A disadvantage with using a yield goal N model is that there is a poor relationship between yield and the EONR because it does not consider (i) water as a nutrient, (ii) N mineralization, and (iii) synergistic relationships between water, N, and the microbial community (Kim et al., 2008). Kim et al. (2008) reported a synergistic relationship between N and water in a corn study. Therefore, these results have been attributed to inaccurate assessment of N mineralization. Also, the excessive application of water in some states causes major N losses through leaching and denitrification. As a result, many states have dropped or modified the yield goal based recommendation is favor of unique recommendation models for each state. For example, (i) the South Dakota and Nebraska models consider yield goals while the Minnesota and Iowa models do not; (ii) the South Dakota model considers 100% of the NO3–N contained in the surface 60 cm, while the Nebraska model considers 50% of the NO3–N contained in the surface 120 cm, and the western Minnesota model considers 60% of the NO3–N in the surface 60 cm; and (iii) only the Nebraska model considers organic matter. Three of the models (Nebraska, western Minnesota, and Iowa) consider the fertilizer cost/corn price ratio in the N recommendation, while the South Dakota model does not.
Validations of regional yield goal based N models for site-specific and whole-field applications have been mixed (Fox and Piekielek, 1995; Lory and Scharf, 2003). These results are attributed to: (i) scaling problems, i.e., using models designed for a specific purpose for applications that have not been validated (Mulvaney et al., 2006); (ii) using simplistic models that do not account for synergistic relationships between N and water (Kim et al., 2008); (iii) using NO3–N as an estimator of the soil N pool; and (iv) using N recommendation models that do not account for local climatic conditions. Five general approaches are available for solving these problems. The first approach is to increase the complexity of the current models. For example, as proposed by Kim et al. (2008), the k constant in Eq.  could be replaced with a variable. The second approach is to develop new mechanistic models that utilize digital soil and climatic data bases. Examples of mechanistic models that could be used include the CERES or Hybrid-Maize models (Jones and Kiniry, 1986; Yang et al., 2006). The third approach is to utilize field-specific N data (e.g., well-fertilized controls that are placed in the field). In the fourth technique, the N recommendations are modified to account for differential N mineralization across the landscape. A fifth technique would use some combination of the four techniques. The objective of this study was to evaluate and test regional N recommendation models from South Dakota, western Minnesota, Iowa, and Nebraska for their suitability to improve South Dakota N recommendations.
MATERIALS AND METHODS
Research was conducted between 2002 and 2004 at Aurora and between 2004 and 2006 at Watertown and Beresford in eastern South Dakota. Experimental details and methods were provided by Bly et al. (2004), Gerwing et al. (2004, 2005), Gelderman et al. (2006), Gelderman and Gerwing (2006), and Kim et al. (2008). The soil series at Aurora was a Brandt silty clay loam (a fine-silty, mixed, superactive, frigid Calcic Hapludoll). The surface horizon contained approximately 110 g sand, 580 g silt, and 310 g clay kg–1. Total N in the 0- to15- and 15- to 60-cm depths was approximately 5.1 and 10.2 Mg N ha–1, respectively. Total C in the 0- to 15- and 15- to 60-cm depths was approximately 44.6 and 78.5 Mg C ha–1, respectively. At Beresford and Watertown, the soils were an Egan silty clay loam (a fine-silty, mixed, superactive, mesic Udic Haplustoll) and a Brookings silty clay loam (a fine-silty, mixed, superactive, frigid Pachic Hapludoll). The parent materials at both sites were glacial till. Total soil organic C at both sites was between 20 and 30 g kg–1 (30–50 g kg–1 organic matter).
