| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
USDA-ARS, National Soil Tilth Laboratory, 2150 Pammel Drive, Ames, IA 50011
* Corresponding author (jaynes{at}nstl.gov).
Received for publication May 5, 2003.
 |
ABSTRACT |
|---|
 TOPNOTES ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
30% decrease for NO3 concentrations in surface water.
Abbreviations: AR, autoregressive • LSNT, late spring nitrate test • MCL, maximum contaminant level
|
INTRODUCTION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
The relationship between NO3 concentrations in tile drainage and N fertilizer management has been extensively studied for continuous corn and corn–soybean [Glycine max (L.) Merr.] production systems and is predominately affected by fertilizer rate and timing. For example, Baker and Johnson (1981) found that increasing the fertilizer rate from 100 to 250 kg ha–1 on corn grown in rotation with either soybean or oat (Avena sativa L.) doubled the NO3 concentration in tile drainage from 20 to 40 mg N L–1. Similarly, for N fertilizer applied to corn in a corn–soybean rotation, Jaynes et al. (2001) found yearly average mass losses of NO3 in tile drainage of 29 and 48 kg N ha–1 for average N fertilizer rates of 62 and 187 kg N ha–1, respectively. Thus, applying the appropriate rate of N fertilizer is critical for minimizing NO3 concentrations in tile drainage.
Timing of N fertilizer application can have a substantial effect on tile drainage NO3 concentration. More than half of the corn land in Iowa and Illinois receives fall application of N, primarily as anhydrous ammonia (Hatfield et al., 1999; Economic Research Service, 1999; Shankar et al., 2000; Dinnes et al., 2002). This practice greatly increases the risk that the applied N will nitrify and leach with fall and spring precipitation. In a study reported by Randall and Mulla (2001), annual losses of NO3 in tile drainage averaged 36% higher with fall application compared with spring application of N for corn production.
Therefore, to reduce N losses in tile drainage from corn production, "the correct rate of N at the optimum time" must be applied (Randall and Mulla, 2001). Combining a split application of N with the pre-sidedress nitrate test (PSNT) proposed by Magdoff et al. (1984) and confirmed for a range of conditions (Blackmer et al., 1989; Roth et al., 1992; Klausner et al., 1993; Sims et al., 1995) is one approach for accomplishing both of these goals. With the PSNT, the quantity of NO3 in the surface 30 cm of soil is measured when the corn is 15 to 30 cm tall. If the soil NO3 content is below a critical level, additional N fertilizer is immediately sidedressed. Currently, the PSNT in the form of the late spring nitrate test (LSNT) is the recommended practice for N fertilization of corn in the state of Iowa (Blackmer et al., 1997) and is suggested as an option in surrounding states as well. A typical scenario for using the LSNT involves applying a nominal rate of N fertilizer before corn emergence followed by measuring residual soil NO3 in the top 30 cm of the soil during early crop growth and sidedressing additional N fertilizer based on soil NO3 concentrations.
Plot-scale and field studies using a PSNT approach have been positive with regard to measured or potential NO3 leaching (Durieux et al., 1995; Bjorneberg et al., 1998; Guillard et al., 1999; Bakhsh et al., 2002). However, the effect on surface water quality at a watershed scale has not been quantified. The objective of this research was to quantify changes in NO3 concentration in surface flow as a result of implementing the LSNT N fertilizer management system across a watershed. Our goal was to modify N management within a watershed to lower NO3 concentrations below the 10 mg N L–1 MCL, while maintaining economically viable crop production levels.
|
MATERIALS AND METHODS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Treatment and control subbasins within Walnut Creek watershed were selected for this study by their similar size, soils, and position within the upper, extensively tiled reaches of the watershed. A 405-ha subbasin (TR in Fig. 1) was selected as the subbasin for implementation of the LSNT N fertilizer management program. Cooperative research agreements were secured with the eight farmers operating sixteen of the fields that were either completely or partially within the subbasin. These fields represented 90% of the area within the subbasin. Subbasins CN1 (491 ha) and CN2 (863 ha) were selected as controls for a paired-watershed research design (Clausen et al., 1996). Farmers within these control subbasins were encouraged to continue following their normal production practices.
