ASABE Logo

Article Request Page ASABE Journal Article

Feasibility of Predicting Subsurface  Drainage Discharge With DRAINMOD Parameterized by Uncalibrated SURRGO  Soil Properties and ROSETTA3

Manal H. Askar1,*, Ehsan Ghane2, Mohamed A. Youssef3, Vinayak S. Shedekar1, Kevin W. King4, Rabin Bhattarai5

Published in Journal of Natural Resources and Agricultural Ecosystems 2(2): 39-52 (doi: 10.13031/jnrae.15735). 2024 American Society of Agricultural and Biological Engineers.

1    Food, Agricultural and Biological Engineering, Ohio State University, Columbus, Ohio, USA.

2    Biosystems and Agricultural Engineering, Michigan State University, East Lansing, Michigan, USA.

3    Biological and Agricultural Engineering, North Carolina State University, Raleigh, North Carolina, USA.

4    Soil Drainage Research Unit, USDA ARS, Columbus, Ohio, USA.

5    Agricultural and Biological Engineering, University of Illinois, Urbana, Illinois, USA.

*    Correspondence: askar.18@osu.edu

Highlights

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 14 July 2023 as manuscript number NRES 15735; approved for publication as a Research Article by Associate Editor Dr. Debasmita Misra and Community Editor Dr. Kati Migliaccio of the Natural Resources & Environmental Systems Community of ASABE on 7 January 2024.

Mention of company or trade names is for description only and does not imply endorsement by the USDA. The USDA is an equal opportunity provider and employer.

Abstract. Implementing hydrologic and water quality (HWQ) models in decision-support tools (DSTs) is essential to increasing their adoption by a wide user base. However, the performance of HWQ models and, consequently, DSTs in predicting site hydrology are highly dependent on the proper representation of site-specific soil properties. The objective of this study was to assess the accuracy of DRAINMOD predictions of subsurface drainage discharge using uncalibrated SSURGO- and ROSETTA3-based soil data. First, the model performance was examined by comparing predicted discharge using uncalibrated soil input data to measured discharge from three sites across the Midwest (Vermillion, IL; Delaware, OH; and Clayton, MI) with a total of 15 site-years of data. A second evaluation of the model performance was conducted by comparing predicted discharge using uncalibrated soil parameters to the model predictions using calibrated soil input parameters. The model performance (i.e., using uncalibrated soil parameters) in predicting drainage discharge compared to measured data ranged from good to excellent, with daily mean Nash-Sutcliffe efficiencies (NSEs) of 0.67 at the Vermillion site, 0.60 at the Delaware site, and 0.82 at the Clayton site. The use of calibrated soil input parameters resulted in better goodness-of-fit between measured and predicted discharge (i.e., monthly NSE range: 0.76 – 0.87) than the uncalibrated soil parameters scenario (i.e., monthly NSE range: 0.65 – 0.86) but was not significantly different (i.e., t-test, p range: 0.59 – 0.97> 0.05). Our results suggest that using SSURGO and ROSETTA3-based soil input data in DRAINMOD is an acceptable approach for representing site-specific hydrologic conditions when soil inputs are unavailable, thereby, presenting a potential for implementing the model into DSTs.

Keywords.Decision-support tools, Modeling, Pedotransfer functions, Soil physical properties.

Hydrologic and water quality (HWQ) modeling is a time- and cost-effective means to analyze and quantify the impact of conservation practices on nutrient loss under different scenarios (Muenich et al., 2016; Nangia et al., 2010) and future climate projections (Gunn et al., 2018; Kujawa et al., 2022; Pease et al., 2017). DRAINMOD (Skaggs, 1982) is one of the widely applied field-scale models used to simulate the hydrology and water quality of subsurface-drained agricultural lands (Cox et al., 1994; Luo et al., 2000; Singh et al., 2006; Skaggs et al., 2012; Tolomio and Borin, 2018; Wang et al., 2006; Youssef et al., 2021). Understanding the theoretical basis, input parameters, and calibration process of DRAINMOD is essential for proper model use and application (Skaggs et al., 2012); however, not all users have the required expertise to apply DRAINMOD. A more simplified and user-friendly DRAINMOD-based framework is needed to extend the model’s use in terms of the number of users as well as the number of applications.

Implementing DRAINMOD in user-friendly, decision-support tools (DSTs) would facilitate the use and application of the model by a wider audience. Such DSTs could assist farmers and stakeholders with limited knowledge of the model in making site-specific decisions related to design (Ghane et al., 2021; Ghane and Askar, 2021), management, or policy. DRAINMOD-based DSTs require soil parameters (e.g., texture, bulk density, and hydraulic conductivity) that could be either measured or acquired from publicly available soil databases such as the Soil Survey Geographic Database (SSURGO) (Soil Survey Staff, 2005). Site-specific soil information, housed within SSURGO, can facilitate the development of web and GIS-based DSTs, allowing for automated extraction of such information for any geo-location of interest in the USA. In-situ measurements of these soil parameters, specifically lateral hydraulic conductivity (Ojha et al., 2017; Van Beers, 1970), can be challenging, time consuming, and cost prohibitive. Thus, previous DRAINMOD-based studies generally used publicly available soil data, such as the SSURGO database, of which select parameters were often calibrated in applications where measured hydrology and water quality data were available (Askar et al., 2020, 2021).

Pedotransfer functions (PTFs) are widely-used predictive functions that translate available soil physical parameters into soil hydraulic properties commonly required for modeling purposes (Odeh and McBratney, 2005; van Genuchten, 1980). The majority of PTFs are based on regression equations that estimate a soil hydraulic property using readily available soil physical properties (Gupta and Larson, 1979; Saxton et al., 1986; Saxton and Rawls, 2006; Wosten et al., 1999). However, neural-network-based approaches have also been applied to improve the performance of PTFs (Minasny and McBratney, 2002; Schaap et al., 1998, 2001). ROSETTA (Schaap et al., 2001) is one of the commonly used PTF tools that adopts artificial neural network approaches and has been implemented in the HYDRUS model (Šimunek and van Genuchten, 2008). ROSETTA has previously been used with DRAINMOD (Adhikari et al., 2020; Pease et al., 2017; Shedekar et al., 2021; Singh et al., 2006) to predict required model parameters such as saturated hydraulic conductivity (Ksat) and soil water retention curve parameters for the van Genuchten equation (van Genuchten, 1980). ROSETTA3 (Zhang and Schaap, 2017) is the most recent and applied version of the model (Cao et al., 2019; Pannecoucke et al., 2020; Thorp, 2020).

Limited studies have investigated the accuracy of DRAINMOD predictions using model inputs based on SSURGO- and ROSETTA-soil properties. For instance, a four-year simulation study in Sweden (Salazar et al., 2008) investigated the variation in DRAINMOD predictions using laboratory measured Ksat compared to ROSETTA Ksat estimates from four available ROSETTA models. The Ksat  

estimates from the four ROSETTA approaches ranged from -62% to +126% of the laboratory measured values for all soil layers in their study. The authors found that predicted subsurface drainage discharge using ROSETTA-based Ksat values was comparable to predictions using measured Ksat with a variation of less than 15%. Similarly, Qi et al. (2015) investigated the performance of DRAINMOD using a combination of measured soil properties, SSURGO soil data, and ROSETTA predicted properties for a silty clay loam soil and concluded that all ROSETTA models, based on SSURGO data, gave reasonable soil-input estimates, resulting in acceptable drainage discharge predictions (monthly Index-of-agreement = 0.92). However, we are not aware of any study that compared field-scale measured drainage discharge to DRAINMOD predictions while parameterizing the model solely with SSURGO and ROSETTA3 generated soil properties in the Corn Belt region of the Midwest USA under different soil types and climatic conditions.

