Publications
12-2017
Simulation of Daily Flow Pathways, Tile-Drain
Nitrate Concentrations, and Soil-Nitrogen
Dynamics Using SWAT
FYRA Engineering
Michelle L. Soupir
Iowa State University, [email protected]
Matthew J. Helmers
Iowa State University, [email protected]
William G. Crumpton
Iowa State University, [email protected]
Jeffrey G. Arnold
United States Department of Agriculture
See next page for additional authors
Follow this and additional works at:
http://lib.dr.iastate.edu/abe_eng_pubs
Part of the
Bioresource and Agricultural Engineering Commons
,
Environmental Indicators and
Impact Assessment Commons
,
Natural Resources Management and Policy Commons
, and the
Water Resource Management Commons
The complete bibliographic information for this item can be found at
http://lib.dr.iastate.edu/
abe_eng_pubs/857
. For information on how to cite this item, please visit
http://lib.dr.iastate.edu/
howtocite.html
.
This Article is brought to you for free and open access by the Agricultural and Biosystems Engineering at Iowa State University Digital Repository. It has been accepted for inclusion in Agricultural and Biosystems Engineering Publications by an authorized administrator of Iowa State University Digital Repository. For more information, please [email protected].
and Soil-Nitrogen Dynamics Using SWAT
Abstract
Tile drainage significantly alters flow and nutrient pathways and reliable simulation at this scale is needed for
effective planning of nutrient reduction strategies. The Soil and Water Assessment Tool (SWAT) has been
widely utilized for prediction of flow and nutrient loads, but few applications have evaluated the model's
ability to simulate pathway-specific flow components or nitrate-nitrogen (NO3-N) concentrations in
tile-drained watersheds at the daily time step. The objectives of this study were to develop and calibrate SWAT
models for small, tile-drained watersheds, evaluate model performance for simulation of flow components and
NO3-N concentration at daily intervals, and evaluate simulated soil-nitrogen dynamics. Model evaluation
revealed that it is possible to meet accepted performance criteria for simulation of monthly total flow,
subsurface flow (SSF), and NO3-N loads while obtaining daily surface runoff (SURQ), SSF, and NO3-N
concentrations that are not satisfactory. This limits model utility for simulating best management practices
(BMPs) and compliance with water quality standards. Although SWAT simulates the soil N-cycle and most
predicted fluxes were within ranges reported in agronomic studies, improvements to algorithms for soil-N
processes are needed. Variability in N fluxes is extreme and better parameterization and constraint, through
use of more detailed agronomic data, would also improve NO3-N simulation in SWAT. Editor's note: This
paper is part of the featured series on SWAT Applications for Emerging Hydrologic and Water Quality
Challenges. See the February 2017 issue for the introduction and background to the series.
Keywords
tile drainage, hydrology, nitrate export, SWAT, denitrification, soil nitrogen
Disciplines
Bioresource and Agricultural Engineering | Environmental Indicators and Impact Assessment | Natural
Resources Management and Policy | Water Resource Management
Comments
This article is published as Ikenberry, Charles D., Michelle L. Soupir, Matthew J. Helmers, William G.
Crumpton, Jeffrey G. Arnold, and Philip W. Gassman. "Simulation of Daily Flow Pathways, Tile‐Drain Nitrate
Concentrations, and Soil‐Nitrogen Dynamics Using SWAT." JAWRA Journal of the American Water
Resources Association 53, no. 6 (2017): 1251-1266. doi:
10.1111/1752-1688.12569
. Posted with permission.
Rights
Works produced by employees of the U.S. Government as part of their official duties are not copyrighted
within the U.S. The content of this document is not copyrighted.
Authors
FYRA Engineering, Michelle L. Soupir, Matthew J. Helmers, William G. Crumpton, Jeffrey G. Arnold, and
Philip W. Gassman
SIMULATION OF DAILY FLOW PATHWAYS, TILE-DRAIN NITRATE CONCENTRATIONS,
AND SOIL-NITROGEN DYNAMICS USING SWAT
1Charles D. Ikenberry, Michelle L. Soupir, Matthew J. Helmers, William G. Crumpton, Jeffrey G. Arnold, and Philip W. Gassman2
ABSTRACT: Tile drainage significantly alters flow and nutrient pathways and reliable simulation at this scale is needed for effective planning of nutrient reduction strategies. The Soil and Water Assessment Tool (SWAT) has been widely utilized for prediction of flow and nutrient loads, but few applications have evaluated the mod-el’s ability to simulate pathway-specific flow components or nitrate-nitrogen (NO3-N) concentrations in tile-drained watersheds at the daily time step. The objectives of this study were to develop and calibrate SWAT mod-els for small, tile-drained watersheds, evaluate model performance for simulation of flow components and NO3 -N concentration at daily intervals, and evaluate simulated soil-nitrogen dynamics. Model evaluation revealed that it is possible to meet accepted performance criteria for simulation of monthly total flow, subsurface flow (SSF), and NO3-N loads while obtaining daily surface runoff (SURQ), SSF, and NO3-N concentrations that are not satisfactory. This limits model utility for simulating best management practices (BMPs) and compliance with water quality standards. Although SWAT simulates the soil N-cycle and most predicted fluxes were within ranges reported in agronomic studies, improvements to algorithms for soil-N processes are needed. Variability in N fluxes is extreme and better parameterization and constraint, through use of more detailed agronomic data, would also improve NO3-N simulation in SWAT. Editor’s note: This paper is part of the featured series on SWAT Applications for Emerging Hydrologic and Water Quality Challenges. See the February 2017 issue for the introduction and background to the series.
(KEY TERMS: tile drainage; hydrology; nitrate export; SWAT; denitrification; soil nitrogen.)
Ikenberry, Charles D., Michelle L. Soupir, Matthew J. Helmers, William G. Crumpton, Jeffrey G. Arnold, and Philip W. Gassman, 2017. Simulation of Daily Flow Pathways, Tile-Drain Nitrate Concentrations, and Soil-Nitrogen Dynamics Using SWAT. Journal of the American Water Resources Association (JAWRA) 53(6): 1251-1266. https://doi.org/10.1111/1752-1688.12569
INTRODUCTION
Artificial subsurface drainage (i.e., tile drainage) allows row crop production and improves crop yields
in poorly drained soils by lowering the water table to limit saturation of the root zone and prevent root aer-ation stress (Hatfield et al., 1998), and by increasing planting and harvest windows during spring and fall, respectively. Streamflow and nutrient transport are
1
Paper No. JAWRA-16-0116-P of the Journal of the American Water Resources Association (JAWRA). Received May 11, 2016; accepted June 23, 2017.© 2017 American Water Resources Association. Discussions are open until six months from issue publication.
2
Senior Engineer (Ikenberry), FYRA Engineering, 100 Court Avenue (Suite 202), Des Moines, Iowa 50309; Associate Professor (Soupir) and Professor (Helmers), Department of Agricultural & Biosystems Engineering, Professor (Crumpton), Department of Ecology, Evolution & Organismal Biology, and Research Engineer (Gassman), Center for Agricultural & Rural Development, Iowa State University, Ames Iowa 50011; and Agricultural Engineer (Arnold), Grassland Soil & Research Laboratory, USDA-Agricultural Research Service, Temple, Texas 76502 (E-Mail/Ikenberry: [email protected]).
significantly impacted by subsurface drainage because tile drains alter the pathways and processes that govern hydrology and nutrient cycling (Schilling and Helmers, 2008). The distribution of water bal-ance components; runoff, lateral flow, shallow groundwater flow, and aquifer recharge; differ in tiled versus nontiled watersheds (Goswami et al., 2008; Sui and Frankenberger, 2008). The presence of tile drainage also impacts water quality processes by reducing surface runoff and associated sheet and rill erosion, increasing soil aeration, thereby increasing mineralization and reducing denitrification, and increasing nitrate-nitrogen (NO3-N) leaching to sur-face water (Dinnes et al., 2002; El-Sadek et al., 2002; Coelho et al., 2012; Boles et al., 2015). Proper identifi-cation and quantifiidentifi-cation of these pathways and pro-cesses is critically important for reliable prediction of nonpoint source pollutant loads (Goolsby et al., 2000) and quantifying nutrient transport to downstream waterbodies (e.g., the Mississippi River and Gulf of Mexico) (Alexander et al., 2008; David et al., 2010; Stenback et al., 2011). Additionally, design and simu-lation of best management practices (BMPs) and strategies to mitigate negative effects of tile drainage require thorough understanding of the underlying hydrologic and water quality processes (Rozemeijer et al., 2010; Yen et al., 2014).
