Hydrologic regimes drive nutrient export behavior in human
impacted watersheds

Abstract. Agricultural watersheds are significant contributors to downstream nutrient excess issues. The timing and magnitude of nutrient mobilization in these watersheds are driven by a combination of anthropogenic, hydrologic, and biogeochemical factors that operate across a range of spatial and temporal scales. However, how, when, and where these complex factors drive nutrient mobilization has previously been difficult to capture with low-frequency or spatially limited datasets. To address this knowledge gap, we analyzed daily nitrate concentration (c) and discharge (Q) data for a four-year period (2016–2019) from five nested, agricultural watersheds in the midwestern United States that contribute nutrient loads to the Gulf of Mexico. The watersheds span two distinct landforms shaped by differences in glacial history resulting in natural soil properties that necessitated different drainage infrastructure across the study area. To investigate nutrient export patterns under different hydrologic conditions, we partitioned the hydrograph into stormflow and baseflow periods and examined those periods separately through the analysis of their concentration-discharge (c-Q) relationships on annual, seasonal, and event time scales. Stormflow showed consistent chemostatic patterns across all seasons, while baseflow showed seasonally dynamic c-Q patterns. Baseflow exhibited chemodyanmic conditions in the summer and fall and more chemostatic conditions in the winter and spring, suggesting that water source contributions during baseflow were nonstationary. Baseflow chemodynamic behavior was driven by low-flow, low-NO3− conditions during which in-stream and near-stream biological processing likely moderated in-stream NO3− concentrations. Additionally, inputs from deeper groundwater with longer residence times and lower NO3− concentration likely contributed to low-NO3− conditions in-stream, particularly in the larger watersheds. Stormflow c-Q behavior was consistent across watersheds, but baseflow c-Q behavior was linked to intensity of agriculture and density of built drainage infrastructure, with more drainage infrastructure associated with higher loads and more chemostatic export patterns across the watersheds. This suggests that how humans replumb the subsurface in response to geologic conditions has implications for hydrologic connectivity, homogenization of source areas, and subsequently nutrient export during both baseflow and stormflow. Our analysis also showed that anomalous flow periods greatly influenced overall c-Q patterns, suggesting that the analysis of high-resolution records at multiple scales is critical when interpreting seasonal or annual patterns.