Understanding the ability of HWQ models such as DRAINMOD to predict subsurface drainage discharge using soil inputs solely based on available soil databases and PTFs is necessary to provide confidence in the approach and extend the use of the model. The objective of this study was to assess the feasibility of using SSURGO- and ROSETTA3-based soil data to parameterize the DRAINMOD model for predicting subsurface drainage discharge. To facilitate this assessment, measured subsurface drainage discharge data as well as calibrated DRAINMOD models from three sites at different locations across the US Midwest were used. This study represents an initial step towards integrating DRAINMOD into a Decision Support Tool (DST) designed to automate model parameterization using SSURGO-based soil data inputs.

Materials and Methods

DRAINMOD Description

DRAINMOD is a widely used field-scale model that was developed to simulate the hydrology of poorly drained soils (Skaggs, 1978). Over the years, the model has been expanded to simulate water quality (Askar et al., 2021; Youssef et al., 2005) and crop yield (Evans et al., 1991; Negm et al., 2014). The hydrologic component of the model simulates key hydrologic processes such as infiltration, subsurface drainage discharge, surface runoff, evapotranspiration, vertical and lateral seepage, soil water distribution in the vadose zone, and water table depth (WTD) (Skaggs et al., 2012). The model inputs include weather, soil, crop, and drainage system parameters. A comprehensive description of soil characteristics is required for each soil layer, with reasonable estimates for soil physical and hydraulic parameters. These parameters include soil lateral hydraulic conductivity (Ksat) and water retention relationships, among others. Parameters that could be estimated by the model’s soil utility program include the drained water volume-WTD relationship, the upward flux-WTD relationship, and the Green-Ampt infiltration parameters (Green and Ampt, 1911). In the case of using ROSETTA, the soil-utility program within DRAINMOD generates the required soil input files using the van Genuchten curve parameters, residual and saturated water content (WC), Ksat, and tortuosity for each soil layer. If not available, these parameters can be obtained from PTF tools such as ROSETTA.

ROSETTA3 Description

ROSETTA is a PTF tool that utilizes artificial neural network analysis to estimate the van Genuchten water retention parameters (Schaap et al., 2001). ROSETTA3, the model’s latest version, includes five different models with different levels of complexity that can be selected according to input data availability (Zhang and Schaap, 2017). The simplest model includes a lookup table with mean values and standard deviations for each USDA soil textural class. As the complexity of the ROSETTA3 models increases, the number of input parameters increases. Model five of ROSETTA3, selected for this study, exhibits the highest complexity and necessitates the definition of bulk density and volumetric water content (WC) at field capacity (33 kPa) and at wilting point (1500 kPa), in addition to soil texture (i.e., percentages of sand, silt, and clay) for each soil layer (Zhang and Schaap, 2017). ROSETTA3 predicts soil hydraulic parameters, including water retention curve inputs for the van Genuchten equation, as well as residual and saturated WC for each soil layer. These outputs can be used in the DRAINMOD soil utility program to generate DRAINMOD soil input files.

Description of Study Sites and  DRAINMOD Calibration

The study sites were selected based on the following criteria: (1) at least one year of daily measured weather and subsurface drainage discharge data were available (Skaggs et al., 2012); (2) the field was under conventional free drainage during the study period as controlled drainage has the potential to increase surface runoff (King et al., 2022; Tan and Zhang, 2011), evapotranspiration (Shedekar et al., 2021; Skaggs et al., 2010), and lateral seepage (Youssef et al., 2018, 2021); (3) the drained field is hydrologically separated from adjacent fields or under the same drainage management practices; (4) the field area is greater than 4.0 ha (10 acres) to minimize the influence of boundary conditions which increases as the size of a field decreases (Skaggs et al., 2010); and (5) DRAINMOD has been previously calibrated and validated at the site thus, model simulations for this study would only require replacement of soil input parameters with SSURGO- and ROSETTA3-based parameters. Three sites fulfilled the selection criteria and were included in this study. These field sites were located in Vermillion, IL; Delaware, OH; and Clayton, MI (fig. 1). Specific characteristics of each site are briefly described next and summarized in table 1.

The Vermillion, IL, site is located in the Little Vermillion River Watershed near Danville, Illinois, and has been previously documented in detail by Singh et al. (2020). The site is comprised of 51% Drummer silty clay loam (fine-silty, mixed, superactive, mesic Typic Endoaquolls), which is classified as poorly drained. Most of the remaining drainage area (41%) was characterized as a Flanagan silt loam (fine, smectitic, mesic Aquic Argiudolls), which is classified as somewhat poorly drained. The field followed a corn (Zea mays L.) and soybean (Glycine max L.) rotation, with annual chisel plowing and disking or field cultivation. DRAINMOD was calibrated at this site using seven years (1992-1998) of measured data (Singh et al., 2020). The site did not have a parallel pattern drainage layout; thus, drain spacing was a calibration parameter. Vertical seepage was considered in the calibrated model, with seepage parameters being adjusted during the calibration process. Other calibration parameters included surface storage, temperature parameters, and soil parameters such as lateral Ksat that were originally obtained from the SSURGO database.

The Delaware, OH, site is located within the Upper Big Walnut Creek watershed in central Ohio, as described by Shedekar et al. (2021). The dominant soil type (> 66% of the area) is Bennington silt loam (fine, illitic, mesic Aeric Epiaqualfs), a somewhat poorly drained soil. Pewamo silty clay loam (fine, mixed, active, mesic Typic Argiaquolls) represents the remainder of the field and is classified as very poorly drained. Field management consisted of a corn-soybean rotation with tillage prior to corn planting. Ten years (2005-2014) of measured data were used to calibrate DRAINMOD at this site with four years of conventional drainage management implemented and six years of controlled drainage (Shedekar et al., 2021). Soil input parameters for the Bennington and Pewamo soil series were based on the SSURGO database along with ROSETTA and adjusted during the calibration process. Other calibrated parameters include surface storage, monthly potential evapotranspiration (PET) correction factors, and soil temperature parameters.

The Clayton, MI, site is a subsurface-drained field, located in the River Raisin watershed within the Western Lake Erie Basin in southeastern Michigan. A detailed description of the study site and data collection was previously documented (Dialameh and Ghane, 2022; Shokrana and Ghane, 2021). The dominant soil type is Blount loam (fine, illitic, mesic Aquic Hapludalfs) which represents 61% of the drained area and is classified as somewhat poorly drained. Glynwood loam (fine, illitic, mesic Aquic Hapludalfs) represents 27% of the field and is classified as moderately well-drained. The field has been managed under a corn-soybean rotation and long-term no-tillage practices (+25 years), with ryegrass planted as a cover crop during the non-growing season. Three years (2019-2021) of measured data (unpublished) were used to calibrate DRIANMOD at this site. Seepage was considered negligible at this site. Soil texture and bulk density were based on measured data. Water contents at field capacity and wilting point were estimated using the ranges outlined by Rawls et al. (1982), Initial values of lateral Ksat were obtained from the SSURGO database and adjusted during the calibration process. Other calibrated parameters include surface storage, PET correction factors, and soil temperature parameters.

Assessment Approach

Physical properties such as depth and vertical Ksat of each soil layer, as well as the depth of the restrictive layer, were obtained from the SSURGO database using the dominant soil types at each location (SM 1). However, DRAINMOD 6.1 requires Ksat for each soil layer in the lateral direction. Thus, lateral Ksat was assumed to be twice the vertical Ksat values reported in SSURGO (Qi et al., 2015; Sadhukhan et al., 2019; Thorp et al., 2009). Depth to restrictive layer was reported as > 200 cm (6.6 ft) in the SSURGO database for dominant soils at the Vermillion and Delaware sites. Thus, we assumed a restrictive layer depth of 305 cm (10 ft) (Anderson et al., 1984). The depth of the restrictive layer at the Clayton site, provided by the SSURGO database, was shallower than the drain depth. Therefore, we used the same assumption of 305 cm depth for the restrictive layer.

