Adjusting corn nitrogen management by including a mineralizableâ•’nitrogen test with the preplant and presidedress nitrate tests

The anaerobic potentially mineralizable N (PMN) test combined with the preplant (PPNT) and presidedress (PSNT) nitrate tests may improve corn (Zea mays L.) N fertilization predictions. Forty‐nine corn N response experiments (mostly corn following soybean [Glycine max (L.) Merr.]) were conducted in the U.S. Midwest from 2014–2016 to evaluate the ability of the PPNT and PSNT to predict corn relative yield (RY) and N fertilizer over‐ and under‐application rates when adjusted by PMN. Before planting and N fertilization, PPNT (0–30, 30–60, and 60–90 cm) and PMN (0–30 cm) samples were obtained. In‐season soil samples were obtained at the V5 development stage for PSNT (0–30, 30–60 cm) in all N rate treatments and PMN (0–30 cm) in only the 0 and 180 kg N ha−1 preplant N treatments. Increasing NO3–N sampling depths beyond 30 cm with or without PMN improved RY predictability marginally (R2 increase up to 0.20) and reduced over‐ and under‐application frequencies up to 14%. Including PMN (preplant only) with PPNT or PSNT improved RY predictability minimally (R2 increase up to 0.10) only for coarse‐ and medium‐textured soils, but N fertilizer over‐ and under‐application frequencies were not substantially reduced (≤12%). These marginal improvements in RY predictability and N fertilizer over‐ and under‐application frequencies, regardless of the variables used (e.g., fertilization, sampling depth, soil texture, and growing degree‐day categories), demonstrate that including PMN with soil NO3–N alone does not improve corn N fertilization need predictions enough to recommend their use. Disciplines Agricultural Science | Agriculture | Soil Science Comments This article is published as Clark, Jason D., Fabián G. Fernández, Kristen S. Veum, James J. Camberato, Paul R. Carter, Richard B. Ferguson, David W. Franzen et al. "Adjusting corn nitrogen management by including a mineralizable‐nitrogen test with the preplant and presidedress nitrate tests." Agronomy Journal (2020). doi: 10.1002/agj2.20228. Authors Jason D. Clark, Fabián G. Fernández, Kristen S. Veum, James J. Camberato, Paul R. Carter, Richard B. Ferguson, David W. Franzen, Daniel E. Kaiser, Newell R. Kitchen, Carrie A. M. Laboski, Emerson D. Nafziger, Carl J. Rosen, John E. Sawyer, and John F. Shanahan This article is available at Iowa State University Digital Repository: https://lib.dr.iastate.edu/agron_pubs/652 Received: 14 January 2020 Accepted: 26 March 2020 DOI: 10.1002/agj2.20228


INTRODUCTION
Improving soil testing and the use of soil tests used in making management decisions can improve corn fertilizer-N rate guidelines (Dinnes et al., 2002). Improved N management guidelines can improve economic profits for farmers and reduce potential negative environmental effects including reduced air quality, global warming, and water pollution (Cavigelli et al., 2012;Helmers, Zhou, Baker, Melvin, & Lemke, 2012;McCasland, Trautmann, Porter, & Wagenet, 2012;Ribaudo et al., 2011;Struffert, Rubin, Fernández, & Lamb, 2016;USEPA, 2018). To be most effective in improving fertilizer-N rate guidelines, soil tests used to make N management decisions will likely need to account for both plant-available inorganic N and N mineralized during the growing season. The preplant (PPNT) and presidedress (PSNT) nitrate tests are commonly used to assess inorganic N, and the anaerobic potentially mineralizable N test (PMN) is commonly used to assess mineralizable N (Magdoff, Ross, & Amadon, 1984;Waring & Bremner, 1964). Using these soil tests together has the potential to improve fertilizer-N rate guidelines.
