This study used Blount silt loam, a non-marginal soil, as base soil for corn (conventional tillage with stover retained, no-till with stover retained, and conventional tillage corn with 70% stover removal), switchgrass and Miscanthus simulations. These cropping management systems were also evaluated in this study on three marginal soils: Miami clay loam, Cincinnati silt loam, and Clermont silt loam in Decatur County in southern Indiana (Figure 1). Southern Indiana was selected because previous researchers identified a need to supply biomass to facilities in southern Indiana [32] . The NRCS’ Land Capability Classification (LCC) was used to determine the marginal quality soils with the Soil Survey Geographic (SSURGO) database (Table 1). As defined by SSURGO metadata, LCC is a system of grouping soils primarily on the basis of their capability to produce common cultivated crops and pasture plants without any deterioration over a long period of time [33] . Each SSURGO soil map unit is assigned a capability class ranging from 1 to 8, and a sub-class assigned by letters to describe limitations or hazards for agricultural purposes. SSURGO LCC 3’s (wetness) and 4’s (erodibility) were considered suitable marginal lands for large-scale perennial grass production.

^[a]Kw: Soil erodibility factor. ^[b]θ33 kPa, is field capacity. ^[c]θ1500 kPa is the wilting point at suction pressure 1500 kPa. ^[d]Porosity, field capacity, and wilting point values for soil layers obtained GLEAMS documentation.

In this study, initial evidence from a GIS analysis suggested that land classes of 5 to 8 would not be suitable for perennial grass production for bioenergy due to restrictions such as rocks or high slopes (25% or greater), which would make in-field operations difficult or impractical. One marginal soil due to wetness (Clermont silt loam, land capability class-3w) and two marginal soils due to high erodibility (Miami clay loam and Cincinnati silt loam, land capability class-4e) were selected based on their abundance in Decatur County, Indiana. At the time of this study, 62% of these soils was cropped with corn (so corn was selected as the base crop), using the 2010 Cropland Data Layer [34] . It is worth noting that land capability class-3w’s can be prime farmlands if properly drained (perhaps because they are for the most part already drained in this county).

The GLEAMS model is a computer program developed to simulate edge-of-field and bottom-of-root-zone loadings of water, sediment, pesticides, and plant nutrients [35] . The major components of the model include hydrology, erosion, nutrient and pesticides. These components can simulate the effects of land management practices on a daily basis utilizing climate, soil, and management data inputs [35] . Erosion prediction in GLEAMS is estimated using the Universal Soil Loss Equation (USLE). The crop management factor (C-fact) drives GLE- AMS erosion computation associated with land management and is not dynamically computed and as such requires user-defined input values. C-factor inputs for management operations (Table 2) are critical for accurate erosion prediction by the model. The RUSLE 2.0 model [36] was used to generate daily C-factor values for corn and perennial management systems, of which the mean annual C-factor values were reported in Table 2. To generate C-factor values for Miscanthus, the switchgrass template was modified to reflect increased plant biomass production and management to obtain estimates for C-factor values required by GLEAMS. The choice of the GLEAMS model for this study was driven by the model capability to simulate edge-of-field hydrology, erosion, nutrients and pesticides and to represent biomass production [37] .

Continuous daily precipitation and temperature were required for climatic inputs in the GLEAMS model [37] (Knisel and Davis 2000). Climatic data were generated for Decatur County, Indiana, using a weather generator known as CLIGEN, version 4.3 [38] [39] . CLIGEN is a stochastic weather generator that statistically derives daily precipitation, temperature, dewpoint temperature, wind, and solar radiation with historic measurements for a given geographic point. The nearest climate stations (39˚21'00"N, 85˚28'48"W) in Greensburg, southeastern Indiana, provided the required input parameters for the weather generator program. CLIGEN is packaged with the Water Erosion Prediction Project (WEPP) model [38] [39] . CLIGEN has a database, with a graphical user-friendly feature, for more than 2600 weather stations in the United States. It is often used also by other hydrologic/water quality models such as the Soil Water Assessment Tool (SWAT) [40] [41] .

