| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
USDA Agricultural Research Service-National Soil Tilth Laboratory, 2150 Pammel Drive, Ames, IA 50010
* Corresponding author (tomer{at}nstl.gov)
Received for publication April 15, 2002.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONTHE WALNUT CREEK WATERSHEDMETHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Abbreviations: DOY, sequential day of year
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONTHE WALNUT CREEK WATERSHEDMETHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Interactions between agricultural practices, watershed dynamics, water quality, and the health of aquatic and marine ecosystems are complex. Recent research has focused on understanding the dynamics of nutrient fluxes in agricultural watersheds (Alexander et al., 1996, 2000; Burkart and James, 1999; David and Gentry, 2000; Pionke et al., 1999; Sauer et al., 2001), on understanding how management practices can affect fluxes through tile lines (Bjorneberg et al., 1998; Bolton et al., 1970; Dinnes et al., 2002; Randall and Gross, 2001), and on ecosystem responses to nutrient loading (Lowery, 1998; Mallin et al., 2001). These kinds of improvements in watershed science require that we better understand the dynamics of nutrient fluxes, and not just concentrations. Policies being developed under the U.S. Clean Water Act of 1972, Section 303(d), are focused on the total maximum daily load (TMDL) that a given water body can accept without impairing its use or ecological function. Therefore, water quality monitoring efforts are increasingly aimed at obtaining both water flow and contaminant concentration data, as both are required to calculate nutrient fluxes and total loads.
Despite the importance of tile drainage and our need to understand nutrient flux dynamics in Midwestern watersheds, there are few long-term data sets that summarize nutrient fluxes from tile-drained watersheds in this region. This is problematic because a long-term record that captures a wide range of hydrologic conditions provides the best opportunity to interpret water quality data. But succinct interpretation of water quality records can be difficult because, in effect, watershed processes uncouple hydrologic flows and solute fluxes. The complexities of water-flow pathways and the timing of their responses to rainfall control a watershed's solute fluxes, even for a conservative solute delivered solely via rainfall (Kirchner et al., 2000). Biogeochemical processes and land management also cause variations in nutrient fluxes, and the sum effect is great variation in hydrologic and nutrient fluxes across the years. Few, if any, watershed monitoring research projects have the opportunity and resources to characterize all the interacting processes affecting these variations at the watershed scale. Here we aim to use simple tools to evaluate key attributes of water flows and nutrient fluxes in a way that captures watershed performance. In small watersheds at least, this may indicate key processes that are occurring, and provide ways to better understand the opportunities and challenges to manage nutrient flows.
As part of the Management Systems Evaluation Area program (Onstad et al., 1991), hydrologic flows and nitrate concentrations have been monitored within the Walnut Creek watershed of central Iowa since 1992 (Hatfield et al., 1999), providing the opportunity to summarize nutrient flows from a tile-drained watershed over a significant time frame. The objectives of this paper include:
|
THE WALNUT CREEK WATERSHED |
|---|
|
TOPABSTRACTINTRODUCTIONTHE WALNUT CREEK WATERSHEDMETHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
The artificial drainage system consists of ditches, subsurface pipes and tile lines, and surface water intakes. Surface water intakes are located within closed depressions and along roadside ditches. These intakes, along with subsurface tile lines installed in varying patterns across the cropped fields, deliver water to drainage mains, which are maintained by local (county) authorities. The drainage mains deliver water to ditches (maintained by the same authorities), which in turn drain to local streams. Installation of these systems at Walnut Creek began nearly 100 years ago, with maintenance and expansion continuing through current times by both local authorities and private landowners. As a result, the subsurface systems are not consistently mapped, and, in essence, impossible to fully characterize by currently available technology.