exploration of these relationships at a time scale previously difficult to observe. For example, high-frequency datasets have been used to investigate c-Q behavior at the event scale, revealing dynamic changes in NO3sourcing and processing at short timescales (Blaen et al., 2017;Bowes et al., 2015;Carey et al., 2014). However, much previous work has focused on a single catchment, and/or data collected over a relatively short period of time. This makes it difficult to determine how the connections between the hydrologic, biogeochemical, and anthropogenic factors, which 75 operate over a range of temporal and spatial scales, influence in-stream NO3concentrations. For example, antecedent moisture conditions, and precipitation timing and intensity reflect changes that occur over hours or days (Rozemeijer et al., 2010), while vegetation dynamics, and on-farm practices such as crop planting and fertilization reflect seasonal changes Royer et al., 2006). Additionally, the influence of these factors are impacted by differences in watershed-specific characteristics, such as topography, soil type, land use practices, and geologic history 80 (Marinos et al., 2020;Moatar et al., 2017). Understanding how these processes and watershed characteristics interact across the relevant spatial and temporal scales in heavily managed watersheds is a crucial step in developing strategies to mitigate downstream impact (Hansen et al., 2018).
Only recently have high-resolution records become sufficiently long and instrumentation sufficiently widespread to examine c-Q relationships under different hydrologic conditions in multiple locations. With this, we are now 85 able to identify how streamflow and NO3concentration relationships vary annually and seasonally across key spatial gradients. Here, we analyze four years of publicly available daily measurements of discharge and NO3concentration from five nested agricultural watersheds in the midwestern United States. Using a semi-autonomous event picking algorithm, we partition the hydrograph into stormflow and baseflow periods to address the following research questions: 90 1) How do c-Q relationships during stormflow and baseflow periods vary by season, and what can that tell us about changes in hydrologic connectivity and nitrogen sources throughout the year?
2) What relationship do NO3concentration, load measurements, and c-Q relationships have to underlying and human-impacted watershed properties?
3) How can high-frequency records be used to identify distinct export regimes and characterize anomalous 95 events that might play a disproportionate role in watershed c-Q behavior? For this study we subdivided the Raccoon River watershed into a series of five nested watersheds shown in Figure 1; the Upstream Sac City (USC) and the Middle Redfield (MRF) on the North Raccoon River, the Upstream 105 Panora (UPN) on the Middle Raccoon River, the Middle Jefferson (MJF) on the South Raccoon River, and the https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License. Downstream Van Meter (DVM), which is below the confluence of the three major tributaries draining the area. The MJF is inclusive of USC; MRF is inclusive of UPN, and DVM is inclusive of the entire Raccoon River watershed. Typical of this area, agricultural productivity is the dominant land use in all five watersheds ranging from 85-92% of land use (Table S2), the vast majority of which is corn (Zea mays L.) and soybeans (Glysine max L.). 110 The Raccoon River watershed is marked by a stark divide in landforms driven by recent glaciations, with the majority of the area underlain by glacial sediments deposited by the Des Moines Lobe during the last glaciation of the region approximately 12,000 years ago (Prior, 1991). These areas are characterized by poorly developed surface drainage networks and ephemeral surface water bodies. As a result, extensive tile drainages, ditches, and canals have been installed and constructed in the latter half of the 20 th century to drain excess water from the subsurface ( Figure  115 1). The southwestern portion of the Raccoon River watershed lies within the Southern Iowa Drift Plain, an area that was shaped by 500,000-year-old glacial advances that extended south into present day Missouri (Prior, 1991). This portion of the watershed is characterized by steeper topography and more naturally well-developed drainage networks, which require less drainage infrastructure such as tile drains, ditches, and canals. UPN, MRF, and DVM drain areas that overlay both the Des Moines Lobe and the Southern Iowa Drift Plain. 120 The Raccoon River watershed is characterized by cold dry winters and warm wet summers, with an average annual precipitation of 850 mm (1981PRISM), the majority of which falls as rain between April and October, aligning with the growing season.

Datasets
We analyzed in situ mean daily NO3concentration (c) and discharge (Q) data from the outlet of each 125 watershed at gaging stations maintained by the U.S. Geologic Survey for USC (05482300), MRF (05483600), MJF (05482500), and DVM (05484500), and from the Iowa Institute of Hydraulic Research (IIHR) for UPN (WQS0032).
To retrieve data, we used the dataRetrieval package in R (v 3.6.0) through the National Water Information System (De Cicco et al., 2018). Data for UPN was obtained directly from the IIHR. We analyzed daily discharge and NO3concentration data from January 2016 to December 2019, during which discharge records were complete for all sites 130 and NO3records had > 88% coverage for all sites except UPN, which had 72% coverage (Table S1). At each gaging station, NO3concentrations were measured at 15-minute resolution (5-min for UPN) using Hach Nitratax plus sc probes (Hach, Loveland, CO) and aggregated to daily average NO3concentration for this study.
To analyze land use characteristics for each watershed, we downloaded land use data from the National Landcover Database 2016 at a 30 m x 30 m resolution (Dewitz, 2019). Land use data were binned into four categories; 135 water/wetlands, developed, forested/barren/shrubs, and crops (including pasture). Data for landforms, drainage infrastructure, and stream network were downloaded from the Iowa Department of Natural Resources. We downloaded daily precipitation data for the four-year period of analysis (2016-2019) for two sites (USC00137312 and USC00136566) within the Raccoon River watershed from the NOAA National Centers for Environmental Information. 140 https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License.

