ASABE Logo

Article Request Page ASABE Journal Article

Evaluation of AnnAGNPS Wetland Nitrogen Removal Component on Farmed Prairie Potholes

Rose Tirtalistyani1,*, Amy L. Kaleita1


Published in Journal of Natural Resources and Agricultural Ecosystems 2(1): 17-28 (doi: 10.13031/jnrae.15614). Copyright 2024 American Society of Agricultural and Biological Engineers.


1    Department of Agricultural and Biosystems Engineering, Iowa State University, Ames, Iowa, USA.

*    Correspondence: roset@iastate.edu, rose.tirtalistyani@ugm.ac.id

The authors have paid for open access for this article. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License https://creative?commons.org/licenses/by-nc-nd/4.0/

Submitted for review on 30 March 2023 as manuscript number NRES 15614; approved for publication as a Research Article by Associate Editor Dr. James Etheridge and Community Editor Dr. Kati Migliaccio of the Natural Resources & Environmental Systems Community of ASABE on 12 October 2023.

Highlights

Abstract. Farmed prairie potholes are common in northern Iowa and have been considered an impediment to agricultural development. Potholes have distinctive wetland characteristics where they might contain a high concentration of soluble nitrogen coming from agricultural upland runoff. With sub-surface drainage underneath them, providing short circuit nutrient transport downstream, Iowa poses a significant threat to nitrogen export to the Gulf of Mexico. Watershed models, such as AnnAGNPS, developed a wetland module allowing users to find the possible BMPs to accomplish nutrient reduction targets. Previous research indicates that AnnAGNPS may be used to mimic inundation depth fluctuations in potholes with moderate success. However, there has been little research on the wetland nitrogen removal components of the model. This study aims to evaluate the AnnAGNPS wetland module's performance in simulating the soluble nitrogen concentration in three farmed prairie potholes by comparing the simulation output with the observed data. The sensitivity analysis of the three key parameters of the model shows high and very high levels of sensitivity, with the temperature coefficient's relative sensitivity ranging from -6.7 to 8.6. Attempts in parameterization on the three targeted parameters indicated that the model poorly simulated the soluble nitrogen concentration. The comparison between observed and simulated data shows that the model fails to predict the high concentration of soluble nitrogen in the early cropping season. This conclusion is consistent with earlier research indicating that AnnAGNPS tends to underestimate the nitrogen concentration during the cropping season.

Keywords.AnnAGNPS wetland module, Farmed prairie pothole, Soluble nitrogen concentration.

In this study, potholes are defined as shallow, ephemerally or semi-permanently inundated depressional areas located in North America (North and South Dakota, Iowa, Minnesota, and Montana) and some provinces in Canada due to the geological development during the Wisconsin glacier retreat more than 10,000 years ago (Green et al., 2019; Richardson et al., 1994; Tiner, 2003). They are spatially frequent within the Prairie Pothole Region (PPR), where the portion that extends into Iowa is named the Des Moines Lobe (DML). In DML, potholes are estimated to account for 7.3% of the total land area (McDeid et al., 2019). Potholes, which are primarily inundated in the DML during the early cropping season, have been seen as an annoyance and an obstacle to agricultural development (Upadhyay et al., 2020). Since the early 1900s, these potholes have been altered by drainage systems following the mass conversion of prairie to row crop agriculture (Crumpton et al., 2012). This drainage technique has caused a loss of more than 95% of Iowa's natural pothole systems (Tiner, 2003). Yet, even with artificial drainage, these potholes still flood regularly for usually short periods of time, classifying them as ephemeral wetlands (Upadhyay et al., 2020).

Numerous studies have been conducted to examine the ecological services offered by potholes. Potholes are well-known for their significance in wetland ecosystems (Mitsch et al., 2015), including habitats for migratory ducks and shorebirds (Drum et al., 2015), groundwater recharge (Winter and Rosenberry, 1995), flood control and sediment entrapment (Hubbard, 1988), and water quality improvement (Crumpton, 2001; Ewel, 1997; Gleason et al., 2011). Despite the fact that many potholes have already been converted into agricultural areas, most research concentrates on pothole behavior in their natural state as wetlands. Farmed prairie potholes differ from non-farmed wetlands in various respects, including the shift of vegetation from prairie to row crops that alter the soil structure through agricultural practices such as tillage (Schilling and Dinsmore, 2018; Upadhyay et al., 2018). Farmed potholes also accumulate more nutrients due to the runoff of applied manure and fertilizer from upland fields (Cambardella et al., 1999). Although nutrient loading to potholes increases potholes' nutrient removal efficiencies (Ross and McKenna, 2022), farmed prairie potholes have greater cropping season nutrient concentrations than non-farmed (Martin et al., 2019b). Additionally, the presence of subsurface drainage and surface inlets, which are typical within these depressional areas, is suspected to increase nitrate-nitrogen (NO3-N) export to surface water (Schilling and Helmers, 2008), contributing nitrogen discharge to the Gulf of Mexico (Baker et al., 1975; Singh et al., 2007; Skaggs et al., 1994).

Several watershed models have been created throughout the years to help understand hydrologic systems and non-point source pollutants loading in wetland systems, such as SWAT (Ikenberry et al., 2017), APEX (McKenna et al., 2020), and AnnAGNPS (Tomer et al., 2013). Such models have the ability to mimic wetlands conditions in an agricultural watershed and are useful for Best Management Practices (BMPs) and Total Maximum Daily Load (TMDL) to accomplish nutrient reduction targets (Ikenberry et al., 2017). The United States Department of Agriculture-Agricultural Research Service (USDA-ARS) Annualized Agricultural Non-point Source Pollutant Loading Model (AnnAGNPS) effectively accommodates land use changes and extended simulations (Nahkala et al., 2021b). AnnAGNPS includes a wetland module that allows the user to simulate the dynamics of pothole systems. In our previous studies, AnnAGNPS has been used with moderate success to model pothole's hydrology (Nahkala et al., 2021a; Upadhyay et al., 2018). In simulating flood volume, AnnAGNPS was calibrated with average daily basis performance with Nash-Sutcliffe Efficiency (NSE) ranged from 0.26 to 0.79, the R2 ranged from 0.37 to 0.80, and the RMSE-observations standard deviation ratio (RSR) ranged from 0.46 to 0.86 (Nahkala et al., 2021a). However, little investigation has been made to model water quality dynamics, especially soluble nitrogen concentration in pothole systems, using AnnAGNPS. Understanding the water quality model for farmed prairie potholes is essential for determining the potential conservation methods to be implemented in the system and limiting the export of surplus nutrients from potholes to the subsurface drainage.