Porosity, field capacity, and wilting point are sensitive parameters in the GLEAMS model [37] . Soil physical properties required to run the model were obtained from the SSURGO database. The available water content (AWC) data in SSURGO were used to compute the wilting point. However, in a few cases, very low AWC val-

ues in SSURGO were not justified by the associated textural class descriptions, especially at soil depths of 5 cm or less on the studied soils. This could impact model estimates for hydrology and water quality. In the absence of justification for low AWC in the SSURGO documentation, GLEAMS description of porosity, wilting point, and field capacity values associated with soil type were used for soil layers deemed invalid.

Prior research and experience of the authors, as well as that of other researchers, has shown that calibration of the GLEAMS model is not required, especially when comparing the relative impacts of scenarios as was done in this paper [42] -[45] . For example, Jetten et al. [44] evaluated six models, including GLEAMS, on seven sites with 73 site-years of erosion data. They concluded that relative model results were more reliable than absolute results and application of the models without calibration is appropriate. Knisel et al. [45] conducted a sensitivity analysis on the hydrology portion of GLEAMS as part of a GLEAMS water balance validation. They found that fine tuning of the four most sensitive parameters (curve number, rooting depth, porosity and field capacity) slightly improved simulated runoff, percolation and evapotranspiration. They concluded that the results of the GLEAMS validation support prior statements that calibration of the model is not required.

The GLEAMS model has been shown to perform well for Indiana conditions in prior work [42] [46] [47] . The GLEAMS model has previously been calibrated and validated for monthly runoff, sediment, NO₃-N, and P under corn-soybean cropping systems for two small watersheds in Allen County (northeast Indiana) [42] , and for pesticide (atrazine) losses on Clermont Silt Loam soil in southern Indiana [46] . The simulation scenarios discussed in this study were run with the calibrated parameters for Allen County watersheds and the Clermont Silt Loam [42] [46] .

The GLEAMS model uses a simple crop growth model driven by leaf area index (LAI) values for fractions of the growing season (Figure 2). Leaf area index is defined as the ratio of the leaf surface area to the soil area the leaf covers [48] . It can be used to describe the canopy structure as well as biomass of a plant [48] .The internal database of the model provides values of LAI for 78 different crops. A user-defined option allows for the simulation of crops that are not included in the model database. Switchgrass and Miscanthus characteristics required to run the model are provided in Table 3. The model allows for pairs of LAI data (NOLAI) with corresponding fraction of the growing season (USRFRC), and user defined LAI (USRLAI) for non-stressed crop [37] . The model dynamically adjusts for actual LAI using crop stress factors. For the model to best simulate crop growth, LAI and crop height (CROPHT) are required for the periods where LAI was provided [37] . Leaf area index data are not sensitive to the water balance computation, in GLEAMS, but are sensitive to soluble NO₃-N and P uptakes [37] . Leaf area index values for established switchgrass and Miscanthus were obtained from data presented in Heaton et al. [9] .

Engel et al. [49] proposed a split simulation approach to represent perennial establishment and full maturity in the SWAT model by modifying an algorithm for simulation of forest trees to represent perennial grasses. The representation of crops with an establishment year and harvesting in subsequent years can be readily performed with the GLEAMS model (Figure 3). However, the simulated data for establishment of the perennial grasses

^[a]Cave-in-Rock, upland switchgrass. ^[b]Warm-season grasses (C₄-C) in the model’s internal database. ^[c]Perennial crop: used by the model to indicate that the root growth does not begin at zero depth after each cutting.

were not discussed in this study. The model’s internal database provides crop characteristics for perennials such as forest, warm-(C₄) and cool-(C₃) season grasses.