Fertilizer management practices can affect the loss of NO3–N from subsurface drainage systems (Randall and Gross, 2001). In the Walnut Creek watershed, applications of fertilizer nitrogen, at rates averaging about 150 kg N ha-1, are frequently applied in the autumn to fields that will be planted to corn the following spring (Hatfield et al., 1999). This is done to ease difficulties in scheduling springtime farming operations that result from wet weather conditions. But the relatively high rates of fertilizer application and asynchronous timing with crop uptake may result in significant leaching of NO3–N. Dinnes et al. (2002) give a comprehensive discussion of N fertilizer management in tile-drained areas. Tiles provide short-circuiting pathways for delivery of shallow soil water and runoff waters to streams, which may minimize opportunities for riparian-zone processes (e.g., biological uptake and denitrification) to reduce NO3–N concentrations.
Given these concerns, the Walnut Creek project was initiated to characterize the hydrology of a tile-drained Midwestern landscape, and its accompanying dynamics of NO3–N and pesticide fluxes. To date, monitoring has provided a long-term data set (approximately nine years) of flow quantities and water quality at the watershed outlet (draining 5130 ha), and from several subbasins defined by subsurface drainage networks (Fig. 1) . County map archives provide reliable information on the location of main drainage lines and associated subbasin areas. The data considered here were collected at three locations, referred to as the 210-tile site, a subbasin draining 493 ha; the 230-tile site, a subbasin draining 863 ha; and the 330-stream site, located at the watershed's outlet (Fig. 1).
|
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONTHE WALNUT CREEK WATERSHEDMETHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Tipping-bucket rain gauges located in the catchment and its subbasins recorded precipitation, also at 5-min intervals. Cumulative precipitation figures included daily snowfall data collected at the Ames (IA) waste treatment plant, about 1.5 km north of the 330-stream site, and outside the watershed.
Water samples were collected using both grab methods, usually at weekly intervals, and by automatic sampling at all three locations. Auto sampling was activated by a change in stage height, or when three days had elapsed since the previous sampling. Water samples were analyzed for nitrate concentration with an autoanalyzer method, as summarized by Hatfield et al. (1999) and detailed by Hatfield and Sauer (1994). The method had a detection limit of 1 mg L-1 for NO3–N concentration.
The flow data were averaged for 30-min periods, and these averages were used in further calculations. The three flow-monitoring stations were all operational as of 18 July 1992, and measurements from then through the year 2000 were included in the analysis. However, frozen conditions during winter prevented reliable operation of the pressure transducers, and therefore much of the record during winter months and spring thaw was omitted. Under ice-free conditions, there were some periods when not all sensors were functioning. These missing data were estimated using relationships with other nearby stations included in the study (Jaynes et al., 1999). Missing tile flows were usually estimated using flows from other tile mains, and missing outlet flows were estimated using flows from an upstream gauging station (Site 310, shown in Fig. 1). Data were plotted, preferably using times near the gap in data, and best-fit expressions were determined. These expressions were most often linear, but in some instances were polynomial, or exponential (as in Jaynes et al., 1999). Correlation coefficients (r) usually exceeded 0.9, with a few minor exceptions. These data gaps were brief (less than three weeks), except for a 190-d gap in the record for the 230-tile site that ended in July 1998. Most of this gap was filled using an exponential relation with the 210-tile site data. But there was also about 18 d of large flows, during which this exponential relation could not reasonably be extrapolated. These 230-tile site flows were estimated using a polynomial expression with the 210-tile site obtained from high flow conditions in 1993.
Measured concentrations of NO3–N were multiplied by discharge to obtain estimates of N fluxes. For each discrete sampling time, a time interval was assigned that went from the midpoint time since the previous sample was collected to the midpoint time before the subsequent sample was collected. Flows during each time interval were accumulated and multiplied by NO3–N concentration to obtain a mass flux. Average fluxes of water and NO3–N were finally calculated, and expressed as mm h-1 for water flux, and kg ha-1 d-1 for NO3–N, to provide a unit-area comparison between sites. Samples collected during frozen stream conditions or early spring thaw were omitted from these calculations.