The Soil and Water Assessment Tool (SWAT) model is a well-established and widely utilized model for simulation of hydrology and pollutant transport in predominantly agricultural watersheds. The model explicitly accounts for both tile drainage and soil nutrient cycling and is under continuous develop-ment/improvement by U.S. Department of Agricul-ture-Agricultural Research Service (USDA-ARS). SWAT has been extensively applied worldwide for many types of water resource problems across a wide spectrum of watershed scales and conditions (Gass-man et al., 2014; Bressiani et al., 2015; Krysanova and White, 2015). Recognizing its extensive use, Arnold et al. (2012) published guidance on the use, calibration, and validation of SWAT models and detailed performance measures and evaluation crite-ria were set forth by Mocrite-riasi et al. (2015).
Reliable models for simulating hydrology and nutrient transport in tile-drained landscapes are crit-ically needed but particularly challenging. Calibra-tion of SWAT and other watershed models often relies heavily on iterative adjustment of a large num-ber of parameters during calibration. Calibration is typically performed to minimize differences between predicted and observed flow and/or pollutant loads at large spatial (greater than HUC-12) and temporal (monthly and annual) scales, and pathway-specific flow and daily data are often not available (Boles et al., 2015). A problem frequently encountered
during the calibration process is that optimized parameter values often produce intermediate pro-cesses/flux values that are unrealistic (Malone et al.,
2015). Even when this problem is recognized,
observed data needed to constrain parameter values and intermediate fluxes are often lacking. As a result, performance criteria for nonpathway specific vari-ables such as streamflow or nutrient loads may appear reasonable, but underlying simulation of sur-face runoff (SURQ), subsursur-face flow (SSF), nutrient transport, and N-dynamics (e.g., denitrification and soil-N levels) may be misrepresented (Yen et al., 2014; Arnold et al., 2015). These challenges can limit the model’s utility for accurately forecasting flow and nutrient transport across spatial scales, through varying weather patterns, with land-use changes, and with implementation of water quality improve-ment strategies. SWAT’s framework makes such com-parative analysis relatively simple, although results may be deceiving if the baseline model is deemed accurate but is right for the wrong reasons.
This study examines the performance of SWAT in simulating hydrology and NO3-N transport in small, tile-drained watersheds (200-1,000 ha) typical of drai-nage districts in north-central Iowa. The goals of this study were to evaluate simulation of hydrology and NO3-N dynamics and to provide deeper insights into model performance. Specific objectives were to (1) develop and calibrate SWAT models for small, tile-drained watersheds, (2) evaluate model performance for pathway-specific flow and NO3-N simulation at daily intervals, and (3) document important interme-diate processes and N-fluxes, such as denitrification, mineralization, crop uptake, and soil-NO3-N storage.
MATERIALS AND METHODS
Study Area
The two watersheds simulated in this study each drain to Conservation Reserve Enhancement Program (CREP) wetlands located in the Des Moines Lobe ecoregion in north-central Iowa (Ikenberry et al., 2017). The 309-ha KS watershed is located in Story County, Iowa, at the headwaters of a first-order tribu-tary to Squaw Creek, a HUC-12 watershed in the South Skunk River basin. The 227-ha AL watershed is located in Kossuth County approximately 120 km northwest of the KS site (Figure 1) and drains to a first-order stream that enters Black Cat Creek, a HUC-12 watershed that discharges to the Des Moines River. Watershed characteristics for both sites are reported in Table 1. All soils in the watersheds are
classified as somewhat poorly drained to very poorly drained, with the exception of Clarion soils, which are moderately well-drained. Therefore, hydrologic response units (HRUs) with Clarion soils were not parameterized with tile drainage, but all other HRUs include tile drain parameters.
The monitoring strategy was designed and imple-mented as part of the CREP wetland monitoring described by Crumpton et al. (2006). This study utilized four years of data collected upstream of the wetland at each site: 2008-2011 for the KS watershed, and 2007-2010 for the AL watershed. NO3-N concentrations were measured using automated samplers that collected daily composite samples during the ice free season sup-plemented by grab samples collected approximately weekly throughout the year. Streamflows were esti-mated from stage-discharge equations developed using a combination of stage recorders, submerged area veloc-ity meters, and point measures of discharge.
A two end-member mixing model based on NO3-N concentrations was used to separate measured dis-charge (Qt) into surface runoff (Qs) and subsurface flow (Qss) end-members (Crumpton et al., in prepara-tion), similar to the approach described by Tomer
FIGURE 1. Map of Watersheds and Sampling Locations. The shaded region within the state boundary is the Des Moines Lobe ecoregion. Watershed/tile flow and water quality sampling were collected at Conservation Reserve Enhancement Program (CREP) wetland inflow sites.
TABLE 1. Characteristics of Study Sites.
Characteristic KS Watershed AL Watershed
Drainage area, DA (ha) 309 227
Row crop (% of DA) 93 80
Continuous corn (% of row crop) 35 14
Poor drainage (% of DA)1 62 77
Annual rainfall (mm)2 1,081 906
Annual water yield (mm)3 395 279
1Row crop areas with slopes
<5% and soils classified as somewhat poor to poorly-drained.
2Average annual rainfall during model simulation period
(2008-2011 for KS Watershed, 2007-2010 for AL Watershed).
et al. (2010). This separation relied on a water bal-ance given by
Qt¼ Qsþ Qss ð1Þ
and a NO3-N mass balance given by
NtQt¼ NsQsþ NssQss ð2Þ
In these equations, N refers to NO3-N concentra-tion, Q refers to the discharge, and the subscripts t, s, and ss refer to the total flow, surface flow, and sub-surface flow, respectively. Combining these equations to solve for the subsurface flow to total flow ratio (r=Qss/Qt) gives
rt¼ ðNt NsÞ=ðNss NsÞ ð3Þ
Over the four-year periods analyzed for these two watersheds, the percent surface runoff estimated from end-member analysis of individual events ran-ged from near 0 to 66% and averaran-ged 12-15% of event flow over a 10-fold range of estimated Ns. These results are consistent with those reported in prior work on Corn Belt systems (Stone and Wilson, 2006; Schilling and Helmers, 2008; Tomer et al., 2010; Vidon et al., 2012).
SWAT Model Development
Watershed delineations were based on the Light Detection and Ranging (LiDAR) data developed for the State of Iowa in 2010. The Iowa Department of Natural Resources—GIS Section aggregated local LiDAR data to a resolution of one square meter, and hydraulically reinforced the data to incorporate cul-verts and bridges that convey water through embank-ments (e.g., roadways). Both watersheds have low topographic relief, with most slopes between 0 and 2% and many enclosed depressions.
Sources of climatic data include the National Cli-matic Data Center Weather Data Library database (NOAA, 2013) and the National Weather Service COOP data available through the Iowa Environmental Mes-onet (Iowa State University, 2014). Weather station data included daily rainfall and maximum and mini-mum daily temperature. The closest weather station to the KS Wetland is located in Ames, Iowa, and data from the weather station in Algona, Iowa, was used for model input in the AL Wetland watershed. Both sta-tions are less than 10 miles from the watersheds. Solar radiation, wind speed, and relative humidity were sim-ulated by the weather generator within SWAT.
The USDA National Agricultural Statistics Service (NASS) cropland data layer (CDL) for the years 2005
through 2010 was obtained and used to assess land use and crop rotations. The 2010 NASS land cover was verified by windshield surveys conducted in early spring 2011. Soils data are from the Soil Survey Geo-graphic Data (SSURGO) database developed by NRCS. Based on the area of land with soils being somewhat poorly, poorly, or very poorly drained, it is estimated that 62% of the KS watershed is tile-drained (Table 1). Hydrologic soil group B/D is dominant, with class B applied to HRUs with tile drainage. Soil data include three or four soil layers, depending on soil type, with layer-specific values for bulk density, saturated hydraulic conductivity, and percent sand/silt/clay. Soils in the KS Wetland watershed include Canisteo, Clarion, Harps, Nicollet, and Webster, with Clarion and Webster soils comprising 67% of the watershed. The AL watershed is more intensely drained, with 77% of soils being at least somewhat poorly drained. Soil classifications include Canisteo, Clarion, Nicollet, Okoboji, Storden, and Webster, with 90% of the water-shed consisting of Canisteo, Nicollet, or Clarion soils.