The PPNT is taken early in the spring before any organic or inorganic N amendments are applied to the soil to measure the amount of NO 3 -N remaining in the soil from the previous season. This soil sample timing is attractive to many corn growers because it can avoid time and labor constraints that occur later in the season. The PSNT measures soil NO 3 -N when corn plants are 15-30 cm tall. The timing of this soil test is important because the value represents the concentration of soil NO 3 -N as an index of available N for corn based on soil and weather conditions just prior to the rapid N uptake period for corn (Magdoff et al., 1984). Studies have shown that both the PPNT and PSNT are positively related to RY (Barbieri, Echeverría, & Saínz Rozas, 2008;Bast, Mullen, Eckert, & Thomison, 2012;Binford, Blackmer, & Cerrato, 1992;Bundy & Andraski, 1995;Bundy et al., , 1999Fox, Roth, Iversen, & Piekielek, 1989;Sainz Rozas, Echeverria, Studdert, & Dominguez, 2000;Schmitt, Randall, & Rehm, 2002;Zebarth, Younie, Paul, & Bittman, 2002). One of the strengths of these soil tests is that they can be used to determine the critical soil nitrate content (CSNC) where values below the CSNC warrant N fertilization, whereas values above require no N input. The utility of these nitrate tests are judged on their capacity to correctly identify the CSNC and separate responsive and nonresponsive sites (Bundy, Walters, & Olness, 1999).
The PPNT has been successfully used primarily in semiarid and arid environments along with semi-humid areas and areas with extended periods of frozen soil when N loss potential is minimal (Bundy & Andraski, 1995;Rehm, Schmitt, & Eliason, 2002;Schmitt & Randall, 1994). However, the PPNT provided less utility in many parts of the U.S. Midwest where excessive rainfall was common after soil sampling, which resulted in large N losses and overestimation of plant-available N (Cela, Berenguer, Ballesta, Santiveri, & Lloveras, 2013). The PSNT was less reliable when soil temperatures were lower than normal up to the time of soil sampling (Andraski & Bundy, 2002) and in areas where N is highly susceptible to loss such as coarse-textured soils and in years where large rain events occurred close to soil sampling (Magdoff, 1991;Yost, Russelle, & Coulter, 2013). Reduced RY predictability with the PPNT and PSNT also occurred when organic amendments were applied recently to the soil or the previous crop was a legume (Andraski & Bundy, 2002;Bundy et al., 1999;Cela et al., 2013).
Reduced RY predictability with the PPNT and PSNT occurs because these tests only account for the inorganic N present at the time of soil sampling and not for N that is mineralized during the remainder of the growing season. Accounting for mineralizable N over the entire growing season is important because mineralization processes can provide anywhere from 20-100% of corn N needs (Broadbent & Hauck, 1984;Morris et al., 2018;Roberts, Ross, Norman, Slaton, & Wilson, 2011;Ros, Temminghoff, & Hoffland, 2011;Yost, Coulter, Russelle, Sheaffer, & Kaiser, 2012). Using a soil test that estimates N mineralization in conjunction with the PPNT and PSNT might improve the predictability of RY and reduce N fertilizer over-and under-application rates.
The PMN test has been used as an N mineralization index in Argentina to divide soils into high and low PMN groups, which resulted in a 12% improvement in RY predictability with the PSNT for low PMN sites (Sainz Rozas, Calvino, Echeverría, Barbieri, & Redolatti, 2008). The inclusion of PMN as a variable with PPNT and PSNT also improved corn grain yield predictability by 5-37% in the control (unfertilized) plots with the greatest improvements coming from fields that had cool early spring temperatures and lower PSNT values (Orcellet, Reussi Calvo, Sainz Rozas, Wyngaard, & Echeverría, 2017). Despite the potential utility shown when coupling the PMN test to the PPNT and PSNT, similar studies are lacking in the U.S. Midwest.
While the PMN test may be useful to improve the PPNT and PSNT in the U.S. Midwest, it may be important to explore a few additional considerations. Nitrogen mineralization indices have been reported to change throughout the growing season depending on variables such as crop rotation, management practices, precipitation, temperature, and soil C to N ratios (Clark et al., 2019;Culman, Snapp, Green, & Gentry, 2013). Nitrogen fertilization has also influenced N mineralization by increasing PMN from in-season soil samples with organic C >21 g kg −1 and clay content <9.5% and decreasing PMN under the opposite conditions . Evaluations have not yet occurred determining whether PMN from different in-season sample timings or those from fertilized soil in conjunction with PPNT and PSNT can improve RY predictability. Accounting for the potential effect of soil sample timing and N fertilizer on PMN may have important practical implications for N fertilizer management. Therefore, the objective of this study was to evaluate RY predictability and frequency of over-and under-applying N fertilizer with the PPNT and PSNT in conjunction with PMN from different soil sample timings and N fertilizer rates under contrasting soil and weather conditions across the U.S. Midwest.