|
|
Soil NO3 results for all the blocks within a given field were averaged and a single N fertilizer application rate computed for the field. Nitrogen fertilizer rates were calculated using the formula:
 | [1] |
Tile Drainage Sampling
The fields within each subbasin were extensively drained by subsurface field tiles that had been installed over the past 120 yr. The field tiles drained into subsurface drainage district pipes that drained each subbasin. The partially submerged district drains were instrumented to measure flow rate as they emptied into Walnut Creek (Fig. 1) by simultaneously measuring water depth and velocity using Flowtote meters (Marsh-McBirney, Frederick, MD). Water samples were taken manually once a week at the flow gage on each subbasin. All water samples were refrigerated until analysis. Nitrate was analyzed by quantitative reduction to NO2 and measuring the NO2 concentration colorimetrically with a Lachat Autoanalyzer (Zellweger Analytics, Lachat Instrument Division, Milwaukee, WI). The method had a quantitation limit of 1.0 mg N L–1 as NO3. Flow-weighted yearly average NO3 concentrations were computed by summing the product of the weekly NO3 concentration and total weekly discharge.
Grain Yield
Grain yield from both corn and soybean were measured with the farmers' combines that were equipped with yield monitors and differential global positioning system (DGPS). Seven of the eight farmers installed Yield Monitor 2000 yield monitors (Ag Leader Technology, Ames, IA) and DGPS systems that used the U.S. Coast Guard differential correctional signal. The remaining farmer installed a GreenStar (Deere and Co., Moline, IL) and DGPS system that used an orbiting satellite source differential correction signal. Each farmer-cooperator was guided in calibration and use of the yield monitors. Yield data were organized with ArcView geographical information system software (ESRI, 2002) for mapping and for preparing further statistical analyses.
Corn yield data were first censored by eliminating data points below 0.63 Mg ha–1 and above 18.9 Mg ha–1. These limits were selected to remove yield monitor source errors due to beginning and ending harvest passes, and factors causing inconsistent grain flow through the combines not due to yield variation (i.e., plugging within the combine platform). Yield data were then mapped by individual data points and organized by field and N treatment for all fields and years of study.
Yield information from the nonlimiting N field-long strips and neighboring combine passes within the LSNT N rate areas were identified by GPS benchmarks of the strip borders and labels applied to the combine harvest passes by the farmers. A single harvest pass was selected from each N rate area to obtain data free of border effects from the different N rates.
Corn yields for the LSNT program were evaluated by comparing the yield within the nonsidedressed and nonlimiting N strips within each field to the yield from the adjacent LSNT yield strip. Comparisons were accomplished by using SAS one-way analysis of variance (ANOVA) and grouping the yields for all fields by year (SAS Institute, 1990). In this paper, we summarize only the basin-wide results for yields from the nonlimiting N strips and adjacent LSNT-recommended N rate. Detailed yield results for individual fields and for the nonsidedressed strips will be presented in a future paper.
Statistical Comparison of Paired Watersheds
Analysis of covariance methods (ANCOVA) have been recommended for examining paired watershed data (Grabow et al., 1998, 1999a, 1999b). Because serial correlation commonly exists in stream monitoring data and model residuals, which can cause underestimation of error variance (Salas, 1993), either time aggregation over hydrological events or autoregressive ANCOVA models have been used as comparison methods. However, Meek et al. (2001) pointed out that several of the underlying assumptions for these approaches, in particular that the covariate is fixed, measured without error, and independent of treatments, are not valid for stream monitoring datasets.
Instead, the method described by Meek et al. (2001) and used by Jaynes et al. (2001) was used to compare between the NO3 concentrations in the LSNT-treated subbasin, TR1, and the concentrations from the control subbasins, CN1 and CN2. In this method, a model was fitted to the difference in paired weekly NO3 concentrations between the treated and each control subbasin. The model had two parts: a trend component and an autoregressive (AR) residual component to correct for the effects of residual autocorrelation. While conceptually the trend component can be almost any known function from a line to intrinsically nonlinear functions, we used rational and logistic polynomials. A nonlinear regression procedure was used to fit an appropriate trend model to the NO3 concentration difference data. Autocorrelation of the residuals was used to select the appropriate lag for the AR component. In all cases, a Lag 1 model was found to be appropriate. A Lag 1 residual AR component was added to the trend model and the combined model simultaneously fitted to the data using an iterative least squares method and Ramsay's weighting function to reduce the effect of outliers (Montgomery and Peck, 1982). Residual lag values were set to zero at the start of the time sequence and after breaks in the time series caused by periods of no drainage flow.
