Antecedent conditions control thresholds of tile-runoff generation and 1 nitrogen export in intensively managed landscapes

33 Threshold changes in rainfall-runoff generation commonly represent shifts in runoff mechanisms 34 and hydrologic connectivity controlling water and solute transport and transformation. In 35 watersheds with limited human influence, threshold runoff responses reflect interaction between 36 precipitation event and antecedent soil moisture. Similar analyses are lacking in intensively 37 managed landscapes where installation of subsurface drainage tiles has altered connectivity 38 between the land surface, groundwater, and streams, and where application of fertilizer has 39 created significant stores of subsurface nitrogen. In this study, we identify threshold patterns of 40 tile-runoff generation for a drained agricultural field in Illinois and evaluate how antecedent 41 conditions—including shallow soil moisture, groundwater table depth, and the presence or 42 absence of crops—control tile response. We relate tile-runoff thresholds to patterns of event 43 nitrate load observed across multiple storm events and evaluate how antecedent conditions 44 control within-event nitrate concentration-discharge relationships. Our results demonstrate that 45 an event tile-runoff threshold emerges relative to the sum of gross precipitation and indices of 46 antecedent shallow soil moisture and antecedent below-tile groundwater moisture deficit, 47 indicating that both shallow soil and below-tile storages must be filled to generate significant 48 runoff. In turn, event nitrate load shows a linear dependence on runoff for most time periods, 49 suggesting that subsurface nitrate export and storage can be estimated using runoff threshold 50 relationships and long-term average nitrate concentrations. Finally, within-event nitrate 51 concentration-discharge relationships are controlled by event size and the antecedent tile flow 52 state because these factors dictate the sequence of flow path activation and tile connectivity over 53 a storm event. 54 55 Plain language summary: 56 Improving nutrient management in intensively managed landscapes requires understanding of 57 how human alterations for agriculture have influenced nutrient transport from and storage within 58 the landscape. In addition to creating large subsurface stores of nitrogen through fertilizer 59 application, humans have altered the drainage structure of intensively managed landscapes by 60 installing subsurface drainage (commonly ‘tiles’) to maintain optimal moisture conditions for 61 crops. Although highly engineered systems, it is unclear how tiles influence the timing and 62 magnitude of water and nutrient export from these landscape. Here, we identify how pre-storm 63 wetness conditions control rapid, nonlinear changes in tile flow (thresholds). We find that a tile 64 flow initiation threshold results from the sequential filling of first a depleted shallow soil storage 65 and then deeper below-tile groundwater storage. Further, nitrate export reflects tile-runoff 66 thresholds, indicating that the factors controlling tile-runoff are also primary controls on tile 67 nitrate export. 68 69

indicating that both shallow soil and below-tile storages must be filled to generate significant 48 runoff. In turn, event nitrate load shows a linear dependence on runoff for most time periods, 49 suggesting that subsurface nitrate export and storage can be estimated using runoff threshold 50 relationships and long-term average nitrate concentrations. Finally, within-event nitrate 51 concentration-discharge relationships are controlled by event size and the antecedent tile flow 52 state because these factors dictate the sequence of flow path activation and tile connectivity over 53 a storm event. 54 55 Plain language summary: 56 Improving nutrient management in intensively managed landscapes requires understanding of 57 how human alterations for agriculture have influenced nutrient transport from and storage within 58 the landscape. In addition to creating large subsurface stores of nitrogen through fertilizer 59 application, humans have altered the drainage structure of intensively managed landscapes by 60 installing subsurface drainage (commonly 'tiles') to maintain optimal moisture conditions for 61 crops. Although highly engineered systems, it is unclear how tiles influence the timing and 62 magnitude of water and nutrient export from these landscape. Here, we identify how pre-storm temperature of 23°C). Monticello receives an annual average precipitation of 1020 mm. 188 Thunderstorms account for 50-60% of annual precipitation (Angel, 2003)  European settlement, the USRB was 90% prairie and 10% forest (IDNR, 1999), with forested 201 portions mainly located in riparian zones. Today 90% of land use in the watershed is row crop 202 agriculture, primarily corn and soybeans, and the majority of cropland is tile-drained. Wetlands 203 historically covered about 40-50% of the land area but now make up less than 2% (IDNR, 1999;204 Rhoads et al., 2016), primarily due to the installation of tile drains and ditches which has 205 artificially lowered the water table. Subsurface flows rather than direct surface runoff are the 206 primary pathway by which water and nutrients enter surface waters in the USRB (Demissie et 207 al., 1996), and subsurface flows are mainly conveyed by tile drains (Botero-Acosta et al., 2018). 208