One of the primary processes in model construction is screening sensitive parameters and statistically examining each parameter's impact on model performance to determine which variables have the most influence on simulation outcomes and what parameter values are the most optimal (Gan et al., 2014). To determine the sensitive parameters, sensitivity analysis is performed to evaluate the influence of model parameters on the watershed model's ability to estimate constituent output for a specific application.(Lenhart et al., 2002; White and Chaubey, 2005). Parameters that have a high degree of influence on the output and that also have a substantial level of uncertainty or variability are commonly fine-tuned in model calibration.

In the AnnAGNPS wetland module, significant physical and chemical processes include the flow of water into and out of the wetland and the removal of pollutants in the water (Bingner et al., 2018). These physical and chemical processes are governed by model equations such as the first-order temperature relationship, where the parameters are edited by the user interface. As a result, while configuring the model parameters, the user may need to make critical parameterization decisions. Therefore, this article discusses several approaches for parameterizing the AnnAGNPS wetland module (AgWET). This study specifically aims to: (1) provide a review on the nitrogen removal component in the wetland module within the Annualized Agricultural Non-point Source Pollution model (AnnAGNPS 5.51); (2) conduct a sensitivity analysis of the key parameters in AnnAGNPS wetland nitrogen removal component; and (3) compare the simulated soluble nitrogen concentration to field measured data from prairie pothole systems. This study does not intend to completely calibrate the model to fit it with our observed data, given the constraints of our limited data. Instead, our attention is directed toward investigating if the model's outputs are within the realm of reasons given the observed data by giving our best effort to effectively parameterize the water-quality relevant model inputs once it has been calibrated for the hydrology.

Methods

Study Area

The primary prairie pothole study area includes ten potholes located on the border of two adjacent HUC-12 watersheds near Ames, Iowa, within the PPR's DML, and a secondary study site with one pothole is located several kilometers away. Various of these potholes have been monitored in a series of related projects since 2016. Ten potholes are located in site one (41.983° N, 93.688° W), situated on the boundary of Story and Boone counties on two neighboring Iowa State University (ISU) fields that span over the Walnut Creek and Worrell Creek HUC-12 basins (fig. 1). The potholes are named Lettuce, Bunny, Plume, Hen, Cardinal, Turkey, Gravy, Yam, Walnut, and Potatoes. One pothole, named Mouth, is situated in site two (42.015° N, 93.743° W), located west of Ames. The potholes in site one are conventionally managed with corn and soybean rotation. On the other hand, Mouth pothole in site two is enrolled in the Conservation Reserve Program (CRP), while its watershed includes mixed managements of traditionally managed crop (corn and soybean) and Miscanthus grass. All of these potholes have been well researched in observed dynamics of water depth and quality in previous studies (Martin et al., 2019a; Martin et al., 2019b) and hydrologically modeled with AnnAGNPS (Nahkala et al., 2021a; Upadhyay et al., 2018).

The potholes were monitored for water level and water quality during periods of inundation during the cropping season, between approximately 15 May and 1 November. In this study, we focus on three potholes – Lettuce, Walnut, and Bunny, from 2016 to 2018, for several reasons. Compared to other potholes, the three chosen potholes had the most water quality samples over the course of three years, which had total rainfall depths during the monitoring periods of 68 cm (27 in.), 54 cm (21 in.), and 90 cm (35 in.), respectively. Furthermore, each of the three potholes had been successfully modeled for water level in prior studies: lettuce was hydrologically calibrated successfully, while Walnut and Bunny were calibrated with moderate success (Nahkala et al., 2021a). These potholes, marked with a red boundary in figure 1, are further referred to as simulated potholes throughout this article. The watershed areas of the simulated potholes varied, ranging from 41.1 to 9.8 ha (table 1). Previous hydrological calibration was performed with the Lettuce pothole having the greatest NSE, RSR, and R2 value and the Bunny pothole having the lowest. The adjusted infiltration rate and Curve Number (CN) value were applied to the model to fit the observed pothole inundated volume (Nahkala et al., 2021a).

Figure 1. Prairie pothole study area (shaded yellow) with three simulated potholes (diagonal stripes): Lettuce, Bunny, and Walnut within enclosed watershed (red lines).

AnnAGNPS Wetland Module

AnnAGNPS is a batch process continuous watershed simulation model that simulates hydrological and non-point source pollution from field-scale agricultural land (Bingner et al., 2018). It is a continuous, daily timestep model built on the AGNPS single-event model. AnnAGNPS 5.51 is, at the time of this writing, the most recent version of this watershed evaluation tool created as part of a USDA-ARS-NRCS collaboration effort. Its goal is to help assess the long-term loading of water and contaminants in ungauged watersheds due to agricultural practices. Several modules that compose AnnAGNPS deal with hydrology, sediment, pesticides, and non-point source pollution as they move through a watershed. It includes AgWET, a GIS-based tool to locate and characterize wetlands at raster grid-scale, and a wetland module with two primary components: nitrogen removal components that simulate the possible nitrogen sink or source, and hydrological features that model water inflow and outflow, including the ponding system in the wetland.

Table 1. Characteristics of simulated potholes.
CharacteristicsFarmed Potholes
BunnyWalnutLettuce
Pothole
characteristics
Pothole area (ha)5.352.602.11
Watershed area (ha)41.19.812.7
Maximum inundation
volume (m3)
29400118008300
Vegetation2016SoybeanSoybeanCorn
2017CornCornSoybean
2018SoybeanSoybeanSoybean
Hydrology
calibration
performance
(daily basis)
NSE0.540.520.78
RSR0.670.690.46
R20.570.520.79

AnnAGNPS AgWET has been increasingly utilized to characterize wetland ponding systems within a watershed using the available database (Yasarer et al., 2018). Several studies have applied the AnnAGNPS wetland module to identify possible wetland locations and determine conservation alternatives to reduce excess agricultural non-point source pollutants. For instance, Tomer et al. (2013) investigated the nitrogen annual nitrogen loss rate from a watershed with several constructed wetlands. Momm et al. (2016) also utilized AnnAGNPS to generate potential wetland locations and characterize wetland characteristics in the Big Bureau Creek watershed, located in north central Illinois. However, limited studies have investigated the AnnAGNPS AgWET nitrogen removal performance in regard to the field monitoring data, and to our knowledge, none have looked at prairie potholes.

