Antecedent flow conditions and nitrate concentrations in the Mississippi River Basin

Both referees stated we needed to provide a more thorough explanation for why we chose to calculate the Qratio using the previous year’s flow. Referee 1 – “The rationale of the method is also missing. For example why is Q ratio calculated for the previous 364 days?...” Referee 2 – “No clear rationale is given why the authors chose one year to describe antecedent flow conditions (p. 11455). Was this decision based on testing conceptual considerations?...” Streamflow is known to vary by season at the eight sites in our study (for example, highest flows typically occur during the spring). Therefore, we used the previous year to calculate antecedent flow conditions primarily because using this method makes Qratio a random variable for which the distribution does not change with time-of-year. To further elaborate on our choice of the Qratio the following paragraph has been added to the methods section: “By using streamflow integrated over the year as a large-basin surrogate for the kinds of hydrologic storage and flux measures that might be used in small-basin-process models we are able to acquire a general measure of basin moisture that is likely related to other physical, chemical and biological processes in a basin that are affected by preceding moisture conditions. The choice of a one year averaging period used to compute the Qratio makes this antecedent flow measure independent of the time of year and season. It is possible that more complex statistical formulations with seasonal terms or an averaging period other than one year would have a stronger statistical association with nitrate anomalies, but model parsimony led us to commit to this simpler formulation. Using Qratio to describe antecedent flows characterizes hydrologic conditions broadly and allows for an initial examination of how nitrate concentration responds following a drought or high flow period. If significant relationships are documented, future studies can help better define the specific hydrologic processes that influence nitrate concentration during and after a drought or high flow period.”


Introduction
Many studies show that antecedent moisture conditions influence nutrient export from river basins (Burt and Worrall, 2009;Garrett, 2012;Macrae et al., 2010;Randall et al., 2003;Soulsby et al., 2003;Vecchia et al., 2008). Commonly, studies document increased nutrient export following a prolonged dry period (Foster and Walling, 1978;5 Macrae et al., 2010), though some studies have observed the opposite effect when considering only more recent antecedent conditions (Creed and Band, 1998;Macrae et al., 2010;Welsch et al., 2001). Most observations concerning the influence of antecedent moisture on nutrient export have been made in small basins with generally homogenous land use, land cover, climate, and geology (e.g., Biron et al., 1999;Burt and Worrall, 2009;Cooper et al., 2007;Foster and Walling, 1978;Lange and Haensler, 2012;Macrae et al., 2010;Welsch et al., 2001), and little attention has been given to how this influence plays out on a large scale. Yet, the degree to which antecedent flows affect nutrient export from large basins may have profound implications for environmental management and policy, particularly for large basins in agricultural regions 15 that contribute substantial masses of nutrients to coastal waters. Nutrient fluxes from the Mississippi River Basin (MRB) are closely related to the spatial extent of the hypoxic zone in the Gulf of Mexico (Donner and Scavia, 2007;Rabalais and Turner, 2001); consequently, the hypoxic zone is often smaller during a drought when low flows from the Mississippi River deliver smaller nutrient loads to the Gulf (Scavia et al., 2003;Turner et 20 al., 2006). However, nitrate and other nutrients may accumulate within the basin during a drought and be subject to flushing by high flows when a drought ends, resulting in higher than normal nitrate concentrations in receiving waters.
The accumulation of nitrate in farm fields is a function of many influences, including weather conditions, soil characteristics, crop type, crop yield, fertilizer application, Introduction Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | low crop yield the previous growing season have elevated soil nitrate concentrations, whereas farms that had average or above average yields have low soil nitrate concentrations (Sawyer, 2013). During a drought, irrigation is often a determining influence for crop yield and thus the amount of nitrate likely to accumulate in the soil (Sawyer, 2013). Most farmland in the MRB is not irrigated ( Table 1) and elevated soil nitrate concen-5 trations are typically anticipated across much of the basin following a drought (Dinnes et al., 2002;Ferguson et al., 2013;Randall et al., 2003;Rehm et al., 2009;Sawyer, 2013).