Event identification
We separated the discharge time series into baseflow and stormflow periods through semi-automating storm event identification using the following criteria: 1) dQ/dt ≥ 1e-4 cfs/second for the rising limb of the event, 2) max(Qevent) ≥ 0.01*max(Qrecord), and 3) the event duration ≥ 3 days. The end of each event was determined when either the event falling limb dQ/dt ≥ 0 or discharge returned to pre-event levels. For some, such as events that showed up as 145 shoulder peaks on larger events, or those with indistinct peaks, visual inspection and subjective decisions were required ( Figure S1). The criteria were derived from similar studies (Dupas et al., 2016;Knapp et al., 2020;Rozemeijer et al., 2010), and exact thresholds for the criteria were tuned and adapted for the structure and dynamics of the watersheds' hydrographs to ensure the selection of peaks. Time periods identified as storm events were classified as stormflow, and all other times were classified as baseflow ( Figure 2). 150 We note that this classification scheme differs from traditional baseflow separation techniques that use graphical, geochemical or isotopic approaches to identify and separate the proportion of the hydrograph that is comprised by baseflow and stormflow (Hooper & Shoemaker, 1986;Klaus & McDonnell, 2013). Baseflow separation techniques have shown that a large fraction of event water is derived from baseflow (e.g. Schilling & Zhang, 2004).
Our goal is not to contradict or supplant this finding, but rather to illustrate how a simple partitioning of the hydrograph 155 based on peaks in discharge allows us to isolate nutrient export dynamics in specific hydrologic regimes.

Characterizing export regimes
Export patterns (chemostatic, dilution, or enrichment) were calculated for stormflow, baseflow, and the full record (herein referred to as stormflow+baseflow) for the full period of analysis and on a seasonal basis. 160 Concentration-discharge relationships for baseflow and stormflow+baseflow periods were calculated by aggregating data for the time period of interest. Stormflow c-Q relationships were calculated in two ways; first by aggregating data from all stormflow events over the time period of interest, and second, by calculating c-Q relationships for each individual storm event and averaging those values over all events ( Figure S4). The former is referred to as bulk stormflow c-Q relationships, and the later as event-averaged c-Q relationships. 165 Seasonal and annual calculations were made based on the water year which begins on October 1 st , and the year was divided seasonally into fall (October, November, December), winter (January, February, March), spring (April, May, June), and summer (July, August, September).

Load estimations 170
Cumulative NO3load estimates were calculated for each hydrologic regime (stormflow, baseflow, stormflow+baseflow) on an annual and seasonal basis as:

175
where ci and Qi are the daily NO3concentration and discharge values, and f is the fraction of data coverage for the period of interest. If data were missing during a period, baseflow and stormflow loads were calculated based on their https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License.
fractional contribution during the periods with data. All annual periods had f > 0.75, but some seasonal periods had low coverage, for seasonal periods where f ≤ 0.75 no load estimate was calculated.

180
3 Results and discussion

Stream flow exhibits strong seasonality
In all five watersheds, 44-52% of the analysis period was classified as stormflow, with an average of 15 unique storm events in each watershed per year (Table 1). While the proportion of stormflow periods was similar between watersheds, the fraction of flow that was partitioned into stormflow and baseflow varied considerably 185 between watersheds. MJF and USC had the highest proportion of stormflow, with 77.0 and 73.4% of annual flow classified as stormflow, respectively, compared to 62.4 and 63.9% in UPN and MRF, respectively (Table 1). This observation is consistent with the higher density of drainage infrastructure (e.g. canals, tile drainage) in MJF and USC, leading to quicker routing of high flows to the stream channel compared to more natural drainage networks in UPN and MRF. 190 Flow in all watersheds exhibited strong seasonality, with an average of 42.9% of total flow delivered in the spring. Summer months contributed the least to overall discharge with an average of 17.3% across all watersheds.
Despite differences in overall flow between the seasons, spring and summer experienced a similar number of stormflow events across all watersheds (average of 5.5 in spring and 4.4 in summer; Student's t-test; p > 0.01), and similar precipitation totals (average 309 mm in spring and 381 mm in summer; Student's t-test; p > 0.01). Increased 195 streamflow in the spring months is likely a result of snow melt, rain on snow events, which can produce excess runoff, and increased crop growth in the summer months leading to more water retention.