Conventional tillage (fall chisel plowing and spring disking) with stover retained, no-till with stover retained, and conventional tillage corn with 70% stover removal practices in a continuous production system were considered for corn production scenarios. However, stover removal was considered only with no-till management practice in conformity with the guidelines of a Department of Energy Study [15] . Thomas and others (2011) modified the GLEAMS algorithm to represent the above surface residue removal following grain harvest. Details of model inputs such as planting date, harvesting date, fertilizer rates, adjustments and timing are provided in Thomas et al. [23] and Thomas [31] . Fertilizer application rates with compensation for nutrient replacement due to stover removal at 70% varied across the four studied soils (Table 4).

Switchgrass is a warm-season crop and as such would require temperatures greater than 25˚C for optimal germination and growth [50] . In the central Great Plains and Midwest, switchgrass can be planted within +/− 3 weeks of the corn planting date [17] . In Indiana, switchgrass planting date is expected to coincide with corn; therefore, the median planting date reported by Thomas et al. [43] was used in this study.

For model simulations, conventional tillage corn with stover retained was assumed as the baseline crop and as such two tillage operations (spring-disk followed by seed-drill) were assumed during establishment years with single yearly harvest for years 2 to 8. Spring-drill operation was performed at a depth of 1.27 cm, which is the maximum recommended for switchgrass. Spring emergence date of established switchgrass stands was assumed to be April 14^th, the expected date for the last spring freeze in southern Indiana. A harvest date of October 18^th, associated with Indiana plot studies, was used for switchgrass management (Jeff Volenec, personal communication). In the model representation, it was assumed that harvest was 80% (8320 kg/ha) for year 2, 90% for year 3 (9360 kg/ha), and maximum yield (100%) for years 4 to 8 (Keith Johnson, personal communication). Perennials such as switchgrass require some level of management to optimize productivity and maintain quality stands for multiple years of harvest [17] . Previous studies reported that switchgrass could be productive for more than 10 years [51] [52] .

General agronomic guidelines for the Midwest region do not recommend nitrogen application during the establishment year to minimize competition with weeds, which could lead to stand or establishment failure [53] . On established switchgrass stands, a single application of nitrogen fertilizer in the form of urea was programed for May 6^th. A nitrogen fertilizer rate of 112 kg/ha was applied to maintain biomass production levels, assuming yearly harvest of upland varieties such as Shawnee or Cave-in-Rock. To maintain maximum yields and to avoid soil mining of P, a single annual application rate of 34 P₂O₅ kg/ha was assumed for model simulations [54] .

Miscanthus is propagated by rhizome division, and no-tilling rhizomes in crop residue would likely be the adopted practice when production reaches its full stage. In United States, researchers reported that Miscanthus yields vary with location, with a range of 27 to 44 t/ha [19] . In Indiana, Miscanthus would likely be established approximately the same date as corn (May 6) with no harvest during the establishment year. A harvest date of October 18^th, associated with Indiana plot studies, was used for Miscanthus management (Jeff Volenec, personal communication). Miscanthus is cold tolerant and as such it was reasonable to assume that spring emergence of established Miscanthus stands was April 1st, 2 weeks earlier than the expected date of last spring freeze in southern Indiana. The first harvest was projected in year 2 at 90% of maximum yield with maximum yields achieved from years 3 to 8 (29.6 t/ha). There is currently limited knowledge on the long-term Miscanthus growth, and re-establishment of Miscanthus stands required to optimize bioenergy production with yearly perennial harvest. Consequently, stand re-establishment was set to occur after the year 8, similar to the production of switchgrass practice assumed in this study. To maintain maximum yields as well as avoiding soil mining of P, a yearly application rate of 34 P₂O₅ kg/ha was assumed for model simulations [54] . Fertilizer application rates of 112 N kg/ha (broadcasted as urea) were applied in production years (Patrick Murphy, personal communication).