The experimental design at Aurora was a randomized split complete block design with four replications. Two main treatments were water and N. Two water regimes (rainfall and rainfall plus irrigation) and four N rates (0, 56, 112, and 168 kg N ha–1) were used. During the growing season from April through August, the natural precipitation was 42.5, 32.9, and 40.1 cm of water in 2002, 2003, and 2004, respectively, and additional irrigation was 10.8, 14.9, and 5 cm of water in 2002, 2003, and 2004, respectively. The experiment conducted at Aurora contained both medium (10 Mg ha–1) and high yield potential (12.54 Mg ha–1) conditions (Kim et al., 2008). The high yield potential environment was created by applying supplemental irrigation water. At Beresford and Watertown, the corn relied on natural precipitation. Yield goals at Watertown and Beresford were 10 Mg ha–1 (Bly et al., 2004; Gerwing et al., 2004, 2005; Gelderman et al., 2006; Gelderman and Gerwing, 2004, 2005, 2006). No manure was applied at any of the sites and the previous crop was either soybean [Glycine max (L.) Merr.], corn, or wheat (Triticum aestivum L.).
Composite soil samples from the surface 60 cm were collected in the spring before corn was planted at all sites. In the experiments described by Bly et al. (2004), Gerwing et al. (2004, 2005), Gelderman et., (2006), and Gelderman and Gerwing (2006), NO3 was extracted from 10 g of soil with 100 mL of 0.01 mol L–1 Al2(SO4)3 and 0.02 mol L–1 H3BO3 and analyzed for NO3–N with an electrode. In the study of Kim et al. (2008), NO3 was extracted from 10 g of soil with 100 mL of 1 mol L–1 KCl and analyzed for NO3–N using Cd reduction (Maynard and Kalra, 1993). In the experiment of Kim et al. (2008), NH4+–N and aboveground N uptake were also measured. At Aurora, grain samples were collected during harvest and analyzed for total N, d15N, and d13C on a ratio mass spectrometer (Clay et al., 2006a). Based on measured grain yields and d13C values, yield losses due to water and N stress were calculated (Clay et al., 2006b).
Nitrogen uptake of the aboveground plant parts was estimated by summing the N contents of grain and stover. Grain N use efficiency (NUE, %) was calculated aswhere Nplant is the N contained in the grain of the fertilized plot, Ncontrol is the N contained in the grain of the unfertilized plot within the block, and N rate is the amount of N applied (Kim et al., 2008).
In the unfertilized control plots, the percentage of soil N used was calculated aswhere the N net balance for medium and high water regimes within a block was calculated as
The net N balance was slightly lower in the medium than the high water regime. The higher N balance under the high water regime was attributed to the irrigation water containing NO3 (15–40 mg NO3––N kg–1) (Kim et al., 2008).
Nitrogen Fertilizer Response and Economic Analysis Calculations
To assess if N recommendations could be improved by using an in-field diagnostic tool, N fertilizer responses (delta yield) were calculated aswhere YEONR is the grain yield at the EONR and Y0N is the grain yield in plots where N was not applied (Kachanoski et al., 1996; Lory and Scharf, 2003).
The EONRs were calculated by fitting the N response data to a second-degree quadratic and plateau model [Yield = a + b(N rate) + c(N rate)2]. The plateau was the maximum yield where further increases in N did not increase yield (Carlson et al., 2003; Scharf et al., 2005). For scenario testing, the EONRs were calculated for three N fertilizer cost/corn sale price ratios (2.80, 5.59, and 11.27) using a corn price of US$118 Mg–1 grain at 15.5% moisture and N fertilizer costs of US$330, US$660, and US$1330 Mg–1 N, respectively. These values were equivalent to the 0.05, 0.10, and 0.20 fertilizer/corn price ratios reported by Sawyer et al. (2006). The EONR is the point where the change in value of the yield [d(value of corn)] equals the change in value of the fertilizer [d(N fertilizer cost)] (Carlson et al., 2003). Mathematically, this expression is
Nitrogen Recommendation Models
The EONRs for corn were calculated for each block and water stress environment using South Dakota, modified South Dakota, western Minnesota, Iowa, and Nebraska N recommendation models. The South Dakota N recommendation model iswhere NR is the N rate recommendation (kg N ha–1), YG is the yield goal (Mg ha–1), k is 21.4 kg N Mg–1 grain, STN is the amount of NO3––N contained in the surface 0 to 60 cm (kg N ha–1), and PCC is the previous crop credit (legume credit, 44 kg N ha–1) (Gerwing and Gelderman, 2005). Irrigation N was determined using the N mass balance approach (Kim et al., 2008). Nitrate-N was estimated using two different approaches. In the first approach, the amount of NO3–N contained in the surface 60 cm was measured (Gerwing and Gelderman, 2005). In the second approach, NO3–N was estimated to be 56 and 100 kg N ha–1 for corn following soybean and corn following corn or wheat. These values were based on soil test average values reported by Gelderman and Gerwing (2004, 2005, 2006). Based on the current South Dakota model (Gerwing and Gelderman, 2005), recommendations were not adjusted for corn or N prices.