In our previous studies, AnnAGNPS was used successfully to simulate the inundation of several farmed prairie potholes in the Des Moines Lobe (Upadhyay et al., 2018; Nahkala et al., 2021a). The potholes were simulated as small wetlands, and their inundation dynamics were assessed using the AgWET module. This module utilizes the extended TR-55 methodology for peak flow rate and the SCS Curve Number (CN) method for runoff depth. In those studies, CN and infiltration rate were calibrated for each simulated pothole and validated against additional years of observed daily inundation. However, these studies did not evaluate model performance in water quality.

Estimates of nitrate removal in the AnnAGNPS AgWET wetland module are derived from the hydraulic loading and nitrate content of the water entering the wetland (Tomer et al., 2013). As explained in the AnnAGNPS technical documentation, the daily wetland nitrogen is computed using pollutant mass balance (Bingner et al., 2018):

        (1)

where

Mi = mass of the pollutant at the end of the day [kg]

M(i-1) = mass of the pollutant at the start of the day [kg]

Minflow = mass of the pollutant added [kg]

Moutflow = mass of the pollutant released [kg]

S = source or sink [kg].

The pollutant added to the system (Minflow) was processed by the AnnAGNPS nitrogen module. The amount of pollutant brought into the wetland as nitrogen discharge coming with water inflow from surface and sub-surface runoff which is influenced by climate factors. It was assumed that there was no pollution released (Moutflow) through the outflow since the potholes were expected to have a closed boundary with no apparent output route, aside from leaching through infiltration.

Wetland systems in AnnAGNPS are assumed to operate as nitrogen sinks. Nitrogen removal is calculated using a temperature dependent first-order model (Crumpton, 2001; Goldsborough and Crumpton, 1998), in which the S is referred to as sink (S) as a function of nitrate-N loss rate (J) [g m-2 day-1].

        (2)

where Awetland is the surface area of the wetland [m2], and the nitrate-N loss rate (J) is determined as seen here:

        (3)

where

k20 = area-based first-order loss rate coefficient for nitrate-N at temperature of 20°C [m day-1]

C = concentration of nitrate-N in the wetland [mg L-1]

? = temperature coefficient for nitrate-N loss

T = water temperature [ºC].

The default value for the temperature coefficient for nitrate-N loss (?) is 1.09. The area-based first-order loss rate coefficient (k20) is a user input with a default coefficient of 0.15. This number can be estimated if the user is aware of the capacity of the wetlands to remove soluble nitrogen or from experiment results of water inundation over the wetland’s sediment (Goldsborough and Crumpton, 1998). Finally, the daily soluble nitrogen concentration (in the form of nitrate-N) in the wetland is calculated in the function of the mass of pollutant at the beginning of the day (Mi-1) per volume of water inflow per unit area (Vi-1) in mm within the wetland area (Awetland). To be consistent with AnnAGNPS technical documentation, the term soluble nitrogen concentration is used throughout this study to express the nitrate-N (NO3-N) concentration.

        (4)

Input File Preparation

Climate Data

AnnAGNPS needs approximately 400 input parameters in 34 data categories, such as, topography, hydrology, land use, wetland input, field management, soil types, and climate (Das et al., 2008). The climate data file is one of the most important input files needed for the AnnAGNPS application. As the primary climatic dataset, the daily weather dataset comprises the beginning day of the desired simulation period through the final day of the simulation period. AnnAGNPS requires the following daily weather parameters: (1) precipitation; (2) maximum air temperature; (3) minimum temperature; (4) dew point; (5) solar radiation; and (6) wind speed. Two sources of climate dataset were used in this study. As in Nahkala et al. (2021a), the Parameter-Elevation Regressions on Independent Slopes Model (PRISM) database was used to obtain daily rainfall data. Daily climate data, including temperatures, solar radiation, dew point, and wind speed, were obtained from the USDA's 'Sustaining the Earth's Watersheds, Agricultural Research Data System' (STEWARDS) project. The data gathered from STEWARDS are combined with PRISM rainfall depths in a single data file and incorporated into the model. This study simulated ten years of weather to obtain nitrogen concentration output for the three-year period from 2016 to 2018. A long-term simulation is required to create sufficient confidence in the projected range of yearly and seasonal precipitation for calculating the pollutant concentration in the potholes. The mass of pollutant inflow into wetlands is heavily influenced by surface and sub-surface flow, which is driven by seasonal precipitation. Thus, the model's output is stabilized throughout the first seven years before the targeted period.

Soil and Land Use Data

Geographic heterogeneity of soils, land use, and topography within a watershed may be simulated in AnnAGNPS by dividing the watershed into numerous user-specified, drainage-area-defined cells. In this study, land cover data were obtained from the National Land Cover Dataset and analyzed with the AnnAGNPS ArcView GIS tool. The USDA Soil Survey Geographic (SSURGO) database, as maintained and disseminated by the National Cartography and Geospatial Center, was used to collect soil information. The NARSIS/SSURGO soil polygon was then intersected by the AnnAGNPS GIS tool. Upadhyay et al. (2018) and Martin et al. (2019a) described further details about agricultural land use and soil types in the study area. The data used in this study are identical to those used in Nahkala et al. (2021a).