Tukey and Bonferroni pairwise comparison tests were used to evaluate differences among mean annual surface runoff, percolation, and losses in erosion, TP, and NO₃-N associated with biofeedstock management scenarios within each soil unit. Prior to hypothesis testing, exploratory data analyses (EDA) were performed to check for independence of variables, normality, and homogeneity of variances. When assumptions were violated, the datasets were log-transformed prior to the analyses (Table 5; [55] ). All the analyses were performed with the SAS/STAT^® software, version 9.2 [56] at a significance level of 0.05.

The simulation scenarios resulted in 1 to 4 mm decrease in annual surface runoff with Miscanthus and switchgrass production compared to corn-based bioenergy production (Figure 4). These results suggested that the production of perennial grasses have no meaningful effects on runoff, which is consistent with the Nyakatawa et al. [57] (2006) plot study comparing runoff from corn and switchgrass. However, other studies reported significant runoff reduction from perennial C4 grasses compared to corn production [22] [58] -[60] . The differences in surface runoff between corn- and grass-based bioenergy systems in this study were not statistically significant (p > 0.05). The similarities in runoff production between corn and perennial grasses in this study (Miscanthus and switchgrass) could be driven by factors influencing soil characteristics after intensive corn and soybean production on these soils. These factors include type and amount of cover, soil structure, and soil moisture content at the beginning of the precipitation season [61] [62] . Under these conditions, the soil physical characteris tics will likely have greater influence on long-term annual runoff estimates than crop production systems would

^[a]See Thomas et al. [23] . ^[b]Stover harvested at 4.2 t/ha, 3.5 t/ha, 5.9 t/ha and 4.7 t/ha for Miami clay loam, Cincinnati silt loam, Clermont and Blount silt loam, respectively.

during similar growing periods. For example, the highest runoff estimates were obtained from the Clermont soil, which is relatively flat, poorly drained, and agronomically marginal with high water content compared to the other two marginal soils, which are moderately well to well drained soils. Even though research efforts exist in the scientific literature regarding the hydrologic benefits of perennial grass production [22] [58] [59] , the specific impacts of Miscanthus and switchgrass production on long-term surface runoff is still not fully examined for various soils (including marginal quality soil).

Surface runoff is partly driven by ET; thus accurate estimation of ET is critical for water balance computations. In this study, annual estimates of ET on the studied marginal quality soils ranged from 721 to 743 mm for Miscanthus, 711 to 744 mm for switchgrass and 679 to 718 mm for corn. The model’s ET estimates were within range of a land use land cover study that included the study area of this project, which estimated 720 mm, 710 mm, and 670 mm for Miscanthus, switchgrass and corn, respectively [13] . Currently, there are knowledge gaps on switchgrass and Miscanthus production in Indiana that need to be explored in order to improve modeling accuracy. Further studies are needed to understand the long-term runoff expected from the production of dedicated perennial grasses.

Percolation is the downward movement of water through the soil profile. Model results showed decrease in percolation below the root zone by 11 to 40 mm for Miscanthus and 11 to 31 mm for switchgrass compared with corn-based bioenergy production (Figure 5). Mean annual percolation estimates associated with corn-based bioenergy on the studied soils ranged from 77 to 187 mm, with highest estimates on Miami clay loam soil (9% slope). McIsaac et al. [58] measured soil moisture at a depth of 94 cm and reported declines in moisture level with Miscanthus and switchgrass compared to corn-soybean rotation. Their study reported a 4-year soil moisture average difference of 45 mm between switchgrass and Miscanthus at the end of the growing season, with little difference during dry years. This would suggest that percolation estimates from fully mature Miscanthus would be lower than that associated with corn and switchgrass, a trend that is consistent with this study. Although the researchers credited high LAI and biomass for lower Miscanthus soil moisture, the establishment period was not represented in the analysis. Corn is an annual crop and as such has greater maximum LAI values than Miscanthus and switchgrass during establishment periods (Figure 2). Consequently, including the establishment year in the analysis would increase soil moisture averages because Miscanthus is slow to establish [6] . In addition, Finch and Riche [63] noted that late in the growing period, Miscanthus intercepts 20% of rainfall, influencing

evaporation from stems and leaves (canopy evaporation). The researchers argued that the quantity intercepted by Miscanthus late in the growing season is more typical of a forest than annual agricultural crops, which could explain measured greater depletion of soil moisture.