SWAT applications typically simulate a large watershed comprised of many subbasins. Because this case study was undertaken to improve tile flow predictions at the drainage-district scale, the water-shed models each have only one small subbasin, which is representative of the local drainage district. Subbasins in SWAT are divided into HRUs that have unique combinations of land use, soil type, and slope classification. Although HRUs represent real-world locations, they are not spatially contiguous and are lumped at the subbasin level within the SWAT framework. Water and pollutants generated in each HRU are aggregated at the subbasin level before being routed in the reach network of the SWAT model.
During HRU development, threshold values were used to filter areas of land use, soil, and slope. Both watershed models included thresholds of 3% for land use, 5% for soil type, and 5% for slope classification. As a result, land uses that comprise less than 3% of a sub-basin were removed and the area was redistributed to the relative percentages of the remaining (non-filtered) land uses in each subbasin. Similarly, soils comprising less than 5% of any land uses were filtered, as well as slopes comprising less than 5% of any soil group. The filtering process resulted in 17 individual HRUs in the KS Wetland watershed with an average area of 18.2 ha. The AL watershed model was filtered to 26 HRUs with an average size of 8.7 ha.
Crop Rotation and Fertilizer Application
Land use was determined using the USDA NASS CDL for the years 2005 through 2010. The majority
of row crop production consists of two-year rotations of corn (Zea mays L.) and soybeans [Glycine max (L.) Merr.], with some continuous corn. Continuous corn was indicated by corn planted in two or more succes-sive growing seasons per historical land use data. Planting and harvest of crops was assumed to occur on May 1 and September 30, respectively. Seventy-five percent of fertilizer-N was applied in the spring prior to planting corn, with the remaining 25% applied in the fall after soybeans. Fertilizer types consisted of anhydrous ammonia, constituting half of applied-N, urea ammonium nitrate, and diammonium phosphate. Fertilizer application rates (Table 2), types, and timing were based on practices typical in the region and are consistent with rates reported in the Iowa Nutrient Reduction Strategy (Iowa State University, 2013).
Hydrologic Parameterization and Calibration
Input parameterization was guided by recommended ranges reported in previous SWAT applications (Dou-glas-Mankin et al., 2010; Arnold et al., 2012), with par-ticular focus on efforts in tile-drained landscapes in the Upper Midwest of the United States (Green et al., 2006; Sui and Frankenberger, 2008; Gassman et al., 2009; Moriasi et al., 2012, 2013; Yen et al., 2014). Selec-tion of tile-drain related parameters was also informed by previous application of the DRAINMOD and RZWQM models to tile-drained field plots in Central Iowa (Thorp et al., 2007, 2009). The spin-up period for both models began in 2000, providing eight years of spin-up for the KS model and seven years for AL. The purpose of this study was to evaluate model behavior and feasibility of calibration at small spatial and tempo-ral scales. Due to limited years of data collection and challenges encountered during calibration, neither spa-tial nor temporal validation was performed.
Table 3 is the list of input parameters that were adjusted during hydrologic calibration. Various com-binations of hydrologic parameter adjustments were made using both manual calibration and the SUFI2 algorithm within the SWAT-CUP software program (Abbaspour, 2011). Simulations were executed using SWAT Version 2012, Revision 634. Performance was assessed using visual assessment of daily time series
data, performance criteria established by Moriasi et al. (2015) for Nash-Sutcliffe Efficiency (NSE) and percent bias (PBIAS) (Table 4), which were applied to all flow pathways, and visual analysis of flow dura-tion curves.
Hydrologic calibration and assessment focused on simulation of daily SURQ, water yield (WYLD), and SSF, with SSF being most critical for NO3-N transport in tile-drained watersheds. SSF is the sum of tile flow, lateral flow, and groundwater flow, with tile flow being the dominant component in both KS and AL watersheds. For both watersheds, better agreement between measured and predicted hydrologic output was obtained, using the Plant ET method to calculate daily CN values. Similarly, model runs using the more recently-incorporated DRAINMOD-based tile equa-tions (Moriasi et al., 2012, 2013) provided more accu-rate hydrologic predictions in both watersheds than the older TDRAIN-based algorithms. Therefore, the Plant ET curve number method and the DRAINMOD-based tile equations were used in final calibration runs.
Nitrogen Parameterization and Calibration
After hydrologic simulations were calibrated and assessed, NO3-N-related variables reported in Table 5 were adjusted during calibration to observed daily NO3-N concentrations using existing SWAT algo-rithms (Calibration A). The calibrated concentration represents the composite concentration of all flow pathways, but the vast majority of simulated and observed NO3-N is transported with tile flow. Perfor-mance assessment focused on daily concentrations rather than monthly loads. Calibration of flows and loads (and not concentrations) can mask performance deficiencies. For example, NO3-N concentrations could potentially be calibrated upwards during peri-ods of flow underestimation in order to improve load predictions, but the appearance of improvement would be artificial and for the wrong reasons. Fur-thermore, calibration of daily concentration provides insights to the suitability of the model for simulation of water quality BMPs and for assessment of water quality standards, which are concentration-based. As with flow, daily concentration predictions were evalu-ated using NSE and PBIAS statistics (Table 4) and concentration duration curves.
After evaluating simulation of daily NO3-N concen-trations using existing algorithms in SWAT, a revised executable version of SWAT (a modified version of Revision 636) was utilized to try and improve simula-tion of NO3-N concentrations. In both the original and revised algorithms, the concentration of NO3-N in the mobile water for a given soil layer is a linear TABLE 2. Estimated Fertilizer-N Application.
Crop Rotation
Watershed
Units
KS AL
Corn years of corn-soybean rotations 170 184 kg-N/ha1
function of the amount of NO3-N (kg/ha) present in the soil layer (NO3ly), expressed as
concNO3;mobile¼ NO3ly 1 exp w 1 h ð Þ SATly =w ð4Þ where w is the amount of mobile water in the layer (mm of H2O),h is the fraction of porosity from which anions are excluded, and SATly is the saturated water content of the soil layer (mm of H2O). To calcu-late the concentration of NO3-N in tile flow the fol-lowing nonlinear function of the amount of NO3-N in the soil profile was developed in the revised algo-rithm.
TNO3ln¼ N LNCO ln TNO3½ ð ÞN LN; ð5Þ where, N_LNCO is a dimensionless coefficient, TNO3 is the total NO3-N in the soil profile (kg/ha), and N_LN is a dimensionless exponent in the nonlinear function. The concentration in the tile flow is then calculated as follows:
concNO3;tile¼ TNO3ln 1 exp ð1w htileÞ SATly
=wtile; ð6Þ where wtileis the amount of tile flow (mm).
In addition to including a nonlinear function for
tile NO3-N in the revised algorithms, the
TABLE 3. Hydrologic Input Parameters Considered during Model Calibration and Assessment.
Parameter ID Description Calibration Range Units
Calibrated Values
KS1 AL2
ICN Daily CN calculation method (0= Soil Moisture,
1= Plant ET)
0,1 — 1 1
CNCOEF Initial SCS curve number for moisture condition II 0.2-1.2 — 0.85 0.50
ESCO Soil evaporation compensation factor 0.8-1.0 — 0.95 1.003
EPCO Plant uptake compensation factor 0.8-1.0 — 0.96 1.003
SURLAG Surface runoff lag coefficient 0.2-4.0 — 1.08 0.27
GW_DELAY Lag time between water that exits soil profile and enters shallow aquifer
0-200 day 77 51
GWQMN Threshold depth of water in shallow aquifer for required return flow to occur
0-2,500 mm 987 1,535
REVAPMN Threshold depth of water in shallow aquifer for revap to occur 500-1,500 mm 1,131 7503
ALPHA_BF Baseflow recession constant 0.04-1.00 day1 0.70 0.0483
TIMP Snow pack temperature lag factor 0-1 — 0.77 1.003
ITDNR Tile drainage simulation method (0= original, 1 = DRAINMOD) 0,1 — 1 1
IWTDN Water table depth algorithm (0= original, 1 = newer) 0,1 — 1 1
DDRAIN Depth to tile drains 900-1,500 mm 1,4464 1,0124
DEP_IMP Depth to restrictive layer 1,575-2,625 mm 1,6574 1,9544
DRAIN_CO Drainage coefficient 5-25 mm/day 24.1 10.0
LATKSATF Multiplier for lateral saturated hydraulic conductivity 0.5-2.5 — 0.55 0.75
SDRAIN Distance/spacing between tile drains/laterals (mm) 18,000-36,000 mm 27,928 23,583
Notes: CN, curve number; ET, evapotranspiration; SCS, Soil Conservation Service (now Natural Resource Conservation Service [NRCS]).