The modified South Dakota N recommendation model iswhere f(k) is a variable that is a function based on the soil productivity level. This function was designed to account for synergistic relationships between water and N (Kim et al., 2008). Kim et al. (2008) developed a conceptual N model that explains this synergistic relationship. For highly productive systems, f(k) is 19.6 kg N Mg–1 grain and for moderately productive systems, f(k) is 21.4 kg N Mg–1 grain. These N recommendations were not adjusted based on corn or N prices.
The western Minnesota model iswhere the maximum return to N (MRTN) is the N fertilizer based on the N cost/crop price ratio and STN is the amount of NO3––N contained in the surface 0 to 60 cm (kg ha–1) (Rehm et al., 2006). In scenario testing, unirrigated corn was assigned an N requirement of 134 kg N ha–1, while irrigated corn was assigned an N requirement of 157 kg N ha–1. Based on the fertilizer cost/corn price ratios, these N recommendations were adjusted (Rehm et al., 2006).
The Iowa N recommendation is based on the MRTN and the previous crop. For corn following soybean, the N recommendations were 145, 123, 95 kg N ha–1 for the 2.8, 5.59, and 11.27 N cost/corn price ratios, respectively, and for corn following corn, the N recommendations were 205, 179, 143 kg N ha–1, respectively (Sawyer et al., 2006). The Iowa N recommendation is not adjusted based on preseason NO3–N.
The equation for the Nebraska model iswhere OM is organic matter content (up to 30 g organic matter kg–1), the soybean credit is 50 kg N ha–1, soil NO3–N is the average concentration (mg kg–1 soil) of soil NO3–N contained in the surface 120 cm, irrigation N was discussed above, fA is a correction factor for application time, and fR is correction factor for the corn/N price ratio (Ferguson et al., 2008). The fR values were 1.19, 1.05, and 0.78 for 0.05, 0.10, and 0.20 N cost/corn prices, respectively (Shapiro et al., 2008). Nitrate-N for the 60- to 120-cm depth was considered to be 3 mg kg–1 soil.
To evaluate the different models, root mean square error [RMSE = S(predicted value – observed value)2/n] and bias [Bias =(predicted value – observed value)/n] values were calculated. Correlation coefficients (r) between the different parameters were determined. A negative bias value indicates that, on average, the predicted recommendation underestimated the N recommendation, while a positive bias indicates that the model overestimated the recommendation. The RMSE values were compared with an F statistic. Significant differences are reported at the 0.05 level. The boundary conditions for the validation were: (i) manure was not applied; (ii) high NO3 levels were not expected; (iii) all sites were located on the eastern side of South Dakota; and (iv) all soils contained moderately high organic matter (>30 g organic matter kg–1 soil).
RESULTS AND DISCUSSION
Environmental Impacts on Yield
Corn yields in the region are typically limited by both N and water availability (Table 1; Clay et al., 2006b; Kim et al., 2008). Adding either N, water, or both had an additive effect on corn yields (Fig. 1). These yield were decreased due to N or water stress. These results were attributed to synergistic relationships between water and N and were attributed to water facilitating the transport of NO3 from the soil to the plant. These findings suggest that the relationship between the water and N cycles will impact the N use efficiency across different landscape positions, where areas with less plant-available water, such as summit and shoulder areas, may have relatively low N use efficiency, while areas with more available water will have higher N use efficiencies. Differential organic mineralization across the landscape may also impact N requirements (Clay et al., 2006a).