NO3concentrations are sensitive to watershed characteristics, season, and hydrologic regime
The outlet of the largest watershed (DVM) showed median NO3concentrations of 7.38±3.07 mg/L. The 200 heavily tile-drained USC watershed showed the highest median NO3concentration (9.23±3.09 mg/L), while MRF, which has the least drainage infrastructure, showed the lowest (6.96±2.51 mg/L; Table S3). This is consistent with observations of increased stream NO3concentrations at the outlets of heavily tile-drained Iowa watersheds compared to those with less built drainage infrastructure (Schilling et al., 2012).
NO3concentrations exhibited pronounced seasonality in both stormflow and baseflow, with annual minima 205 during the summer and maxima during the spring (Figure 3). Summer baseflow NO3concentrations correlated well with watershed area as the outlet of the largest watershed experienced the lowest concentration ( Figure 3A). Low NO3concentrations in summer are often associated with lower flow periods which may have increased contributions from groundwater flow paths with longer residence times, and more streambed-water interaction, both positively associated with watershed area (Peralta-Tapia et al., 2015). In addition, summer periods have warmer temperatures, 210 which promote biological nitrogen uptake activity (e.g. denitrification and assimilation) that can lower NO3concentrations (Moatar et al., 2017;Rode et al., 2016). Weakened correlations between baseflow NO3concentrations and watershed area during the rest of the year suggest that other processes may be more effective at driving NO3concentrations at other times of the year. https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License.
Maximum NO3concentrations were observed in the spring during both baseflow and stormflow periods 215 ( Figure 3). During stormflow periods, NO3concentrations correlated positively with drainage infrastructure density during all seasons, but the correlation was strongest during the spring months when the NO3concentrations were highest ( Figure 3B). During spring precipitation events, water infiltrates rapidly through relatively bare soils, encountering accumulated nitrogen stocks in the shallow subsurface from previous years or early season fertilizer application and is routed off the landscape through tile drains Royer et al., 2006). High flow 220 periods can also reduce the ability of biological processes to alter NO3concentrations (Rodríguez-Blanco et al., 2015;Royer et al., 2006). This seasonality of NO3concentration has been previously observed in the Raccoon River watershed (K. Schilling & Zhang, 2004), as well as other agricultural catchments in the Midwest (Dupas et al., 2017;Van Meter et al., 2020;Pellerin et al., 2014).

Baseflow c-Q patterns reveal seasonally shifting nitrate processing and sources
Concentration-discharge relationships showed a difference between baseflow and bulk stormflow periods, with baseflow periods exhibiting generally more chemodynamic c-Q slopes (Figure 4). Enriching chemodynamic export patterns (c-Q slope > 0.2) were observed during baseflow periods in all watersheds annually, with UPN showing the strongest enrichment signal (c-Q slope = 0.79) and USC showing the weakest (c-Q slope = 0.21) ( Figure 4A). 230 Baseflow c-Q slopes were seasonally dynamic with the fall and summer experiencing generally higher c-Q slopes (blue and red triangles, respectively; Figure 4A), and winter and spring c-Q slopes closer to zero (green and yellow triangles, respectively; Figure 4A). There is a negative correlation between seasonal baseflow c-Q slope and drainage infrastructure density, which is strongest during the spring months (R 2 = 0.85; Table S4). During these months, positive baseflow c-Q slopes are driven by low flow, low NO3concentration periods, which are less prevalent in the 235 watersheds with a higher density of drainage infrastructure (USC, MJF, and DVM). The lack of low NO3concentration periods in these watersheds results in chemostatic c-Q slopes as the built drainage infrastructure serves to homogenize baseflow sources.
These human impacts can be highlighted by comparing the two end member watersheds in our dataset. MRF, which has the lowest density of drainage infrastructure (0.37 km/km 2 ), experienced chemodynamic enriching c-Q 240 slopes across all seasons during baseflow, ranging from 0.34 in the winter to 0.75 in the summer. This suggests highly heterogenous source regions contributed to baseflow throughout the year. In contrast, USC, which has the highest density of drainage infrastructure (1.11 km/km 2 ) experienced chemodynamic conditions only in the summer (c-Q slope = 0.29), and chemostatic conditions across the other seasons. This suggests there were consistent, homogeneous sources producing stable NO3concentrations across a range of flow conditions throughout the year. 245 The strongest chemodynamic enrichment patterns occurred in the summer across all watersheds, while the most chemostatic season was generally the spring (Figure 4). This pattern is exemplified in DVM, which integrates the signal from the other four upstream watersheds ( Figure 2E). The summer baseflow period in DVM is strongly enriching (c-Q slope = 0.75), while in spring, baseflow is chemostatic (c-Q slope = 0.08). This dynamic shift is driven by differences in baseflow NO3concentrations from spring to summer, suggesting differences in the sourcing or 250 internal processing of baseflow from one season to the next (Richardson et al., 2020). https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License.