1KS Wetland watershed parameter values (final calibration). 2AL Wetland watershed parameter values (final calibration). 3Default value.
4DDRAIN and DEP_IMP input only in hydrologic response units (HRUs) with subsurface tile drains.
TABLE 4. Performance Evaluation Criteria.1
Statistic Output Time Scale
Performance Criteria
Very Good Good Satisfactory Not Satisfactory
NSE Flow D-M-A NSE> 0.80 0.70< NSE ≤ 0.80 0.50< NSE ≤ 0.70 NSE≤ 0.50
NO3-N M NSE> 0.65 0.50< NSE ≤ 0.65 0.35< NSE ≤ 0.50 NSE≤ 0.35
PB Flow D-M-A PB< 5 5 < PB ≤ 10 10 < PB ≤ 15 PB≥ 15
NO3-N D-M-A PB< 15 15 < PB ≤ 20 20 < PB ≤ 30 PB≥ 30
Notes: NSE, Nash-Sutcliffe efficiency; PB= PBIAS, percent bias (%); D, daily; M, monthly; A, annual.
concentration was also lagged using a digital filter technique of the form
concNO3;tilei ¼ concNO3;tilei1 1 Nð LAGÞ þ concNO3;tilei N LAG
ð7Þ where the subscripts i and i1 indicate concentra-tions on the current and previous day, respectively, and N_LAG is a recession constant used to lag tile NO3-N concentrations. These algorithms were intro-duced in order to evaluate impact of smoothing tem-poral variations in NO3-N transported from the soil profile to tile flow, and the model was recalibrated using the new NO3-N equations (Calibration B). The goal of revising the algorithms was not to physically represent individual processes in the transport of NO3-N to tile drains, but to evaluate and document the need to further refine soil-N processes in the model. The parameter names, descriptions, and val-ues of all NO3-N calibration variables are listed in the lower section of Table 5.
RESULTS AND DISCUSSION
Evaluation of Hydrologic Simulation
All NSE and PBIAS values for both daily and monthly WYLD meet the evaluation criteria of satis-factory or better set forth by Moriasi et al. (2015) (Table 6). NSE values for daily SSF were not satisfac-tory for either watershed (using Moriasi criteria for
total flow), although PBIAS is very good in the KS model and satisfactory for AL. Simulation of SURQ is not satisfactory at either time step for the AL model. Average runoff in the AL watershed was only 30 mm/ yr from 2007-2010, and SWAT was unable to repli-cate these extremely low runoff conditions. The over-all water balance of both models matched observed data reasonably well. Observed SSF in the KS water-shed accounted for 75% of the measured flow, with simulated SSF equal to 73% of total WYLD. Observed SSF in the AL watershed comprised 89% of total flow, whereas simulated SSF made up 85% of the simu-lated WYLD. Simulations of monthly WYLD were good (0.70 ≤ NSE ≤ 0.80) for both watersheds. It is noteworthy that the maximizing model agreement with observed data required calibration parameters unique to each watershed, likely due to different degrees of drainage intensity, along with other dis-tinct watershed characteristics.
Time series plots illustrate the challenges of accu-rately simulating daily SSF in SWAT. The model cap-tures the general trends/directions in SSF, but consistently underestimates peak flows and fails to reflect hydrograph recession in both the KS (top por-tion of Figure 2) and AL (top porpor-tion of Figure 3) watersheds. Flow duration curves (Figure 4) reveal that both models failed to simulate low flows well, although the magnitude of these flows is so small that they have little effect on the annual water bal-ance and mass transport of NO3-N. Potential parame-terization errors that contribute to deviations between observed and simulated SSF could stem from differences in localized precipitation patterns not cap-tured by available weather stations located well TABLE 5. Nitrogen-Related Parameters Considered during Model Calibration and Assessment.
Parameter ID Description Calibration Range Units1
Calibrated Values
KS2 AL3
NO3-N calibration using existing soil and tile NO3-N algorithms (Calibration A)
NPERCO Nitrate percolation coefficient 0.1-0.8 — 0.204 0.204
ANION_EXCL Fraction of porosity (void space) from which anions are excluded 0.1-0.8 Fraction 0.11 0.27
CDN Denitrification exponential rate coefficient 0-3 — 1.26 1.24
SDNCO Denitrification threshold water content 1.0-1.3 Fraction 1.18 1.19
NO3-N calibration using modified algorithms with lagging parameters (Calibration B)
N_LN New dimensionless exponent for nonlinear tile NO3-N
concentration function
0.5-2.5 — 1.5 1.5
N_LNCO New dimensionless coefficient for nonlinear tile NO3-N
concentration function
0.5-2.5 — 1.5 1.5
N_LAG New dimensionless lag coefficient for NO3-N in tile flow 0.1-1.0 — 0.25
4
0.254
ANION_EXCL Fraction of porosity (void space) from which anions are excluded 0.1-0.8 Fraction 0.40 0.50
CDN Denitrification exponential rate coefficient 0-3 — 0.46 0.21
SDNCO Denitrification threshold water content 1.0-1.3 Fraction 1.29 1.27
1
Units are dimensionless except ANION_EXCL (fraction of porosity) and SDNCO (fraction of field capacity).
2
KS Wetland watershed parameter values (final calibration).
3
AL Wetland watershed parameter values (final calibration).
4
outside the watersheds. Additionally, the presence of surface inlets is not parameterized, and there is uncertainty associated with characteristics of local tile drainage infrastructure and other hydrologic inputs.
The conceptual framework and simplifications of the SWAT model may also limit accuracy of tile flow simulation at these small scales. Such simplifications include the lumped nature of HRUs, which does not allow mechanistic routing of subsurface flow through subbasins, and simplified groundwater routines noted by Pfannerstill et al. (2014), who developed a modi-fied version of SWAT in which groundwater storage was split into fast and slow contributing aquifers to improve hydrograph recession and low flow simula-tion. Additionally, the static value of many
hydrologi-cal input parameters, combined with varying
temporal sensitivity of hydrologic parameters, almost certainly limits model performance (Pfannerstill et al., 2015). This limitation has been evaluated and documented by others, including Guse et al. (2016), Haas et al. (2016), Herman et al. (2013), and Yilmaz et al. (2008).
NO3-N Simulation (Calibration A)
Simulation of NO3-N concentration prior to modifi-cation of algorithms (i.e., Calibration A) was more problematic than prediction of daily flow components, with concentrations falling steeply in June/July and remaining near zero through the end of the growing season in both KS (bottom portion of Figure 2) and AL (bottom portion of Figure 3) models in all years. Simulated NO3-N is depleted from the soil too quickly, possibly due to misrepresentation of soil-N cycle and/or NO3-N transport algorithms. This rapid depletion does not appear to be driven by hydrologic predictions, since the models do not significantly
overestimate SSF peaks or volumes prior to and dur-ing soil-NO3 depletion (Figure 2 and 3). Prior to depletion of soil-N, simulated concentration varied with flow, showing more short-term fluctuation than observed concentration. Additionally, there are sev-eral instances of sharp increases in simulated
concen-tration concurrent with declines in observed
concentration. This occurs in June 2008 in both watersheds, and again in July 2010 for AL, at times when both SSF and SURQ increase and SSF NO3-N concentrations are overestimated.
Although model performance for NO3-N concentra-tion was not satisfactory (Table 7), the proporconcentra-tion of NO3-N carried by SSF relative to SURQ was reason-able, with SSF concentrations consistently far exceed-ing runoff concentrations. Simulated flow-weighted average (FWA) NO3-N concentrations in SURQ were less than 1 mg/L for both watersheds, while simu-lated FWA NO3-N concentrations in SSF were over 10 mg/L. This compares well to previous measure-ments of NO3-N in surface runoff in Iowa, including a study by Zhou et al. (2014), in which NO3-N con-centrations in surface runoff in the Walnut Creek
watershed in Jasper County averaged 1.08 mg
nitrate-N/L across all treatments and years (ranging from 0.04-3.7 mg N/L).