In this paper, we explore the influence of antecedent flow conditions on nitrate anomalies in the MRB and identify the contemporaneous flow conditions in which an-10 tecedent flows are most influential. Nitrate anomalies are the unexplained variability in nitrate concentration after filtering out season, long-term trend, and contemporaneous flow effects. Our objective is to quantify these relationships for eight sites in the MRB ( Fig. 1) using data collected over three decades and across a range of contemporaneous flow conditions. Introduction semi-monthly to monthly frequency (e.g., 9-18 samples per year). Nitrate data were collected across a range of streamflow conditions including base and peak flows.

Methods
In the main channel of the Mississippi River and in several of its major tributaries, nitrate concentrations have been related to season, long-term trend over time, and contempo-5 raneous daily mean flow . The remaining unexplained variability in nitrate concentration may be related in part to antecedent flow conditions. In this study, a statistical model is used to quantify the unexplained variability in nitrate concentration after filtering out these effects. This unexplained variability is the deviation of the observed log nitrate concentration from the log nitrate concentration predicted by 10 a statistical model (based on contemporaneous daily mean flow, season, and trend), herein referred to as nitrate anomalies (Vecchia et al., 2008). If antecedent flows influence nitrate concentration, a statistically significant relationship (p ≤ 0.05) between antecedent flows, expressed in terms of a hydrologic statistic, and nitrate anomalies should be observed. 15 In this study, we define antecedent flow as a ratio between mean daily flow of the previous year and mean daily flow of the period of record, for a given site (Q ratio). The Q ratio (Qr i ) for day i is calculated as where Qyr i is the mean daily flow for the previous year (day i through the previous 20 364 days), and Q POR is the mean daily flow for the period of record. The Q ratio serves as a surrogate for overall basin wetness or dryness the previous year, and Q ratios likely relate to other physical, chemical and biological processes in a basin that are affected by preceding moisture conditions. The calculation of Q ratio is straightforward pc i = β 0 + β 1 t + β 2 ln (Q) + β 3 sin (2 π t) + β 4 cos (2 π t) where ln is the natural log, β 0 , β 1 , . . . , β 4 , are fitted coefficients, t is time, and Q is daily mean streamflow (Hirsch et al., 2010). Nitrate anomaly (CA i ) for day i is defined as where c i is the observed nitrate concentration on day i , and pc i is the predicted log 5 nitrate concentration on day i . By using WRTDS, nitrate anomalies can be conceptualized as the portion of the concentration signal that is not accounted for by contemporaneous discharge, season or long-term trend. Thus, a positive nitrate anomaly indicates higher-than-anticipated observed concentration; a negative anomaly indicates a lower-than-anticipated concentration. For details on WRTDS and the modeling of Nonparameteric statistical methods were used to explore the influence of antecedent flows on nitrate anomalies because the Q ratio data are positively skewed and contain outliers (Fig. 3). The strength of the correlation between nitrate anomaly and Q ratio was determined using Kendall's tau, and the relationship was quantified using the 15 Kendall-Theil robust line (Helsel and Hirsch, 2002). The robust line describes the response of nitrate anomaly to Q ratio and is defined as where CA i is the nitrate anomaly for day i , Qr i is the flow ratio on day i , and β o and β 1 are the fitted coefficients for the intercept and slope, respectively. Rather than using 20 ordinary least squares to estimate the coefficients, the slope is based on the median slope of all pairwise slopes between CA i and Qr i values, and the intercept is backcalculated using this median slope and a point defined by the median of all CA i values and the median of all Qr i values (Helsel and Hirsch, 2002 Additionally, to identify the contemporaneous flow conditions in which concentrations are most sensitive to antecedent flows, data at each site were divided into flow classes according to the daily mean flow on the day of sample collection, and robust lines were fit to each site and flow class. Contemporaneous flow classes consist of four percentile ranges based on the period of record: low (< 25th percentile), mid-low (> 25th and 5 < 50th percentile), mid-high (> 50th and < 75th percentile), and high (> 75th percentile) contemporaneous flows.