Fertilizer Application and Data

According to the field manager, Urea Ammonium Nitrate (UAN) fertilizer with 32% UAN nitrogen was administered in the fall prior to the corn cropping season, with fertilizer rates varying spatially depending on field management (K.Berns, personal communication). Lawrence et al. (2021) reported that before planting corn in the Walnut pothole, a total of 179 kg N ha-1 of Urea Ammonium Nitrate (UAN) fertilizer was applied before the 2017 season and 168 kg N ha-1 before the 2019 season. However, fertilizer information for other potholes is unavailable. Lawrence et al. (2021) also reported that continuous corn cultivation or soybean-corn rotation systems in the Midwest may use 120 to 256 kg N ha-1 of synthetic fertilizer, depending on the region and management system. In this study, although the fertilizer application schedule was recorded by the field management team, the specific rate of fertilizer application within each watershed during the monitoring period is unavailable. Therefore, we considered the amount of fertilizer applied in this study to be subject to parameterization.

Field Monitoring and Laboratory Analysis

The observed inundation depth of each pothole was monitored hourly with a Solinst Level logger pressure transducer positioned at the pothole's bottom elevation. The level loggers were installed within a PVC stilling well dug into the earth at each position. To ensure no damage or disturbance of the transducers from machinery, they were installed just after planting and removed just prior to harvest. The quality of the standing water in the potholes was monitored throughout the cropping season. A one liter sample of ponded water was taken within 24 hours after rainfall occurrences and continued on a daily basis, when potholes were flooded by 10 cm or more. Water samples could not be obtained when the pothole inundation depth was less than 10 cm because of sediment resuspension during water collection. The water samples were then brought for laboratory analysis to find the Nitrate-N concentration with the EPA-103-A Rev 10 method. From the three simulated potholes, a total of 124 samples were collected. The number of water quality samples collected annually from 2016 to 2018 was 57, 5, and 62 samples, respectively. In 2017, few samples were obtained because the potholes were rarely flooded due to precipitation 10% below the 30-year average (Martin et al., 2019a).

Model Sensitivity Analysis

Although AnnAGNPS needs a large number of input parameters, many of them are derived from the DEM processed modules (Das et al., 2008) and thus cannot be analyzed for sensitivity. There is hardly any published research on the specific parameter for sensitivity analysis in the AnnAGNPS wetland nitrogen removal component. Therefore, in this study, the sensitivity analysis included factors described as part of nitrogen removal calculation in the wetland module (Bingner et al., 2018). Other hydrological parameterizations were kept unchanged, as it was assumed that they had been adequately calibrated in prior research. Two modeling parameters and one input value are included in this sensitivity analysis: the area-based nitrogen loss rate coefficient (k20), the temperature coefficient (?), and the fertilizer rate (table 2), using the one-factor-at-a-time (OAT) strategy (Pianosi et al., 2016). Area-based nitrogen loss rate coefficient (k20) and temperature coefficient (?) are user-supplied parameters required to compute nitrogen removal using a temperature dependent first-order equation. Fertilizer rate, on the other hand, is an input value where the user may enter their field measured fertilizer rate. Because the exact fertilizer supplied to the field is unknown and the possible fertilizer application may vary greatly depending on agricultural management, this input value is subjected to parameterization in this research.

The sensitivity of each parameter was estimated using relative sensitivity (Sr) (James and Burges, 1982). This method has been applied in quantifying AnnAGNPS parameter sensitivity for evaluating watershed phosphorus loading (Yuan et al., 2005) and nitrogen loading (Yuan et al., 2003), as well as simulating runoff and sediment (Chahor et al., 2014). The calculation of Sr is expressed as follows:

        (5)

where I12 is the base parameter value and O12 is the corresponding projected daily soluble nitrogen concentration in ponding water. I1, I2, and O1, O2 equal 10% of the initial parameter and its corresponding output values, respectively. The base parameters are a default value of the model (k20 and ?) or the most likely value based on the literature (fertilizer rate). As the output of the model is in daily base units, the Sr was analyzed and presented in daily basis instead of yearly.

Sr offers a platform for comparing across input variables because it is dimensionless (Yuan et al., 2003). The higher the Sr, the more sensitive the model output variable was to that specific parameter. Sr denotes the ratio of a normalized variation in output to a normalized variation in input. An index of one implies that the output varies to the same extent that the input distributes around the average output. Input and output are reversely connected if the Sr value is negative. Based on Lenhart et al. (2002), the sensitivity index can be classified into four categories based on the absolute value of Sr: less than 0.05: low sensitivity; 0.05-0.2: medium sensitivity; 0.2-1.0: high sensitivity; and more than 1: very high sensitivity.

Table 2. Nitrogen loss rate (k20), fertilizer rate, temperature coefficient (?) for sensitivity analysis.
ParametersValues
I1I12I2
Fertilizer rate (kg ha-1)180200220
Nitrogen Loss Rate (k20)0.1350.150.165
Temperature Coefficient (?)0.9811.091.199

Model Performance in Variabilities of Parameterization

Paired parameterization was applied to the simulation in two steps. In the first step, we parameterized combinations of fertilizer and area-based nitrogen loss rate (k20), and from this step, we determined a final fertilizer rate to carry to the second step. In the second step, we parameterized the two most important components of nitrogen removal, k20 and the temperature coefficient (?). These steps were conducted in consideration of the sensitivity of each parameter, as described below.

Various nitrogen loss rates (k20) and fertilizer rates were entered during the initial examination of the parameterizations in order to better comprehend the program's output. The nitrogen loss rates (k20) are set between 0.05 and 0.25, with increments of 0.05. On the other hand, the fertilizer rates are set between 100 and 250 kg N ha-1, with a 50 kg N ha-1 increase (table 3). Using Python v3.10 scripts, models were parameterized step-by-step, iteratively, for both ranges of values within the selected parameter ranges, similar to the approaches reported by Nahkala et al. (2021a).

Table 3. Scenario coded for variabilities of fertilizer rate and k20.
NoScenario
Code
Rate
(kg ha-1)
K20
(m day-1)
1f1k11000.05
2f2k11500.05
3f3k12000.05
4f4k12500.05
5f1k21000.1
6f2k21500.1
…..…..
23f3k52000.25
24f4k52500.25

Table 4. Sensitivity analysis results.
ParameterSensitivity Index (Sr)
Fertilizer rateranges from 0.6 to 0.99
Temperature coefficient (?)ranges from -6.7 to 8.6
First order loss rate (k20)ranges from -3 to -0.01

Similar to the initial investigation, multiple pairings of k20 and ? parameters were set to evaluate the model's output in relation to the observed data. The k20 was set in the same range as before but with smaller increments of 0.025, considering the result from the sensitivity analysis, and ? was set between 1.0 and 1.4 with 0.1 increments. The performance result is displayed as Relative Error (RE), which compares the model output to our observed data. The Relative Error (RE) is computed as follows:

        (6)

where O is the observed soluble nitrogen concentration and P is the predicted soluble nitrogen output from the model. The RE is further expressed in the average value of error based on the collected data.