Stormflow c-Q patterns show stationarity in seasonal NO3sources
Bulk stormflow periods generally exhibited more chemostatic behavior than baseflow periods (Figure 4).
The observation that low flow periods were more chemodynamic than high flow periods is consistent with other 255 studies that have partitioned the hydrograph seasonally (Ehrhardt et al., 2019), by breakpoint analysis (Marinos et al., 2020), or by median discharge (Moatar et al., 2017), suggesting that this is a general feature of watershed hydrologic routing. Bulk stormflow c-Q slope exhibited subtle seasonality with a slight dilution trend in winter c-Q slopes in several watersheds, and a slight enrichment trend in spring and summer ( Figure 4B). Fall bulk stormflow c-Q slopes were chemostatic to weakly chemodynamic for all watersheds except UPN, which showed a c-Q slope of 0.71. This 260 higher c-Q slope was driven by two anomalous, low NO3concentration events discussed in further detail in Section 3.5.
Although tile drained watersheds show higher stormflow NO3concentrations ( Figure 3B), there does not appear to be a systematic effect on stormflow c-Q slopes ( Figure 4B). This indicates that the nitrate sources activated during stormflow periods are transport-limited across all watersheds. That is, regardless of season, storms contribute 265 flow to streams generally through shallow, quick flow paths that intersect high-NO3stores in these agriculturally intensive landscapes (Buda & DeWalle, 2009;Mellander et al., 2012).
Analysis of individual storm events reveals that event-averaged c-Q slopes form a narrow distribution around zero across all seasons ( Figure 5A). Although many individual events could be classified as strongly chemodynamic if considered in isolation, examining the events in aggregate shows that there is a tendency towards chemostatic 270 behavior across all watersheds ( Figure 5B and S4). The comparison of bulk stormflow c-Q slopes ( Figure 4B) and event averaged c-Q slopes ( Figure 5B) highlights the importance of c-Q event analysis at multiple temporal scales. If, for example, three events each showed a chemostatic response but at different NO3concentration, they could be interpreted as chemodynamic when grouped together. Both methods of analysis could be useful in determining the nutrient export behavior of stormflow events which has been observed to be highly non-linear and hysteretic (Carey 275 et al., 2014;Lloyd et al., 2016).