Experimental design
This study was conducted across eight U.S. Midwestern states: Illinois, Indiana, Iowa, Minnesota, Missouri, Nebraska, North Dakota, and Wisconsin. Two or three experimental sites were established in each state each year in 2014-2016, resulting in 49 site-years of data. Kitchen et al. (2017) contains detailed descriptions of the research protocol, agronomic practices, and the soil properties of each site-year. All experimental sites used the same randomized complete block Core Ideas • Including an N mineralization estimate plus a soil NO 3 -N measurement improved RY predictability for only coarse-and medium-textured soils. • Including an N mineralization estimate with soil NO 3 -N did not reduce N fertilizer over-and underapplication rates. • Partitioning soils by texture or temperature improved relative yield (RY) predictability. • Critical soil nitrate content varied substantially depending on soil texture and temperature.
design with four replications, N fertilizer treatments, and soil, plant, and weather data collection methods. Eight N rate application treatments created a complete grain yield response to single-N applications by applying N rates at planting from 0 to 315 kg ha −1 in 45 kg ha −1 increments. Nitrogen fertilizer was broadcast on the soil surface using ammonium nitrate (340 g N kg −1 ). Ammonium nitrate is no longer a commonly used fertilizer; however, results show when different forms of N fertilizers are applied correctly, the response of corn is similar (Fernández, Nafziger, Ebelhar, & Hoeft, 2009). Further, ammonium nitrate was used as it was expected to be suitable for surface application, provide a uniform broadcast application allowing for soil NO 3 -N and NH 4 -N evaluation shortly after application, and perform more similarly across the environmental conditions in our study region (Kitchen et al., 2017).

Soil sampling and analysis
Soil characterization sampling was completed before planting at each experimental site by obtaining a 90-cm deep soil core and dividing it by horizons. These soil samples were evaluated for bulk density, soil texture, total organic C by dry combustion (Nelson & Sommer, 1996), soil organic matter by loss-on-ignition (Soil Survey Staff, 2014), and total N using methods described in Kitchen et al. (2017). Weighted averages of these soil measurements were calculated for three depth increments (0-30, 30-60, and 60-90 cm) using the depth of each horizon within each depth increment. Soil samples for NO 3 -N content were obtained from each replication before planting and fertilization (PPNT) and at the V5 ± 1 corn development stage from each 0 kg N ha −1 plot (PSNT) and each N-fertilized plot (45-315 kg N ha −1 ). Preplant soil samples were obtained using a ten-core composite soil sample (1.9-4.0 cm i.d.; 0-30, 30-60, and 60-90 cm depths) and V5 soil samples were obtained using a six-core composite soil sample (1.9-cm i.d.; 0-30 and 30-60 cm depths). These preplant and V5 soil samples were dried (≤32 • C) and ground to pass through a 2-mm sieve. Nitrate-N was extracted using 0.2 mol L −1 KCl (Saha, Sonon, & Biswas, 2018) and NO 3 -N concentration was quantified using the cadmium reduction method with a modified Technicon AutoAnalyzer (SEAL Analytical, Inc., Fareham, UK) (Gelderman & Beegle, 2015). Bulk density values from each replication and 30-cm depth increment were used to convert soil NO 3 -N concentrations (mg NO 3 -N kg −1 to kg NO 3 -N ha −1 ). The soil available N content was calculated as PPNT plus N fertilizer rate applied to each plot [PPNT + N rate from 0-315 kg ha −1 (PPNT N )] and as soil NO 3 -N at V5 from the zero-N and N fertilized plots (PSNT N ) following the approach used by Cela et al. (2013).
The PMN test was run on a subset of soil samples (0-30 cm) that included (1) preplant soil sampling with 0 kg N ha −1 , (2) V5 soil sampling with 0 kg N ha −1 , and (3) V5 soil sampling with 180 kg N ha −1 as described in Clark et al. (2019). Briefly, PMN was quantified by determining extractable NH 4 -N in the soil by 2 mol L −1 KCl and subtracting it from the extractable NH 4 -N after the soil was incubated in an anaerobic environment for 7 d at 40 • C (i.e. PMN = NH 4 -N after incubation − NH 4 -N before incubation) (Bundy & Meisinger, 1994). Bulk density values from the 0-30 cm depth were used to convert PMN concentrations (mg kg −1 to kg ha −1 ).