Finally, to quantify the effect of antecedent flow on nitrate concentration, as opposed to nitrate anomaly, the percent difference in nitrate concentration relative to a previous year that had average daily flows (Q ratio = 1) was determined using the following equation, where β 1 is the slope coefficient for a given site and flow class (see Tables 2 and 3), 15 Qr avg = 1 (the Q ratio value for a hypothetical day that had average daily flows the previous year), and Qr i is the Q ratio for day i . Because the denominator in Eq. (5) gives the expected nitrate concentration following a year with average flow conditions, the resulting percent difference from this equation gives the anticipated increase or decrease in nitrate concentration for a given antecedent flow condition (Qr i ). Four hypothetical 20 Q ratio values (0.5, 0.75, 1.25 and 1.5) were applied using Eq. (5). These results are anticipated to parallel those quantified by the robust line relationships (Eq. 4) but apply directly to nitrate concentration instead of nitrate anomaly. 10,2013 Antecedent flow conditions in the Mississippi River Basin

Nitrate anomaly across all contemporaneous flows
When all contemporaneous flows at each site are considered together, the upper Mississippi River (CLIN) and the major tributaries (WAPE, VALL, and GRCH), except the Missouri River (HERM), exhibit statistically significant relationships (p ≤ 0.05) between 5 Q ratio and nitrate anomaly (Fig. 4), though tau is small, ranging from −0.13 to −0.17 depending on the site ( Table 2). All sites have negative slopes and the steepest slope occurred in the upper Mississippi River (CLIN). Downstream Mississippi River sites (GRAF, THEB, and MISS-OUT) and the Missouri River (HERM) do not demonstrate significant relationships across the observed range of flows ( Fig. 4). In general, the strength of the relationships shown here (Table 2) are weaker than those reported elsewhere for smaller basins (e.g., Biron et al., 1999;Burt et al., 1988;Foster and Walling, 1978;Macrae et al., 2010;Welsch et al., 2001), which is not necessarily surprising given the complexity of solute behavior in large rivers (Webb and Walling, 1984). In this analysis, the Q ratio describes previous flow conditions in a basin and also 15 serves as a proxy for changes to other physical, chemical and biological processes that are affected by inter-annual variation in the overall moisture of a basin. Grouped into two broad categories, variations in antecedent flow conditions often coincide with changes to: (1) the mass and availability of nitrate in soil (supply), and (2) hydrologic processes that move nitrate through the basin to the stream (transport). Many pro-20 cesses control the accumulation of available nitrate in the soil during a drought, and most are closely related to soil moisture conditions. These may include increased plant stress resulting in low nitrate uptake and low crop yields (Groves and Bailey, 1997), decreased microbial processes resulting in more limited denitrification (Ashby et al., 1998;de Klein and van Logtestijn, 1996) and decreased runoff and leaching (Emmerich and 25 Heitschmidt, 2002; Stites and Kraft, 2001). The timing of fertilizer application before or after a rainfall or irrigation event also influences the amount of available nitrate in the soil (Aulakh and Bijay-Singh, 1997 typically coincide with lowered water tables, decreased hydrologic storage, and decreased hydrologic connectivity, all of which inhibit nitrate transport to streams (Bernal and Sabater, 2012;Detty and McGuire, 2010;Macrae et al., 2010). Wetter antecedent conditions can cause these supply and transport limiting processes to have the opposite effect of minimizing the accumulation of nitrate in the soil through denitrification, 5 crop uptake and other processes, while also increasing hydrologic connectivity and the frequency with which nitrate is transported to groundwater or a stream. Although, supply and transport limiting processes interact to encourage or inhibit nitrate export, the varying influence of these processes can result in inconsistent relationships between antecedent flow conditions and nitrate concentration among different basins (Macrae 10 et al., 2010) and even over time within a single basin (Burt and Worrall, 2009;Burt and Worall, 2007). The statistically significant negative relationships (p ≤ 0.05) between Q ratio and nitrate anomaly (Fig. 4) exhibited in the upper Mississippi River (CLIN), Iowa River (WAPE), Illinois River (ILLI) and Ohio River (GRCH) indicate dry hydrologic conditions 15 the previous year relate to higher nitrate anomalies and wet hydrologic conditions the previous year relate to lower nitrate anomalies. These four sub-basins are likely the most homogenous in the study area. At these sites, it appears soil nitrate that accumulates during dry periods increases the supply of nitrate, which may influence nitrate export later in the year. The remaning sites further downstream on the Mississippi 20 River (GRAF, THEB and MISS-OUT) and the Missouri River (HERM) do not provide evidence that nitrate anomalies respond to previous antcedent flow conditions, at least when