|Yield potential of the soil|
|N rate, kg ha–1|
Generally increasing the fertilizer cost lowered the EONR. The EONR values were sensitive to the fertilizer cost/corn price ratio and were not influenced by the soil yield potential (Table 2). The different soil yield potentials were attributed to the irrigation water transporting N to the crop. Nitrogen fertilizer and soil-derived N use efficiency may be higher in areas with more rather than less available water. The soil N use efficiency was higher in the irrigated environment (high yield potential) than the dryland environment (moderate yield potential). For example, the soil N use efficiency was 67.7 and 61.6% in the high and moderate yield potential soils and the fertilizer use efficiency was 48 and 44%, respectively (Kim et al., 2008).
|Yield potential of the soil||EONR
|Year||2.80 ratio||5.59 ratio||11.27 ratio|
|kg N ha–1|
Predicting Nitrogen Recommendations
Historically, yield goal based approaches have been the basis for many N recommendations in the central portion of the United States. Several recent studies have questioned the value of these models (Derby et al., 2005; Lory and Scharf, 2003). Lory and Scharf (2003) suggested that the delta yield technique might be an alternative. The strength of the relationships among the corn’s response to N fertilizer, the EONR, and the yield at the EONR were impacted by the fertilizer/corn price ratio (Table 3). The N responseEONR values were poorly correlated with the EONR for the 2.80 and 5.59 fertilizer/corn price ratios and highly correlated with the EONR for the 11.27 fertilizer/corn price ratio. These results suggest that in-field assessment tools, such as the potential N response model, need further study. Others have reported that this approach can be used to estimate N fertilizer responses. Lory and Scharf (2003) reported that N responses were highly correlated with the EONR and that a linear equation could be used to describe data collected from five states (Illinois, Minnesota, Missouri, Pennsylvania, and Wisconsin). Kachanoski et al. (1996) and Braum et al. (1999) reported that within-field variation in the EONR for corn was related to potential N responses. Kachanoski et al. (1996) reported that in Canada the delta yield were correlated with the EONR and that delta yield can be used to estimate maximum economic N rates. Kim et al. (2008) showed that the yield goal approach can be improved by replacing a constant with a variable (Fig. 1.)
||Yield at EONR
|Parameter||2.80 ratio||5.59 ratio||11.27 ratio||2.80 ratio||5.59 ratio||11.27 ratio||2.80 ratio||5.59 ratio||11.27 ratio|
|Delta yield, 5.59 ratio||1.00||1.00|
|Delta yield, 11.27 ratio||0.98||0.99||1.00|
|EONR, 2.80 ratio||–0.09||–0.10||–0.14||1.00|
|EONR, 5.59 ratio||0.11||0.10||0.07||0.97||1.00|
|EONR, 11.27 ratio||0.56||0.57||0.57||0.64||0.80||1.00|
|Yield at EONR, 2.80 ratio||0.73||0.71||0.66||–0.25||–0.13||0.21||1.00|
|Yield at EONR, 5.59 ratio||0.72||0.71||0.67||–0.26||–0.14||0.20||1.00||1.00|
|Yield at EONR, 11.27 ratio||0.70||0.69||0.67||–0.30||–0.18||0.18||0.98||0.99||1.00|
Comparisons among Nitrogen Recommendation Models
The RMSE and bias values were influenced by the fertilizer cost, corn price, and recommendation model. The RMSE values were directly related to the differences between the measured and predicted values (Table 4). A negative bias indicates that the recommendation was lower than the actual requirement. The South Dakota model generally underestimated the N recommendation for the 2.89 fertilizer/corn price ratio and overestimated the N requirement for the 11.27 fertilizer/corn price ratio. The best results were observed for the 5.59 ratio. These results were expected because the model did not adjust the recommendation based on the fertilizer cost/corn price ratio.
|Fertilizer/corn price ratio||South Dakota model||Western Minnesota model||Iowa model||Nebraska model||Modified South Dakota model|
|2.80||1871 (–16)†||1955 (–38)||3670 (48)||2042 (–21)||1935 (–27)|
|5.59||1414 (–5)||1832 (–36)||2107 (29)||1560 (–20)||1309 (–13)|
|11.27||2205 (17)||2316 (–36)||1629 (18)||1780 (–28)||1765 (9)|
|Replacing measured NO3–N with a constant|
|2.80||1456 (–11)||1972 (–34)||1555(–19)||1287 (–21)|
|5.59||1006 (1)||1742 (–33)||1091(–16)||670 (–8)|
|11.27||1812 (23)||2082 (–32)||1608(–28)||1140 (15)|
Using long-term estimated NO3 concentrations rather than measured NO3–N concentrations either improved or did not impact N recommendations. These results were attributed to NO3 being only one of several plant-available N pools (NH4, mineralizable N, and NO3) and the fact that a portion of the NO3–N can be sorbed onto exchange sites (Clay et al., 2004). The N mass balances from data from the Aurora site showed that NO3–N represented <30% of the available N. Different results would be expected in soils containing less organic matter and/or greater amounts of NO3–N.