Results and Discussion

Sensitivity Analysis

The results of the sensitivity analysis (table 4) reveal that the model outcomes are highly or very highly sensitive to all investigated factors, with the temperature coefficient being the most sensitive, followed by the nitrogen loss rate and the fertilizer rate. The fertilizer rate is positively correlated with the soluble nitrogen concentration in the water, with a sensitivity index of daily soluble nitrogen concentration output ranging from 0.6 to 0.99, which is regarded as high. The nitrogen loss rate (k20) is inversely proportional to nitrogen concentration, with daily relative sensitivity ranging from -3 to -0.01, indicating that as k20 increases, nitrogen concentrations decrease. On the other hand, as a function of temperature within the first-order temperature model, the coefficient of temperature has a positive or negative correlation with the daily nitrogen concentration output. AnnAGNPS uses the average air temperature as the assumption for the water temperature, resulting in a number of large deviations from the base temperature (20°) as the air temperature fluctuates throumayghout the cropping season. Figure 2 illustrates an example of temperature coefficient parameterizations for a simulated Bunny pothole from 22 September to 30 September 2016. The ? was adjusted from 1.0 to 2.5 with 0.5 increments. Other parameters were left unaltered. It is evident that mean temperatures below 20°C result in a highly variable output as the temperature coefficient increases.

Model Performance Under Various Parameterizations

The parameterization steps and increments considered the results of the sensitivity analysis. Figure 3 depicts the initial exploration of parameterization confluence between fertilizer rate and k20 during the cropping season of 2016 in Bunny pothole. The average annual soluble nitrogen concentration during ponding from all types of scenarios ranges from 0.01 to 0.14 mg N L-1. The changes of soluble nitrogen concentration are coherent with how sensitive the parameter is. For example, increasing the fertilizer rate by 50 kg N ha-1 raises the mean annual soluble nitrogen content by 0.026 mg N L-1. In contrast, a 0.05 increase in k20 decreases the mean annual soluble nitrogen concentration by 60%. Out of all scenarios, the maximum concentration of 1.5 mg N L-1 was simulated during the highest setting of fertilizer rate and lowest k20. This number is far below the maximum nitrogen concentration found in the observed data, which is 15.6 mg N L-1.

The most apparent difference observed in this investigation across different values of k20 is the decrease in the non-zero output as k20 increases. For instance, between f1k1 and f1k2, 25% of the output of the simulated soluble nitrogen concentration becomes zero, despite earlier showing a substantial number of more than 0.5 mg N L-1. It indicates that the soluble nitrogen concentrations in the inundated water have been totally removed. This result suggests that AnnAGNPS may not have adequate data resolution for simulated farmed prairie pothole nitrogen, as it occasionally returns a value of zero. From our simulations, the lowest non-zero output value is approximately 0.01 mg N L-1, which is tenfold higher than the observed data.

As highly sensitive parameters, the standard error between observed and simulated data varies widely among the three simulated potholes during k20 and ? parameterization (fig. 4). Even though we failed to determine a fertilizer rate and k20 that follow the nitrogen concentration pattern of our observed data, we decided to keep the fertilizer of 200 kg N ha-1 based on the average fertilizer application rate in the Midwest (Lawrence et al., 2021). The Bunny pothole has the greatest range of average daily RE, which is caused by the model's inability to accurately forecast the high concentration of soluble nitrogen in the beginning of cropping season in 2016. The model also does not account for the low nitrogen concentrations that can be found in the field. In a number of instances, our observed data are two to three orders of magnitude below the output, generating a huge discrepancy

Figure 2. Example of soluble nitrogen concentrations output with several temperature coefficient (?) parameterizations and air temperature fluctuations at Bunny pothole in September 2016.
Figure 3. Simulated soluble nitrogen concentration in varieties of k20 and ? parameterizations during the cropping season of 2016 in Bunny pothole. The scenario codes represent the fertilizer rate (f) and the k20 (k) used in the simulation. For instance, f1k1 denotes a fertilizer rate of 100 kg N ha-1 and k20 of 0.05. An increase by one in the scenario codes indicates raising f and k by one level, which is 50 kg N ha-1 and 0.05, respectively.

between predicted and actual data. Therefore, the low nitrate concentration we discovered in the field, which is employed as the denominator in the RE computations, is responsible for RE values of more than a thousand. The RE in the Walnut pothole are, in general, lower than those found in the other two simulated potholes; however, this does not necessarily imply that the model fits the observed data better. As explained previously, the model tends to have poor resolution of output when it comes to a very small concentration. Thus, the output of the model can abruptly become zero, exhibiting a smaller standard error (RE) than when we observed an extremely low nitrogen content.

Comparison of Predicted and Observed Nitrogen Concentration

Three potholes are simulated with constant parameters of 200 kg N ha-1 for fertilizer rate before corn year, 0.075 for k20, and 1.1 for ?. Those numbers were selected after long consideration of the best possible representative output with enough amount of data (non-zero) to be compared with the observed field data. Although fertilizer rates vary in the field depending on field management and soil nutrient demands, in this part of the study, the fertilizer rate was kept constant at 200 kg N ha-1 for all potholes considering the average nitrogen fertilizer application in the Midwest (Lawrence et al., 2021). This decision was made due to our inability to identify the fertilizer rate that produces the output that most closely resembles the observed data.

(a)
(b)
(c)
Figure 4. Relative error (%) of the soluble nitrogen concentration between observed and simulated data rate in (a) Lettuce, (b) Bunny, and (c) Walnut potholes with the varieties of loss rate (k20) and temperature coefficient (?) under constant fertilizer rate of 200 kg N ha-1.