In addition to computing the model parameters, a 95% confidence interval was computed for each model. While many formal statistical tests are possible, for practical purposes, we assumed that there was a significant difference between the NO3 concentrations in two subbasins whenever the 95% confidence intervals for the modeled differences no longer included the 0 line (i.e., rejected the null hypothesis that the difference in concentrations was 0). This graphical presentation used ideas suggested in Tufte (1983) and made the method of analysis easy to interpret. Models were fitted independently for each subbasin comparison and for the pretreatment (1992–1996) and treatment time periods (1997–2000). All modeling was conducted using SAS Version 6 software (SAS Institute, 1990).
|
RESULTS AND DISCUSSION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
During the treatment period, 2000 was also a very dry year with the monthly precipitation falling below the 30-yr average for the last quarter of 1999 and most of 2000. The low precipitation levels caused the tile drains to cease flowing and even Walnut Creek to stop flowing for much of the summer of 2000. None of the years within the treatment period could be termed flood years although precipitation in June of 1998 resulted in extensive flooding of the local potholes and may have affected the efficacy of the sidedressed N (discussed later). The greater than average November precipitation in 1996 also affected the study because the farmers were for the most part unable to apply their N that fall due to wet soil conditions and deferred N application until the spring of 1997. Changing from fall to spring N application for 1997 may have reduced NO3 leaching losses that year (Randall et al., 2003) and reduced the difference between the LSNT treated and control subbasins. Overall, the marked variability of temperature and precipitation illustrates why watershed studies need to be conducted over extended periods of time. For this study, both the pretreatment and treatment periods had similar temperature regimes and included both wet and dry growing seasons.
During the treatment period of 1997–2000, corn was grown on 49% of Subbasin CN1, 44% of Subbasin TR1, and 45% of Subbasin CN2. For the same period, soybean was grown on 42% of Subbasin CN1, 46% of Subbasin TR1, and 51% of Subbasin CN2. Compared with the pretreatment period, the percent areas planted to corn and soybean during the treatment period were very similar for Subbasins CN1 and CN2. Corn was grown on relatively less area on average for the treatment Subbasin TR1 and soybean on relatively more area during the treatment period. This was primarily due to a single 32-ha field that switched from a corn after corn rotation in 1992–1993 to a corn after soybean rotation for the remainder of the study. Overall, corn was grown on nearly half of the subbasin areas during the treatment period and corn and soybean accounted for about 90% of the area within each subbasin.
Nitrogen Fertilizer Rates
In 1997, the LSNT program did not reduce overall N application rates compared with the farmer-cooperators' normal N programs (Table 2). However, the LSNT program dramatically reduced N rates in 1998 and 2000, and moderately reduced overall N rates in 1999, compared with what the farmers would have applied. The farmer-cooperators increased their N fertilization rates by about 20 kg ha–1 during the second half of the 4-yr period. The increased N fertilization rates were a result of the farmers' perception that they were losing a large amount of the fall-applied N fertilizer because of the wetter than average spring weather and they wanted to compensate for this loss.
|
Actual N fertilizer rates applied to corn within Subbasin TR1 were different than recommended by the LSNT in 1999 and 2000. In 1999, a farmer mistakenly applied N in the fall to one of the fields at a rate of 168 kg ha–1. Again in 2000, N fertilizer rates applied were greater than the LSNT recommendation because a farmer mistakenly applied N to one field in the fall and a second farmer withdrew from the program and also applied N in the fall. Thus, the experimental design was compromised slightly from the original plan because not all corn fields within the treatment subbasin received N as a spring split application at the LSNT-recommended rate. These deviations would have tended to diminish the effect of the LSNT program on NO3 concentrations within the treatment subbasin.
Yield
Average corn yields for 1997 through 2000 are shown in Table 2. Yields from the LSNT-treated strips were not significantly different (P = 0.05) than the nonlimiting N treatment strips in any year. The lowest mean LSNT yield relative to the mean of the nonlimiting N strips yields was in 1998. In this year, a warm, dry spring caused little leaching or denitrification of the NO3 that had mineralized. This produced relatively high soil NO3 concentrations and correspondingly low recommended rates of sidedressed N. The sidedressing in June was followed by a 3-wk period when precipitation was 150 mm greater than average (Table 1). Much of the applied urea ammonium nitrate (UAN) may have leached below the root zone during this period, thus causing the lower yields. Deep leaching of N would agree with observations by Jaynes et al. (1992) that solute leaching may be enhanced when irrigation immediately follows application. Kluitenberg and Horton (1990) attributed this to enhanced preferential flow for solutes that were recently applied and had not diffused or been incorporated into the soil pore matrix away from the preferential pathways for leaching. Overall, the LSNT produced insignificantly lower yields in all four years.
