days preceding the storm event (AP 5 ) and volume of precipitation for each gage for the storm event (P)) shown in Table 1.Volumetric hydrologic response variables have been normalized by the precipitation volume of the associated storm within the contributing area of the sensor location

4. 5
to 36.3 mm/h.Storm 5 on 16 August produced the highest intensity rainfall event for all sites during the study period in 2015 (17.5-30.2mm/h), but due to having a shorter duration it only produced between 12 and 63 mm-km 2 of cumulative volume.In contrast, Storm 4 on 8 July only had moderate precipitation intensities (9.9-11.5 mm/ h) but with storm durations lasting more than 10 h and significant preceding precipitation, produced the highest cumulative volume at all of the site locations (51-209 mm-km 2 ).These two storms were the most significant of the summer and have been noted in other work in Skin

F
I G U R E 5 Photograph of channel erosion upstream of sensor 8 taken at the time of sensor installation in June 2015.The red dashed line shows the estimated bank line prefire.Sensor site 8 had the second highest amount of local channel erosion (after sensor site 9) with À1436 m 3 change during the postfire period of 2012-2015.See Figure 1 for in-channel volume change of all the sensor sites.
noted that net volume change (NVC) has a strong negative (or positive for erosion) correlation with High B.S. (R 2 = 0.52) and a strong T A B L E 2 (Continued)

Q
negative for erosion) relationship with Mod.B.S. (R 2 = 0.65), indicating a significant link between percentage of contributing area with high burn severity on likely hood of increased localized erosion.High percentage of contributing area with high burn severity may therefore be indirectly correlated with an increase in pre-event water in the late season through erosion, rather than a direct correlation.
correlation through net volume change (NVC).High percentages of high burn severity area (High B.S.) is correlated with pre-event water (V pe /P) in the same way that net volume change (NVC); is a higher percentage of moderate burn severity (Mod B.S.); is correlated with a decrease in pre-event water; and both have similar strength of correlations.High B.S. and Mod.B.S. are themselves correlated with NVC, which suggests that sensor locations that had a contributing area with a higher percentage of High B.S. are more likely to have significant local erosion (or negative NVC), and areas with more Mod.B.S. are less likely to have erosion.

Table 2
provides an overview of the precipitation data for each site and storm.Total storm precipitation volumes ranged from 12 to 209 mm-km 2 and peak 15-min precipitation intensities ranged from table, so the near-stream groundwater table gradient becomes locally high, effectively replicating the more widespread steep near-stream groundwater table of the early season.Because the early season already has a steeper gradient of groundwater table to the stream, local erosion has no effect on the groundwater table gradient and therefore the amount of pre-event water that becomes runoff per precipitation.Confining geology, soil types and many other factors will influence this result, and we acknowledge that this result is sitespecific and may not be applicable in other seemingly similar basins.
Table3).Pre-event water (V pe /P) and total runoff (RR) decrease with antecedent precipitation, but only when the groundwater table gradient is low (late season).Higher precipitation intensity increases the volume of event water (V e /P) and total runoff (RR) when the groundwater table gradient is low (late season).When the groundwater table gradient is high (early season), only the relative peaks of the event water runoff (Q e /P) are affected by precipitation intensity, and no volumes are correlated.Pre-event water (V pe /P) and total runoff (RR) are also increased by local erosion of sediment in the stream corridor, but only when the existing groundwater table is low, suggesting that the erosion of the stream bed increases the local slope of the groundwater table to the stream.