The average simulated annual nitrogen concentration during cropping season in Bunny, Lettuce, and Walnut potholes is 6.37, 8.78, and 1.24 mg N L-1, respectively. In 2016, high levels of soluble nitrogen were detected at the beginning of the cropping season, which the model failed to simulate (fig. 5). As 2017 was a dry year with 52 cm (21 in.) of total precipitation during the cropping season, limited samples of water quality were collected. The soluble nitrogen concentration pattern from the model output corresponded well with the observed concentration in Bunny pothole in 2018 but was considerably lower than the observed soluble nitrogen concentration in Lettuce and Walnut potholes (fig. 5). In 2018, the majority of potholes were planted with soybean (table 1), and thus no N fertilizer was applied (K.Berns, personal communication). Furthermore, 2018 was a wet year with 32% higher precipitation than the 30-year average, resulting in a lower soluble nitrogen concentration than the previous two years.

The poor capability of AnnAGNPS in estimating nutrient loading has been reported in several other studies (e.g., Yuan et al., 2003, 2005, 2011). According to Yuan et al. (2011), despite AnnAGNPS giving a satisfactory result in simulating the monthly runoff, it tends to underpredict nitrogen loading from the watershed. The authors suggested several possible explanations, including that more nitrogen fertilizer was applied than was specified in the model input, and the likelihood that an underestimation of runoff might result in an underestimation of nutrient loading. Our finding, however, shows that even with the variety of fertilizer rates, the model is still unable to follow the daily pattern of soluble nitrogen concentration.

Figure 5. Timeseries of the simulated soluble nitrogen (solid black line), simulated inundation dynamics (dashed red line), and observed soluble nitrogen concentration in three simulated potholes: Bunny (square), Lettuce (round), and Walnut (triangle) during 3 year monitoring period. Note that the y-axis scale is different for each graph.
Figure 6. Comparison between observed and simulated soluble nitrogen concentration in three simulated potholes: Bunny (square), Lettuce (round), and Walnut (triangle).

Our evaluation of the comparison between observed data and simulation output also indicates that a better hydrologically calibrated pothole does not always mean the soluble nitrogen content is simulated more accurately (fig. 6). For example, Lettuce pothole with the highest RSR and NSE score among all three potholes in terms of water volume fitness (Nahkala et al., 2021a) does not give a better fit of simulated nitrogen concentration to the observed data. The big difference between observed and predicted nitrogen concentration can especially be seen in the beginning of the cropping season. Yuan et al. (2003) previously reported that AnnAGNPS tends to overestimate nitrogen loss throughout the winter due to denitrification, and underestimate nitrogen loadings throughout the crop season due to leaching loss. The underestimation of the nitrogen loading to the wetland led to an underestimation of the soluble nitrogen within the potholes. The simplicity of the model in analyzing the nitrogen processes and the variability of input parameters might be the cause of the poor performance of the model in determining the nitrogen loadings. Further, although AnnAGNPS output is on a daily time scale, the model was initially designed for long-term analysis of agricultural practices, which may result in less accuracy and resolution in daily output compared to longer temporal scales such as monthly or yearly. However, for systems like the farmed prairie potholes, which inundate for short periods at a time, monthly or yearly aggregated data are not possible. Therefore, model enhancements such as improved daily resolution output, expanded functionality, and refined mathematical formulation, of the AnnAGNPS nitrogen removal component might be valuable in enhancing the accuracy of daily soluble nitrogen for simulating pothole systems.

Conclusion

The soluble nitrogen concentrations in three potholes were estimated using AnnAGNPS. The AnnAGNPS 5.51 model incorporates a wetland module that predicts daily soluble nitrogen concentration using a temperature dependent first-order model. Hydrologically calibrated models were employed in this study to further analyze the wetland nitrogen removal component. The sensitivity analysis of three critical nitrogen removal parameters in AnnAGNPS demonstrates a high and very high level of sensitivity, resulting in significant output variations. Our investigations in multiple parameterizations indicate that AnnAGNPS simulate soluble nitrogen poorly. The model typically fails to predict the high nitrogen concentration at the beginning of the cropping season. The limited ability of AnnAGNPS to predict nitrogen loading may result in an underestimate of soluble nitrogen within the potholes.

Our results align with prior research about nitrogen and phosphorus loading to wetland using AnnAGNPS, demonstrating the model limited capability to simulate nitrogen processes during cropping season in watershed and wetland systems. This study demonstrates how AnnAGNPS AgWET's nitrogen removal component may provide a wide range of output with parameterizations to three coefficients. Even though AnnAGNPS provides daily time step data, the model outputs do not correspond with our daily water quality data from the field, which raises the possibility that the model does not accurately reflect accuracy in daily resolution data. The simplicity of the model may contribute to the poor performance in nitrogen loading and removal mechanisms. Model improvements such as enhanced resolution, functionality, and mathematical formulation of the AnnAGNPS nitrogen removal component might be useful for better simulating daily soluble nitrogen in potholes systems.

It is particularly difficult to simulate nitrogen removal in wetland systems because variations in pothole soluble nitrogen concentrations are highly dependent on complicated spatial and temporal trends. Pothole topographic location, as isolated transient wetlands within a watershed, necessitates a thorough modeling of two interconnected systems, the watershed and the wetlands. As hotspots for nutrient routing downstream, however, the ability to simulate water quality in the prairie potholes environment can be advantageous for future conservation practices. Given the increasing utilization of models in the examination of Total Maximum Daily Load (TMDL) and Best Management Practices (BMP) for the purpose of attaining nutrient reduction and creating policy, it is important to ensure the accuracy of the nutrient transport and removal components within the model. Thus, it is necessary to conduct further investigation into the possibility of modeling water quality in pothole systems and incorporating conservation practices to achieve the nutrient reduction targets.

Acknowledgments

Part of this project was supported by a Fulbright Program grant sponsored by the Bureau of Educational and Cultural Affairs of the United States Department of State and administered by the Institute of International Education.

References

Baker, J. L., Campbell, K. L., Johnson, H. P., & Hanway, J. J. (1975). Nitrate, phosphorus, and sulfate in subsurface drainage water. J. Environ. Qual., 4(3), 406-412. https://doi.org/10.2134/jeq1975.00472425000400030027x

Bingner, R. L., Theurer, F., & Yuan, Y. (2018). AnnAGNPS TECHNICAL PROCESSES - Documentation. Version 5.5.