Thirty to 60% decrease in erosion occurred with Miscanthus production in comparison with continuous corn with tillage and conventional tillage corn plus 70% stover removal (Figure 6). No-till corn without stover harvesting had lower erosion losses than switchgrass production, with differences ranging from 0.1 to 2.4 t/ha (Figure 6). Large amounts of plant residue associated with no-till corn and low residue with harvested switchgrass could influence the amount of soil cover that could mitigate erosion. However, Miscanthus, which was no-tilled when established, produced lower erosion than no-till corn by at least 30% on the studied soils (Figure 6).

Miscanthus and switchgrass production on highly erodible lands such as Miami clay loam (9% slope) and Cincinnati silt loam (9% slope) is believed to be suitable in mitigating erosion risks associated with production practices. Successful production of switchgrass and Miscanthus could reduce erosion below the soil loss tolerance (T factor) of 6.72 t/ha on Miami clay loam. Cincinnati silt loam has higher soil erodibility factor (K_w = 0.49) and lower T factor value (4.48 t/ha or 2 t/ac) than Miami clay loam. Consequently, all bioenergy scenarios, except for Miscanthus, produced erosion losses that were greater than the soil loss tolerance value for Cincinnati clay loam. Regardless of the bioenergy production on marginally wet Clermont silt loam soil, erosion losses were far below the soil loss tolerance value of 8.96 t/ha, even though it has an erodibility factor (K_w = 0.55), that is greater than that of all studied soils. This is likely due to the gentle slope (0% to 1%) of Clermont silt loam soils compared to the other soils.

Total P losses from switchgrass and Miscanthus production ranged from 2.3 to 7.6 kg/ha, and 1.5 to 3.5 kg/ha, respectively (Figure 7). No-till corn and corn grain plus stover removal produced high TP losses, ranging from 4.2 to 9.8 kg/ha, and 7.0 to 18.7 kg/ha, respectively (Figure 7). Differences in application rates, application methods (e.g. surface application with no-till practices), and soil erosion associated with management practices on the studied soils were reflected in TP losses for conventional tillage corn with stover removal, switchgrass and Miscanthus (Figure 7). Except on the relatively flat Clermont silt loam, TP losses were dominated by P trans-

ported with sediments (erosion).

Overall, there were higher TP losses from no-till corn practices compared to the production of switchgrass and Miscanthus, which generated higher TP losses than conventional tillage corn across all the studied soils (Figure 7). This pattern could be explained by the method used to apply P₂O₅ with no-till. The general practice with corn under fall chisel/spring disk is to incorporate P fertilizer, which means that lower P would be available for transport with sediments, which is the dominant form in which P losses from agricultural lands occur. However, P losses should, in principle, be higher for conventional tillage than for no-till corn and switchgrass. Model results also showed that P losses represent 25% - 30% of the P fertilizer applied each year (Figure 7), suggesting that there is no buildup of P applied to the surface over time. Another plausible explanation is that the model indicates soil P build up near the surface over time, instead of leaching it to subsurface layers. Thus, these results should not be considered conclusive without further assessment of the dynamics of soil P and perennial grass production. Currently, there are no known long-term monitoring studies that evaluate P annual loads for no-till corn systems.