Figure 1. Location of the study sites in Illinois, Michigan, and Ohio that fulfilled the selection criteria and were used in this study.

Model five in ROSETTA3 was used in our study to predict soil hydraulic parameters such as water retention curve parameters of the van Genuchten equation for each soil layer (Zhang and Schaap, 2017). Input parameters required for model five (i.e., soil texture, bulk density, and volumetric WC at field capacity and wilting point for each soil layer) were obtained from the SSURGO database. Soil hydraulic parameters, predicted by ROSETTA3, along with Ksat values from the SSURGO data(table 2), were used in the DRAINMOD soil utility program to generate the required soil input files.

Table 1. Selected characteristics of the sites, including the number of monitored years, drainage, soil, and management data.
SiteVermillion, ILDelaware, OHClayton, MI
Period of analysis8 years (1992-1999)4 years (2005-2008)3 years (2019-2021)
Long-term (1991-2020) average annual precipitation (cm)103.9105.291.9
Drainage area (ha)4.914.914.7
Drainage system
Drain depth (m)1.10.90.9
Drain spacing (m)36.512.213.0
Soil type
(contributing area %)		Drummer silty clay (51%),
Flanagan silt loam (42%)		Bennington silt loam (66%),
Pewamo silty clay loam (34%)		Blount loam (61%),
Glynwood loam (27%)
Drainage class
(primary, secondary soil)		Poorly,
somewhat poorly		Somewhat poorly,
very poorly		Somewhat poorly,
moderately well
Cropping systemCorn-Soybean rotation Corn-Soybean rotationCorn-Soybean rotation
ManagementChisel plowed + disked
or field cultivated		Tillage only prior to cornNo-till (+25 years)

DRAINMOD simulations were initialized one year prior to the period of interest to minimize the impact of initial conditions on drainage discharge predictions (Askar et al., 2020; Thorp et al., 2009). Except for the soil input parameters, all model input parameters (i.e., weather, crop, and drainage system) were obtained from previously calibrated DRAINMOD studies at each location. DRAINMOD simulations were conducted for each dominant soil at each location using SSURGO- and ROSETTA3-based soil parameters. Then, the area-weighted average of daily drainage discharge was calculated using the ratio of the corresponding drainage area of each soil type (eq. 1).

        (1)

where

Qtotal = representative area-weighted discharge from the two dominant soils (cm)

A1 and A2 = areas of the two dominant soils (ha)

Q1 and Q2 = DRAINMOD-predicted discharges using soil inputs corresponding to the two dominant soils (cm)

Atotal = total area of the two dominant soils (i.e., A1 + A2, ha).

Model Evaluation Criteria  and Statistical Analysis

DRAINMOD performance using SURRGO- and ROSETTA3-based soil parameters was first evaluated graphically and statistically by comparing predicted and measured drainage discharges on daily, monthly, and annual time scales at each site. This analysis enables us to investigate the model performance using uncalibrated soil parameters (i.e., other model input parameters are calibrated) against measured data. Based on common model performance statistics reviewed by Moriasi et al. (2015), the model performance was assessed using the Nash-Sutcliffe efficiency (NSE) and normalized percent error (NPE) statistical measures following the criteria outlined by Skaggs et al. (2012). The NSE is a useful indicator to evaluate goodness-of-fit between measured and predicted drainage discharge values, but it is very sensitive to large discrepancies as the error is squared (Moriasi et al., 2015). Daily and monthly predicted drainage discharges were assessed using the NSE. Additionally, monthly predictions were also evaluated using scatter plots and box plots to provide visual insights of model performance. The NPE measure was used to indicate if the model systematically overpredicted or underpredicted drainage discharge on an annual basis. The performance criteria used for evaluating predicted drainage discharge have been previously outlined by Skaggs et al. (2012) and can be summarized as satisfactory (daily NSE > 0.4, monthly NSE > 0.5, and NPE < ±25%), good (daily NSE > 0.6, monthly NSE > 0.7, and NPE < ±15%), or excellent (daily NSE > 0.75, monthly NSE > 0.8, and NPE < ±5%). Trends in seasonal prediction were also examined by calculating the summation of discharge in the fall (Oct–Dec), winter (Jan–Mar), spring (Apr–Jun), and summer (Jul–Sep), then calculating the seasonal average across years at each site.

Table 2. Statistics of measured and predicted drainage discharge using both uncalibrated and calibrated soil parameters at the Vermillion, Delaware, and Clayton sites at the daily, monthly, and annual temporal scales, along with a two-tailed Student paired t-test comparing predicted discharge using calibrated vs. uncalibrated soil parameters.
Site/YearPrecipitation
(cm)		Measured
Drainage
discharge
(cm)		Uncalibrated soil parameters predictions
(SSURGO-ROSETTA3)		Calibrated soil parameters predictions
Drainage
discharge
(cm)		Daily
NSE(a)		Monthly
NSE		Annual
NPE(b)
(%)		Drainage
discharge
(cm)		Daily
NSE(a)		Monthly
NSE		Annual
NPE(b)
()%
Vermillion, IL199298.120.418.90.770.90-7.720.80.780.941.9
1993129.751.553.90.650.704.755.10.700.736.9
199484.521.514.60.740.84-32.314.70.680.78-31.6
199574.211.910.10.670.93-15.811.70.810.97-2.1
1996100.927.518.90.760.92-31.420.90.780.92-24.1
199778.719.89.90.530.63-50.012.10.570.73-38.8
1998126.024.332.60.450.7834.231.00.710.8927.4
199969.14.53.40.260.66-24.06.8-0.29-0.6253.1
(1992-1999)95.222.720.280.670.86-10.621.60.720.87-8.4
Delaware, OH2005114.729.936.50.610.4417.628.50.690.64-4.7
2006104.624.329.00.650.3915.622.90.570.57-5.9
2007107.833.233.30.790.89-1.624.30.620.90-26.9
200899.731.033.50.350.73-1.122.10.500.73-28.5
(2005-2008)106.729.633.10.600.706.924.50.600.76-17.4
Clayton, MI2019111.751.547.50.850.87-7.747.90.860.87-7.0
2020100.840.934.70.780.85-15.235.70.870.87-12.8
2021100.743.740.00.810.75-8.637.90.840.79-13.5
(2019-2021)104.445.440.70.820.82-10.240.50.860.84-10.8

    (a)    NSE = Nash-Sutcliffe efficiency.

    (b)    NPE = normalized percent error.

The model performance was also examined by comparing predicted drainage discharge using uncalibrated soil parameters (i.e., SURRGO- and ROSETTA3-based soil parameters) to predictions using calibrated soil parameters obtained from the previously mentioned DRAINMOD studies at each location (table SM 2). This comparison enables us to investigate the extent to which the accuracy of predicted discharge declines when using uncalibrated soil parameters instead of calibrated ones. In other words, quantifying the loss in model accuracy in terms of discharge prediction due to uncalibrated soil input parameters. The previously mentioned statistical measures (i.e., NSE and NPE) were used for conducting this comparison, as well as the same performance criteria outlined by Skaggs et al. (2012). Additionally, differences between predicted monthly discharges using uncalibrated versus calibrated soil parameters were evaluated using a two-tailed Student paired t-test at a significance level of 0.05. The goodness-of-fit between the calibrated and uncalibrated drainage predictions was also evaluated by calculating the coefficient of determination (R-square) to investigate the correlation between the two scenarios. Additionally, the percent bias (PBIAS) was examined to test if the model tends to systematically over- or underpredict drainage discharge (Moriasi et al., 2015). A negative PBIAS value indicates that uncalibrated predictions are greater than calibrated ones.