209
The farm is about 60 ha total, but the monitored tile network drains an estimated 10 ha based on 210 analysis of aerial photography (Kratt et al., 2020). The drainage network consists of five 211 individual 10-cm diameter perforated pipes, each about 400 m long and spaced 30 m apart, 212 draining into a 10-cm diameter main that empties into a surface drainage ditch. The tiles are 213 about 1-1.2 m below the land surface. The field is not irrigated, so the only water input is 214 precipitation. An annual crop rotation of corn-soybean, a common practice in the Midwestern 215 U.S., is used. During the study period, corn was planted in 2016, 2018, and 2020, and soybean 216 was planted in 2017 and 2019. Prior to the monitoring period, anhydrous ammonia was applied 217 in the fall of 2015 (Table 1). During the monitoring period, 32% urea and ammonium nitrate 218 solution (UAN) was applied in the spring when corn was planted. In spring 2016, 2018, and 219 2020, 32% UAN was applied as an herbicide carrier in April prior to crop planting. In spring 220 2018 and 2020, 32% UAN was side-dressed in May after emergence. Each spring, the field was 221 cultivated. During the fall after corn was planted, the field was chisel-plowed to cut and 222 incorporate stalk residue into the soil to preserve soil organic matter and protect against erosion.

Modeling Methods 248
To supplement field observations with a mechanistic simulation, we used the coupled surface-249 subsurface flow and soil-vegetation-atmosphere interaction model Dhara (Le and Kumar, 2017;250 Woo and Kumar, 2019) to simulate the hydrologic and biogeochemical dynamics of a parcel of 251 tile-drained land. The model is used here as a heuristic, providing a basis for interpretation of the 252 processes that are likely to underlie our field observations. The model was previously calibrated 253 and validated for a tile-drained site in DeLand, Illinois that has similar soils, topography, and 254 drainage infrastructure to Allerton Trust Farm and is also located within the USRB. A corn-255 soybean crop rotation is used at the site, and this rotation is also employed in model simulations. 256 Compared to observed tile flow, simulated flow was muted during high peak flows, which also 257 affected the accuracy of nitrogen loads at high flow. However, tile flow and nitrate loads 258 captured the patterns of the observed data well overall, providing confidence in the use of the 259 simulation results for process investigation. Refer to Woo and Kumar (2019) for a detailed 260 description of the parameters and equations governing the model. A schematic diagram of Dhara 261 is provided in Supplemental Information ( Figure S1). A small number of alterations were made 262 to the model application of Woo and Kumar (2019)

Storm event selection and hydrograph separation 278
In order to identify relationships between event tile-runoff and antecedent catchment wetness, we 279 first defined a procedure for selecting the tile-runoff volume, gross precipitation, and antecedent 280 moisture conditions associated with discrete storm events. Event selection followed one of two 281 methods, depending on whether precipitation resulted in tile flow, and the same procedures were 282 used for field and model data. If precipitation initiated a tile response, storm event runoff 283 included the period between an initial increase in discharge until either discharge returned to 284 approximately the initial value or increased in response to a different storm. Compound storm 285 events (i.e., those with significant hydrograph overlap between multiple events) were omitted 286 from the runoff threshold analyses. However, compound events were included in the within-287 event nitrate c-Q analyses, in which we investigated the influence of antecedent tile flow state on 288 event-scale concentration dynamics and flow paths. Events in which snowmelt was expected to 289 contribute to stormflow were also omitted in runoff threshold analyses due to uncertainties in the 290 amount and timing of inputs. While tile flow at the site mainly consisted of stormflow, tiles 291 contributed some baseflow to the drainage ditch during wetter periods. As such, stormflow 292 volumes were determined using the constant slope hydrograph separation method (Hewlett and 293 Hibbert, 1967). For storms that resulted in a tile response, gross event precipitation was defined 294 as the total precipitation that occurred up to one day prior to the initial tile storm response until 295 the end of the tile storm response. For storms that did not initiate tile flow, gross event 296 precipitation was calculated as total precipitation that occurred over a day or over consecutive 297 days with precipitation. Gross precipitation over the considered time period had to exceed 1mm 298 to be included in the analysis. Soil moisture values immediately preceding the considered 299 precipitation time period were used to determine an antecedent soil moisture index (ASI), 300 calculated as the total soil water content within the surface soil layer expressed as depth (mm). 301 For this study, we consider the surface soil layer to be 0-0.3 m depth as an indicator of 302 antecedent soil moisture conditions largely independent of groundwater dynamics. ASI is 303 calculated as: 304 (1) 305 where VWCi is the volumetric water content (mm/mm) in the i th sublayer, and D is the layer 308 thickness (mm). We used n = 2 sublayers, with the VWC for 0-5 cm soil depth estimated from 309 the sensor at 5cm depth and VWC for 5-30 cm soil depth estimated from the sensor at 20 cm 310 depth. 311 312 313