The application of commercial fertilizer with perennial grass production could result in edge-of-field losses (Figure 8). On the studied marginal soils, NO₃-N losses with surface runoff range from 3.7 to 6.2 kg/ha for switchgrass, and 3.8 to 6.6 kg/ha for Miscanthus (Figure 8). Statistical analysis indicated that NO₃-N losses from switchgrass and Miscanthus were not significantly different (p > 0.05) from each other. Owing to higher fertilizer application rates, no-till and conventional tillage corn (except no-till corn plus stover removal) produced high NO₃-N losses to the edge-of-field (Figure 8). Clermont silt loam, considered marginal due to wetness, produced the highest NO₃-N losses with estimates of 10.1, 7.9 and 6.0 kg/ha for tillage corn, no-till corn, and no-till corn with stover removal, respectively. When compared with conventional tillage corn, corn under no-till practice (with or without stover removal) generated lower annual NO₃-N with runoff because of lower mineralization and accumulation of nutrients in crop residues. Tillage operations (fall chisel/spring disk) with corn production influence the decomposition and release of nitrogen in crop residues. The timing of tillage occurs within high rainfall periods that could make released NO₃-N available for runoff transport as reported by Thomas et al. [23] who provided detailed discussions on effects of corn stover removal on nitrate losses with surface runoff.

The simulated scenarios suggested that Miscanthus would produce significantly lower NO₃-N (approximately 90% lower; p < 0.05) leached than switchgrass and corn-based production systems (Figure 9), primarily due to differences in fertilizer application rates. Among the three studied soils, Clermont silt loam produced high losses in annual NO₃-N leached, with highest estimates associated with conventional corn grain (31.8 kg/ha) followed by no-till corn plus stover removal (19.5 kg/ha) (Figure 9), and no-till corn (without stover removal) with a range of 6.0 to 16.4 kg/ha (Figure 9). In Indiana, Kladivko et al. [64] conducted a 15-year study on a Clermont silt loam for conventional corn and soybean-corn rotation production. The authors observed average annual NO₃-N losses of 23, 30 and 41 kg/ha, for 3 different subsurface drain spacings. Although the effects of tillage and no-till

options (included in the management for the study period), were not separately reported in this study, it appeared that a combined average of the 3 drain spacings (5-, 10- and 20-m) would suggest a NO₃-N loss of 31.3 kg/ha for the 15-year period, which is within ranges of the model estimates (31.8 kg/ha) on Clermont silt loam (Figure 9).

While the simulated percolation results showed that Miscanthus and switchgrass had higher percolation estimates than corn production systems (Figure 5), NO₃-N leached from grass-based cropping systems was lower than the corn-based production systems. This result could be explained by the differences in the amount of fertilizer applied across the various production systems.

The results from this study showed that simulated runoff, movement of water below the root zone, erosion rates, and nutrient losses were lower for perennial grasses compared to corn systems for bioenergy production. While the relative comparisons can be accepted with confidence, the absolute values predicted by the model should be treated with caution because there were no observed hydrologic and water quality data to enable model calibration for switchgrass and Miscanthus for the study sites. The GLEAMS model has been calibrated at the watershed level and may not completely represent in detail all processes that drive the differences among scenarios at the plot scale. The perennial grasses examined in this study were modeled as established stands over the 32-year period (i.e. without taking into consideration detailed soil water dynamics of the stands over a long period of time). Generally, standing perennial grass stubble would slow down soil erosion in the winter relative to a harvested corn field, especially with a conventional till corn field. The soil tilth of perennial grasses could support greater infiltration compared to a cropped site, and grasses such as Miscanthus and switchgrass extract more soil water over the years than corn because of greater LAI and a longer growing season. The longer growing season will enable these grasses to intercept and hold more rainfall than corn (if corn is harvested and shredded after 110 to 140 days after planting), allowing less water to reach the soil surface. These processes and their effects on soil water were not explicitly represented in the GLEAMS model during scenario simulations to account for greater infiltration and greater soil water extraction. Additional studies (with perhaps detailed process modeling) are therefore needed to better understand the water-use efficiency of large-scale perennial grass production in rain-fed regions.