Data Summarization
Precipitation (mm), flow (mm h-1), and NO3–N load (kg ha-1) were summed, and flow-weighted NO3–N concentrations calculated, for each calendar year. We then summarized the flow, NO3–N concentration, and NO3–N flux data by plotting duration curves and calculating univariate statistics for each monitoring station. Duration curves were plotted using the individual event data. In doing so, we extended the concept of flow-duration curves, which hydrologists use to characterize variations in stream flow records (Prakash et al., 1996), to also plot duration curves for nitrate concentration and flux. The entire sample record was then sorted according to sampling date and summed by date; these date-sorted totals were plotted to examine the seasonality of the data.
Univariate statistics were calculated on weekly averaged data, which provided equal weighting of observations. The weekly interval was selected because it was the largest interval between samples. We calculated product-moment statistics of mean, standard deviation, skewness, and kurtosis.
Data Analyses: Flow–Nitrate Nitrogen Flux Relationships
In this evaluation, we used a simple expression to summarize the flow–NO3–N flux relationships for these three monitoring stations, as follows. If water flow (Q, mm h-1) is plotted against flux of a given solute (L, kg ha-1 d-1) on a log–log plot, then a line fitted to this plot is given by:
 | [1] |
If the concentration of the solute never changes in the record, all points would plot exactly on a straight line with a slope (b) of 1, and the intercept's value (a) would be determined by the concentration (and unit-conversion constants). Of course, water quality data do not behave this way; concentrations often change in a way that partly depends on Q. Large flows may flush contaminants and exhibit an increased concentration, and/or in-stream processes may more effectively remove contaminants at low flows. Under these scenarios the slope's value (b) would exceed 1. In other locations, or under a different land use, large flows may lead to diluted concentrations, leading to b < 1. Minimal values of b would be associated with those constituents associated with baseflow contribution to stream flow.
The term b in Eq. [1] was named the elasticity coefficient by Alexander et al. (1996), in a study of annual flows and N loads discharged from U.S. rivers to the Atlantic Ocean. Among these rivers, values of b ranged from 0.05 to 1.59, with a median of 0.93. Annual data for developed watersheds of the USA show an elasticity coefficient of 0.86 (Sauer et al., 2001), suggesting smaller N concentrations in basins yielding more flow. Here we apply this concept to detailed data from three single stations, undertaking statistical procedures to account for measurement errors and autocorrelation, described as follows.
Coefficients for Eq. [1] were determined for a summary data set that was calculated to reduce autocorrelation effects. To obtain this summary data set, the discharge and NO3–N flux data were first averaged across 7-d periods, and then evaluated to determine the time interval (number of weeks) at which flows became independent. This is known as the scale of fluctuation, or correlation scale (Vanmarcke, 1983; Cressie, 1993), which is denoted as 
and defined as:
 | [2] |
In Eq. [2], m is the time interval separating a pair of observations, and 
(m) is the autocorrelation function (Vanmarcke, 1983; Cressie, 1993) defined as:
 | [3] |
Meek (2001) presents a semiparametric, iterative method of estimating using SAS macros (SAS Institute, Cary, NC), which was applied to data from the 210-tile, 230-tile, and 330-stream sites. Briefly, (m) and were determined on the weekly data, for both Q and L, and for regression residuals from fitting of Eq. [1] using ordinary least squares (OLS). Selection of was done considering integration results (Eq. [2]) for these three variates (i.e., Q, L, and OLS regression residuals), and at all three sites. A common for the three monitoring stations was selected to suggest a block of time with minimal autocorrelation effects between values, and the weekly data were further averaged across durations of weeks. This time block was selected to be small enough to retain as large a number of points as possible, while large enough to effectively compensate for autocorrelation. Regression analyses were then run on the -week averaged data.
To ascribe confidence intervals to regression results, the OLS approach assumes measurement without error for the X variate (in this case Q). Neglect of this assumption may lead to an attenuated estimate of slope unless the error is considered in the analysis (Kempthorne and Allmaras, 1986). It is not possible to measure hydrologic flows without error. To finalize our estimates of Eq. [1] coefficients for the three monitoring stations, we used two techniques that consider measurement error in the predictor. First, a reduced major axis (RMA) method (Draper, 1991; Mann, 1987) was used, which calculates the geometric mean of the two slopes calculated by OLS regression of X on Y and of Y on X. This method does not require knowledge of the measurement error in X, but cannot provide confidence intervals for the regression coefficients. The second technique, the method of moments (Fuller, 1987), provides confidence intervals but the measurement error in X must be known or estimated. However, ascribing the measurement error inflates the r2 value accordingly. For this method we estimated a standard error of 5% for flow measurements (a reliability ratio of 0.95 for untransformed data), based on instrument calibration and rating curve data (unpublished data, 1991–1995).