considering all contemporaneous flows together. Interestingly, the GRAF site, located on the Mississippi River below the confluence with the Illinois River ( Fig. 1), has relatively similar climate and basin characteristics as CLIN, WAPE and VALL (Table 1), 25 yet does not show a statistically significant relationship between Q ratio and nitrate anomaly when all contemporaneous flows are considered. The lack of an apparent influence of antecedent flow conditions at HERM, THEB or MISS-OUT is necessarily surprising. The Missouri River Basin extends from the Rocky Mountains in the most Introduction western portion of the basin, through the semi-arid Great Plains and into the humid corn belt in the most eastern portion of the basin (Fig. 1), thus making it the most heterogeneous sub-basin in this study. The wide range of climates and terrains throughout the Missouri River Basin can cause parts of the basin to experience markedly differently hydrologic conditions simultaneously, which may lead to distinct hydrologic processes 5 in this basin compared to others in this study. Further downstream in the Mississippi River, the THEB site ( Fig. 1) is primarily a mix of Missouri River water (approximately 39 %) and upstream Mississippi River water (approximately 54 %), neither of which exhibit statistically significant relationships. A significant relationship was not anticipated at the outflow of the Mississippi River (MISS-OUT, Fig. 4) because it is a mix of diverse 10 inputs including Ohio River water (43 %), Missouri River water (14 %), and other water from upper (19 %) and lower (24 %) portions of the basin ( Table 1). The travel time of water from different locations in the MRB can take weeks to months to reach MISS-OUT (Nolan et al., 2002), thus the influence of antecedent flows observed at upstream and more homogenous tributaries is likely smeared as water moves downstream and 15 mixes with other water sources.

Nitrate anomaly by contemporaneous flow class
In most cases, the relationship between Q ratio and nitrate anomaly is stronger when the flow condition on the day of sample collection (contemporaneous flow) is considered. Robust line coefficients and tau are typically greater in magnitude for specific 20 contemporaneous flow classes (Table 3) as compared to those derived using all contemporaneous flow data together (Table 2).

Storm response
At the highest contemporaneous flows (> 75th percentile) Q ratio and nitrate anomaly are negatively related (p ≤ 0.05) at three (CLIN, WAPE and VALL) of the eight sites falling limbs of major storms within a basin. Also, at mid-high contemporaneous flows (> 50th and < 75th percentile), nitrate anomalies are negatively related to the Q ratio at three of the eight sites (VALL, THEB and GRCH) and positively related at one site (HERM). For these sites, mid-high flows include all or portions of the rising and falling limbs of a hydrograph. During periods with generally elevated flows (during the spring, 5 for example), mid-high flows typically occur near the beginning and end of a storm event. For smaller events or events that occur during a generally lower flow period (during the summer, for example), the mid-high flow range may encompass the entire event, including its peak flow.
In total, six of the eight sites (including GRAF, though the relationship is not statistically significant (p = 0.06)) show negative relationships between Q ratio and nitrate anomaly when contemporaneous flows were greater than the 50th percentile of flow (Fig. 5), suggesting a flushing response occurs during storm events that follow extended dry antecedent conditions. This process has been explored extensively in the literature for forested and agricultural basins (Biron et al., 1999;Burt et al., 1988;Fos-15 ter and Walling, 1978;Hornberger et al., 1994;Macrae et al., 2010;Walling and Foster, 1975), and is primarily attributed to the rapid movement of nitrate during a storm when the water table intersects soil horizons that have accumulated elevated stocks of nitrate during periods of low moisture. Our results suggest that a flushing response, previously documented for small, relatively homogenous basins, is also observable at 20 a regional scale. Conversely, wetter antecedent conditions at these sites result in lower nitrate anomalies during storms possibly because the mass of stored nitrate has been depleted by increased export from the basin and uptake by plants earlier in the year. Noticeably, the flushing response at the highest flows (> 75th percentile) is evident only for the smallest basins (< 250 000 km 2 ) and no statistically significant relationships oc-25 cur at the highest flows for basins larger than 250 000 km 2 (Fig. 5). With the exception of GRAF (Fig. 1 influence of antecedent flow conditions (Creed and Band, 1998) in larger study basins, whereas dilution in smaller, more intensely farmed basins appears less common. Contrary to other sites in the MRB, nitrate anomaly is positively related to the Q ratio in the Missouri River (HERM) during mid-high contemporaneous flows (Fig. 5). This observation directly contradicts the flushing response model described for other sites.