Tile-runoff relationships: calculations and data analysis
We analyzed relationships between storm event tile-runoff and wetness metrics to identify how 315 antecedent wetness controls event tile-runoff. For field data, analyzed wetness metrics included 316 gross precipitation (Pgross), ASI, and the sum of gross precipitation and ASI (Pgross + ASI). Model 317 analysis included an additional metric of a below-tile groundwater moisture deficit (GWdef), 318 calculated as the depth-equivalent unsaturated pore volume below the tile (mm). In other words, 319 GWdef represents the depth of water needed to raise the water table to the tile elevation and is 320 calculated as: 321 (2) 322 where VWCi is the modeled volumetric water content (mm/mm) of the i th layer beneath the tile, 325 VWCS is the volumetric water content of the soil at saturation (0.56 mm/mm), and D is the layer 326 thickness (100 mm). GWdef has a negative value and decreases the overall wetness metric 327 because it indicates a lack of moisture that must be overcome to initiate tile-runoff. 328

329
In the absence of field observed below-tile moisture data to explore the effect of bottom-up 330 controls, the antecedent groundwater table position was inferred from tile flow conditions and 331 gross precipitation over the days leading up to the event. Similarly, previous investigations of 332 nonlinear rainfall-runoff response have used proxies for inferring antecedent water storage when 333 soil moisture observations were unavailable, including the duration of inter-storm dry periods 334 (Graham and McDonnell, 2010) and the amount of water input required for runoff to initiate (Ali 335 et al., 2015). Here, storm events were categorized as "GWdef low," indicating that the 336 groundwater table was expected to be near the tile elevation such that the antecedent below-tile 337 moisture deficit was near zero, if antecedent conditions met either of the following criteria: gross 338 precipitation for the day prior to the event (i.e., 2 days prior to initial tile storm response) 339 exceeded 20 mm or tile flow volume during the 6 days prior to the event exceeded 10 m 3 . This 340 procedure was implemented to exclude events in which the antecedent groundwater deficit was 341 high but direct percolation to the tile resulted in a small amount of tile flow. If an event did not 342 meet the above criteria, it was categorized as "GWdef high," and the groundwater table was 343 expected to be significantly lower than the tile such that antecedent below-tile moisture deficit was high. We also investigated how the presence or absence of crops affects event tile-runoff by 345 categorizing storm events as occurring either during the growing season or during the non-346 growing season. Growing season events occurred when crops were present and water uptake was 347 largest, during the months of June, July, August, or September. We expected that large seasonal 348 fluctuations in water uptake and interception of precipitation in IMLs due to presence or absence 349 of crops could pose an additional top-down moisture control on tile-runoff generation. During 350 the growing season, a larger Pgross + ASI value would be needed to initiate tile flow due to 351 greater water uptake and interception by crops. Therefore, the runoff initiation threshold relative 352 to Pgross + ASI would be larger than during the non-growing season. . We expected that these time periods could 374 reveal differences in nitrate concentration resulting from yearly/seasonal management decisions (e.g., fertilization, crop type). All statistical analyses were conducted in MATLAB, and we use a 376 significance threshold of 0.05. We performed the ANCOVA using the anovan function, 377 including discharge as a covariate. This procedure enabled analysis of differences between time 378 periods after the effects of discharge were removed. We followed with a Tukey post hoc test 379 using the multcompare function to analyze the main effect of time period. We explored the effect 380 of heteroscedasticity and deviations from normality by performing statistical analyses on log10-381 transformed data and found no change in results (Table S1). As such, we report results of 382 analyses performed on the non-transformed data. Based on the Tukey post-hoc test, we grouped 383 time periods and performed linear regressions to determine the relationship between discharge 384 and nitrate concentration. Slopes not significantly different from zero would support chemostasis 385 over those time periods. Linear regressions were fit to total event nitrate load and event tile-386 runoff based on these groupings. We also performed linear regressions on modelled nitrate loads 387 and event tile-runoff for comparison with field data.