Time series comparisons (Figures 2 and 3) indicate that while SSF predictions may contribute to some of the deviation between observed and simulationed concentrations, hydrology alone does not explain errors in NO3-N predictions. Concentration errors often occur during times of relatively good SSF pre-diction, especially well into the growing season after NO3-N levels have dropped markedly. Further, the concentration duration curves (Figure 5) reveal that there is substantial deviation in concentration between the 30th and 60th duration intervals com-pared to the SSF duration curve.
NO3-N Simulation Using Modified Algorithms (Calibration B)
Due to problems simulating daily NO3-N concentra-tions, modifications were made to the SWAT source code to improve NO3-N loss from the soil profile by smoothing temporal variations in NO3-N export via tile drains using the N_LAG, N_LN, and N_LNCO parameters. The updated models were calibrated by adjusting ANION_EXCL, CDN, and SDNCO, along with the new lagging parameters (Table 5). The addi-tion of the new algorithms improved simulaaddi-tion of daily NO3-N fluctuations for both the KS (Figure 2) and AL (Figure 3) watersheds, but there remained periods of significant divergence between simulated and observed NO3-N concentrations.
TABLE 6. Performance Statistics for Pathway-Specific Flow Com-ponents.
Daily Monthly
NSE PBIAS1 NSE PBIAS1
KS Watershed WYLD 0.68 [S] 2.7 [VG] 0.79 [G] 5.0 [G] SURQ 0.55 [S] 10.0 [S] 0.87 [VG] 11.1 [S] SSF 0.36 [NS] 0.3 [VG] 0.55 [S] 2.9 [VG] AL Watershed WYLD 0.51 [S] 9.2 [G] 0.71 [G] 9.2 [G] SURQ 0.25 [NS] 21.5 [NS] 0.10 [NS] 21.5 [NS] SSF 0.46 [NS] 12.9 [S] 0.66 [S] 12.9 [S]
Notes: VG, very good; G, good; S, satisfactory; NS, not satisfactory.
1
FIGURE 2. Daily Subsurface Flow (SSF) and NO3-N Concentration for the KS Watershed. Observed and simulated SSF shown in top half of
plot with solid gray and dashed black lines, respectively. In bottom half of plot, blue lines represent observed NO3-N concentrations, red
dashed lines show simulated concentrations (Calibration A), and green dashed lines illustrate concentrations simulated using the new lagging algorithms (Calibration B).
FIGURE 3. Daily SSF and NO3-N Concentration for the AL Watershed. Observed and simulated SSF shown in top half of plot with solid
gray and dashed black lines, respectively. In bottom half of plot, blue lines represent observed NO3-N concentrations, red dashed lines show
simulated concentrations (Calibration A), and green dashed lines illustrate concentrations simulated using the new lagging algorithms (Calibration B).
Performance statistics with the new algorithms (Table 8) were improved significantly by lagging the release of NO3-N from the soil profile compared with predictions obtained, using the original SWAT algo-rithms (Table 7). Times series results (green-dashed lines in bottom portion of Figure 2 and Figure 3) illus-trate this improvement, with delayed reduction in predicted NO3-N concentrations later into the growing season and elimination of sharp peaks of NO3-N con-centrations predicted using the original SWAT equa-tions. Substantial improvement was obtained in the distribution of NO3-N concentrations, as illustrated by the concentration duration curves (Figure 5).
Despite improved predictions using the new equa-tions, NSE remained unsatisfactory for daily concen-trations but was satisfactory for daily loads in both models. Concentration PBIAS was very good for the KS model but not satisfactory for AL. Simulated con-centrations did not drop as sharply in mid-summer
months as with the original equations, but short-term fluctuations continued to exceed fluctuations in observed concentration, and overall, SWAT still underestimated NO3-N loss from these small water-sheds. Despite challenges in simulating daily concen-trations and loads, monthly statistics are categorized as “good” or better for all performance criteria except PBIAS in the AL model. If model calibration and assessment had focused on monthly NO3-N loads, it would have been possible to obtain good results, despite poor performance at the daily time step.
Evaluation of Soil NO3-N Processes
To better understand and document intermediate N-related processes, simulated soil-NO3-N concentra-tions for several soils in corn-soybean rotaconcentra-tions are plotted for the KS model (Figure 6) and AL model (Figure 7). For comparison, measured data for similar soils in Central Iowa (Cambardella et al., 1999) are shown with the simulated soil-NO3 levels. Simulated soil-NO3-N considers the depth of the soil layers in model, which extends 1,524 mm into the soil profile. Cambardella et al. (1999) measured soil-N to a depth of 1,050 mm. Although the depths differ, the compar-ison of soil-NO3-N levels is reasonable because soil-NO3-N levels are highest in the upper layers and decrease significantly with depth. Another distinction between the Cambardella et al. (1999) study and the model is the timing of N-application. In the study, all FIGURE 4. SSF Duration Curves for the KS (top graph) and AL (bottom graph) Watersheds.
TABLE 7. Performance Statistics for NO3-N Simulation
(Calibration A).
Watershed
Daily Concentration Daily Load Monthly Load
NSE PBIAS NSE PBIAS NSE PBIAS
KS 1.90 50.6 0.23 42.9 0.37 [S] 41.4
AL 2.35 66.3 0.15 53.7 0.14 53.6
Note: All performance criteria are not satisfactory (NS) per Moriasi et al. (2015) unless otherwise indicated.
N was applied in the fall as anhydrous ammonia, but in the model, various forms of N fertilizer were uti-lized, and application was split between spring and fall.
The trend for simulated soil-NO3-N is similar to the measured pattern using both original (Calibration A) and modified (Calibration B) algorithms with an important distinction: in Calibration A, soil-NO3-N is fully depleted by mid-summer in both corn and soy-bean years, whereas Calibration B and measured mid-summer residual levels off at 30-40 kg-NO3/ha in corn years and remains steady at approximately 45 kg-NO3/ha in soybean years. The increase in soil-NO3-N from fertilizer application and/or mineraliza-tion in the spring and after soybean harvest in the fall is reflected by the models and the observed data. Simulated soil-NO3-N levels are much lower in the Canisteo soil than the Webster or Clarion soils in the KS watershed, but this difference is not observed in the AL model.
Modeled corn yields using the original algorithms (Calibration A) were 8,713 kg/ha (139 bu/ac) in the KS watershed and 10,335 kg/ha (164 bu/ac) in the AL watershed, which are 16% and 10% lower than reported countywide yield data, respectively (ISU, 2015). Yields were little-changed with modified algo-rithms in Calibration B. The fact that simulated yields are higher in the AL watershed than KS is consistent with the countywide yield data. Simulated depletion of soil-NO3-N levels to zero in the middle of the growing season may be partly responsible for lower than expected corn yields, based on the number of N-stress days in model output. However, this depletion occurs in both wet and dry years and in years in which simulated denitrification is zero. This suggests that simulation of N mineralization may also be problematic and partially responsible for errors in prediction of NO3-N loss. Model prediction of crop growth processes may also contribute to over-depletion of soil. Nair et al. (2011) noted the FIGURE 5. Concentration Duration Curves for the KS (top graph) and AL (bottom graph) Watersheds. Observed concentrations represented
by blue line, simulated concentrations shown using red dotted line (Calibration A), and concentrations simulated using the new lagging equations are illustrated with the green dashed line (Calibration B).
TABLE 8. Performance Statistics for Modified Algorithm NO3-N Simulation (Calibration B).
Watershed
Daily Concentration Daily Load Monthly Load
NSE PBIAS NSE PBIAS NSE PBIAS
KS 0.20 [NS] 8.9 [VG] 0.40 [S] 19.0 [G] 0.72 [VG] 17.3 [G]
FIGURE 6. Simulated Soil Profile NO3-N for Corn-Soybean Rotations in the KS Watershed Using Existing (Calibration A) and Modified
(Calibration B) Algorithms. Dashed green line is soil-NO3-N for Canisteo soil, dotted purple line is for Webster, and solid brown is for
Clarion. Observed and simulated SSF is shown in the top portion of each graph. Squares and circles represent soil-NO3-N measured in
similar Central Iowa soils in corn and soybean years, respectively (Cambardella et al., 1999).