Cambardella, C. A., Moorman, T. B., Jaynes, D. B., Hatfield, J. L., Parkin, T. B., Simpkins, W. W., & Karlen, D. L. (1999). Water quality in walnut creek watershed: Nitrate-nitrogen in soils, subsurface drainage water, and shallow groundwater. J. Environ. Qual., 28(1), 25-34. https://doi.org/10.2134/jeq1999.00472425002800010003x

Chahor, Y., Casalí, J., Giménez, R., Bingner, R. L., Campo, M. A., & Goñi, M. (2014). Evaluation of the AnnAGNPS model for predicting runoff and sediment yield in a small Mediterranean agricultural watershed in Navarre (Spain). Agric. Water Manag., 134, 24-37. https://doi.org/10.1016/j.agwat.2013.11.014

Crumpton, W. G. (2001). Using wetlands for water quality improvement in agricultural watersheds; the importance of a watershed scale approach. Water Sci. Technol., 44(11-12), 559-564. https://doi.org/10.2166/wst.2001.0880

Crumpton, W. G., Helmers, M. J., Stenback, G. A., Lemke, D. W., & Richmond, S. (2012). Integrated drainage-wetland systems for reducing nitrate loads from Des Moines Lobe watersheds - Agricultural and Biosystems Engineering Technical Reports and White Papers. Iowa State University. Retrieved from http://lib.dr.iastate.edu/abe_eng_reports/13

Das, S., Rudra, R. P., Gharabaghi, B., Gebremeskel, S., Goel, P. K., & Dickinson, W. T. (2008). Applicability of AnnAGNPS for Ontario conditions. Can. Biosyst. Eng., 50(1), 1-11.

Drum, R. G., Loesch, C. R., Carrlson, K. M., Doherty, K. E., & Fedy, B. C. (2015). Assessing the biological benefits of the USDA-Conservation Reserve Program (CRP) for waterfowl and grassland passerines in the Prairie Pothole Region of the United States: Spatial analyses for targeting CRP to maximize benefits for migratory birds - Final Report for USDA–FSA Agreement 12-IA-MRE-CRP-TA. 1-98. Prairie Pothole Joint Venture. Retrieved from https://www.fsa.usda.gov/Assets/USDA-FSA-Public/usdafiles/EPAS/PDF/drumetal2015_crp_prr_final.pdf

Ewel, K. C. (1997). Water quality improvement by wetlands. In G. C. Daily (Ed.), Nature’s services: Societal dependence on natural ecosystems. Washington, DC: Island Press.

Gan, Y., Duan, Q., Gong, W., Tong, C., Sun, Y., Chu, W.,... Di, Z. (2014). A comprehensive evaluation of various sensitivity analysis methods: A case study with a hydrological model. Environ. Model. Softw., 51, 269-285. https://doi.org/10.1016/j.envsoft.2013.09.031

Gleason, R. A., Euliss Jr., N. H., Tangen, B. A., Laubhan, M. K., & Browne, B. A. (2011). USDA conservation program and practice effects on wetland ecosystem services in the Prairie Pothole Region. Ecol. Appl., 21(sp1), S65-S81. https://doi.org/10.1890/09-0216.1

Goldsborough, L. G., & Crumpton, W. G. (1998). Distribution and environmental fate of pesticides in prairie wetlands. Great Plains Res., 8(1), 73-95. Retrieved from https://www.jstor.org/stable/24156335

Green, D. I., McDeid, S. M., & Crumpton, W. G. (2019). Runoff storage potential of drained upland depressions on the Des Moines Lobe of Iowa. JAWRA - J. Am. Water Resour. Assoc., 55(3), 543-558. https://doi.org/10.1111/1752-1688.12738

Hubbard, D. E. (1988). Glaciated prairie wetland functions and values: A synthesis of the literature (Vol. 88). U.S. Department of the Interior, Fish and Wildlife Service, Research and Development.

Ikenberry, C. D., Crumpton, W. G., Arnold, J. G., Soupir, M. L., & Gassman, P. W. (2017). Evaluation of existing and modified wetland equations in the SWAT model. JAWRA - J. Am. Water Resour. Assoc., 53(6), 1267-1280. https://doi.org/10.1111/1752-1688.12570

James, L. D., & Burges, S. (1982). Chapter 11: Selection, calibration, and testing of hydrologic models. In C. T. Haan, H. P. Johnson, & D. L. Brakensiek (Eds.), Hydrologic modeling of small watersheds - Hydrologic modeling monograph. St. Joseph, MI: American Society of Agricultural Engineers.

Lawrence, N. C., Tenesaca, C. G., VanLoocke, A., & Hall, S. J. (2021). Nitrous oxide emissions from agricultural soils challenge climate sustainability in the US Corn Belt. Proc. Natl. Acad. Sci., 118(46), e2112108118. https://doi.org/10.1073/pnas.2112108118

Lenhart, T., Eckhardt, K., Fohrer, N., & Frede, H. G. (2002). Comparison of two different approaches of sensitivity analysis. Phys. Chem. Earth., Parts A/B/C, 27(9), 645-654. https://doi.org/10.1016/S1474-7065(02)00049-9

Martin, A., Kaleita, A. L., & Soupir, M. L. (2019a). Inundation patterns of farmed pothole depressions with varying subsurface drainage. Trans. ASABE, 62(6), 1579-1590. https://doi.org/10.13031/trans.13435

Martin, A., Soupir, M. L., & Kaleita, A. L. (2019b). Seasonal and intra-event nutrient levels in farmed prairie potholes of the Des Moines Lobe. Trans. ASABE, 62(6), 1607-1617. https://doi.org/10.13031/trans.13414

McDeid, S. M., Green, D. I., & Crumpton, W. G. (2019). Morphology of drained upland depressions on the Des Moines Lobe of Iowa. Wetlands, 39(3), 587-600. https://doi.org/10.1007/s13157-018-1108-4

McKenna, O. P., Osorio, J. M., Behrman, K. D., Doro, L., & Mushet, D. M. (2020). Development of a novel framework for modeling field-scale conservation effects of depressional wetlands in agricultural landscapes. J. Soil Water Conserv., 75(6), 695-703. https://doi.org/10.2489/jswc.2020.00096