Plant sampling and analysis
All plants in the middle two rows of each experimental unit were harvested to determine grain yield (adjusted to 155 g kg −1 moisture). Relative yield was calculated for each site by dividing the yield of each N rate treatment by the mean yield of the N rate treatment that yielded the highest and multiplying the value by 100 to express the result as a percentage of the yield from the highest yielding N rate treatment.

Weather
A Hobo U30 automatic weather station (Onset Computer Corporation, Bourne, MA, USA) was used to collect daily minimum, maximum, and mean temperatures ( • C) and cumulative precipitation. These data were quality checked using Multi-Radar/Multi-Sensor precipitation data (The National Severe Storms Lab, NOAA) as described in Kitchen et al. (2017). Irrigation water applied to eight of the experimental sites as part of normal management practices was treated as natural rainfall in the precipitation equations. Cumulative growing degree-days (GDD) were calculated from the first day of the year where a GDD was accumulated (first-GDD) to the preplant soil sample timing and from the preplant soil sample timing to the V5 corn development stage as described in Clark et al. (2019).

Statistical analysis
Statistical evaluations were completed across all sites and after soils were grouped into three textural categories (coarse, medium, and fine) at the replication level and two GDD categories (high and low) at the site level because of the influence of soil texture and temperature on PPNT, PSNT, and PMN. Soils were grouped by texture categories following the approach used by Tonitto, David, and Drinkwater (2006) and Tremblay et al. (2012). There were 34 replications that had coarse-textured soils, 88 that had medium-textured soils, and 74 that had fine-textured soils. Sites were also grouped into GDD categories using the classifications from the Nutrient Star TED framework tool (Nutrient Star, 2018), as developed by Van Wart et al. (2013). Experimental sites were classified as high GDD when TED GDD units were ≥2,222 (4,000 using • F) and low GDD when TED GDD units were <2,222 (4,000 using • F). Nineteen of the 49 experimental sites were in the low GDD category (Minnesota, North Dakota, Wisconsin, and 2014 Mason City, Iowa) and 30 were in the high GDD category (all other experimental sites in Iowa along with Illinois, Indiana, Missouri, and Nebraska).
All statistical analyses were completed with SAS software version 9.4 (SAS Institute Inc., Cary, NC). The means and standard deviations of RY, PMN, soil properties, and weather measurements were determined using the MEANS procedure. Linear and quadratic regressions were performed using the REG procedure and linear-plateau and quadratic-plateau models using the NLIN procedure. We used these regressions to evaluate the change in predictability of RY when including PMN with PPNT or PSNT (zero-N plots only) and soil available N (PPNT N or PSNT N using the zero-N plus N-fertilized plots). The CSNC was calculated using the linear-plateau and quadratic-plateau models as the point between the linear or quadratic part of the model and the plateau portion. The linear-plateau model was used because it correctly identified sites as responsive or non-responsive to N correctly 9-12% more often than the quadratic-plateau model. The CSNC was the amount of soil available N (e.g. soil inorganic N with and without PMN plus N fertilizer) above which no additional increase in grain yield was predicted. The fit of these models was evaluated by comparing lack of fit F-tests, coefficients of determination (Cerrato & Blackmer, 1990), and visually assessing the model fit to the data.
We employed a similar approach as used by Bundy et al. (1999) to evaluate the statistical models used to calculate the CSNC and determine whether N fertilizer should be applied. Briefly, each graph relating RY as a function of soil available N (e.g. soil inorganic N with and without PMN plus N fertilizer) was separated into four quadrants with the line dividing the Y-axis at a RY of 90% and the X-axis at the CSNC. Values in the upper-left quadrant represented experimental units that were incorrectly categorized as responsive to N fertilization, resulting in an over-application of N while values in the lower-right quadrant represented experimental units that were incorrectly categorized as nonresponsive to N fertilization, resulting in an under-application of N. The percent of sites in these two categories was the total misapplication frequency. Values in the lower-left quadrant represented experimental units that were correctly categorized as responsive to N and values in the upper-right quadrant represented experimental units that were correctly categorized as nonresponsive to N. The percent of sites in these two categories was the total success frequency. The strength of the relationship between RY and soil N measurements as determined by model R 2 was also compared.