We considered several additional issues through our analysis. The first was the effect of flooding during 1993. Annual precipitation that year in parts of Iowa (including Walnut Creek) exceeded 1200 mm, which may have been a 1000-yr return event (Pitlick, 1997). In this context, we deemed it appropriate to carry out the analyses with and without the 1993 data. The second issue was the effect of nondetectable concentrations (<1.0 mg NO3–N L-1). A concentration of 0.5 mg NO3–N L-1 was assumed for nondetects (half the detection limit). The effect of this assumption was estimated by also assuming a concentration of zero. This was only necessary for the 330-stream station, as only one sample collected at the tile stations showed a nondetectable concentration.
Denitrification in a Hypothetical Wetland Receiving Tile Drainage
Given the observed flows and NO3–N concentrations, we assessed the potential removal of NO3–N from the tile drainage (210- and 230-tile sites) waters through denitrification within constructed wetlands. We used the individual-sample data in this analysis. Tile-drained areas of Iowa are being targeted for installation of constructed wetlands, in part for nutrient removal, using subsidies available under USDA programs (e.g., the Conservation Reserve Enhancement Program, or CREP; see www.fsa.usda.gov/dafp/cepd/crep.htm [verified 2 Dec. 2002]). This was a general assessment, of denitrification alone, based on the following three assumptions.
The first assumption was to size the wetland; a ratio of 0.02 between contributing area and receiving wetland area was assumed, with an average water depth of 0.3 m. Criteria established for Iowa's CREP target this as a maximum area ratio, and limit the extent of water depths greater than 0.9 m (T. Isenhart, personal communication, 2001).
The second assumption was to estimate denitrification rate as a function of temperature. We used data reported by Xue et al. (1999), in which constructed-wetland denitrification rates in the Midwest (Illinois) were measured and shown to be consistent with several other studies (e.g., Christensen and Sorensen, 1986; Lindau et al., 1990). An exponential expression was fit between denitrification rate (mg N m-2 h-1) and temperature (°C) to the data given in Table 1 of Xue et al. (1999), by analogy estimating denitrification rate as a temperature-governed, first-order process. We omitted two data points reported for dates when nitrate concentrations were nondetectable; nitrate availability rather than temperature could have limited the denitrification rate at these times. Given the small size of the remaining data set (n = 5), an iterative procedure was used to obtain the following fitted equation:
 | [4] |
|
 | [5] |
The temperatures obtained using Eq. [5] were between -2.0 and 25.5 for the days of year with recorded flow. We assumed D = 0 for T < 0 and D given by Eq. [4] for above-freezing temperatures, that is, between 1.1 and 12.7 mg N m-2 h-1.
Given these three assumptions, we examined the effect of detention of drainage waters in this hypothetical wetland employing two alternate scenarios, namely nondetention (outflow = inflow for all flows, and constant storage volume), and indefinite detention (denitrification proceeds at its temperature-regulated potential at all times). Under nondetention, denitrification rates were limited by the supply rate of NO3–N in tile waters when the temperature-regulated (loss) rate was estimated to exceed that supply rate. Under the indefinite detention, we assumed that nitrate supply has no influence on denitrification, and simply summed the rates calculated from inferred water temperatures across the entire flow record. These two NO3–N detention scenarios provided a range of denitrification estimates under the above-given assumptions of wetland sizing, temperature control of denitrification rates, and ambient temperature domain. Actual detention storage would vary with the design of the wetland and its outflow structure. However, constructed wetlands are typically constrained to a given area, so that storage increases result in greater water depth, which would tend to decrease physical contact of waters with carbon-rich substrates and thereby decrease denitrification. Nevertheless, the two detention scenarios provide reasonable bounds to estimate the potential effect of a constructed wetland on NO3–N load, through denitrification.