Water Quality
Nitrate concentrations in the weekly water samples from the three subbasins ranged from 1 to 25 mg N L–1 (Fig. 3)
. Nitrate concentrations exhibited marked seasonal patterns from 1995 through 2000 with concentrations rising from late fall to mid-summer and then dropping to lows in late summer. This pattern has been reported for watershed data across the Midwest (Fenelon and Moore, 1998; Moog and Whiting, 2002). Increasing concentrations are probably due to the flushing by snow melt and spring rains of NO3 mineralized from organic matter and from N fertilizer. Decreasing concentrations in late summer are probably due to removal of NO3 from soil by growing crops. The seasonal pattern is not as obvious in the 1993 and 1994 data. In 1993, the excessive rain and runoff probably diluted the NO3, lowering the concentration. In 1994, little springtime leaching occurred because of low rainfall and because much of the soil NO3 had been flushed from the system in 1993. The high concentrations in 1995 and 1996 may reflect leaching of soil NO3 that accumulated in the soil from fertilizer application and mineralization in 1994 but had not leached that year due to low rainfall. Higher concentrations in tile drainage in years following dry years have been observed elsewhere in the Midwest (Randall and Iragavarapu, 1995).
|
|
To compare the three subbasins during the pretreatment period, a combined trend and AR model was fitted to the time series of the paired differences in weekly measured NO3 concentrations in each subbasin. Using the criterion that there was no significant difference in NO3 concentrations if the 95% confidence limits for the fitted model included 0, there were no significant differences between NO3 concentrations in Subbasins TR1 and CN1 other than a 2-mo period in the fall of 1995 and a week in the fall of 1996 (data not shown). Likewise, the 95% confidence limits for the model fitted to the time series for the difference in NO3 concentration between Subbasin CN2 and TR1 showed no time period when the difference was significantly different than 0 (data not shown). Thus, the NO3 concentrations in the drainage water leaving the three subbasins were consistent and not significantly different during the pretreatment period.
Conversely, the time series of paired NO3 concentration differences exhibited markedly different behavior during the LSNT treatment years. Comparing Subbasins TR1 to CN1, the NO3 concentration difference time series exhibited an initial period when the concentrations were higher in TR1 than CN1. This was followed by a period of increasingly greater NO3 concentrations in CN1 that continued to the end of the treatment period (Fig. 4)
. A rational quadratic polynomial model with an AR residual component was fitted to the concentration difference series giving:
 | [2] |
TR1–CN1 is the NO3 concentration in Subbasin TR1 minus the concentration in Subbasin CN1 (mg N L–1), 
is the residual or error, and i is an index representing the number of weeks from 1 Jan. 1997. The model fit the data well (R2 = 0.86) and describes a nearly constant period in the NO3 concentration difference for about 33 weeks or until mid-August 1997 after which the NO3 concentration in the water coming from the LSNT-treated subbasin started to decrease in relation to the water coming from Subbasin CN1. This decrease continued, with the mean difference being about 8 mg L–1 by the end of 2000. In the later half of 1998, the 95% confidence bands for the model no longer included the null hypothesis that the concentration difference was zero and thus, the NO3 concentration coming from LSNT-treated subbasin can be considered significantly lower than the NO3 concentration coming from the control subbasin after this time.
|
 | [3] |
TR1–CN2 is the difference between the weekly NO3 concentrations coming from Subbasins TR1 and CN2 (mg N L–1). The NO3 concentration in Subbasin TR1 was initially lower than that in Subbasin CN2 and, similar to the Subbasin CN1 comparison, remained constant through much of 1997. The logistic model showed the NO3 concentration difference to become increasingly negative starting about week 50 (mid-December 1997) but then leveling off early in 1998 to an average difference of about 4.5 mg N L–1. Using the 95% confidence limits of the model as the criteria, the NO3 concentration in Subbasin TR1 was significantly lower than that in Subbasin CN2 by early 1998.