Results and Discussion

Model Performance Using Uncalibrated Soil Input Parameters Versus Measured Data

Vermillion Site, IL

Daily predicted drainage discharge at the Vermillion site was in good agreement with measured values (i.e., NSE = 0.67, table 2). A general overprediction of drainage discharge at peak flow events was observed during some periods, including the wet summer of 1993 (fig. 2). Additionally, predicted hydrograph receding curves were relatively more rapid than measured values (e.g., the period from December 1992 to February 1993, fig. 2), indicating a faster transport of predicted water flow through the soil profile, which resulted in relatively larger peak flows and less base flow. This could be partially explained by greater hydraulic conductivity input values that could contribute to more than 80% of the uncertainty in predicted subsurface drainage discharge (Skaggs et al., 2012). Previous studies showed that DRAINMOD predictions are sensitive to lateral hydraulic conductivity (Golmohammadi et al., 2016), particularly within the bottom layer and the layer containing the subsurface drainage system (Haan and Skaggs, 2003). Other factors that might have contributed to the discrepancy between measured and predicted drainage discharge include greater ET predictions (average = 65.2 cm) compared to the range (50.3-54.9 cm) reported in previous DRAINMOD studies in Illinois (Youssef et al., 2018). Additionally, the poor model performance in some years was attributed to inaccurate predictions of a single or a few events. For instance, in 1998, the model predicted a snowmelt event on 17 March that was 4 days earlier than observed values and resulted in decreasing the NSE by 0.3 for this year (fig. 2). Furthermore, the two dominant soils at the Vermillion site represented only 76% of the total area, which implies that soil properties in 24% of the area have not been considered in our simulations and could have affected discharge predictions.

Figure 2. Daily measured and predicted drainage discharge using uncalibrated soil parameters along with daily precipitation at the Vermillion, IL, site (1992-1999).

Monthly predicted drainage discharge at the Vermillion site was in excellent agreement with measured values (i.e., NSE = 0.86, table 2); however, the model generally underpredicted discharge in months with low flow (fig. 3a). During the eight-year simulation period, the model predicted zero flow in seventeen months, while measured drainage discharge ranged from 0.2 to 2.3 cm. In contrast, the model overpredicted monthly drainage discharge during years with intense precipitation or snowmelt events, resulting in wider interquartile ranges compared to measured values (fig. 3b). As previously mentioned, this discrepancy could be attributed to our initial observation of underpredicted base flow and overpredicted peak flow at this location, possibly due to inaccurate hydraulic conductivity values (Haan and Skaggs, 2003), uncertainty in drainable porosity, and the focus on only the top two dominant soils in our simulations. Monthly prediction efficiencies were slightly better than daily efficiencies (table 2), likely because the underprediction of base flow was balanced in some instances by the overprediction of peak flow.

The mean annual drainage discharge at the Vermillion site was underpredicted by 11% (table 2), indicating good model performance (i.e., NPE < ±15%). However, overpredictions and underpredictions in discharge were noted across individual years. The underprediction of annual discharge was generally associated with dry years (i.e., below 30-year NOAA average precipitation, table 1), while the overprediction was associated with wet years (i.e., above 30-year NOAA average precipitation, table 1) (table 2). For instance, the large underprediction of annual discharge in 1994 (-32%) and 1997 (-25%) took place when the annual precipitation was less than the long-term mean annual precipitation (103.9 cm) by 19% and 25%, respectively. In contrast, the model overpredicted annual discharge in 1993 (5%) and 1998 (34%), two exceptionally wet years during the simulation period. This was partially due to the overprediction in peak flow during large events that took place in wet years and the general underprediction in base flow that was dominant in dry years. Additionally, the larger predicted ET likely contributed to the general underpredicted discharge.

Figure 3. (a) A scatter diagram of measured and predicted monthly drainage discharge using uncalibrated soil parameters, and (b) a box and whiskers plot of monthly measured and predicted drainage discharge at the Vermillion, IL, site (1992-1999). The line in each box represents the median drainage discharge, the edges of the box represent the 25th and 75th percentiles, and the whiskers represent the minimum and maximum values.

Delaware Site, OH

Daily predicted drainage discharge at the Delaware site was in good agreement with measured values (i.e., NSE = 0.60, table 2). Predicted hydrographs followed measured discharge values, with some instances of overprediction or underprediction in peak flow events (fig. 4). For example, the model predicted zero discharge on 8 February 2005, while measurements showed a value of 1.4 cm. The discrepancy between measured and predicted discharge for this event is most likely related to the model’s freezing and thawing input parameters, which predicted earlier snowmelt, as indicated by the 6.9 cm (64%) overprediction in drainage discharge over the first two weeks of January 2005. In another example, the model considerably underestimated a significant snowmelt event on 1 March 2008, predicting only 0.06 cm compared to the measured 3.8 cm. For this particular event, the underprediction in discharge was primarily due to uncertainty in the drainable porosity used in the model, particularly during the winter season when the soil is partially frozen, which resulted in predicting 3.2 cm of surface runoff. However, there was no instrumentation at the site for measuring surface runoff to verify predicted volumes. Similar discrepancies due to snowmelt input parameters have been reported in previous DRAINMOD studies (Luo et al., 2010; Negm et al., 2017). Additionally, wind drift snow is not represented in DRAINMOD, which might have contributed to underpredicting snowmelt during some events (Luo et al., 2001). Overpredicted peak flow events highlight hydraulic conductivity values that might have been greater than the actual values. Additionally, average annual seepage (1.8 cm) was less than what has been reported in the literature for similar soil types (Thorp et al., 2009) and could have resulted in shallower predicted WTD and consequently greater discharge.

Monthly predicted drainage discharge at the Delaware site was in good agreement with measured values (i.e., NSE = 0.70, fig. 5a), however, monthly NSE in 2006 (NSE = 0.39) was unsatisfactory (table 2). As previously mentioned, the NSE is very sensitive to extreme values due to the squared error (Moriasi et al., 2015), and in some cases, the model’s poor performance can be attributed to a single or a few events. For example, the snowmelt overprediction in January 2005 resulted in a considerable underprediction (67%) of drainage discharge in February and reduced the NSE by 57% for that year. Similarly, underpredicting snowmelt in March 2008 reduced monthly NSE by 0.18 in that year. The discrepancies are likely due to the impact of freezing and thawing input parameters that affected the timing and magnitude of snowmelt events (Luo et al., 2000, 2001). Thus, these events contributed to the wider predicted interquartile ranges for the months of January (range: 4.1-12.2 cm) and February (range: 1.3-4.1 cm) and the narrower interquartile range for March (range: 4.3-7.4 cm) compared to measured values (fig. 5b).

Mean annual drainage discharge at the Delaware site was overpredicted by 7% (table 2), indicating good model performance (i.e., NPE < ±15%). The overprediction in discharge only occurred in 2005 (16%) and 2006 (17%). In 2005, most of the overprediction in discharge occurred during January, indicating an error in the timing of snowmelt prediction. This could be explained by the approaches used in DRAINMOD to predict snowmelt, which occurs when the temperature on a specific day rises above a user defined snowmelt base temperature (Luo et al., 2000, 2001). However, hourly air temperatures can greatly vary over a day, resulting in snowmelt during warmer daytime temperatures that could be underpredicted by the model if the average daily temperature is below the snowmelt base temperature. Additionally, the relatively low predicted seepage might have contributed to the general overprediction of drainage discharge.