Controls of antecedent conditions on tile-runoff: field data
Time series of tile discharge, shallow soil moisture, and precipitation data were used to 434 investigate how antecedent conditions control event tile-runoff. A total of 157 storm events were 435 analyzed, 45 of which resulted in tile-runoff. We found that event runoff depth correlated with 436 gross precipitation (r 2 = 0.28, Figure 2b) but not antecedent soil moisture (Figure 2a). When 437 gross precipitation and antecedent soil moisture were summed (Pgross + ASI), a threshold 438 relationship emerged, and the above-threshold correlation was larger (r 2 = 0.39, Figure 2c We performed additional analyses to explore whether antecedent below-tile moisture deficit and 443 the presence of crops pose additional controls on tile-runoff response. If below-tile moisture 444 deficit was an important control, we expected that GWdef low events would have a strong linear 445 correlation above the Pgross + ASI threshold, but GWdef high events would be overestimated by 446 the above-threshold trendline for GWdef low events. Overall, we found this to be the case: GWdef 447 low events showed a strong correlation above the Pgross + ASI threshold (r 2 = 0.79, Figure 3a), 448 whereas GWdef high events showed more spread (r 2 = 0.13) and tended to be overestimated by 449 the GWdef low trendline. These data indicate that information on available below-tile storage is 450 needed to predict storm event tile-runoff. We also expected that the presence of annual crops 451 would pose an additional control on event tile-runoff. However, both growing season and non-452 growing season data showed considerable spread around the trend line (Figure 3b, "no crops" r 2 = 0.42 and "crops" r 2 = 0.08). The presence or absence of crops does not contribute additional 454 information to ASI in explaining tile-runoff response. 455

Controls of antecedent conditions on tile-runoff: model data 457
In addition to field observations, hydrologic simulations of a tile-drained agricultural site 458 provided 20 years of tile hydrologic response and additional below-tile soil moisture information 459 to investigate how antecedent conditions control tile-runoff. We found that event runoff depth 460 correlated with gross precipitation (r 2 = 0.82) but not ASI or GWdef alone (Figure 4a, b, d). A 461 threshold relationship emerged relative to Pgross + ASI, with an above-threshold correlation of r 2 462 = 0.85 (Figure 4c). Similar to field data, the above-threshold correlation for GWdef low events 463 improved relative to all data (GWdef low r 2 = 0.94 and all data r 2 = 0.85; Figure S2a). On average, 464 the GWdef low linear trend overestimated runoff for GWdef high events. We expected that adding 465 the numeric below-tile groundwater moisture deficit (GWdef) to the catchment wetness metric would result in a clearer threshold trend with event tile-runoff. Indeed, we found that the runoff 467 relationship with Pgross + ASI + GWdef increased the above-threshold correlation (r 2 = 0.90) 468 relative to Pgross + ASI. The above-threshold correlation relative to Pgross + GWdef was similar to 469 Pgross + ASI (r 2 = 0.85). Thus, considering either GWdef or ASI improves our ability to predict 470 event tile-runoff using a threshold relationship. However, the strongest above-threshold trend 471 emerges relative to an antecedent wetness metric which includes both ASI and GWdef, indicating 472 that both are strong controls on tile-runoff initiation. 473  Figure S3). ANCOVA results showed that there is a highly significant interaction between discharge and seasonal time period on nitrate concentration at the 95% 478 confidence interval, F(4,782) = 6.0, p < .001 ( we fit a linear regression through all data excluding Y1 Corn Spring/Summer, which was fit with 486 a separate regression line (Figure 5a). The first trend line has an intercept of 9.3 ppm and small 487 slope (m = -0.001), which is not meaningfully different than zero and indicates a chemostastic 488 response at the interannual timescale. The fit through Y1 Corn Spring/Summer has a higher 489