However, wetter antecedent conditions have been related to increased nitrate export in other studies, though in these studies antecedent conditions were typically considered over time periods shorter than a year and in basins smaller than those considered in this study (e.g., Welsch et al., 2001;Macrae et al., 2010).
A possible conceptualization of this relationship in the Missouri River (HERM) is that 10 the supply of exportable nitrate is reduced by irrigation or other processes during a drought. Approximately 25 % of cropland in the Missouri River Basin is irrigated making it the most irrigated basin in this study (Table 1). During droughts, irrigation may remove nitrate from the soil horizon by leaching, denitrification, or uptake by crops (Aulakh and Bijay-Singh, 1997;Dinnes et al., 2002). Leached nitrate typically moves 15 downward below the active root zone, leading to elevated nitrate concentrations in groundwater (Burkart and Stoner, 2008;Stites and Kraft, 2001). Increased denitrification occurs with irrigation because elevated soil moisture conditions increase microbial activity (de Klein and van Logtestijn, 1996; Groves and Bailey, 1997). Which process dominates during a drought is debatable and may depend on soil properties, fertilizer 20 application rates, and climate (Aulakh and Bijay-Singh, 1997;Brown et al., 2011). In the Missouri River Basin, a recent modelling effort found that increases in irrigation relate to decreases in total nitrogen export on a regional scale (Brown et al., 2011). Irrigation likely occurs at a higher rate when the weather is drier than average, according to a study in Illinois (Bowman and Collins, 1987), therefore, lower nitrate anomalies in the 25 Missouri River (HERM) following a drought may occur because processes associated with irrigation do not allow for the accumulation of nitrate during drier than average climatic conditions. However, the supply-limiting influence of irrigation does not account for the higher nitrate anomalies observed following wetter antecedent conditions. Interestingly, the Missouri River Basin also has the greatest number of dams and the largest relative storage of any basin ( Table 1). The reservoirs in this basin hold approximately 1.89 times the annual flow of the Missouri River at HERM which is more than twice the relative storage of any other site in this study (Table 1). Therefore, flow conditions at HERM, and low flows in particular, are not just the result of natural hydro-5 logic conditions but are also influenced by release decisions made by dam operators. The confounding processes of irrigation and dam storage in addition to the geophysical and climatological heterogeneity of the Missouri River Basin make even rudimentary interpretation problematic. 10 Only the Ohio River (GRCH) and Mississippi outflow (MISS-OUT) demonstrate a significant negative response (p ≤ 0.05) to the previous year's flow at mid-low (> 25th and < 50th percentile) or low (< 25th percentile) contemporaneous flows (Fig. 5, Table 3). These flow ranges generally occur between storm events and represent baseflow conditions. For other sites, the lack of significant relationships during baseflow suggests 15 the groundwater system is not closely influenced by surface conditions, at least over a time span of one year. Among all tributaries to the Mississippi River, the Ohio River contributes about 43 % of flow to the Mississippi River (Table 1); therefore if flow at GRCH is low, flow at MISS-OUT is likely to also be low. Since low flow conditions at GRCH and MISS-OUT are closely related, it is likely any interpretation about the influence 20 of antecedent flows on nitrate anomalies for GRCH also applies to the low flow response observed at MISS-OUT. However, insight into the influence of previous drought on baseflow conditions and nitrate concentration at very large scales is limited and the relationships documented at GRCH and MISS-OUT are not readily interpretable.