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONTHE WALNUT CREEK WATERSHEDMETHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
The results presented in this paper are from the period of record given in the upper portion of Table 1. Because of this, and because of a minimal reliance on estimation techniques for periods of missing record, and use of 0.5 h flow data to calculate sample-interval loads, annual loads given here are slightly different than those reported elsewhere (e.g., Jaynes et al., 1999), where the flow record consists of longer-period averages, and a somewhat greater portion of the record is derived from estimated values. We consider these differences an acceptable consequence of using techniques appropriate to the task at hand.
Results under the three stated objectives are discussed separately in the following three sections.
Data Summary (Objective 1)
Cumulative flows of water and NO3–N, expressed on an area-weighted basis, showed variations between the stations (Table 1). The tile outlets generally had lesser water flow, but greater nitrate concentrations relative to the 330-stream site, and this led to larger estimates of NO3–N loads for the 210- and 230-tile sites (Table 1). Nitrate N concentrations at the tile sites were predominantly greater than the 10 mg L-1 drinking-water standard, when expressed on either a time-weighted or a flow-weighted basis, but only about one-third of the flows from the watershed outlet (330-stream site) exceeded this standard (Table 1). Year-to-year differences in precipitation, flows, and concentration data were important features of the data (Fig. 2)
. In fact, differences in cumulative flows between stations are largely due to differences that occurred during the 1993 flood year. The largest flows in 1993 overwhelmed the subsurface drainage systems; some flood waters bypassed the tile outlets, but then were measured at the 330-stream site. The 1993 floods also had the apparent effect of flushing stored soil N from the catchment, as evidenced by smaller NO3–N concentrations in 1993 and 1994.
|
Stream-flow export of NO3–N changed according to differences in flow across the years, with small loads in the low-flow years of 1994 and 2000 (Fig. 2D). Loads exceeded 20 kg ha-1 yr-1 at both tile outlets in four of the nine years (1993, 1995, 1996, 1999), but only two of the nine years at the 330-stream site (1993, 1998), coinciding with the years of flooding. The 1993 floods had a large influence on the cumulative data (Fig. 2), accounting for about 19% of the precipitation received during the period of record, but about 35% of the water flows and 25% of the N loads for the tile sites (210- and 230-tile sites), and about 47% of the water flows and 35% of the N loads at the watershed outlet (330-stream site).
Most flows of water and NO3–N during the year occurred during the second quarter (i.e., DOY 92–183; see Fig. 3) . Row crops in Iowa are planted usually beginning after DOY 120; therefore, the crop is either not present or not fully established during most of this period. Flows decreased after the second quarter because evapotranspiration increases as the crops become fully established, and because late summer and early autumn tends to be dry. Nitrate N concentrations also decreased, generally, after the second quarter (Fig. 4) , which is attributed to leaching and crop uptake of available N as the growing season progresses. After removing the 1993 data, flow-weighted NO3–N concentrations increased at all three stations given the remaining record, from 11.3 to 13.1 mg L-1 at the 210-tile site, from 13.4 to 15.5 mg L-1 at the 230-tile site, and from 9.2 to 10.8 mg L-1 at the watershed outlet (330-stream site). This occurred because many of the large flows with small NO3–N concentrations were recorded in 1993 (Fig. 4).
|
|
|
|
|
|
The highest flows at the 330-stream site frequently had relatively small (<10 mg L-1) concentrations (Fig. 6). However, most of the data points with high flow and low concentration at this site were recorded in the flood year of 1993. A similar pattern might be anticipated for high flows from the tile drains, because these large flows included those surface runoff waters entering into tile inlets. However, no dilution of large flows is evident for the two tile mains.