The western Minnesota model had slightly higher RMSE values and more negative bias than the South Dakota N recommendation model. Associated with the more negative bias were lower N fertilizer recommendations (Table 5). The relatively high RMSE value and negative bias may be associated with identifying these soils in the medium yield potential category.
|Fertilizer/corn ratio||EONR||South Dakota model||Western Minnesota model||Iowa model||Nebraska model||Modified South Dakota model|
|kg N ha–1|
|Replacing measured NO3–N with a constant|
The Iowa N recommendation model, when compared with the South Dakota N recommendation model, had higher RMSE and bias values. The results for both the western Minnesota and Iowa N recommendation models might be related to these models providing an empirical fit to the synergistic relationships that occur between N and water, resulting in a different amount of N fertilizer being required to produce a corn crop in moderate and high yield potential soils. For example in a moderate yield potential environment, 9.2 kg of N might be required per megagram of grain, while in a high yield potential environment, 7.7 kg of N fertilizer might be required per megagram of grain.
The Nebraska N recommendation model is a yield goal model that also accounts for organic matter contents. Such an N recommendation model might be better for site-specific N recommendations. In many landscapes across the Midwest and Great Plains, organic C and N contain strong spatial structures, with footslope areas often having more organic matter than summit and shoulder areas. The Nebraska N recommendation model, when compared with the South Dakota model, had numerically higher RMSE values for the 2.80 and 5.59 ratios and a numerically lower value for the 11.27 ratio. Using measured NO3–N concentrations did not improve the Nebraska N recommendation.
The modified South Dakota N model had RMSE values either lower than or similar to the current South Dakota N model. We attribute this to the ability of the modified model to reduce the amount of N fertilizer required under high yield conditions. One factor contributing to this is the synergistic relationship between N and water. For example, Bauer et al. (1965) showed that in North Dakota water availability impacts N use by wheat. Clay et al. (2001) and Li et al. (2003) had similar results and reported that wheat N use efficiency in Montana was indirectly related to water stress. O’Neill et al. (2004) reported that in 13 experiments conducted in Nebraska, corn N use efficiency was numerically higher in high compared with moderate yield potential soils. Derby et al. (2005) reported that in North Dakota yield goal-based N recommendations overestimated the N requirement in high yield potential soils. Kim et al. (2008) developed a conceptual N model that could explain N and water synergistic relations. The model predicts that because NO3 is transported to the plant in the water transpiration stream, the amount of inorganic N transported to the plant is related to the amount of water transpired. The second factor that could result in reduced N fertilizer need is increased N mineralization in high yield environments.
The regional N recommendation models from South Dakota, western Minnesota, Iowa, and Nebraska and the modified South Dakota model were tested to see whether the current South Dakota model can be improved. The current South Dakota model could be improved by replacing the constant (k) in the South Dakota yield goal equation with a variable related to water availability. The modified South Dakota N recommendation model considers synergistic relationships between yield-limiting factors. The non-yield-goal models such as those in Iowa and western Minnesota most likely are the result of synergistic relationships occurring between water and N in those states; the amount of precipitation during the growing season is higher in Iowa and Minnesota than in South Dakota. These results suggest that more N fertilizer is needed in low-yield environments than high-yield environments in South Dakota. Different results would be expected in systems where high NO3–N concentrations exist or organic matter contents are lower (e.g., high manure applications or drought).
We believe that a better approach for a regional N recommendation model is to utilize a function that describes the mechanism as a function based on the soil productivity level rather than a block value. Further study may be needed to evaluate the models from other states where similar boundary conditions exist.