Potential effect on nitrate concentration
For each statistically significant relationship (p ≤ 0.05, Tables 2 and 3), Eq. (5) and the appropriate slope coefficient were used to translate nitrate anomalies to the percent change in nitrate concentration that would occur following a wet or dry year (Qr i > 1 or Qr i < 1, respectively) relative to the nitrate concentration expected following a year with 5 average flows (Qr avg = 1). For example, the three different Q ratio values for the Illinois River (VALL) in Fig. 2 Table 3 for VALL at mid-high flow conditions. The resulting calculations indicate concentration anomalies are expected to be positive (0.29), negative (−0.18) and near zero (0.10), respectively, for these three dates. To put the results into terms of percent change in concentration, Eq. (5) 15 was used to estimate that nitrate concentrations on these three dates will be +23 % different, −23 % different, or indistinguishable (+2 %), respectively, from nitrate concentrations expected following an average flow year.

All contemporaneous flows
Rather than apply Eq. (5) to each day in the period of record at each site, four hypotheti-20 cal Q ratio scenarios were used to describe the potential response of nitrate concentration to different antecedent flow conditions. Hypothetical Q ratios and Eq. (5) were applied by site and only for the flow conditions that had significant robust line relationships (p ≤ 0.05, Table 4 (WAPE, and VALL). In the upper Mississippi River (CLIN) and Ohio River (GRCH), the difference in nitrate concentration could be expected to be as much as 27 % higher or 21 % lower than expected (Table 4).

Storm response
Analogous to the nitrate anomaly results, nitrate concentration responds more strongly 5 to antecedent flow conditions when contemporaneous flow data are subdivided into flow classes. For contemporaneous flow classes that capture all or part of a storm event (contemporaneous flows > 50th percentile), when the previous year's flow is 25 % drier than average, nitrate concentration may be about 6 to 10 % higher than expected, for most sites where nitrate anomaly is negatively related to the Q ratio (  (Table 4). At these sites, differences in nitrate concentration 15 are slightly smaller in magnitude and negative when the previous year is wetter than average (Table 4). In the Missouri River, percent differences in nitrate concentration are similar in magnitude to those at other sites but opposite in direction; when the previous year was 25 or 50 % drier than average, nitrate concentration is 8 or 16 % lower than expected, respectively. With the exception of HERM, these patterns at mid-high and 20 high contemporaneous flow conditions are consistent with the conceptual model of soil nitrate flushing during storm events following a drought.

Baseflow response
Nitrate concentration appears to be more sensitive to changes in antecedent flow during low and mid-low contemporaneous flows in the Ohio River (GRCH) and Mississippi However, while the relationships between Q ratio and nitrate anomaly at mid-low and low flows at GRCH and MISS-OUT are statistically significant (p ≤ 0.05), they do not appear as visually strong as those at other sites or higher contemporaneous flow conditions (Fig. 5). For mid-low and low flow classes, when the previous year's flow is 25 % drier than average, nitrate concentration may be about 9 to 20 % greater than ex-5 pected. As antecedent flow conditions become increasingly dry (50 % of average flow) nitrate concentration can be 19 to 44 % higher than expected (Table 4). Similarly, during baseflow conditions when the previous year's flow is 25 and 50 % wetter than average, nitrate concentration can be between 8 and 30 % lower than expected. 10 Nitrate sensors deployed in several Iowa rivers during the spring of 2013 provide some empirical support for the results presented in this study. From May 2012, through February 2013, much of the central United States experienced moderate to extreme hydrologic drought. By the following spring (2013), much of the State of Iowa (Fig. 1) had recovered and was moderately to very wet (National Oceanic and Atmospheric Admin-15 istration, 2013). For example, peak discharge between early-October and mid-June of 2013 would rank as the 5th highest annual peak discharge in the 111-yr flood record at the WAPE site on the Iowa River. At this site, daily mean flow from March through May was predominately mid-high (50th to 75th percentile) to high (greater than 75th percentile) and the mean of the daily Q ratios over this period was 0.49, indicating that flow 20 during the previous year was approximately 50 % lower than average. Considering the contemporaneous flow conditions, we used Eq. (5) and observed (though provisional) daily streamflow data to predict the concentration differences for each day during this 3-month period. These predictions indicate nitrate concentration at WAPE was likely between 7 to 19 % higher (13 % higher on average) during these months than would