Periods of anomalous flow and NO3concentrations can alter overarching riverine c-Q characteristics
During baseflow and stormflow periods, episodes of anomalous flow and NO3concentrations had a significant effect on c-Q slope analysis. In UPN, two events, during low flow periods in October 2017, had low NO3 -280 concentrations (average 1.34 and 0.46 mg/L; 4 th and 2nd NO3concentration percentile across the whole study period, respectively). Individually, the events had c-Q slopes of -0.50 and 0.45. Inclusion of these events in the calculation of fall bulk stormflow c-Q behavior resulted in fall bulk stormflow c-Q slope of 0.71 ( Figure 4B). However, with the removal of these events, the same calculation yields a slope of 0.09, much more in line with the other watersheds for the fall season. These events were included in our analysis, as they met the criteria for event selection, however their 285 ability to skew the bulk analysis is notable as they represent < 1% of annual flow and NO3load.
Similarly, during baseflow in MJF, a period of anomalously low flow (mean = 96 cfs; < 0.1 flow percentile) and low nitrate concentration (mean = 0.05 mg/L; < 0.1 NO3concentration percentile) from 07/26/2017-10/19/2017 https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License. had a dramatic impact on the baseflow c-Q relationship ( Figure S3D). Inclusion of the data from this period resulted in an annual baseflow c-Q slope of 1.42, indicating very strong enrichment behavior. Removal of the data from this 290 anomalous period decreased the slope to 0.42. Data from this time period may be influenced by biofouling, as such we do not include this period in further discussion of nutrient export behavior, but we do include it in our estimates of annual and seasonal nitrate load, though it has little effect on our overall load estimates as the amount of nutrient export during this period is low.
The ability of a single anomalous period to influence the overall characterization of a hydrologic system 295 highlights the difficulty of representing nutrient export behavior based on a single parameter fit across several seasons and flow regimes (Diamond & Cohen, 2018;Dupas et al., 2017;Marinos et al., 2020). This also highlights the need for high-frequency data collection activities that allow researchers and water quality practitioners to observe anomalous events during periods of the year that are not traditionally targeted by discrete or synoptic sampling campaigns. 300

Seasonal patterns in nitrate load across watersheds
Annual average NO3export across the study watersheds ranged from 4216±768 kg-N/km 2 /yr in USC to 2222±371 kg-N/m 2 /yr in MRF. Partitioning the hydrograph into seasonal stormflow and baseflow periods allows the identification of periods which contribute disproportionately to annual watershed NO3export magnitudes ( Figure 6). 305 Spring stormflow periods accounted for the largest contribution to annual load across all watersheds, with an average of 37.5±11.5% for all years. Spring stormflow contributions displayed a large spatiotemporal range, from 19.7% in UPN in 2016 to 59.8% in DVM in 2017. Summer stormflow loads also showed considerable variation, with an average contribution of 9.4% of annual load, but ranging from < 1% (19.4 kg-N/km 2 /yr) in 2017 to 18.3% (711 kg-N/km 2 /yr) in 2018. 310 These ranges in NO3loads are largely driven by observed variation in summer stormflow events. For example, in the summer of 2017, which had an anomalously low NO3load, there were fewer stormflow events than average. Specifically, there was an average of 1.8 events across the watersheds with zero events identified in USC and MJF. In contrast, there was an average of 6.0 events across all five watersheds in summer 2018, which has anomalously high nitrate load. Additionally, the identified events in summer 2017 were approximately 22% the size 315 of the events in summer 2018. This variability highlights the difficulty in predicting loads across seasons, hydrologic regimes and watersheds.
Baseflow loads showed considerable variability seasonally, although they consistently made up ≤ 15% of the annual load in each watershed. Baseflow loads typically peaked in the spring months, likely due to a seasonally high water table, which increased shallow groundwater contribution to streams (Jiang et al., 2010;Molenat et al., 2008). 320 Additionally, spring fertilizer application and plowing can increase surface leaching, increasing the nitrate pool in the shallow subsurface (Royer et al., 2006). That said, there were some discrepancies within individual watersheds; UPN had generally higher export in fall baseflow and MRF had similar fall and spring loads ( Figure 6A). 325 https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License.