RESULTS AND DISCUSSION
Corn grain yield varied across all experimental site-years with a range of 1.4-18.6 Mg ha −1 and a mean of 12.2 Mg ha −1 ( Table 1). The mean grain yield varied minimally between coarse-(12.3 Mg ha −1 ), medium-(12.5 Mg ha −1 ), and fine-textured soils (11.9 Mg ha −1 ). Classifying sites into high (12.7 Mg ha −1 ) and low (11.4 Mg ha −1 ) GDD categories resulted in larger mean differences. These differences in yield when evaluating by texture and GDDs may be important because sites with differences in yield potential can alter the CSNC (Fox et al., 1989). Coarse-textured soils generally had the greatest increase in grain yield with added N fertilizer with a mean increase of 6.6 Mg ha −1 followed by 5.8 Mg ha −1 for medium-and 5.2 Mg ha −1 for fine-textured soils. High GDD sites had a greater increase in mean yield (6.4 Mg ha −1 ) with added N fertilizer compared to low GDD sites (4.5 Mg ha −1 ). Greater grain yield responses to N fertilizer for some of these categories may be the result of less N supplied to the corn crop from mineralization (Lory & Scharf, 2003), as N mineralization potential and soil organic matter concentration are related to soil texture and temperature (Cabrera, Kissel, & Vigil, 2005;Clark et al., 2019;Kuzyakova, Turyabahika, & Stahr, 2006). Likewise, our study showed that grain yield response to N decreased as the mean potential for N mineralization (PMN) increased among categories within the textural and GDD groupings (Table 1). These results indicate there is a potential benefit to further investigating the use of PMN as a tool to improve N need predictions.

Calculating critical soil nitrate content
When using only the zero-N plots, RY generally increased with soil N (Figures 1 and 2). The RY was better predicted by PPNT (R 2 = 0.25-0.26) or PSNT alone (R 2 = 0.35-0.43) than when combined with PMN from any of the three soil sampling methodologies (preplant with 0 kg N ha −1 , V5 with 0 kg N ha −1 , and V5 with 180 kg N ha −1 applied preplant) (R 2 = 0.03-0.24). Only 13% of the zero-N plots had ≥90% RY; thus, we could not calculate a CSNC for an optimal RY. However, when zero-N and N fertilized plots were included, RY increased with soil available N (PPNT N and PSNT N ) with and without PMN included until yield plateaued (P ≤ 0.05) (data not shown). The strength of the relationship between RY and soil available N with and without PMN (R 2 = 0.19-0.66) varied by soil grouping and soil NO 3 -N sampling depth (Table 2). A similar conclusion was reported in Northeast Spain (Cela et al., 2013). These results indicate that plantavailable N early in the season and the amount of N that may be mineralized during the season for corn after soybean was normally inadequate to obtain optimal yield. This highlights the importance of N fertilizer application in corn after soybean rotations to obtain optimal yield in the U.S. Midwest.
Generally, the PPNT and PSNT have been most successful when there was little to no N fertilizer applied, in situations where N carried over from the previous season, or where diverse cropping systems or recent manure applications led to greater N mineralization for the current growing season. (Bundy & Andraski, 1995;Bundy et al., 1999;Magdoff, 1991;Magdoff et al., 1984;Mulvaney, Khan, Hoeft, & Brown, 2001;Rehm et al., 2002;Schmitt & Randall, 1994). The low strength of the model R 2 between PPNT or PSNT and RY determined from our study support these other findings as the soybean-corn rotations used in this study and in the U.S. Midwest do not have a strong chance of carrying over N from the previous season. This low N carry over potential is due to the large N requirements of corn and substantial potential for residual N loss with excess precipitation, especially in early spring (Bakhsh et al., 2000;Jokela & Randall, 1989;Randall, Vetsch, & Huffman, 2003). Furthermore, our experimental sites had not received any recent manure applications that would have increased their mineralization potential. Therefore, these results indicate that in a primarily corn-soybean rotation, N fertilized fields need to be included to make sure yields will optimize when calculating a CSNC using soil N measurements regardless of soil groupings.