Mitsch, W. J., Bernal, B., & Hernandez, M. E. (2015). Ecosystem services of wetlands. Int. J. Biodivers. Sci. Ecosyst. Serv. Manag., 11(1), 1-4. https://doi.org/10.1080/21513732.2015.1006250

Momm, H. G., Bingner, R. L., Yuan, Y., Kostel, J., Monchak, J. J., Locke, M. A., & Gilley, A. (2016). Characterization and placement of wetlands for integrated conservation practice planning. Trans. ASABE, 59(5), 1345-1357. https://doi.org/10.13031/trans.59.11635

Nahkala, B. A., Kaleita, A. L., & Soupir, M. L. (2021a). Assessment of input parameters and calibration methods for simulating prairie pothole hydrology using AnnAGNPS. Appl. Eng. Agric., 37(3), 495-503. https://doi.org/10.13031/aea.14399

Nahkala, B. A., Kaleita, A. L., & Soupir, M. L. (2021b). Characterization of prairie pothole inundation using AnnAGNPS under varying management and drainage scenarios. Agric. Water Manag., 255, 107002. https://doi.org/10.1016/j.agwat.2021.107002

Pianosi, F., Beven, K., Freer, J., Hall, J. W., Rougier, J., Stephenson, D. B., & Wagener, T. (2016). Sensitivity analysis of environmental models: A systematic review with practical workflow. Environ. Model. Softw., 79, 214-232. https://doi.org/10.1016/j.envsoft.2016.02.008

Richardson, J. L., Arndt, J. L., & Freeland, J. (1994). Wetland soils of the prairie potholes. In D. L. Sparks (Ed.), Advances in agronomy (Vol. 52, pp. 121-171). Academic Press. https://doi.org/10.1016/S0065-2113(08)60623-9

Ross, C. D., McKenna, O. P. (2022). The Potential of Prairie Pothole Wetlands as an Agricultural Conservation Practice: A Synthesis of Empirical Data. Wetlands, 43(1), 5. https://doi.org/10.1007/s13157-022-01638-3

Schilling, K. E., & Dinsmore, S. (2018). Monitoring the wildlife, hydrology and water quality of drained wetlands of the Des Moines Lobe, Northern Iowa: Introduction to special feature. Wetlands, 38(2), 207-210. https://doi.org/10.1007/s13157-017-0989-y

Schilling, K. E., & Helmers, M. (2008). Effects of subsurface drainage tiles on streamflow in Iowa agricultural watersheds: Exploratory hydrograph analysis. Hydrol. Process., 22(23), 4497-4506. https://doi.org/10.1002/hyp.7052

Singh, R., Helmers, M. J., Crumpton, W. G., & Lemke, D. W. (2007). Predicting effects of drainage water management in Iowa’s subsurface drained landscapes. Agric. Water Manag., 92(3), 162-170. https://doi.org/10.1016/j.agwat.2007.05.012

Skaggs, R. W., Brevé, M. A., & Gilliam, J. W. (1994). Hydrologic and water quality impacts of agricultural drainage. Crit. Rev. Environ. Sci. Technol., 24(1), 1-32. https://doi.org/10.1080/10643389409388459

Tiner, R. W. (2003). Geographically isolated wetlands of the United States. Wetlands, 23(3), 494-516. https://doi.org/10.1672/0277-5212(2003)023[0494:GIWOTU]2.0.CO;2

Tomer, M. D., Crumpton, W. G., Bingner, R. L., Kostel, J. A., & James, D. E. (2013). Estimating nitrate load reductions from placing constructed wetlands in a HUC-12 watershed using LiDAR data. Ecol. Eng., 56, 69-78. https://doi.org/10.1016/j.ecoleng.2012.04.040

Upadhyay, P., Kaleita, A. L., & Soupir, M. L. (2020). Assessment of USGS DEMs for modelling pothole inundation in the prairie pothole region of Iowa. Geocarto Int., 35(9), 1018-1032. https://doi.org/10.1080/10106049.2019.1573852

Upadhyay, P., Pruski, L. O., Kaleita, A. L., & Soupir, M. L. (2018). Evaluation of AnnAGNPS for simulating the inundation of drained and farmed potholes in the Prairie Pothole Region of Iowa. Agric. Water Manag., 204, 38-46. https://doi.org/10.1016/j.agwat.2018.03.037

White, K. L., & Chaubey, I. (2005). Sensitivity analysis, calibration, and validations for a multisite and multivariable swat model. JAWRA - J. Am. Water Resour. Assoc., 41(5), 1077-1089. https://doi.org/10.1111/j.1752-1688.2005.tb03786.x

Winter, T. C., & Rosenberry, D. O. (1995). The interaction of ground water with prairie pothole wetlands in the Cottonwood Lake area, east-central North Dakota, 1979-1990. Wetlands, 15(3), 193-211. https://doi.org/10.1007/BF03160700

Yasarer, L. M., Bingner, R. L., & Momm, H. G. (2018). Characterizing ponds in a watershed simulation and evaluating their influence on streamflow in a Mississippi watershed. Hydrol. Sci. J., 63(2), 302-311. https://doi.org/10.1080/02626667.2018.1425954

Yuan, Y., Bingner, R. L., & Rebich, R. A. (2003). Evaluation of AnnAGNPS nitrogen loading in an agricultural watershed. JAWRA - J. Am. Water Resour. Assoc., 39(2), 457-466. https://doi.org/10.1111/j.1752-1688.2003.tb04398.x

Yuan, Y., Bingner, R. L., Theurer, F. D., Rebich, R. A., & Moore, P. A. (2005). Phosphorus component in AnnAGNPS. Trans. ASAE, 48(6), 2145-2154. https://doi.org/10.13031/2013.20100

Yuan, Y., Mehaffey, M. H., Lopez, R. D., Bingner, R. L., Bruins, R., Erickson, C., & Jackson, M. A. (2011). AnnAGNPS model application for nitrogen loading assessment for the future midwest landscape study. Water, 3(1), 196-216. https://doi.org/10.3390/w3010196