Nutrient export is driven by the spatial distribution of land use types and hydrologic infrastructure
There is a systematic trend toward higher NO3load in watersheds with a higher density of built drainage infrastructure (Figure 7), consistent with other studies (Basu et al., 2010;Musolff et al., 2015;Schilling & Zhang, 2004). The slope of the relationship between NO3load and drainage infrastructure density is much shallower for baseflow than for stormflow, given the greater range in observed stormflow load across the watersheds (Kennedy et 330 al., 2012). Drainage structures and tile drains route water from high NO3source areas directly to the stream, decreasing travel time and bypassing riparian areas that are highly active in nutrient processing (Dosskey et al., 2010).
These structures are common features in agricultural landscapes and show strong correlation to the amount of cropped area across the five watersheds analyzed (R 2 = 0.95).
The short circuiting of subsurface flow paths and increased cropped area drives watershed nutrient export 335 patterns towards chemostatic behavior by homogenizing the source regions and limiting nutrient cycling during transport (Marinos et al., 2020;Musolff et al., 2015;Thompson et al., 2011). These patterns are most clear during both baseflow and stormflow periods in the spring months, when tile drains likely have their greatest influence on hydrologic routing. During these periods, stormflow NO3loads are strongly correlated with drainage infrastructure density (R 2 = 0.88 and 0.88, respectively; Table S4) and stormflow export regimes are chemostatic (average c-Q slope 340 = 0.15 for stormflow and 0.18 for baseflow).
In contrast, summer baseflow periods showed the strongest chemodynamic enrichment patterns with an average c-Q slope of 0.73 across all watersheds. The NO3load during these periods is most strongly correlated with the percentage of cropped area within 100 m of the stream (R 2 = 0.94; Table S4). This suggests that summer chemodynamic regimes are driven by low flow, low NO3periods where source areas that are proximal to the stream 345 are contributing more significantly to discharge (Molenat et al., 2008). Lower density of agricultural activity in riparian areas (Table S2) leads to more heterogeneous source regions, which promotes low NO3load and the observed chemodynamic behavior.
Seasonal and annual c-Q slopes across all hydrologic regimes show only weak correlations with watershed area suggesting that drainage infrastructure and the distribution and intensity of agriculture are the dominant drivers 350 of NO3export regime in these watersheds. This is consistent with a recent study of 33 agricultural watersheds in the Midwest (Marinos et al., 2020). Our results show that both conditions that lead to high NO3loads, whether hydrologic (i.e. stormflow) or landscape (i.e. increases in drainage infrastructure and agricultural intensity) are associated with chemostatic behavior. This trend is in line with the idea that landscapes with such agricultural intensity are a saturated solute source, whose delivery is flow-limited (Thompson et al., 2011). 355

Conclusions
Detailed analysis of event, seasonal, and annual NO3export showed that all five heavily agricultural watersheds showed similar temporal patterns of NO3load with highs in spring stormflow and lows in summer baseflow. Stormflow across all seasons was largely chemostatic and spring stormflow accounted for ~40% of annual 360 loads. In contrast, baseflow periods exhibited seasonality in export regimes, with low summer flows driving periods of chemodynamic enrichment and winter and spring driving more chemostatic behavior in the winter and spring. The https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License. differences in c-Q behavior between stormflow and baseflow suggests that the systems dynamically, but predictably, shift between NO3export patterns in response to hydrologic forcing. There was a systematic trend toward more chemostatic behavior and higher NO3loads with increasing density of drainage infrastructure and agricultural land 365 use across the five watersheds. These anthropogenic controls on NO3export in these watersheds are driven by disparate glacial histories across the watersheds that necessitate different flow routing infrastructure. During baseflow conditions, land use near the stream has a large impact on NO3loads, indicating that buffer strips or other near-stream management practices may be effective management practices for reducing loads during these periods.
Analysis of specific low-flow periods demonstrated that anomalous periods have the power to significantly 370 affect our classification of export patterns and influence our understanding of watersheds as a whole. This highlights the dynamic nature of these systems and argues for event, seasonal, and longer-term analyses of nutrient export, particularly when attempting to measure the efficacy of management practices such as reductions in fertilizer application or near-stream buffer strips. High-resolution hydrochemical observations allow the detailed characterization of storm events which facilitate more accurate estimates of NO3loads that have been previously 375 measured using regression-based techniques with sparse sample resolution. This study demonstrates the utility of high spatial and temporal resolution water quality sampling to disentangle the key factors controlling watershed nutrient export as well as the important role of state and federal water quality monitoring programs in addressing important water quality issues.  https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License.  https://doi.org/10.5194/hess-2020-562 Preprint. Discussion started: 5 November 2020 c Author(s) 2020. CC BY 4.0 License.