Conclusions
Except for the Missouri River (HERM), our results show a negative relationship between antecedent flow conditions and nitrate anomaly during mid-high and high contemporaneous flows for the major tributaries and two of the four Mississippi River mainchannel sites (Fig. 5). In general, when the previous year was drier than average, nitrate concentration is higher than expected relative to nitrate concentrations following a year with average flow conditions. This response is likely due to the accumulation 15 of soil nitrate during a drought and subsequent flushing with moderately high to high flow events when the drought ends. When the previous year was wetter than average nitrate concentrations are lower than expected because more nitrate is likely taken up by crops, removed from the system through denitrification, or transported with greater frequency (at lower concentrations) to the stream and groundwater earlier in the year. 20 The positive relationship observed in the Missouri River (HERM) during mid-high contemporaneous flow conditions (Fig. 5), indicates the influence of antecedent flow on nitrate anomaly not only varies by contemporaneous flow class but also regionally. The heterogeneity of the Missouri River Basin coupled with high levels of irrigation and dam storage ( Due to the large scale of these basins and their inherent complexity, the flushing response observed at these sites is dampened (Tables 2 and 3) compared to observations from smaller basins (e.g., Biron et al., 1999;Burt et al., 1988;Foster and Walling, 1978;Macrae et al., 2010;Welsch et al., 2001). Yet, nitrate concentrations following a drier or wetter than average year appear to be up to 27 or −21 % different from 5 nitrate concentrations expected following a year with average flow (Table 4). These percent differences in nitrate concentrations typically increase in magnitude when contemporaneous flows are considered (Table 4) and can be as much as 34 % different from expected during storm events and high flows (CLIN) or up to 44 % different from expected during baseflows (GRCH). How higher-than-expected nitrate concentrations 10 following a drought will affect the hypoxic zone in the Gulf of Mexico is debatable and is likely influenced by factors such as, the timing of delivery to the Gulf (during the spring versus the fall, for example), the magnitude of flows transporting nitrate through the basin, the spatial and temporal variability of sub-basins experiencing drought and flushing, and changes to nutrient management practices throughout the basin. 15 While this study identifies significant relationships between antecedent flow conditions and nitrate concentration, it does little to explain the cause of these relationships, thus we propose several questions to encourage future studies on this topic at similar scales.
-What are the controlling influences for relationships between antecedent flow con-20 ditions and nutrient export, and how do these relationships change based on climate, basin characteristics, and management practices?
-Do relationships between antecedent flows and nitrate export change over time, as documented in other basins with long temporal records (Burt and Worrall, 2009;Burt and Worall, 2007)?

25
-Which specific aspects of drought conditions (such as the magnitude and duration of low flows, and the timing of low and high flows) most influence nitrate accumulation in an agricultural basin and its subsequent flushing to a stream? -Based on these results might it be possible to develop a better statistical model of nitrate export that simultaneously uses both current and antecedent flow conditions to estimate concentration?
-How would one go about using new, high frequency nitrate sensor data to improve understanding on how antecedent flows influence solute concentration? Will these 5 new, richer data sets facilitate understanding of storage, transport, and processing of nitrogen within watersheds at this scale?
The results of our analysis suggest that nitrate transport in the Mississippi River Basin is not a simple product of current hydrologic conditions and nitrate concentrations, but rather an integration of current conditions with past inputs of water and 10 changes in nitrate supply that vary regionally and with contemporaneous flow classes. Therefore, an improved understanding of the evolving pattern of nitrate fluxes from the entire Mississippi River Basin will require detailed analysis of the diverse patterns of nitrate export from the various sub-basins and their interaction with similarly variable spatial and temporal patterns of climate and management practices. As a result, the 15 evaluation of progress in nutrient management will benefit from consideration of antecedent influences. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Scavia, D., Rabalais, N. N., Turner, R. E., Justic, D., and Wiseman, W.