|
Examining Table 3 and Fig. 3 it is apparent that much of the difference in NO3 concentrations between the LSNT and control subbasins was not because of decreases in NO3 concentration coming from the LSNT subbasin. The flow-weighted annual average NO3 concentration in the LSNT-treated subbasin decreased about 1 mg N L–1 during the treatment period compared with the pretreatment period. Conversely, the NO3 concentration coming from the control subbasins tended to increase during the four years and be at or above the highest concentrations observed during the pretreatment years. Whether the increases in NO3 concentrations in the control watershed were due to variations in weather, increases in N fertilization rates, or changes in other cultural practices in the control subbasins cannot be determined from this study. Nevertheless, if the LSNT program had been adopted throughout the Walnut Creek watershed, a considerable reduction in NO3 concentration leaving the watershed would have been realized.
|
CONCLUSIONS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Despite the water quality advantages, there are drawbacks to using the LSNT approach. The test is often unable to identify soils not responsive to N fertilizer (Bundy et al., 1999). The factor of 8 used in Eq. [1] is at best an approximation and should be modified as additional information for specific fields is acquired. Also, crop yields may suffer when using the LSNT approach depending on the timing between sidedressing and rainfall. Adopting the LSNT approach also increases the risk to the farmer. The window of opportunity for sidedressing N using standard equipment is fairly short and coincides with typically rainy weather in the Midwest. Wet soil conditions could potentially delay sidedressing and prevent timely application of needed N. Development of high-clearance application equipment may help reduce this risk in the future. For these reasons, the LSNT approach used here should be considered only an initial start, to be replaced or refined as new research emerges.
The 30% relative reduction in NO3 concentration observed at the end of this study is comparable with the suggested decrease for N discharges from the Mississippi River to manage hypoxia in the northern Gulf of Mexico (Mississippi River/Gulf of Mexico Watershed Nutrient Task Force, 2001). The reduction was achieved by reducing the application of N fertilizer within the subbasin by 23% as compared with the cooperating farmers' N program in the last 2 yr. This gives a ratio for the percent reduction in NO3 concentration to percent reduction in fertilizer N of about 1.3. Randall and Mulla (2001), working with small continuous-corn plots, also found a ratio of 1.3 when N fertilizer rates were reduced and application was switched from fall to spring. The similar ratios found at the two different spatial scales gives us encouragement that similar responses could be realized by applying the LSNT approach at even larger spatial scales.
In the past, criticisms have been directed at best management practices promoted by public institutions that have not been proven to function adequately on an applied scale such as a farm or watershed. This project extended over 400 ha comprising eight different farmers' production fields and 11 different soil types. Our results show that implementing an intensive N management strategy across this agricultural watershed resulted in lower NO3 concentrations leading to improved water quality.
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
REFERENCES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Related articles in JEQ:
This article has been cited by other articles:
|
|
 |
|
  V. Nangia, P. H. Gowda, D. J. Mulla, and G. R. Sands Water Quality Modeling of Fertilizer Management Impacts on Nitrate Losses in Tile Drains at the Field Scale J. Environ. Qual., March 1, 2008; 37(2): 296 - 307. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. W. Simpson, A. N. Sharpley, R. W. Howarth, H. W. Paerl, and K. R. Mankin The New Gold Rush: Fueling Ethanol Production while Protecting Water Quality J. Environ. Qual., March 1, 2008; 37(2): 318 - 324. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
X. Hu, G. F. McIsaac, M. B. David, and C. A. L. Louwers Modeling Riverine Nitrate Export from an East-Central Illinois Watershed Using SWAT J. Environ. Qual., May 25, 2007; 36(4): 996 - 1005. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. B. Jaynes and T. S. Colvin Corn Yield and Nitrate Loss in Subsurface Drainage from Midseason Nitrogen Fertilizer Application Agron. J., October 3, 2006; 98(6): 1479 - 1487. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. R. Ball Coelho, R. C. Roy, and A. J. Bruin Optimization of Liquid Swine Manure Sidedress Rate and Method for Grain Corn Agron. J., August 17, 2005; 97(5): 1322 - 1332. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. L. Karlen, D. L. Dinnes, D. B. Jaynes, C. R. Hurburgh, C. A. Cambardella, T. S. Colvin, and G. R. Rippke Corn Response to Late-Spring Nitrogen Management in the Walnut Creek Watershed Agron. J., June 17, 2005; 97(4): 1054 - 1061. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. J. Kladivko, J. R. Frankenberger, D. B. Jaynes, D. W. Meek, B. J. Jenkinson, and N. R. Fausey Nitrate Leaching to Subsurface Drains as Affected by Drain Spacing and Changes in Crop Production System J. Environ. Qual., September 1, 2004; 33(5): 1803 - 1813. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||