T A B L E 1 Range and mean for various corn grain yield calculations, soil parameters, and weather variables across 49 site-years (All) or partitioned by soil texture (coarse, medium, and fine) or growing degree-day (GDD) categories (high and low)

Soil NO 3 -N sampling depth
For the PPNT N alone or combined with PMN, sampling soil NO 3 -N beyond 30 cm in low GDD sites improved RY predictability by an R 2 of as much as 0.07, but for other categories increases in R 2 were <0.03 (Table 2). This result likely occurred because of greater soil NO 3 -N deeper in the soil profile for low GDD sites compared to other soil categories (Table 1). However, the improved RY predictability with deeper soil samples still did not substantially reduce (≤2%) the misapplication frequency in any category (Table 2). Others also reported that RY predictability improvements from increasing PPNT sampling depths beyond 30 cm were insufficient to justify the increased time and cost of obtaining the deeper soil samples (Binford et al., 1992;Bundy & Andraski, 1995;Cela et al., 2013). In contrast to the minimal reduction in over-application frequency we observed (≤2%), Bundy et al. (1999) reported a greater (8%, on average) reduction in N over-application frequency with deeper sampling. Our results highlight that a shallow 0-30 cm sampling depth is sufficient for the PPNT N , and the only exception may be in soils that have large amounts of NO 3 -N in the deeper soil depths, as pointed out by Bundy et al. (1999). For the PSNT N alone or combined with PMN, increasing soil NO 3 -N sampling depth from 30 to 60 cm across all soil categories improved RY predictability by an R 2 of as much as 0.20. Other research has reported that the cost of deeper soil sampling is normally only compensated by improvements in explained variability T A B L E 2 Model R 2 and misapplication type and frequency from linear-plateau regressions using different combinations of soil available N from various soil sampling depths with and without anaerobic potentially mineralizable N (PMN) from three sampling methodologies across 49 site-years or partitioned by soil texture (coarse, medium, and fine) or growing degree-day (GDD) categories (high and low)  of RY in permeable soils that have a greater chance of N leaching past the 30-cm depth such as coarse-textures soils (Magdoff et al., 1984;Vinten, Vivian, Wright, & Howard, 1994). However, RY predictability using PSNT from 0-60 cm compared to 0-30 cm soil samples in our fine-textured soils improved similarly or more (0.09-0.14 improvement in R 2 ) than that in the coarse-textured soils (0.10-0.16 improvement in R 2 ). These results indicate that less permeable soils such as fine-textured soils can benefit from deeper PSNT sampling similarly to coarse-textured soils.

Including PMN with PPNT N to improve N management
The PPNT N alone across soil categories accounted for 49-65% of the variability in RY (Table 2). Partitioning soils into texture categories marginally improved RY predictability (<0.04) by PPNT N alone only in medium-textured soils and altered the total misapplication frequency ≤5%. Across soil texture categories, including PMN with PPNT N improved RY predictability by an R 2 between 0.07-0.10, but only when using PMN from preplant and in coarse-and mediumtextured soils. In fine-textured soils, including PMN, with PPNT N regardless of sampling methodology substantially reduced RY predictability by an R 2 between 0.08-0.24 and increased the total misapplication frequency ( Table 2). The PMN test may have underestimated mineralizable-N in fine-textured soils because the greater organic matter and clay content of these soils (Table 1) produce more NH 4 -N during incubations, resulting in suppressed mineralization and more NH 4 -N fixed to clay surfaces (Russell, Dunn, Batten, Williams, & Angus, 2006). This underestimation of PMN is likely the reason for the weaker relationships with RY in fine-textured soils. Others observed improvements in yield predictability when including PMN from preplant with PPNT when soils with similar soil textures, topographies, and climate conditions were evaluated together (Nyiraneza, N'Dayegamiye, Chantigny, & Laverdière, 2009;Orcellet et al., 2017;Reussi Calvo, Rozas, Echeverría, & Berardo, 2013;Sainz Rozas et al., 2008). Our findings and these of others indicate that soil texture is an important variable to consider when using soil mineralization tests. Further, our results show that the improved predictability of N fertilizer needs by including PMN is soil texture dependent and for some soils including PMN will not improve N management. Partitioning soils into GDD categories improved RY predictability by an R 2 between 0.04-0.08 by PPNT N alone for high GDD sites but reduced it by an R 2 between 0.02-0.07 for low GDD sites ( Table 2). The total misapplication frequency (20-23%) remained similar to when all sites were evaluated together even with the changes in RY predictability when grouping sites by GDDs. The lack of change in total misapplication frequency was because the reduction in one misapplication category (under-or over-application) was offset by the increase in the other category. For both GDD categories, including PMN from any of the three sampling methodologies with PPNT N did not improve RY predictability or reduce any of the misapplication frequency categories, similar to the results from fine-textured soils. This lack of improvement in RY predictability when PMN was included with PPNT N may be because the GDD categories did not separate out the influence of greater clay and soil organic matter content on PMN as discussed earlier for the analysis of soils partitioned by texture.