Figure 4. Daily measured and predicted drainage discharge using uncalibrated soil parameters along with daily precipitation at the Delaware, OH, site (2005-2008).
Figure 5. (a) A scatter diagram of measured and predicted monthly drainage discharge using uncalibrated soil parameters and (b) a box and whiskers plot of monthly measured and predicted drainage discharge at the Delaware, OH, site (2005-2008). The line in each box represents the median drainage discharge, the edges of the box represent the 25th and 75th percentiles, and the whiskers represent the minimum and maximum values.

Clayton Site, MI

Daily predicted drainage discharge at the Clayton site was in excellent agreement with measured values (i.e., NSE = 0.82, table 2). Predicted peak flow events were in good agreement with measured values in terms of timing and magnitude, but in some instances, they were largely underpredicted or completely missed by the model (fig. 6). For example, the flow event (1.3 cm) on 18 August 2020 was largely underpredicted (-83%) while the model predicted 7.9 cm of surface runoff on the same day. This can partially be attributed to the volume drained–WTD relationship that resulted in predicting deeper WTD than the actual conditions. Another possibility could be the existence of macropore flow, which can be an important pathway for drainage discharge in fine textured soils (Williams et al., 2016), particularly in no-till fields (Buczko et al., 2006; Cullum, 2009; Schwen et al., 2011), as present at the Clayton study site. Macropore flow has a greater impact during the spring and summer seasons as a result of root development and dryer conditions (Schwen et al., 2011). However, the macropore flow component is not represented in DRAINMOD 6.1 used in this study.

The monthly predicted drainage discharge at the Clayton site was in excellent agreement with measured values (i.e., NSE = 0.82, fig. 7a). Additionally, monthly predicted discharge medians were comparable to measurements with a median difference within ± 1.1 cm for all months (fig. 7b). The underestimated discharge of the flow events that took place during October 2021 decreased the monthly NSE (0.75) for this year by 0.2 and the upper limit of the predicted interquartile range for the month of October by 2.6 cm. This could be partially attributed to deeper WTD predictions and the development of macropores over the summer period (Williams et al., 2016), particularly when conditions were relatively dry.

Figure 6. Daily measured and predicted drainage discharge using uncalibrated soil parameters along with daily precipitation at the Clayton, MI, site (2019-2021).
Figure 7. (a) A scatter diagram of measured and predicted monthly drainage discharge using uncalibrated soil parameters and (b) a box and whiskers plot of monthly measured and predicted drainage discharge at the Clayton, MI, site (2019-2021). The line in each box represents the median drainage discharge, the edges of the box represent the 25th and 75th percentiles, and the whiskers represent the minimum and maximum values.

The mean annual drainage discharge at the Clayton site was underpredicted by 10% (table 2), indicating good model performance (i.e., NPE < ±15%). The underprediction of discharge varied across years and ranged from about 8% in 2019 to 15% in 2020. On average, about 31% of underpredicted discharge occurred over the summer period (i.e., July–September). Factors that may have influenced daily and aforementioned monthly predictions are also relevant to annual model performance. These include predicting deeper WTD and possible macropore flow (Williams et al., 2016) in the absence of tillage (Buczko et al., 2006; Cullum, 2009; Schwen et al., 2011). Additionally, the predicted average annual ET (55 cm) was on the high side compared to values reported in the literature (range: 45-54 cm) for DRAINMOD studies in MI (Youssef et al., 2018).

Model Performance Using Uncalibrated  Versus Calibrated Soil Input Parameters

Vermillion Site, IL

The daily model performance at the Vermillion site using uncalibrated soil parameters was not significantly different (i.e., p-value = 0.33 at 95% confidence level, R-square = 0.93) from the calibrated soil parameters scenario (table 2, fig. 8a). This is not surprising given that all the drainage, weather, and crop parameters were the same under both scenarios. Additionally, except for Ksat, all soil characteristics and parameters used in Singh et al.’s (2020) study were based on the SSURGO database as well. However, the uncalibrated model had a general tendency to overpredict high flows (PBIAS for values > 75th percentile = -18%) (table 2, fig. 8a), which can be attributed to multiple factors. The Ksat values used in the uncalibrated scenario were approximately two times greater than the values used by Singh et al. (2020) (tables SM 1 and SM 2), thus accelerating water movement within the soil profile and resulting in slightly larger drainage volumes (fig. 8a). This indicates that vertical Ksat, reported in SSURGO, was considered representative of lateral Ksat in Singh et al. (2020) rather than multiplying the values by a factor of two as done in this study. Additionally, in the uncalibrated scenario, the depth to the restrictive layer was at 305 cm depth compared to 260 cm in the calibrated model, which provides more storage in the soil profile and can result in deeper WTD as well as less discharge. Finally, Singh et al. (2020) used only one representative soil for the entire field in their study, which could have affected the discharge predictions (Singh et al., 2020).

Figure 8. A scatter diagram of daily predicted drainage discharge at (a) Vermillion, IL, (c) Delaware, OH, and (e) Clayton, MI sites and monthly drainage discharge at (b) Vermillion, IL, (d) Delaware, OH, and (f) Clayton, MI sites using calibrated versus uncalibrated soil parameters.

Monthly drainage discharge predictions of DRAINMOD using uncalibrated soil parameters were not significantly different (i.e., p-value = 0.77 at 95% confidence level, R-square = 0.97) from predictions by the model with calibrated soil parameters (table 2, fig. 8b). Additionally, the overprediction in daily drainage discharge that was previously observed in high flow events was not observed on the monthly time-scale, indicating a better model performance on the monthly scale compared to the daily scale. This could be explained by the overpredicted discharge in high flow events being balanced by the underprediction in base flow events. Our results highlight that the model performance was comparable at both timescales but exhibited a better performance on the monthly timescale. On the annual timescale, mean annual discharge using uncalibrated parameters was comparable to the calibrated scenario (table 2).

Delaware Site, OH

The daily model performance at the Delaware site using uncalibrated soil parameters was not significantly different (i.e., p-value = 0.15 at 95% confidence level, R-square = 0.92) from the calibrated soil parameters scenario (table 2, fig. 8c). The insignificant difference between the two scenarios aligns with our results for the Vermillion site. As previously mentioned, except for soil parameters, all other model input parameters in the two scenarios were the same. Additionally, soil input parameters in the calibrated model were based on the SSURGO database and the ROSETTA model, with some changes made during the calibration process. However, an overprediction in high flow discharge using uncalibrated soil parameters compared to the calibrated scenario was observed (PBIAS for values > 50th percentile = -16.18%, fig. 8c). Similar to the Vermillion site, Ksat values for most soil layers used in the uncalibrated scenario were about two times greater than the calibrated scenario. Larger Ksat values, particularly for the top layer, resulted in accelerated water movement and larger discharge volumes during peak flow events (fig. 8c). Additionally, the corresponding areas for each soil type were different in the calibrated scenario (i.e., Bennington = 54% and Pewamo = 46%) than the values used in the present study (i.e., Bennington = 66% and Pewamo = 34%), most likely due to delineation approaches. This could have resulted in an overprediction of discharge in our simulations, as Ksat values for deeper soil layers of the Pewamo soil are 2.5 times greater than the values used in Shedekar et al. (2021).

The model’s monthly performance using uncalibrated soil parameters was not significantly different (i.e., p-value = 0.59 at 95% confidence level, R-square = 0.97) from the calibrated soil parameters scenario (table 2, fig. 8d). Similar to the Vermillion site, monthly model predictions using uncalibrated soil parameters had closer agreement with the calibrated scenario than the daily values, indicating better performance on the monthly timescale. The balance between overprediction and underprediction of discharge could be due to the use of a deeper restrictive layer (305 cm) in the uncalibrated scenario, which allows for more water storage within the soil profile compared to the calibrated model's shallower layer (215 cm). However, the general overprediction in high flow events resulted in overpredicting mean annual discharge by 6.9% using uncalibrated parameters, compared to an underprediction of 17.4% in the calibrated scenario.