FIGURE 7. Simulated Soil Profile NO3-N for Corn-Soybean Rotations in the AL Watershed Using Existing (Calibration A) and Modified
(Calibration B) Algorithms. Dashed green line is soil-NO3-N for Canisteo soil, dotted purple line is for Webster, and solid brown is for
Clarion. Observed and simulated SSF is shown in the top portion of each graph. Squares and circles represent soil-NO3-N measured in
importance of crop yields when simulating nitrogen transport in SWAT, and it is likely that plant growth parameters for corn in the SWAT plant database are outdated and do not reflect current crop genetics.
Simulated soil-N dynamics for a tile-drained Web-ster soil are reported in Table 9. With the exception of several zero-nitrification years, the magnitude of
simulated fluxes were generally within ranges
reported in regional guidance and literature, but these fluxes are highly variable and there is large uncertainty associated with estimates of N-fixation and denitrification (Christianson et al., 2012). In Webster soil HRUs, the average simulated denitrifica-tion was 28 kg-N/ha/yr1 for the KS model and 20 kg-N/ha/yr for AL, using the original tile NO3-N algorithms (Calibration A). David et al. (2009) simu-lated denitrification rates ranging from 3.8 to 21 kg-N/ha/yr1using a variety of models to estimate deni-trification rates in a tile-drained corn and soybean rotation in Illinois. In well-drained Clarion soils in the KS and AL models, the simulated denitrification rate was zero and large magnitudes of NO3-N were lost to deep seepage because of the absence of a restrictive soil layer. N-fixation by soybeans was somewhat higher than reported in other studies in Iowa (Jaynes et al., 2001; Christianson et al., 2012) and near the upper-end of fixation rates summarized in a meta-analysis of published data (Salvagiotti
et al., 2008), and N-uptake was near or above the high end of rates estimated for high yielding corn crops in Iowa (ISU, 2006).
Soil-NO3-N levels simulated using the modified algorithms (Calibration B) were more representative of Central Iowa soil data (Cambardella et al., 1999). However, calibration using modified algorithms elimi-nated denitrification in these HRUs, which is not realistic and resulted in much higher NO3-N losses via SSF (Table 9) and deep seepage in soils without tile drainage (e.g., Clarion soils). While the modified algorithms and subsequent calibration improved pre-dictions of NO3-N concentrations and loads compared with the original algorithms, the basis for the modifi-cations is not well established and problems simulat-ing important N processes and NO3-N transport remain. Nevertheless, the modifications provide insight to the possible causes of error and reveal the need for improved soil-N and crop growth processes in the simulation of NO3-N transport in tile-drained watersheds. Our work suggests simulations of nitrifi-cation, mineralization, and denitrification need
fur-ther evaluation and more physically-based
modifications than the empirical lagging algorithms presented here.
Pohlert et al. (2007) identified similar limitations of SWAT, and found improvement in the prediction of NO3-N transport using lysimeter data after the TABLE 9. Simulated Soil-NO3Dynamics for Webster Soil HRUs with Tile Drainage.
Soil/Crop
Positive Fluxes (kg-N/ha1)1 Negative Fluxes (kg-N/ha1)2
Appl3 Atmos Fix Min Denit Uptake Runoff SSF Seep
Simulated using original NO3-N algorithms (Calibration A)
KS 2008 Soy 49 13 276 113 27 313 <1 22 0 2009 Corn 122 9 0 109 55 205 <1 29 0 2010 Soy 49 13 227 101 28 271 <1 6 0 2011 Corn 122 8 0 134 0 239 <1 39 0 AL 2007 Corn 135 10 0 121 12 254 <1 34 0 2008 Soy 49 8 293 131 31 338 <1 27 0 2009 Corn 135 8 0 155 0 302 <1 15 0 2010 Soy 49 9 232 134 36 301 <1 20 0
Simulated using new NO3-N algorithms (Calibration B)
KS 2008 Soy 49 13 274 129 0 313 <1 67 0 2009 Corn 122 9 0 125 0 251 <1 43 0 2010 Soy 49 13 230 116 0 271 <1 60 0 2011 Corn 122 8 0 137 0 239 <1 29 0 AL 2007 Corn 135 10 0 132 0 277 <1 55 0 2008 Soy 49 8 291 142 0 338 <1 33 0 2009 Corn 135 8 0 165 0 321 <1 8 0 2010 Soy 49 9 224 143 0 301 <1 45 0
1Inputs: Appl, fertilizer-N; Atmos, rainfall-N; Fix, N-fixation; Min, mineralization of organic-N.
2Outputs: Denit, denitrification; Uptake, plant uptake; Runoff and SSF, N lost to surface water; Seep, N lost to deep aquifer via seepage. 3Fertilizer application occurs in fall after soybean harvest and in spring in corn years.
integration of a detailed biogeochemical model into SWAT. However, the updated model is not publically available, and to our knowledge, has not been applied at the watershed scale. The time-varying nature of the sensitivity of nitrate-related model parameters has been documented by others (Haas et al., 2015, 2016). Evaluation of this phenomenon as it relates to simulating soil-N dynamics may facilitate the devel-opment of more physically descriptive modifications to the algorithms revised for the AL and KS water-sheds studied here.
CONCLUSIONS
Model assessment revealed that it is possible to meet generally accepted performance criteria (Moriasi et al., 2015) for simulation of monthly WYLD, SSF, and NO3-N loads in both case study watersheds, while not accurately capturing the daily fluctuation of pathway specific flows or NO3-N concentrations. For the KS and AL watersheds, NSE values were 0.79 and 0.71, respectively, for monthly WYLD; 0.55 and 0.66 for monthly SSF; and 0.72 and 0.60 for monthly NO3-N load (using the modified NO3-N algo-rithms). Simulation of daily SURQ and SSF proved more challenging and were generally not satisfactory (NSE < 0.50). Simulation of daily NO3-N concentra-tion was not satisfactory even after modifying NO3-N algorithms to lag NO3-N transport from the soil pro-file via tile drainage, with the KS watershed NSE of 0.20 and AL watershed NSE value of 1.12.
Differences in hydrology and NO3-N transport between watersheds were not reflected by the model, as evidenced by distinct calibration parameters and parameter values. This suggests that parameteriza-tion may not be transferable across watersheds with similar characteristics, and also that models cali-brated at larger scales may not accurately reflect hydrology and nutrient transport at small watershed (e.g., drainage district) scales, as noted by Baffaut et al. (2015). These limitations are especially impor-tant in cases where the model is intended to help locate, design, and/or estimate NO3-N removal capa-bilities of water quality BMPs, as indicated by impacts on NO3-N simulation wetlands at the outlet of these case study watersheds (Ikenberry et al., 2017).
Investigation of intermediate N processes
revealed SWAT has the capability to simulate vari-ous N fluxes and soil-NO3 levels, but overestimates depletion from the soil during summer months. Simulated mineralization and plant uptake rates are generally reasonable compared to literature
values; however, these fluxes are highly variable in space and time and heavily influence NO3-N trans-port via tile drainage. Soil-N fluxes such as miner-alization and denitrification should therefore be evaluated and reported as standard practice when applying the SWAT model for simulation of NO3-N transport, and more physically-based improvements to soil-N algorithms than those presented here is warranted. Additionally, better parameterization methods and supporting data for model inputs related to these processes are needed to improve and appropriately constrain soil-N fluxes and associ-ated prediction of NO3-N transport.
ACKNOWLEDGMENTS
This study was made possible through funding by EPA’s Initia-tive Enhancing State and Tribal Wetland Programs grant program (Grant Agreement Number CD-97723301-0). Flow and water qual-ity monitoring data used for model testing and evaluation were col-lected as part of the Iowa CREP program and supported in part by funding from the Iowa Department of Agriculture and Land Ste-wardship. Assistance in data compilation by Greg Stenback in the Department of Ecology, Evolution, and Organismal Biology (EEOB) at Iowa State University, is greatly appreciated. SWAT model FORTRAN code modification and testing was conducted in collabo-ration with Jeff Arnold and Nancy Sammons in the Grassland, Soil, and Water Research Laboratory of USDA-ARS in Temple, Texas. The Iowa Department of Natural Resources Watershed Improvement Section was also instrumental in the completion of this research.
LITERATURE CITED
Abbaspour, K.C., 2011. SWAT-CUP2: SWAT Calibration and Uncertainty Programs Manual Version 2. Eawag. Swiss Federal Institute of Aquatic Science and Technology, Department of Sys-tems Analysis, Integrated Assessment and Modeling (SIAM), Duebendorf, Switzerland, p. 106.