Results of regression analyses (Table 3) indicate similarity in the relationships between water flow and NO3–N flux for the two tile outlet sites (210- and 230-tile sites). The slopes (b) of these relationships are not greatly different than 1, indicating that a 1% change in flow, on average, was consistent with about a 1% change in NO3–N flux. This results from concentration being essentially independent of flow, and indeed, regression of flow versus concentration (individual sample data) shows that less than 5% of the variation in concentrations is explained by flow at the two tile mains, either by linear or log-linear expression. However, the stream outlet (330-stream site) shows a slope greater than one, resulting from smaller concentrations at low flows. At this site concentration is related to ln(Q) for the individual sample data with an r2 of 0.16. This apparently contrasts results given by Jaynes et al. (1999) that, based on the 1992–1995 record at the 330-stream site, NO3–N concentrations were more likely to increase when flow decreased and more likely to decrease when flow increased. However, across this longer record any such trend was weak. Across all individual sample intervals, concentrations increased in 54% of the intervals when flow decreased, and decreased in 48% of the intervals when flow increased. A zero change in concentration was measured in 13% of the intervals, whether flow increased or decreased.
These comparative plots (Fig. 6) highlight the change that can occur in flow–nutrient flux relationships at different points within a small (second-order) agricultural watershed. Similar plots could be used to compare different locations or perhaps paired watershed studies. Paired watershed assessments have often focused on assessing event–response relationships, under different land management practices. But analysis of flow–nutrient flux relationships considers the full range of event and baseflows, and may be a useful alternative for analysis in a number of studies considering patterns of water quality within watersheds.
Regression results were applied to the data with nondetectable concentrations (<1.0 mg L-1) assigned a value of 0.5 mg L-1. This applied to 81 of the samples from the 330-stream site (3.6% frequency), but only one of the tile-site samples. Nondetects tended to occur during low-flow conditions. Therefore, decreasing these assigned concentrations to zero resulted in a slight increase in the slope of the regression lines, from about 1.2 to about 1.27. Confidence intervals for b (Table 3) shifted to similarly larger values.
These regressions were also run without data from the 1993 flood year to assess the influence of that year's data on the results. Because 1993 data were dominantly large flows with small concentrations, their exclusion resulted in greater slopes and intercepts being calculated for Eq. [1] parameters, though these greater values were within confidence intervals established using all the data (Table 3). Regardless of inclusion or exclusion of the 1993 data, the slopes of the tile drain sites were close to 1.0, whereas the 330-stream site had a slope greater than 1.0.
Treatability of Nitrate Nitrogen Fluxes Using Constructed Wetlands (Objective 3)
Two scenarios were used to estimate possible denitrification of tile NO3–N fluxes in a hypothetical constructed wetland. Under the scenario of indefinite detention of nitrate within a wetland, potential denitrification rates were calculated using Eq. [4] and [5], and summed for the entire flow record. Given these assumptions, wetland denitrification could have removed 34% (59 kg) of the cumulative NO3–N flux from the 210-tile site, and 28% (65 kg) of the cumulative flux from the 230-tile site during the period of monitoring. The difference in mass occurs because of the difference in length of the flow record.
Under the other scenario of no detention storage of nitrate in the wetland, denitrification was not permitted to exceed the inflow rate of nitrate. Given this scenario, denitrification could have removed 21% (36 kg) of the cumulative NO3–N flux at the 210-tile site and 18% (41 kg) of the cumulative flux at the 230-tile site. If the 1993 flows are neglected, denitrification in a constructed wetland could have removed between 22 and 40% (29 to 53 kg) of the NO3–N flux from the 210-tile site, and between 19 and 33% (33 to 57 kg) of the NO3–N flux at the 230-tile site, depending on which detention scenario is used. We view the zero-detention estimates as being realistic, with the indefinite detention scenario providing estimates that are optimistic, but perhaps feasible. But we have ignored the fact that water depth increases under high flows, and this can limit availability of carbon that is needed for denitrification. The estimates are for denitrification only, and neglect other wetland sinks such as seepage and assimilation. However, nitrogen assimilated by wetland plants could be released (become a source of N in the long term) unless the biomass is harvested. Also, seepage losses would vary considerably depending on the hydrogeologic setting. Xue et al. (1999) suggested that seepage was an important loss pathway in their study wetlands, but subsurface hydraulic conditions were not measured. Significant seepage losses through the clay till sediments dominant in the upland drainage ditches of Walnut Creek watershed would not be anticipated; very slow permeability is the reason these areas have been artificially drained. Regionally, however, some tile outlets discharge to channels underlain by alluvial sediments that could allow significant surficial-aquifer recharge.