Including PMN with PSNT N to improve N management
The PSNT N alone across soil categories accounted for 45-62% of the variability in RY (Table 2). Partitioning soils into texture categories improved RY predictability by PSNT N alone 0.02-0.05 for coarse-and medium-textured soils and reduced RY predictability by an R 2 between 0.03-0.05 for fine-textured soils (Table 2). Only a minimal improvement in RY predictability (≤0.03 increase in R 2 ) resulted from including PMN with PSNT N for coarse-and medium-textured soils but only when using PMN from the preplant sampling and using PSNT N at the 0-60 cm soil sampling depth. The reduced improvement in RY predictability by including PMN with PSNT N compared with PPNT N was likely because the PSNT N accounts for the net N mineralization potential of the soil up to the V5 development stage, as discussed earlier (Magdoff et al., 1984). Whereas, the PPNT N accounted for little to no net N mineralization potential because it was performed early in the spring before substantial N mineralization and N loss processes occurred. The improvements in RY predictability with PMN from preplant combined with PSNT N for coarse-and medium-textured soils resulted in an increase in the over-application frequency similar to including PMN from preplant with PPNT N ( Table 2). No further improvement in RY predictability or reduction in misapplication frequency resulted from including PMN from either of the V5 N rates for coarse-and medium-textured soils or the inclusion of PMN from any sampling methodology for fine-textured soils.
Partitioning sites by GDD also improved RY predictability using PSNT N alone; however, the improvements in predictability did not reduce the total misapplication frequency, which was similar to using PPNT N alone ( Table 2). The total misapplication frequency did not change because when one of the misapplication categories increased (underor over-application), the other misapplication frequency increased similarly. Including PMN from any sampling methodology with PSNT N did not improve the RY predictability or reduce any of the misapplication frequencies for either GDD category, similar to evaluating fine-textured soils alone. Therefore, including PMN as a mineralization estimate with PSNT N was not an effective method to improve N management when soils were partitioned by GDDs alone.
Overall, including the PMN test as an estimate of N mineralization with PPNT N or PSNT N was not able to improve RY predictability substantially or minimize rates of N fertilizer over-and under-fertilization. There are several reasons this may have occurred. First, Bundy and Meisinger (1994) point out that the PMN test is done under anaerobic conditions where microbial respiration is much slower than aerobic conditions, which may result in underestimating actual N mineralization. Second, since anaerobic conditions kills aerobic organisms, the PMN test may be mainly measuring the death and lysing of N from aerobic microorganisms and not mineralizable N. However, the PMN test was evaluated in this study because it was reported in other studies that the use of the PMN test improved N management (Williams et al., 2007;Nyiraneza et al., 2009;Orcellet et al., 2017;Reussi Calvo et al., 2013;Sainz Rozas et al., 2008). In addition, we measured PMN from the 0-30 cm soil depth; however mineralization occurs below 30 cm and also including mineralization estimates from deeper soil depths may improve relationships between N mineralization estimates and RY. Future studies could evaluate the effect of including other N mineralization estimates such as the anaerobic mineralization test from surface and subsurface soils with PPNT or PSNT on RY predictability.