Alexander, R.B., R.A. Smith, G.E. Schwarz, E.W. Boyer, J.V. Nolan, and J.W. Brakebill, 2008. Differences in Phosphorus and Nitrogen Delivery to the Gulf of Mexico from the Mississippi River Basin. Environmental Science & Technology 42(3):822-830. https://doi.org/10.1021/es0716103.
Arnold, J.G., D.N. Moriasi, P.W. Gassman, K.C. Abbaspour, M.J. White, R. Srinivasan, C. Santhi, R.D. Harmel, A. van Griens-ven, M.W. Van Liew, N. Kannan, and M.K. Jha, 2012. SWAT: Model Use, Calibration, and Validation. Transactions of the ASABE 55(4):1491-1508. https://doi.org/10.13031/2013.42256. Arnold, J.G., M.A. Youssef, H. Yen, M.J. White, A.Y. Sheshukov,
A.M. Sadeghi, D.N. Moriasi, J.L. Steiner, D.M. Amatya, R.W. Skaggs, E.B. Haney, J. Jeong, M. Arabi, and P.H. Gowda, 2015. Hydrological Processes and Model Representation: Impact of Soft Data on Calibration. Transactions of the ASABE 58 (6):1637-1660. https://doi.org/10.13013/trans.58.10726.
Baffaut, C., S.M. Dabney, M.D. Smolen, M.A. Youssef, J.V. Bonta, M.L. Chu, J.A. Guzman, V.S. Shedekar, M.K. Jha, and J.G. Arnold, 2015. Hydrologic and Water Quality Modeling: Spatial and Temporal Considerations. Transactions of the ASABE 58 (6):1661-1680. https://doi.org/10.13031/trans.58.10714.
Boles, C.M.W., J.R. Frankenberger, and D.N. Moriasi, 2015. Tile Drainage Simulation in SWAT2012: Parameterization and Eval-uation in an Indiana Watershed. Transactions of the ASABE 58 (5):1201-1213. https://doi.org/10.13031/trans.58.10589.
Bressiani, D.D.A., P.W. Gassman, J.G. Fernandes, L.H.P. Gar-bossa, R. Srinivasan, N.B. Bonuma, and E.M. Mendiondo, 2015. A Review of Soil and Water Assessment Tool (SWAT) Applica-tions in Brazil: Challenges and Prospects. International Journal of Agricultural & Biological Engineering 8(3):9-35. https://doi. org/10.3965/j.ijabe.20150803.1765.
Cambardella, C.A., T.B. Moorman, D.B. Jaynes, J.L. Hatfield, T.B. Parkin, W.W. Simpkins, and D.L. Karlen, 1999. Water Quality in Walnut Creek Watershed: Nitrate-Nitrogen in Soils, Subsur-face Drainage Water, and Shallow Groundwater. Journal of Environmental Quality 28:25-34.
Christianson, L., M. Castellano, M. Helmers, and M. Helmers. 2012. Nitrogen and Phosphorus Balances in Iowa Cropping Sys-tems: Sustaining Iowa’s Soil Resource. Iowa State University in collaboration with the Iowa Department of Agriculture and Land Stewardship.
Coelho, B.B., R. Murray, D. Lapen, E. Topp, and A. Bruin, 2012. Phosphorus and Sediment Loading to Surface Waters from Liq-uid Swine Manure Application under Different Drainage and Tillage Practices. Agricultural Water Management 104:51-61. https://doi.org/10.1016/j.agwat.2011.10.020.
Crumpton, W.G., G.A. Stenback, B.A. Miller, and M.J. Helmers, 2006. Potential Benefits of Wetland Filters for Tile Drainage Systems: Impact on Nitrate Loads to Mississippi River Sub-basins. Final Report to the U.S. Department of Agriculture (USDA). Ames, Iowa.
David, M.B., S.J. Del Grosso, X. Hu, E.P. Marshall, G.F. McIsaac, W.J. Parton, C. Tonitto, and M.A. Youssef, 2009. Modeling Den-itrification in a Tile-Drained, Corn and Soybean Agroecosystem of Illinois, USA. Biogeochemistry 93:7-30. https://doi.org/10. 1007/s10533-008-9273-9.
David, M.B., L.E. Drinkwater, and G.F. McIsaac, 2010. Sources of Nitrate Yield in the Mississippi River Basin. Journal of Envi-ronmental Quality 39:1657-1667. https://doi.org/10.2134/jeq2010. 0115.
Dinnes, D.L., D.L. Karlen, D.B. Jaynes, T.C. Kaspar, J.L. Hatfield, T.S. Colvin, and C.A. Cambardella, 2002. Nitrogen Management Strategies to Reduce Nitrate Leaching in Tile-Drained Midwest-ern Soils. Agronomy Journal 94:153-171. https://doi.org/10.2134/ agronj2002.0153.
Douglas-Mankin, K.R., R. Srinivasan, and J.G. Arnold, 2010. Soil and Water Assessment Tool (SWAT) Model: Current Develop-ments and Applications. Transactions of the ASABE 53(5):1423-1431. https://doi.org/10.13031/2013.34915.
El-Sadek, A., J. Feyen, W. Skaggs, and J. Berlamont, 2002. Eco-nomics of Nitrate Losses from Drained Agricultural Land. Jour-nal of Environmental Engineering-ASCE 128(4):376-383. https://doi.org/10.1061/(ASCE)0733-9372(2002) 128:4(376). Gassman, P.W., M. Jha, and S. Mickelson. 2009. Application of the
SWAT2005 Alternative Runoff Curve Number Method for the Boone River Watershed in North Central Iowa, United States. Presented at the Soil and Water Assessment Tool-Southeast Asia (SWAT-SEA) Conference, January 7-8, Chiang Mai, Thailand. Gassman, P.W., A.M. Sadeghi, and R. Srinivasan, 2014.
Applica-tions of the SWAT Model Special Section: Overview and Insights. Journal of Environmental Quality 43(1):1-8. https://d oi.org/10.2134/jeq2013.11.0466.
Goolsby, D.A., W.A. Battaglin, B.T. Aulenbach, and R.P. Hooper, 2000. Nitrogen Flux and Sources in the Mississippi River Basin. Science of the Total Environment 248:75-86. https://doi.org/10. 1016/S0048-9697(99)00532-X.
Goswami, D., P.K. Kalita, R.A. Cooke, and M.C. Hirschi, 2008. Estimation and Analysis of Baseflow in Drainage Channels in
Two Tile-Drained Watersheds in Illinois. Transactions of the ASABE 51(4):1201-1213. https://doi.org/10.13031/2013.25238 Green, C.H., M.D. Tomer, M. Di Luzio, and J.G. Arnold, 2006.
Hydrologic Evaluation of the Soil and Water Assessment Tool for a Large Tile-Drained Watershed in Iowa. Transactions of the ASABE 49(2):413-422. https://doi.org/10.13031/2013.20415. Guse, B., M. Pfannerstill, A. Gafurov, N. Fohrer, and H. Gupta,
2016. Demasking the Integrated Information of Discharge: Advancing Sensitivity Analysis to Consider Different Hydrologi-cal Components and Their Rates of Change. Water Resources Research 52:8724-8743. https://doi.org/10.1002/2016WR018894. Haas, M.B., B. Guse, M. Pfannerstill, and N. Fohrer, 2015.
Detec-tion of Dominant Nitrate Processes in Ecohydrological Modeling with Temporal Parameter Sensitivity Analysis. Ecological Mod-elling 314:62-72. https://doi.org/10.1016/j.ecolmodel.2015.07.009. Haas, M.B., B. Guse, M. Pfannerstill, and N. Fohrer, 2016. A
Joined Multi-Metric Calibration of River Discharge and Nitrate Loads with Different Performance Measures. Journal of Hydrol-ogy 536:534-545. https://doi.org/10.1016/j.jhydrol.2016.03.001. Hatfield, J.L., J.H. Prueger, and D.B. Jaynes, 1998. Environmental
Impacts of Agricultural Drainage in the Midwest. In: Drainage in the 21st Century: Food Production and the Environment. Proceedings of the 7th Annual Drainage Symposium, Orlando, Florida. 8–10 March 1998. ASAE, St. Joseph, Michigan. Herman, J.D., P.M. Reed, and T. Wagener, 2013. Time-Varying
Sensitivity Analysis Clarifies the Effects of Watershed Model Formulation on Model Behavior: Time-Varying Sensitivity of Watershed Models. Water Resources Research 49:1400-1414. https://doi.org/10.1002/wrcr.20124.