Figure 9. Seasonal average measured and predicted drainage discharge using uncalibrated soil parameters along with precipitation data at the three study sites (i.e., Vermillion, Delaware, and Clayton).

Clayton Site, MI

The daily model performance using uncalibrated soil parameters was not significantly different (i.e., p-value = 0.94 at 95% confidence level, R-square = 0.92) from the calibrated soil parameters scenario (table 2, fig. 8d). Again, this is attributed to all input parameters in the two scenarios being the same except the soil parameters. In the calibrated scenario, soil texture and bulk density for the top three layers were based on measured data. Water content at saturation, field capacity, and wilting point were estimated using the ranges outlined by Rawls et al. (1982). The Ksat values were initially obtained from the SSURGO database and adjusted during the calibration process. Variations between predicted discharge using calibrated and uncalibrated soil parameters were observed. Factors discussed earlier for the Vermillion and Delaware sites are also relevant to the Clayton site and could have contributed to differences in predicted discharge for both scenarios. These include using only one representative soil for the site, having a shallower depth to restrictive layer, and having lower Ksat values for the top and bottom soil layers in the calibrated scenario compared to the uncalibrated scenario of the present study. Despite the differences between the calibrated and uncalibrated scenarios, the performance of the model was similar under these two scenarios.

The model monthly performance using uncalibrated soil parameters was not significantly different (i.e., p-value = 0.97 at 95% confidence level, R-square = 0.96) from the calibrated soil parameters scenario (table 2, fig. 8f). Monthly model predictions of the uncalibrated scenario were almost the same as the calibrated scenario (PBIAS = -0.66%), with a few incidences of overprediction or underprediction. Following the Vermillion and Delaware sites, predicted monthly discharge using uncalibrated soil parameters was in closer agreement with the calibrated scenario in contrast to daily predictions. Additionally, the mean annual discharge for the uncalibrated scenario (40.7 cm) was almost identical to the calibrated scenario (40.5 cm).

Cross-Site Comparison

The three sites considered for this study varied in climatic conditions, soil characteristics, and management practices, with drainage areas ranging from 4.9 to 14.9 ha (table 1). This is indicated by the precipitation variation across all sites (table 2). The soils at the sites ranged from loams to silty clay and were generally classified as poorly to somewhat poorly drained soils. Management practices differed among sites, with tillage being implemented at the Vermillion and Delaware sites while no-tillage was adopted at the Clayton site. From a seasonal perspective, a considerable amount of the discharge at the Vermillion site took place during the spring season, which represented 45% of the measured discharge during the simulation period (1992-1999) (fig. 9). For the four-year simulation period at the Delaware location, about 54% of the flow occurred during the winter season. In contrast, the Clayton site was draining during each of the winter, spring, and fall seasons, with only limited amounts during the summer. Therefore, the data show that the seasonality of flow was variable across seasons and sites.

Despite the differences across sites, the model performance in predicting drainage discharge using uncalibrated soil parameters can generally be rated as good (Skaggs et al., 2012). However, the performance at the Clayton site was better than the other two sites. This is largely attributed to the good performance of the calibrated model at the Clayton site. In contrast, the lowest model performance was observed at the Delaware location, a site with soils that have a drainage classification of somewhat poorly to very poorly. Our results show that we did not lose a lot of accuracy in terms of predicting discharge using uncalibrated soil parameters at each of the three sites.

Conclusion

The reliability of DRAINMOD in predicting drainage discharge with uncalibrated soil input parameters sourced entirely from the SSURGO database, in conjunction with ROSETTA3, was investigated across three sites in the U.S. Midwest region. The model’s performance in predicting drainage discharge across sites could be rated as good (daily NSE = 0.60 and monthly NSE = 0.7). Predicted annual drainage discharges at all locations were within the NPE of ±15%, indicating an overall good model performance. However, biased predictions with a tendency to overpredict daily peak flows compared to the calibrated scenario were observed at two locations. The accuracy of the predictions was generally affected by saturated hydraulic conductivity and depth to restrictive layer values. In contrast, the model performed relatively well on the monthly timescale with no detectable bias.

It should be noted that DRAINMOD’s performance in this study was affected by the use of uncalibrated soil parameters along with other calibrated weather, crop, and drainage parameters. Therefore, our results do not imply that the current model’s performance is representative of DRAINMOD’s performance if the model were to be implemented into a DST under a completely uncalibrated scenario. Rather, this study informs the users of the expected decline in the accuracy of DRAINMOD predictions that is caused by the use of uncalibrated SSURGO-ROSETTA3 soil inputs. Overall, our results show that SSURGO-based soil data, along with ROSETTA3, can be used to obtain an acceptable representation of field conditions and predict drainage discharge using DRAINMOD.

Supplemental Material

The supplemental materials mentioned in this article are available for download from the ASABE Figshare repository at: https://doi.org/10.13031/25075922

Acknowledgments

The authors would like to thank Yousef Abdalaal for his help with obtaining the SSURGO soil data at the three sites. Additionally, the authors are grateful to the anonymous reviewers for their valuable feedback that helped to improve the manuscript. This work was funded by the USDA-NRCS Classic Conservation Innovation Grant (Award No. NR213A750013G030).

References

Adhikari, N., Davidson, P. C., Cooke, R. A., & Book, R. S. (2020). DRAINMOD-linked interface for evaluating drainage system response to climate scenarios. Appl. Eng. Agric., 36(3), 303-319. https://doi.org/10.13031/aea.13383

Anderson, L., Scheib, W., Bruns, E., Lucas, P., Allred, E., & Weiberg, E. (1984). Minnesota Drainage Guide. USDA, Soil Conservation Service. Retrieved from https://drive.google.com/file/d/1u8JWGmtFiObHzZGtlBCuIQ6Qmbylmhgg/view

Askar, M. H., Youssef, M. A., Chescheir, G. M., Negm, L. M., King, K. W., Hesterberg, D. L.,... Skaggs, R. W. (2020). DRAINMOD Simulation of macropore flow at subsurface drained agricultural fields: Model modification and field testing. Agric. Water Manag., 242, 106401. https://doi.org/10.1016/j.agwat.2020.106401

Askar, M. H., Youssef, M. A., Hesterberg, D. L., King, K. W., Amoozegar, A., Skaggs, R. W.,... Ghane, E. (2021). DRAINMOD-P: A model for simulating phosphorus dynamics and transport in drained agricultural lands: II. Model testing. Trans. ASABE, 64(6), 1849-1866. https://doi.org/10.13031/trans.14510

Buczko, U., Bens, O., & Hüttl, R. F. (2006). Tillage effects on hydraulic properties and macroporosity in silty and sandy soils. Soil Sci. Soc. Am. J., 70(6), 1998-2007. https://doi.org/10.2136/sssaj2006.0046

Cao, B., Quan, X., Brown, N., Stewart-Jones, E., & Gruber, S. (2019). GlobSim (v1.0): Deriving meteorological time series for point locations from multiple global reanalyses. Geosci. Model Dev., 12(11), 4661-4679. https://doi.org/10.5194/gmd-12-4661-2019

Cox, J. W., Mcfarlane, D. J., & Skaggs, R. W. (1994). Field-evaluation of DRAINMOD for predicting waterlogging intensity and drain performance in South-Western Australia. Aust. J. Soil Res., 32(4), 653-671. https://doi.org/10.1071/SR9940653

Cullum, R. F. (2009). Macropore flow estimations under no-till and till systems. CATENA, 78(1), 87-91. https://doi.org/10.1016/j.catena.2009.03.004