Controls of antecedent conditions on nitrate concentration dynamics within events 512
Of the 18 events analyzed for c-Q relationships, 50% exhibited counterclockwise hysteresis, 17% 513 exhibited clockwise hysteresis, and 33% were non-hysteretic ( Figure 6). We did not observe a 514 clear control of ASI or GWdef on HI, as would have been exhibited by a trend between HI and 515 ASI or GWdef. However, hysteretic behavior grouped by runoff event size and antecedent tile 516 flow state (Figure 7). Larger events (> 150 m 3 d -1 peak tile flow) which occurred when there was 517 little to no tile flow at the onset exhibited strong counterclockwise hysteresis (events 3, 4, 11, 23 518 in Figures 6 and 7). Small events (< 150 m 3 d -1 peak tile flow) tended to exhibit weak 519 counterclockwise hysteresis to non-hysteretic behavior (events 6, 7, 9, 10, 14 -17 in Figures 6  520   and 7). Larger events which occurred when the tile was still flowing from a previous event (i.e., 521 a storm occurred on the falling limb of another event) exhibited weak counterclockwise 522 hysteresis to non-hysteretic behavior (events 1, 2, 5, 8, 12, 18 in Figures 6 and 7). 523 In our empirical and modeling studies, we find evidence for both top-down and bottom-up tile-548 runoff generation mechanisms. Our analysis of field-observed tile discharge, shallow soil 549 moisture, and precipitation, in conjunction with modeled output including below-tile soil 550 moisture, demonstrates that tile-runoff at the study site is a function of gross precipitation and 551 both below-and above-tile storage controls. Tile-runoff response displays a threshold behavior 552 similar to that observed in forested hillslopes, whereby runoff increases linearly with increasing 553 Pgross + ASI after a threshold value is exceeded; prior to the threshold, little runoff is produced in 554 response to rainfall, resulting in an overall relationship reminiscent of a hockey stick shape. 555 However, the above-threshold correlation is not as strong as has been observed in some forested 556 catchments (Detty and McGuire, 2010;Farrick and Branfireun, 2014). This is potentially due to 557 variation in the observational data set (e.g., number of storm events, available sensor data), 558 intrinsic properties of the system, or the ability of the analysis to capture all relevant storage and 559 runoff generation mechanisms in the tile-drained landscape. We also find that a similar tile-560 runoff threshold emerges relative to Pgross + GWdef. Moreover, including both ASI and  IMLs the water table  579 boundary defines available bottom-up storage and varies temporally. Although the landscape 580 structure and associated runoff generation mechanisms of low-gradient, tile-drained IMLs differs 581 from that of steep, bedrock hillslopes, the conceptual filling and spilling of landscape storages 582 and resultant threshold runoff behavior are similar. Further, fill-and-spill was recently proposed 583 as a framework to more broadly describe runoff generation processes by which landscape 584 storages become progressively filled and connected (McDonnell et al., 2021). Another 585 comparable bottom-up mechanism explaining threshold runoff response in untiled, minimally 586 managed hillslopes is "transmissivity feedback" (Bishop, 1991;Kendall et al., 1999). Initially 587 observed in till soils, this describes the process by which rapid lateral flow occurs when the groundwater table rises and encounters surficial soil layers of increasing hydraulic conductivity, 589 often due to the presence of macropore networks. In intensively managed landscapes, the tile 590 elevation threshold controlling lateral subsurface water transmission is analogous to the 591 transmissivity feedback mechanism observed to generate nonlinear runoff response in some 592 forested catchments. Although the water table in tile-drained landscapes is typically constrained 593 too deep to encounter high conductivity shallow soil layers, tiles themselves impart a similar 594 threshold runoff response. 595 Figure 9. Conceptual tile-runoff generation model for a scenario in which both top-down and bottom-up moisture controls are present. Brown indicates the soil matrix, the white box a preferential flow path, gray a tile drain, and light blue groundwater. Red indicates soil water and dark blue event water. A soil moisture threshold in the shallow subsurface must be met prior to significant transport to greater depths. Initially, water infiltrates the soil matrix and macropores at the beginning of the event, and a small amount of event water reaches the tile drain via preferential flow paths. Once the soil moisture threshold is reached, soil matrix water is mobilized and enters preferential flow paths. If a groundwater deficit is present, below-tile storage must be filled to raise the water table to the elevation of the tile to generate significant runoff. Counterclockwise nitrate c-Q hysteresis, observed during large events with little/no antecedent tile flow, reflects a shift from dilute event water in early runoff to nitrate-laden pre-event water after a soil moisture threshold is exceeded.