Ikenberry, C.D., W.G. Crumpton, J.G. Arnold, M.L. Soupir, and P.W. Gassman, 2017. Evaluation of Existing and Modified Wet-land Equations in the SWAT Model. Journal of the American Water Resources Association 53(6):1267-1280. https://doi.org/10. 1111/1752-1688.12570.
ISU (Iowa State University), 2013. Iowa Nutrient Reduction Strategy. http://www. nutrientstrategy.iastate.edu, accessed February 2014. ISU (Iowa State University), 2014. Iowa Environmental Mesonet.
http://mesonet.agron.iastate.edu/, accessed March 2014.
ISU (Iowa State University), 2015. Historical Corn and Soybean Yields. Iowa Ag Decision Maker. https://www.extension.iasta te.edu/agdm/, accessed July 2014.
Jaynes, D.B., T.S. Colvin, D.L. Karlen, C.A. Cambardella, and D.W. Meek, 2001. Nitrate Loss in Subsurface Drainage as Affected by Nitrogen Fertilizer Rate. Journal of Environmental Quality 30:1305-1314. https://doi.org/10.2134/jeq2001.3041305x. Krysanova, V. and M. White, 2015. Advances in Water Resources
Assessment with SWAT—An Overview. Hydrological Sciences Journal 60(5):771-783. https://doi.org/10.1080/02626667.2015. 1029482.
Malone, R.W., G. Yagow, C. Baffaut, M.W. Gitau, Z. Qi, D.M. Ama-tya, P.B. Parajuli, J.V. Bonta, and T.R. Green, 2015. Parameter-ization Guidelines and Considerations for Hydrologic Models. Transactions of the ASABE 58(6):1681-1703. https://doi.org/10. 13031/trans.58.10709.
Moriasi, D.N., M.W. Gitau, N. Pai, and P. Daggupati, 2015. Hydro-logic and Water Quality Models: Performance Measures and Evaluation Criteria. Transactions of the ASABE 58(6):1763-1785. https://doi.org/10.13031/trans.58.10715.
Moriasi, D.N., P.H. Gowda, J.G. Arnold, D.J. Mulla, S. Ale, J.L. Steiner, and M.D. Tomer, 2013. Evaluation of the Hooghoudt and Kirkham Tile Drain Equations in the Soil and Water Assessment Tool to Simulate Tile Flow and Nitrate-Nitrogen. Journal of Environmental Quality 42:1699-1710. https://doi.org/ 10.2134/jeq2013.01.0018.
Moriasi, D.N., C.G. Rossi, J.G. Arnold, and M.D. Tomer, 2012. Evaluating Hydrology of the Soil and Water Assessment Tool (SWAT) with New Tile Drain Equations. Journal of Soil and
Water Conservation 67(6):513-524. https://doi.org/10.2489/jswc. 67.6.513.
Nair, S., K.W. King, J.D. Witter, B.L. Sohngen, and N.R. Fausey, 2011. Importance of Crop Yield in Calibrating Watershed Water Quality Simulation Tools. Journal of the American Water Resources Association 47(6):1285-1297. https://doi.org/10.1111/j. 1752-1688.2011.00570.x.
NOAA (National Oceanic and Atmospheric Administration), 2013. National Climatic Data Center. https://www.ncdc.noaa.gov/, accessed January 2013.
Pfannerstill, M., B. Guse, and N. Fohrer, 2014. A Multi-Storage Groundwater Concept for the SWAT Model to Emphasize Nonlin-ear Groundwater Dynamics in Lowland Catchments. Hydrologi-cal Processes 28:5599-5612. https://doi.org/10.1002/hyp.10062. Pfannerstill, M., B. Guse, D. Reusser, and N. Fohrer, 2015.
Tempo-ral Parameter Sensitivity Guided Verification of Process Dynamics. Hydrology and Earth System Sciences Discussion 12 (2):1729-1764. https://doi.org/10.5194/hessd-12-1729-2015. Pohlert, T., J.A. Huisman, L. Breuer, and H.G. Frede, 2007.
Inte-gration of a Detailed Biogeochemical Model into SWAT for Improved Nitrogen Predictions—Model Development, Sensitiv-ity, and GLUE Analysis. Ecological Modelling 203:215-228. https://doi.org/10.1016/j.ecolmodel.2006.11.019.
Rozemeijer, J.C., Y. van der Velde, F.C. van Geer, M.F.P. Bierkens, and H.P. Broers, 2010. Direct Measurements of the Tile Drain and Groundwater Flow Route Contributions to Surface Water Contamination: From Field-Scale Concentration Patterns in Groundwater to Catchment-Scale Surface Water Quality. Envi-ronmental Pollution 158(12):3571-3579. https://doi.org/10.1016/ j.envpol.2010.08.014.
Salvagiotti, F., K.G. Cassman, J.E. Specht, D.T. Walters, A. Weiss, and A. Dobermann, 2008. Nitrogen Uptake, Fixation and Response to Fertilizer N in Soybeans: A Review. Field Crops Research 108:1-13. https://doi.org/10.1016/j.fcr.2008.03.001. Schilling, K.E. and M. Helmers, 2008. Effects of Subsurface
Drai-nage Tiles on Streamflow in Iowa Agricultural Watersheds: Exploratory Hydrograph Analysis. Hydrological Processes 22 (23):4497-4506. https://doi.org/10.1002/hyp.7052.
Stenback, G.A., W.G. Crumpton, K.E. Schilling, and M.J. Helmers, 2011. Rating Curve Estimation of Nutrient Loads in Iowa
Rivers. Journal of Hydrology 396(1–2):158-169. https://doi.org/ 10.1016/j.jhydrol.2010.11.006.
Stone, W.W. and J.T. Wilson, 2006. Preferential Flow Estimates to an Agricultural Tile Drain with Implications for Glyphosate Transport. Journal of Environmental Quality 35:1825-1835. https://doi.org/10.2134/jeq2006.0068.
Sui, Y. and J.R. Frankenberger, 2008. Nitrate Loss from Subsur-face Drains in an Agricultural Watershed Using SWAT2005. Transactions of the ASABE 51(4):1263-1272. https://doi.org/10. 13031/2013.25243.
Thorp, K.R., R.W. Malone, and D.B. Jaynes, 2007. Simulating Long-Term Effects of Nitrogen Fertilizer Application Rates on Corn Yield and Nitrogen Dynamics. Transactions of the ASABE 50(4):1287-1303. https://doi.org/10.13031/2013.23640.
Thorp, K.R., M.A. Youssef, D.B. Jaynes, R.W. Malone, and L. Ma, 2009. DRAINMOD-N II: Evaluated for an Agricultural System in Iowa and Compared to RZWQM-DSSAT. Transactions of the ASABE 52(5):1557-1573. https://doi.org/10.13031/2013.29144. Tomer, M.D., C.G. Wilson, T.B. Mooreman, K.J. Cole, D. Heer, and
T.M. Isenhart, 2010. Source-Pathway Separation of Multiple Contaminants During a Rainfall-Runoff Event in an Artificially Drained Agricultural Watershed. Journal of Environmental Quality 39:882-895. https://doi.org/10.2134/jeq2009.0289. Vidon, P., H. Hubbard, P.E. Cuadra, and M. Hennessey, 2012.
Storm Flow Generation in Artificially Drained Landscapes of the US Midwest: Matrix Flow, Macropore Flow, or Overland Flow. Water 4:90-111. https://doi.org/10.14294/WATER.2012.8. Yen, H., R.T. Baily, M. Arabi, M. Ahmadi, M.J. White, and J.G.
Arnold, 2014. The Role of Interior Watershed Processes in Improving Parameter Estimation and Performance of Water-shed Models. Journal of Environmental Quality 43(5):1601-1613. https://doi.org/10.2134/jeq2013.03.0110.
Yilmaz, K.K., H.V. Gupta, and T. Wagener, 2008. A Process-Based Diagnostic Approach to Model Evaluation: Application to the NWS Distributed Hydrologic Model. Water Resources Research 44(9):1-18. https://doi.org/10.1029/2007WR006716.
Zhou, X., M.J. Helmers, H. Asbjornsen, R. Kolka, M.D. Tomer, and R.M. Cruse, 2014. Nutrient Removal by Prairie Filter Strips in Agricultural Landscapes. Journal of Soil and Water Conserva-tion 69(1):54-64. https://doi.org/10.2489/jswc.69.1.54.