Dialameh, B., & Ghane, E. (2022). Effect of water sampling strategies on the uncertainty of phosphorus load estimation in subsurface drainage discharge. J. Environ. Qual., 51(3), 377-388. https://doi.org/10.1002/jeq2.20339

Evans, R. O., Skaggs, R. W., & Sneed, R. E. (1991). Stress day index models to predict corn and soybean relative yield under high water table conditions. Trans. ASAE, 34(5), 1997-2005. https://doi.org/10.13031/2013.31829

Ghane, E., & Askar, M. H. (2021). Predicting the effect of drain depth on profitability and hydrology of subsurface drainage systems across the eastern USA. Agric. Water Manag., 258, 107072. https://doi.org/10.1016/j.agwat.2021.107072

Ghane, E., Askar, M. H., & Skaggs, R. W. (2021). Design drainage rates to optimize crop production for subsurface-drained fields. Agric. Water Manag., 257, 107045. https://doi.org/10.1016/j.agwat.2021.107045

Golmohammadi, G., Rudra, R. P., Prasher, S. O., Madani, A., Goel, P. K., & Mohammadi, K. (2016). Modeling the impacts of tillage practices on water table depth, drain outflow and nitrogen losses using DRAINMOD. Comput. Electron. Agric., 124, 73-83. https://doi.org/10.1016/j.compag.2016.03.031

Green, W. H., & Ampt, G. A. (1911). Studies on soil phyics. J. Agric. Sci., 4(1), 1-24. https://doi.org/10.1017/S0021859600001441

Gunn, K. M., Baule, W. J., Frankenberger, J. R., Gamble, D. L., Allred, B. J., Andresen, J. A., & Brown, L. C. (2018). Modeled climate change impacts on subirrigated maize relative yield in Northwest Ohio. Agric. Water Manag., 206, 56-66. https://doi.org/10.1016/j.agwat.2018.04.034

Gupta, S. C., & Larson, W. E. (1979). Estimating soil water retention characteristics from particle size distribution, organic matter percent, and bulk density. Water Resour. Res., 15(6), 1633-1635. https://doi.org/10.1029/WR015i006p01633

Haan, P. K., & Skaggs, R. W. (2003). Effect of parameter uncertainty on DRAINMOD predictions: I. Hydrology and yield. Trans. ASAE, 46(4), 1061. https://doi.org/10.13031/2013.13968

King, K. W., Hanrahan, B. R., Stinner, J., & Shedekar, V. S. (2022). Field scale discharge and water quality response, to drainage water management. Agric. Water Manag., 264, 107421. https://doi.org/10.1016/j.agwat.2021.107421

Kujawa, H., Kalcic, M., Martin, J., Apostel, A., Kast, J., Murumkar, A.,... Scavia, D. (2022). Using a multi-institutional ensemble of watershed models to assess agricultural conservation effectiveness in a future climate. JAWRA, 58(6), 1326-1340. https://doi.org/10.1111/1752-1688.13023

Luo, W., Sands, G. R., Youssef, M., Strock, J. S., Song, I., & Canelon, D. (2010). Modeling the impact of alternative drainage practices in the northern corn-belt with DRAINMOD-NII. Agric. Water Manag., 97(3), 389-398. https://doi.org/10.1016/j.agwat.2009.10.009

Luo, W., Skaggs, R. W., & Chescheir, G. M. (2000). DRAINMOD modifications for cold conditions. Trans. ASAE, 43(6), 1569-1582. https://doi.org/10.13031/2013.3057

Luo, W., Skaggs, R. W., Madani, A., Cizikci, S., & Mavi, A. (2001). Predicting field hydrology in cold conditions with DRAINMOD. Trans. ASAE, 44(4), 825. https://doi.org/10.13031/2013.6247

Minasny, B., & McBratney, A. B. (2002). The Neuro-m method for fitting neural network parametric pedotransfer functions. Soil Sci. Soc. Am. J., 66(2), 352-361. https://doi.org/10.2136/sssaj2002.3520

Moriasi, D. N., Gitau, M. W., Pai, N., & Daggupati, P. (2015). Hydrologic and water quality models: Performance measures and evaluation criteria. Trans. ASABE, 58(6), 1763-1785. https://doi.org/10.13031/trans.58.10715

Muenich, R. L., Kalcic, M., & Scavia, D. (2016). Evaluating the impact of legacy P and agricultural conservation practices on nutrient loads from the Maumee River Watershed. Environ. Sci. Technol., 50(15), 8146-8154. https://doi.org/10.1021/acs.est.6b01421

Nangia, V., Gowda, P. H., Mulla, D. J., & Sands, G. R. (2010). Modeling impacts of tile drain spacing and depth on nitrate-nitrogen losses. Vadose Zone J., 9(1), 61-72. https://doi.org/10.2136/vzj2008.0158

Negm, L. M., Youssef, M. A., & Jaynes, D. B. (2017). Evaluation of DRAINMOD-DSSAT simulated effects of controlled drainage on crop yield, water balance, and water quality for a corn-soybean cropping system in central Iowa. Agric. Water Manag., 187, 57-68. https://doi.org/10.1016/j.agwat.2017.03.010

Negm, L. M., Youssef, M. A., Skaggs, R. W., Chescheir, G. M., & Kladivko, E. J. (2014). DRAINMOD-DSSAT simulation of the hydrology, nitrogen dynamics, and plant growth of a drained corn field in Indiana. J. Irrig. Drain. Eng., 140(8), 04014026. https://doi.org/10.1061/(ASCE)IR.1943-4774.0000738

Odeh, I. O., & McBratney, A. B. (2005). Pedometrics. In D. Hillel (Ed.), Encyclopedia of soils in the environment (pp. 166-175). Oxford: Elsevier. https://doi.org/10.1016/B0-12-348530-4/00020-5

Ojha, R. P., Verma, C. L., Denis, D. M., Singh, C. S., & Kumar, M. (2017). Modification of inverse auger hole method for saturated hydraulic conductivity measurement. J. Soil Water Conserv., 16(1), 47-52. https://doi.org/10.5958/2455-7145.2017.00011.X

Pannecoucke, L., Le Coz, M., Freulon, X., & de Fouquet, C. (2020). Combining geostatistics and simulations of flow and transport to characterize contamination within the unsaturated zone. Sci. Total Environ., 699, 134216. https://doi.org/10.1016/j.scitotenv.2019.134216

Pease, L. A., Fausey, N. R., Martin, J. F., & Brown, L. C. (2017). Projected climate change effects on subsurface drainage and the performance of controlled drainage in the Western Lake Erie Basin. J. Soil Water Conserv., 72(3), 240-250. https://doi.org/10.2489/jswc.72.3.240

Qi, Z., Singh, R., Helmers, M. J., & Zhou, X. (2015). Evaluating the performance of DRAINMOD using soil hydraulic parameters derived by various methods. Agric. Water Manag., 155, 48-52. https://doi.org/10.1016/j.agwat.2015.03.019

Rawls, W. J., Brakensiek, D. L., & Saxtonn, K. E. (1982). Estimation of soil water properties. Trans. ASAE, 25(5), 1316-1320. https://doi.org/10.13031/2013.33720

Sadhukhan, D., Qi, Z., Zhang, T., Tan, C. S., Ma, L., & Andales, A. A. (2019). Development and evaluation of a phosphorus (P) module in RZWQM2 for phosphorus management in agricultural fields. Environ. Model. Softw., 113, 48-58. https://doi.org/10.1016/j.envsoft.2018.12.007

Salazar, O., Wesström, I., & Joel, A. (2008). Evaluation of DRAINMOD using saturated hydraulic conductivity estimated by a pedotransfer function model. Agric. Water Manag., 95(10), 1135-1143. https://doi.org/10.1016/j.agwat.2008.04.011