596
In addition to analyzing how antecedent wetness controls tile-runoff response patterns, we 597 examined how distinct landcover regimes in IMLs influence runoff response. In agricultural landscapes dominated by annual crops, vegetation is typically present for periods that coincide 599 with the growing season, resulting in large seasonal fluctuations in evapotranspiration (Sacks and 600 Kucharik, 2011;Shaw, 1963). Therefore, we expected that vegetation could impart an additional 601 top-down control on subsurface runoff in IMLs via fluctuations in water uptake and interception, 602 with peak water use corresponding to critical crop growth stages (Al-Kaisi, 2000). Further, there 603 is evidence that in natural systems ecology and hydrology co-evolve in response to climate, 604 establishing equilibrium conditions between vegetation and water availability to avoid water 605 shortages (Eagleson, 1982;Gao et al., 2014;Troch et al., 2015). Due to these linkages between 606 vegetation and root zone soil moisture, soil moisture runoff thresholds may closely reflect 607 vegetation controls in minimally managed systems. In contrast, vegetation patterns in IMLs 608 reflect continuous human manipulation and could act as an independent control on runoff 609 patterns. However, we found no evidence that the presence or absence of crops contributes 610 additional information to ASI in explaining tile-runoff response. This suggests that the influence 611 of crop presence on tile-runoff thresholds is already reflected within the soil moisture metric. 612 Our field data analysis, though, is limited due to the small number of events which produced 613 large runoff during the growing season. In a study of forested headwater catchments at the 614 Coweeta Hydrologic Laboratory, Scaife and Band (2017) similarly found little evidence that the 615 Pgross + ASI runoff threshold value differed between the dormant and growing season. 616 Nonetheless, their data demonstrate that runoff thresholds vary interannually, largely due to 617 variation in runoff initiation thresholds between growing seasons, and they conclude that 618 interannual runoff thresholds are influenced by ecohydrologic feedbacks with forest 619 evapotranspiration rates. 620 overwrite dominant patterns during extreme periods, such as rain shortly after fertilizer 654 application, which is particularly troublesome given that most nitrogen mass is mobilized during 655 a relatively small number of these events (Royer et al., 2006). 656 657 In addition to controlling nitrate loading to downstream waterways, tile-runoff thresholds 658 modulate the accumulation of nitrate in groundwater. Tiles reduce recharge of high nitrate 659 concentration soil water to deeper groundwater by providing direct flow paths to streams that bypass deeper groundwater (Rodvang and Simpkins, 2001). While the mere presence of tiles is 661 expected to influence spatial variations in groundwater contamination across IMLs (Power and 662 Schepers, 1989), emergent runoff thresholds within drained landscapes reveal conditions leading 663 to nitrate storage versus export. For example, a below-threshold event which mobilizes soil water 664 and nitrate but does not raise the groundwater table to intersect the tile would primarily result in 665 storage of nitrate in groundwater. Conversely, an above-threshold event with low antecedent 666 groundwater deficit would result in greater nitrate export. Thus threshold relationships could 667 provide a tool for predicting both the storage and delivery of water and nitrate in IMLs. 668 consists of mainly soil water. While soil water that reaches the saturated zone likely mixes with a 686 small amount of older groundwater, we expect the shallow saturated zone is stratified (Fenelon 687 and Moore, 1998; Jiang and Somers, 2009) such that tile flow resembles recent soil water. 688 689 In our data, the transition from event to soil water is reflected by strong counterclockwise 690 hysteresis during large events which occurred when there was little to no tile flow at the event onset ( Figure 9). We expect that during small events, the threshold of soil water mobilization 692 was not reached so c-Q shows weak to no counterclockwise hysteresis ( Figure S5). Likewise, 693 large events that occur when the tile is already flowing (i.e., when the tile is initially connected to 694 the water table) do not reflect the transition from event to soil-derived water because tile water is 695 already composed of primarily pre-event water at the beginning of an event. Thus, tile flow 696 exhibits non-hysteretic behavior or weak clockwise hysteresis. Although the tight coupling 697 between tile flow and nutrient load observed in this study indicates that nitrate dynamics were 698 primarily transport-limited, the latter behavior may indicate nitrate source exhaustion when 699 consecutive storm events occurred. Further, while small events observed in this study tended to 700 occur when there was little to no tile flow, we expect that small events which occur when the tile 701 is flowing prior to the event would similarly exhibit weak to no hysteresis, following the same 702 rationale described above. 703