Influence of soil texture and growing degree-days on critical soil nitrate content
The CSNC calculated using soil available N (including both zero-N and N fertilized plots) increased with deeper soil NO 3 -N sampling depths and when including PMN for all soil categories (Table 3). For the PPNT N model, the CSNC values across soil categories ranged between 122-175 kg ha −1 (31-39 mg kg −1 ) for the top 30 cm, 143-190 kg ha −1 (17-21 mg kg −1 ) for the top 60 cm, and 157-207 kg ha −1 (13-15 mg kg −1 ) for the top 90 cm. These CSNC values from PPNT N models were normally lower than those found by Cela et al. (2013) where the CSNC for the 0-30-cm depth was 172-243 kg N ha −1 and for the 0-90-cm depth CSNC was 208-302 kg N ha −1 . These greater CSNC values from the Cela et al. (2013) study may be due to the greater mean PPNT values of their study sites (0-30 cm: 60-69 kg N ha −1 ; 0-90 cm: 119-180 kg N ha −1 ) compared to ours (0-30 cm: 23-29 kg N ha −1 ; 0-90 cm: 52-62 kg N ha −1 ). Our lower PPNT values were likely the result of lower N carryover potential in our study region as discussed earlier.
There were distinct differences between CSNC values within soil texture and GDD groupings for the PPNT N model (Table 3). Across sampling depths, the CSNC values using PPNT N alone were 175-207 kg ha −1 for coarse-textured soils, 145-183 kg ha −1 for medium-textured soils, and 128-178 kg ha −1 for fine-textured soils. The decrease in CSNC with finer textures likely occurred because mineralization supplied more N, as indicated by greater PMN for fine-textured soils (Table 1). For the coarse-and medium-textured soils where including PMN with PPNT N improved RY predictability, the CSNC from medium-textured soils (287-322 kg ha −1 ) became greater than that of coarse-textured soils (256-295 kg ha −1 ) because PMN from preplant soil was greater for medium-textured soils ( Table 1).
The CSNCs using soil NO 3 -N alone were 171-207 kg ha −1 for high GDD sites and 122-157 kg ha −1 for low GDD sites (Table 3). Likely, the lower temperatures and GDDs (Table 1) in the low GDD sites limited N mineralization early in the season and reduced the PPNT and CSNC.
There were also distinct differences between CSNC values within soil texture and GDD groupings for the PSNT N alone model (Table 3). The CSNC from PSNT N for both coarse-(88 and 157 kg ha −1 ) and medium-textured soils (92 and 150 kg ha −1 ) were similar while fine-textured soils were still in a lower range (81 and 135 kg ha −1 ), which is opposite of what we observed for the PPNT N model. The CSNC for high GDD sites were greater (118 and 188 kg ha −1 ) than the low GDD sites (60 and 113 kg ha −1 ). The lower temperatures for the low GDD sites before the time of PSNT N sampling (Table 1) likely decreased the quantity of N mineralized (Ma & Wu, 2008) and reduced PSNT N and the subsequent CSNC. Others also reported similar results (Andraski & Bundy, 2002;N'Dayegamiye et al., 2015;Sainz Rozas et al., 2008). These results indicated that CSNC was influenced by soil texture and temperature and that these variables should be considered when determining and using CSNC values to manage N fertilizer needs.

CONCLUSIONS
Using only the zero-N plots, RY was not >90% in enough of them for CSNC to be calculated using only PPNT and PSNT. However, using the zero-N and N fertilized plots (PPNT N and PSNT N ) resulted in sufficient plots yielding greater than 90% RY and enabled CSNC to be calculated. In future studies, other cropping rotations besides the primarily cornsoybean rotations used in this study and other management practices that increase N mineralization potential should be included to more completely evaluate mineralization potential and calculate CSNC using PPNT and PSNT from only zero-N plots.
For PPNT N and PSNT N based models, partitioning soils into textural or GDD categories improved RY predictability marginally for some categories but did not reduce the total misapplication frequency (under-plus over-application of N). The inclusion of PMN with PPNT N or PSNT N only improved RY predictability for coarse-and medium-textured soils and only with PMN from preplant. However, including PMN with PPNT N or PSNT N did not substantially reduce over-application frequencies and only minimally reduced under-application frequencies. Therefore, this study demonstrated that including PMN with PPNT N or PSNT N to account for N mineralization had little utility to improve N management regardless of soil NO 3 -N sampling depth, PMN sampling methodology, and soil categorization evaluated. However, this study showed that CSNC varied extensively within soil texture and GDD categories, indicating that these parameters may have potential to improve CSNC calculations to reduce over-and under-applications of N fertilizer.