Saxton, K. E., & Rawls, W. J. (2006). Soil water characteristic estimates by texture and organic matter for hydrologic solutions. Soil Sci. Soc. Am. J., 70(5), 1569-1578. https://doi.org/10.2136/sssaj2005.0117

Saxton, K. E., Rawls, W. J., Romberger, J. S., & Papendick, R. I. (1986). Estimating generalized soil-water characteristics from texture. Soil Sci. Soc. Am. J., 50(4), 1031-1036. https://doi.org/10.2136/sssaj1986.03615995005000040039x

Schaap, M. G., Leij, F. J., & van Genuchten, M. T. (1998). Neural network analysis for hierarchical prediction of soil hydraulic properties. Soil Sci. Soc. Am. J., 62(4), 847-855. https://doi.org/10.2136/sssaj1998.03615995006200040001x

Schaap, M. G., Leij, F. J., & van Genuchten, M. T. (2001). ROSETTA: A computer program for estimating soil hydraulic parameters with hierarchical pedotransfer functions. J. Hydrol., 251(3), 163-176. https://doi.org/10.1016/S0022-1694(01)00466-8

Schwen, A., Bodner, G., Scholl, P., Buchan, G. D., & Loiskandl, W. (2011). Temporal dynamics of soil hydraulic properties and the water-conducting porosity under different tillage. Soil Tillage Res., 113(2), 89-98. https://doi.org/10.1016/j.still.2011.02.005

Shedekar, V. S., King, K. W., Fausey, N. R., Islam, K. R., Soboyejo, A. B., Kalcic, M. M., & Brown, L. C. (2021). Exploring the effectiveness of drainage water management on water budgets and nitrate loss using three evaluation approaches. Agric. Water Manag., 243, 106501. https://doi.org/10.1016/j.agwat.2020.106501

Shokrana, M. S., & Ghane, E. (2021). An empirical V-notch weir equation and standard procedure to accurately estimate drainage discharge. Appl. Eng. Agric., 37(6), 1097-1105. https://doi.org/10.13031/aea.14617

Šimunek, J., & van Genuchten, M. T. (2008). Modeling nonequilibrium flow and transport processes using HYDRUS. Vadose Zone J., 7(2), 782-797. https://doi.org/10.2136/vzj2007.0074

Singh, R., Helmers, M. J., & Qi, Z. (2006). Calibration and validation of DRAINMOD to design subsurface drainage systems for Iowa’s tile landscapes. Agric. Water Manag., 85(3), 221-232. https://doi.org/10.1016/j.agwat.2006.05.013

Singh, S., Bhattarai, R., Negm, L. M., Youssef, M. A., & Pittelkow, C. M. (2020). Evaluation of nitrogen loss reduction strategies using DRAINMOD-DSSAT in east-central Illinois. Agric. Water Manag., 240, 106322. https://doi.org/10.1016/j.agwat.2020.106322

Skaggs, R. W. (1978). A water management model for shallow water table soils. Technical Report No. 134. Raleigh, NC: North Carolina State University, Water Resources Research Institute. Retrieved from https://repository.lib.ncsu.edu/handle/1840.4/1618

Skaggs, R. W. (1982). Field evaluation of a water management simulation model. Trans. ASAE, 25(3), 666-674. https://doi.org/10.13031/2013.33592

Skaggs, R. W., Youssef, M. A., & Chescheir, G. M. (2012). DRAINMOD: Model use, calibration, and validation. Trans. ASABE, 55(4), 1509-1522. https://doi.org/10.13031/2013.42259

Skaggs, R. W., Youssef, M. A., Gilliam, J. W., & Evans, R. O. (2010). Effect of controlled drainage on water and nitrogen balances in drained lands. Trans. ASABE, 53(6), 1843-1850. https://doi.org/10.13031/2013.35810

Soil Survey Staff. (2005). Soil Survey Geographic (SSURGO) Database. Natural Resources Conservation Service, United States Department of Agriculture. Retrieved from https://websoilsurvey.sc.egov.usda.gov/

Tan, C. S., & Zhang, T. Q. (2011). Surface runoff and sub-surface drainage phosphorus losses under regular free drainage and controlled drainage with sub-irrigation systems in southern Ontario. Can. J. Soil Sci., 91(3), 349-359. https://doi.org/10.4141/cjss09086

Thorp, K. R. (2020). Long-term simulations of site-specific irrigation management for Arizona cotton production. Irrig. Sci., 38(1), 49-64. https://doi.org/10.1007/s00271-019-00650-6

Thorp, K. R., Youssef, M. A., Jaynes, D. B., Malone, R. W., & Ma, L. (2009). DRAINMOD-N II: Evaluated for an agricultural system in Iowa and compared to RZWQM-DSSAT. Trans. ASABE, 52(5), 1557-1573. https://doi.org/10.13031/2013.29144

Tolomio, M., & Borin, M. (2018). Water table management to save water and reduce nutrient losses from agricultural fields: 6 years of experience in North-Eastern Italy. Agric. Water Manag., 201, 1-10. https://doi.org/10.1016/j.agwat.2018.01.009

Van Beers, W. F. (1970). The Auger-hole method: A field measurement of the hydraulic conductivity of soil below the water table. Wageningen, The Netherlands: Veenman, International Institute for Land Reclamation and Improvement. Retrieved from https://library.wur.nl/WebQuery/wurpubs/fulltext/60453

van Genuchten, M. T. (1980). A closed-form equation for predicting the hydraulic conductivity of unsaturated soils. Soil Sci. Soc. Am. J., 44(5), 892-898. https://doi.org/10.2136/sssaj1980.03615995004400050002x

Wang, X., Mosley, C. T., Frankenberger, J. R., & Kladivko, E. J. (2006). Subsurface drain flow and crop yield predictions for different drain spacings using DRAINMOD. Agric. Water Manag., 79(2), 113-136. https://doi.org/10.1016/j.agwat.2005.02.002

Williams, M. R., King, K. W., Ford, W., Buda, A. R., & Kennedy, C. D. (2016). Effect of tillage on macropore flow and phosphorus transport to tile drains. Water Resour. Res., 52(4), 2868-2882. https://doi.org/10.1002/2015WR017650

Wösten, J. H., Lilly, A., Nemes, A., & Le Bas, C. (1999). Development and use of a database of hydraulic properties of European soils. Geoderma, 90(3), 169-185. https://doi.org/10.1016/S0016-7061(98)00132-3

Youssef, M. A., Abdelbaki, A. M., Negm, L. M., Skaggs, R. W., Thorp, K. R., & Jaynes, D. B. (2018). DRAINMOD-simulated performance of controlled drainage across the U.S. Midwest. Agric. Water Manag., 197, 54-66. https://doi.org/10.1016/j.agwat.2017.11.012

Youssef, M. A., Liu, Y., Chescheir, G. M., Skaggs, R. W., & Negm, L. M. (2021). DRAINMOD modeling framework for simulating controlled drainage effect on lateral seepage from artificially drained fields. Agric. Water Manag., 254, 106944. https://doi.org/10.1016/j.agwat.2021.106944

Youssef, M. A., Skaggs, R. W., Chescheir, G. M., & Gilliam, J. W. (2005). The nitrogen simulation model, DRAINMOD-N II. Trans. ASAE, 48(2), 611-626. https://doi.org/10.13031/2013.18335

Zhang, Y., & Schaap, M. G. (2017). Weighted recalibration of the Rosetta pedotransfer model with improved estimates of hydraulic parameter distributions and summary statistics (Rosetta3). J. Hydrol., 547, 39-53. https://doi.org/10.1016/j.jhydrol.2017.01.004