704
In addition to hysteretic behavior, we also analyzed nitrate flushing or dilution over the rising 705 limb. Although the majority of events had an overall flushing effect (FI > 0.1), rising limbs often 706 exhibited periods of both dilution and flushing. This is evident in the three events with strong 707 counterclockwise hysteresis (i.e., those capturing the transition from event to pre-event water) in 708 which high sampling resolution was achieved over the rising limb (events 4, 11, and 13; Figures  709   8 and S4). An initial period of nitrate dilution is followed by a period of flushing. The decrease 710 in nitrate concentration corresponds with an increase in soil moisture prior to both reaching an 711 inflection point. This relationship suggests that the source of tile drain water shifted once a water 712 storage threshold was exceeded, further supporting interpretation of counterclockwise hysteresis 713 as the result of a soil moisture mobilization threshold. For all events, the inflection point 714 occurred when shallow soil water content exceeded 31-32% soil water content. We expect that 715 this soil water content represents the threshold of soil water mobilization within soils at the site. 716 The initial decrease in nitrate concentrations may result from event water depleting nitrate stored 717 within preferential flow paths or on the soil surface. Another potential explanation for the initial 718 decrease in concentration is that water was transported faster than nitrate could be dissolved or 719 mobilized. After the soil moisture threshold is reached, soil matrix water and associated nitrate 720 mobilize, resulting in a rapid increase in nitrate. The threshold of soil water mobilization 721 occurred prior to peak tile discharge, 1-2 hours after the initial increase in tile discharge. 722 723 724

Conclusions 725
In this study, we investigated how antecedent conditions control thresholds of tile-runoff 726 generation and nitrate loads between events, as well as nitrate c-Q relationships within events. 727 First, we expected a tile-runoff threshold would emerge relative to the sum of gross precipitation 728 and an antecedent catchment wetness index reflecting either shallow soil moisture, indicating 729 top-down runoff generation, or below-tile groundwater moisture deficit, indicating bottom-up 730 runoff generation. Instead, we found that the most distinct runoff threshold and linear response 731 emerged as a combination of both top-down and bottom-up controls, quantified as the sum of 732 gross precipitation, antecedent soil moisture index (ASI), and below-tile groundwater moisture 733 deficit (GWdef). Moreover, our results demonstrate a simple additive effect of below-and above-734 tile storage in determining the threshold of tile-runoff initiation. 735 736 Next, we expected that event nitrate load would reflect runoff threshold relationships. We found 737 this to be the case for most of the study period, with the exception of a two-month period when 738 wet conditions directly followed fertilizer application and led to elevated nitrate export. 739 Therefore, although interactions between management and hydroclimatic variables can overwrite 740 dominant patterns, under most conditions export of accumulated nitrate is controlled by the same 741 factors controlling tile-runoff and can be accurately predicted using runoff threshold 742 relationships. Finally, we expected that antecedent wetness conditions would control within-743 event nitrate c-Q relationships. While we did not observe a clear control of ASI or GWdef on HI, 744 we found that hysteretic behavior grouped by antecedent tile flow state and runoff event size. 745 Our results suggest that these factors are the dominant controls on event-scale nitrate c-Q 746 because they determine the sequence of flow path activation and tile connectivity over a storm 747 event. Further, the relationship between nitrate concentration and soil water content timeseries 748 indicate a threshold of soil water mobilization, a key mechanism underpinning event-scale nitrate 749 dynamics.