Most of the estimated reductions in NO3–N loads would occur during periods of low flow, given the scenario of zero detention (Fig. 7) . At the 210-tile site, only about 7 kg ha-1 of the total estimated denitrification of 36 kg ha-1 would have occurred when flows exceeded 0.1 mm h-1. These large flows, however, contained 52% of the water and 48% of the NO3–N that flowed past this site. Half of the denitrification would have occurred during low flows (less than about 0.026 mm h-1), when only 14% of the total water flow and 15% of the NO3–N load was delivered. If 1993 data are omitted, then 10% of the 30 kg ha-1 of denitrification would have occurred when flow rates exceeded 0.1 mm h-1, with these flows carrying 37% of the water discharge and 39% of the NO3–N load.
|
|
SUMMARY AND CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONTHE WALNUT CREEK WATERSHEDMETHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Across the record, large flows with small NO3–N concentrations were predominantly from measurements made during the flood year of 1993. Flows occurring during the second quarter of the year, before full establishment of the crop, carried most of the NO3–N load. Flux rates exceeding 0.1 kg ha-1 d-1 occurred 23 to 31% of the time, but these flows account for 74 to 79% of the NO3–N loads at the three sites. Similarly, large water flows (>0.1 mm h-1) occurred 6 to 10% of the time, but accounted for 37 to 51% of the cumulative flow volume.
At the outlet, small flows tended to have smaller NO3–N concentrations, but no such trend was clear for either subbasin. We established a log-linear relationship between flow and NO3–N flux for each of the three stations. For the subbasins, this relationship had a slope, or "elasticity" coefficient, of about 1.0, whereas the outlet had a coefficient of about 1.2. We employed specific methods to account for autocorrelation and measurement error in the analysis that allowed us to calculate confidence intervals for coefficients defining the flow–flux relationship. This approach should be generally useful in comparing water quality records between watersheds or between locations within a watershed, provided a sufficient record is available. Flow–flux relationships can change within a small, second-order drainage basin, with a downstream increase in the elasticity coefficient demonstrated in the case of this watershed dominated by tile drainage. Results show that nitrate concentrations were not typically diluted by large flows, with the 1993 floods being the major exception.
We also estimated how effective constructed wetlands might be in denitrifying NO3–N fluxes from the tile-drained subbasins. Depending on detention of nitrate in the wetland, and the subbasin flows input to this hypothetical wetland, denitrification could remove 18 to 34% of the NO3–N fluxes, assuming the wetland meets criteria for USDA support (i.e., 0.3 m deep and 0.02 ha in area per hectare of contributing cropland). Because NO3–N concentrations are not diluted by large flows, much of the NO3–N load is delivered with large flows. Constructed wetlands cannot effectively denitrify NO3–N delivered with large flows, and thus may not achieve water quality goals with out additional measures. Therefore, practices to optimize nitrogen management within fields should also be encouraged. In particular, spring applications of N to target crop requirements, rather than large, agronomically conservative applications in autumn, should be beneficial. Research conducted on this topic within Walnut Creek will be reported in a future paper that is under preparation.
|
ACKNOWLEDGMENTS |
|---|
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONTHE WALNUT CREEK WATERSHEDMETHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Related articles in JEQ:
This article has been cited by other articles:
|
|
 |
|
  C. K. Keller, C. N. Butcher, J. L. Smith, and R. M. Allen-King Nitrate in Tile Drainage of the Semiarid Palouse Basin J. Environ. Qual., March 1, 2008; 37(2): 353 - 361. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. D. Nance, L. R. Gibson, and D. L. Karlen Soil Profile Nitrate Response to Nitrogen Fertilization of Winter Triticale Soil Sci. Soc. Am. J., June 29, 2007; 71(4): 1343 - 1351. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
